Even though BCs and SCs are the two stereotyped MLIs, the classification of MLIs is still under debate. There is a contrasting idea to the traditional classification of BCs and SCs; all MLIs originate from a single type of interneurons with continuous variation in morphology and different target locations on PCs (Rakic, 1972; Sultan and Bower, 1998; Bower, 2010). A transcriptome study identified two distinct types of MLIs, MLI1s and MLI2s, characterized by different molecular expressions, such as Sorcs3 in MLI1s and Nxph1 in MLI2s (Figure 2A; Kozareva et al., 2021). Because both MLI1s and MLI2s showed similar continuum in morphological properties, particularly an SC-like morphology in the outer one-third of the ML, they do not align with the traditional BC and SC classification. However, further analyses revealed at least two distinct morphological aspects between MLI1s and MLI2s (Lackey et al., 2024). First, MLI1 somas are spiny, while MLI2 somas are smooth. Second, MLI1s in the inner two-third of the ML contribute to pinceau formations, whereas all MLI2s, regardless of soma location, retain classical SC morphologies without contributing to pinceau. Supporting this second observation, another study identified two discrete morphological types of MLIs, canonical BCs and SCs with extensive heterogeneity, which are distinguished by the morphological features of axons rather than their soma positions in the ML (Wang and Lefebvre, 2022). This study confirmed the dissociations between morphological and transcriptomic types, but showed some relationships between them. Sorcs3-positive MLIs, namely MLI1s, are mostly BCs and SCs with long axons, further distinguished by the combined expression levels of Grm8 and Cacna1e. Nxph1-positive MLIs, namely MLI2s, are largely SCs with short axons and partly SCs with long axons located in the superficial layer.

In general, neuronal cell-type classification considers not only transcriptional and morphological properties, but also electrophysiological and connectional properties (Zeng and Sanes, 2017). MLI2s were found to be more excitable than MLI1s, and the expression of Gjd2, the gene encoding the dominant gap junction protein, connexin 36, was detected in MLI1s, but not in MLI2s (Kozareva et al., 2021), indicating electrical differences in the two transcriptionally identified cell types. In terms of connectional properties, a subpopulation of MLIs located in the inner one-third of the ML was found to receive inhibitory inputs from PCs (Witter et al., 2016; Halverson et al., 2022), which may contribute to the disinhibition of other PCs. Further studies are required to clarify whether these MLIs have other characteristic properties and can be categorized into a specific cell type based on the functional uniqueness.

The anatomical organization of MLIs and their circuits, which affect cell type classification, appears to rely on process occurring during the postnatal developmental period. MLIs are gradually generated from precursors in the white matter (WM), which originate in the ventricular zone (Yamanaka et al., 2004). Immature postmitotic MLIs radially migrate toward the ML, where they find their final positions through a complex migratory route involving radial and tangential migration (Simat et al., 2007a; Cameron et al., 2009; Park et al., 2021). This complex migration is regulated by several pathways, such as excitatory and inhibitory synaptic transmission (Wefers et al., 2017). Specifically, the blockade of synaptic transmission from PF bundles resulted in abnormal distributions of MLIs (Park et al., 2019), indicating that PF inputs likely regulate the migration that determines the final positions of MLIs. The final position of MLIs also depends on their timing of birth; early-and late-born MLIs are positioned in the inner and outer parts of the ML, respectively, meaning that BCs are generally born earlier than SCs (Yamanaka et al., 2004; Weisheit et al., 2006; Leto et al., 2008; Schilling et al., 2008; Sudarov et al., 2011; Consalez and Hawkes, 2012; Sotelo, 2015), although their positions are not strictly defined. Despite differences in morphology, transcriptional and electrophysiological properties, or connectivity in mature MLIs, MLI precursors before leaving the WM are indistinguishable in terms of their genetic markers (Maricich and Herrup, 1999; Leto et al., 2006; Carter et al., 2018). This suggests that each MLI acquires its characteristic phenotypes after migration begins. It has been proposed that an additional tangential migration, supported by premigratory GCs in the EGL, is required for the synaptic differentiation of SCs (Cadilhac et al., 2021). Additionally, a study linking morphological and transcriptomic types demonstrated that BC and SC morphological identities diverged during early phases of migration, prior to the expression of transcriptomic markers (Wang and Lefebvre, 2022). Thus, the diverse characteristics of MLIs may arise from the impact of their migration process itself or the environmental situations encountered during migration.

As described above, PFs are well-known excitatory inputs to MLIs. With the revelation of the heterogeneous properties of GCs and PFs, different patterns of short-term synaptic plasticity (STP) were observed in PF synapses of the two types of MLIs, namely, BCs and SCs. The PF-SC synapses demonstrated persistent facilitation, whereas PF-BC synapses showed depression following initial paired-pulse facilitation (Bao et al., 2010). This implies that PF inputs trigger firing in BCs first, followed by delayed firing in SCs. Given their target locations on PCs, PF inputs may sequentially hyperpolarize PCs from soma to dendrites (Bao et al., 2010; Blackman et al., 2013). A recent study further categorized STP in unitary PF-MLI synapses into four types, attributing this diversity to the heterogeneous expression of the key STP molecule, synapsin II, in GCs, but found no clear MLI type-dependent STP patterns (Dorgans et al., 2019). Although it is unclear how to reconcile these conflicting results in terms of cell types, the diversity of PF-MLI synaptic functions may contribute to increased coding patterns, potentially allowing efficient separation of different contexts in the cerebellum.

In contrast to the clear excitatory synaptic connections from PFs to MLIs, excitatory inputs from CFs to MLIs have long remained controversial. Although anatomical studies of labeled CFs suggested possible synaptic connections between CFs and MLIs (Palay and Chan-Palay, 1974; Sugihara et al., 1999), no conventional synaptic architecture was found between them in electron microscopy studies (Hámori and Szentágothai, 1980; Kollo et al., 2006). Nevertheless, functional effects of CFs were detected in MLIs (Eccles et al., 1966; Jörntell and Ekerot, 2003). It became clear that CF-MLI connections were mediated uniquely, through glutamate spillover (Figure 2B; Szapiro and Barbour, 2007; Mathews et al., 2012; Coddington et al., 2013). CF synapses to PCs have a very high release probability and typically show multivesicular release (Wadiche and Jahr, 2001; Foster et al., 2002; Harrison and Jahr, 2003), so that CF activation may reasonably lead to glutamate spillover. The involvement of N-methyl-D aspartate (NMDA) receptors in the spillover-mediated CF-MLI connections is also plausible, considering that MLIs have NMDA receptors on their dendrites (Carter and Regehr, 2000; Clark and Cull-Candy, 2002). Thus, it is now generally accepted that MLIs receive CF inputs via glutamate spillover. This spillover-mediated, CF-dependent regulation of MLIs has been shown to be involved in sensory processing in vivo (Arlt and Häusser, 2020).

As commonly seen in inhibitory interneurons of the mammalian brain (Galarreta and Hestrin, 2001; Pereda, 2014; Coulon and Landisman, 2017), MLIs are also connected with each other via electrical synapses (Mann-Metzer and Yarom, 1999). The electrical synapses between MLIs rely on the gap junction protein connexin 36, which is differentially expressed according to their locations in the ML (Alcami and Marty, 2013). MLIs in the inner ML, presumably BCs, have higher levels of connexin 36 than those in the outer ML. Consistently, electrophysiological recordings have demonstrated that electrical connections are more frequent and stronger in BCs than in SCs (Figure 2C; Alcami and Marty, 2013). The difference in electrical connections is clear between MLI1s and MLI2s, as the expression of Gjd2 and electrical coupling were found only in MLI1s, but not ML2s (Kozareva et al., 2021; Lackey et al., 2024). Given that BCs are likely part of MLI1s (Wang and Lefebvre, 2022; Lackey et al., 2024), these results regarding cell type-dependent electrical connections appear consistent. Furthermore, studies using multiple patch-clamp recordings and optogenetic mapping identified that electrical coupling occurs in the sagittal plane (Kim et al., 2014; Rieubland et al., 2014). Such structured organization of electrical connections between MLIs contributes to the regulation of feedforward inhibition of PCs, by generating spatiotemporally controlled synchronization of MLI activity (Hoehne et al., 2020).

MLIs are well known to receive GABAergic inhibitory synaptic inputs from neighboring MLIs (Llano and Gerschenfeld, 1993; Häusser and Clark, 1997; Mittmann et al., 2005), and this inhibitory connection is presumably important for fine-tuning the temporal precision of spiking in PCs (Mittmann et al., 2005). Similar to the electrical connections between MLIs, their inhibitory synaptic connections were also found to be oriented in the sagittal plane, although not as strongly biased as electrical coupling (Rieubland et al., 2014). Additionally, transitive patterns of MLI-MLI synaptic connections were demonstrated to be organized from the outer to the inner part of the ML (Figure 2D; Arlt and Häusser, 2020). This raises the possibility that such MLI microcircuit organization, together with the location of activated PFs, may be utilized for the regulation of cerebellar information processing – e.g., additional activation of PFs in the outer ML may lead to greater excitation of PCs through the reduction of MLI-dependent inhibition of PCs. Cell type-dependent innervation patterns have recently been demonstrated: MLI1s mainly innervate PCs to inhibit them, while MLI2s mainly innervate MLI1s to disinhibit PCs (Lackey et al., 2024). In other words, many MLI1s would receive relatively strong inhibitory inputs from MLI2s, whereas MLI2s may have little or no inhibitory inputs from other MLIs.

In addition to the inhibitory synaptic inputs from MLIs, early anatomical studies have suggested other possible inputs from PCs and Lugaro cells (LuCs), the latter of which are a type of inhibitory interneuron in the GL (Hámori and Szentágothai, 1968; Larramendi and Lemkey-Johnston, 1970; O’Donoghue et al., 1989; Lainé and Axelrad, 1996, 1998). While functional synaptic connections from LuCs to MLIs are not yet clarified, functional inhibitory inputs from PCs to MLIs were identified by recording synaptic currents evoked by the optogenetic stimulation of PCs (Figure 2E; Witter et al., 2016; Halverson et al., 2022). Optogenetically evoked synaptic currents were detected in approximately 16 to 30% of MLIs, indicating these are not extremely rare synaptic connections. Reciprocally connected pairs of PCs and MLIs have not been found so far, thus MLIs receiving inputs from PCs are considered to send inhibitory inputs to other PCs, suggesting that this pathway from PCs to MLIs contributes to synchronization of PCs through disinhibition of MLIs (Halverson et al., 2022). Given that inhibitory LuCs also receive inputs from PCs (Witter et al., 2016), PC population activity may rely on the organization of connections among PCs, LuCs, and MLIs, if LuCs functionally inhibit MLIs. Direct connections from PCs to MLIs are considered to mediate synchronization of neighboring PCs (Witter et al., 2016; Halverson et al., 2022), while indirect connections from PCs to MLIs through LuCs may contribute to desynchronization of other PC populations, as seen between aldolase C/zebrin II-positive and-negative PC populations (Tsutsumi et al., 2015). In addition, whereas glutamate spillover from CFs primarily exerts excitatory influence to MLIs, as described above, other MLIs located outside the glutamate diffusion limit have been shown to be inhibited upon CF activation via disynaptic inhibition (Coddington et al., 2013). Thus, CF-mediated functional segregation of MLIs may also contribute to the distinctive patterns of activity of different PC populations. Alternatively, segregation of MLIs could be along the perpendicular direction to the layer, as CF inputs were reported to activate many MLIs but inhibit MLIs in the deep ML that have strong inhibitory impacts onto PCs (Arlt and Häusser, 2020).

While PC axons project outside the cerebellar cortex, PC axon collaterals enable the modulation of information processing within the cerebellar cortex by sending feedback signals. Early studies already described PC axon collaterals as forming synapses onto other PCs and neurons of the ML (Eccles et al., 1967; Larramendi and Lemkey-Johnston, 1970; Chan-Palay and Palay, 1971). Although they were anatomically observed in both young and adult rodents (Larramendi and Lemkey-Johnston, 1970; Chan-Palay and Palay, 1971; Bishop, 1982; Hawkes and Leclerc, 1989; Bernard et al., 1993), their functional synaptic transmission was first confirmed only in young animals (Orduz and Llano, 2007; Watt et al., 2009; Hirono et al., 2012). These studies demonstrated that PC axon collaterals form functional synapses with other PCs and globular cells, small inhibitory interneurons located in the GL. A computational model and experimental confirmation demonstrated that PC axon collaterals generated waves of PC activity traveling along chains of connected PCs in the developing cerebellum, whereas these waves were not observed in the mature cerebellum (Watt et al., 2009). This led to an idea that PC axon collaterals play a crucial role only in cerebellar circuit formation during postnatal development. In contrast, studies using optogenetic stimulation with patch-clamp recording identified functional synaptic transmissions from PC axon collaterals in adult animals (Figure 3A; Guo et al., 2016; Witter et al., 2016), although these synaptic connections were restricted to the parasagittal plane and were present only near the PCs sending the axon collaterals. GABAergic inhibitory synaptic transmissions were recorded from PCs, MLIs, LuCs, and candelabrum cells (neurons situated in the PCL), but not from GoCs, upon optogenetic stimulation of PCs or spontaneous PC firing (Witter et al., 2016; Halverson et al., 2022; Osorno et al., 2022). While there were variabilities depending on the lobule, PC axon collaterals also projected to the GL, and inhibited GCs and a subset of unipolar brush cells (UBCs) (Guo et al., 2016; Guo et al., 2021). These studies suggest that PC axon collaterals may have additional roles in the mature cerebellum. It would be interesting to differentiate properties of collateral axons in the cerebellar cortex from axons projecting to the CN, as this may lead to techniques to selectively manipulate collateral axons.

Signal processing and transmission in the nervous system involve the flow of current across neuronal cell membranes, which alters the electrical potential in extracellular regions. Through these alterations in extracellular potential, neurons can interact with adjacent neurons without forming the specific cell-to-cell contacts required for chemical and electrical synaptic transmission. These electrical interactions, described as early as the 1940s (Katz and Schmitt, 1940; Arvanitaki, 1942), are now called ephaptic transmission or coupling (Anastassiou and Koch, 2015; Faber and Pereda, 2018). In the well-established circuit structures where PFs provide direct excitatory inputs and disynaptic inhibitory inputs through BCs or SCs onto PCs, recordings from PCs during PF volleys were expected to show a sequence of excitation followed by inhibition. However, an initial study revealed that a brief inhibition occurred before this expected sequence (Korn and Axelrad, 1980). Due to its short latency, this brief inhibition was proposed to be electrically generated by nearby BCs. BCs form pinceau that wrap around and terminate on the AIS of PCs (Sotelo, 2015). While BC axons make GABAergic synaptic contacts onto PC somas, the pinceau themselves are largely devoid of chemical and electrical synapses (Ichikawa et al., 2011), raising a question about their functions. The dense plexuses of pinceau appear to be well-suited for generating local electric fields, and indeed, one study demonstrated that the pinceau could induce rapid ephaptic inhibition of PC axons (Figure 3B; Blot and Barbour, 2014). This study proposed that pinceau-dependent ephaptic inhibition allows GCs to instantaneously inhibit PCs. Given the variability in pinceau size (Zhou et al., 2020), this inhibitory regulation may differ depending on PC zones identified by specific molecular markers, such as aldolase C/zebrin II (Apps and Hawkes, 2009; Cerminara et al., 2015).

In addition to the ephaptic coupling between the pinceau in BCs and the AIS in PCs, two other types of ephaptic coupling between PCs have been identified (Han et al., 2018; Han et al., 2020). One type of ephaptic coupling observed in the AIS of PCs led to the synchronization of spontaneous PC firing (Figure 3C; Han et al., 2018). Another type of coupling, occurring between PC dendrites, resulted in the inhibition of multiple neighboring PCs when synaptically connected PCs were activated by CF inputs (Figure 3D; Han et al., 2020). This study proposed that this type of ephaptic coupling allows CF inputs to efficiently activate CN neurons through the simultaneous inhibition of many PCs. Although these two types of ephaptic coupling between PCs may seem contradictory—one causing excitation and the other inhibition of neighboring PCs—they likely contribute to the spatially distinct responses of neighboring PCs upon CF inputs.

The diversity of PCs has been well characterized by different expression levels of marker molecules, such as aldolase C/zebrin II (Cerminara et al., 2015). The expression patterns of these molecules form longitudinally striped compartments known as microzones. PCs in different microzones (e.g., AldoC+ and AldoC-microzones) exhibit distinct firing patterns and project to specific sets of CN or vestibular nuclei, linking each microzone to distinct functions (De Zeeuw, 2021). For instance, fast and slow eye movements are mediated by distinct PC subpopulations in the flocculus, known as 9- and 9+ PCs, respectively (Blot et al., 2023). Apart from these modules visualized by molecular markers, the diversity of PC morphology has been gradually recognized. PCs typically have a characteristic morphology with a large, planar, highly branched dendritic tree, and the diversity is observed in the dendritic shapes (Nedelescu and Abdelhack, 2013; Nedelescu et al., 2018; Chang et al., 2020; Busch and Hansel, 2023; Magnus et al., 2023). In zebrafish, four types of PCs with different morphological and firing properties were classified, and they were active at different phases during swimming, suggesting that different types of PCs have different functions (Chang et al., 2020). In mice, morphological types were specifically considered based on the number of primary dendrites (Nedelescu and Abdelhack, 2013; Nedelescu et al., 2018; Busch and Hansel, 2023). Most PCs in the anterior and central lobules have single primary dendrites, but some PCs, particularly those in the posterior lobules, have two primary dendrites. Furthermore, a recent study (Busch and Hansel, 2023) demonstrated that PCs with two or multiple primary dendrites, called as poly PCs, are a near-universal morphological feature in humans and are functionally different from normative PCs in terms of the number of CF innervations. While PCs usually receive inputs from a single CF, each of the two primary dendrites of poly PCs receives discrete CF inputs (Figure 3E), suggesting that poly PCs could integrate different information from independent CFs. Since poly PCs are more prevalent in the posterior lobules, this information integration would be more beneficial for functions involving the posterior lobules, such as fear-evoked freezing behavior (Koutsikou et al., 2014).

Updates have also been made to our understanding of PC axonal morphology. Axonal swellings, known as torpedoes, were historically associated with pathological conditions (Louis et al., 2009; Babij et al., 2013; Louis et al., 2014; Redondo et al., 2015). However, axonal swellings have also been observed in healthy young animals and are thought to serve a different function from those associated with disease (Ljungberg et al., 2016). In fact, axonal swellings in healthy animals have been shown to enhance action potential fidelity and are linked to cerebellum-dependent motor leaning (Lang-Ouellette et al., 2021). Thus, these swellings appear to represent a form of axonal structural plasticity.

In addition to influences from traditional synaptic inputs, axon collaterals of other PCs, and ephaptic coupling, PCs have been reported to be affected by dopamine through D2 receptors (Cutando et al., 2022). D2 receptor-mediated dopaminergic signaling appears to play a role in regulating social behaviors. Dopamine release may originate from dopaminergic projection neurons outside the cerebellum, as early studies showed sparse dopaminergic innervation in the cerebellum (Panagopoulos et al., 1991; Ikai et al., 1992; Nelson et al., 1997). Another potential source could be a subset of PCs that express tyrosine hydroxylase, the rate-limiting enzyme in the classical dopamine synthesis pathway (Takada et al., 1993; Abbott et al., 1996). However, a recent study demonstrated that dopamine is synthesized in PCs via an alternative pathway involving a member of the cytochrome P450 superfamily, CYP2Ds (Li et al., 2023). This study also revealed a role for D1 receptors in Bergmann glial cells, contributing to motor and social behaviors. Understanding how dopamine released from PCs acts in a coordinated manner on D2 receptors in PCs and D1 receptors in Bergmann glial cells would be an intriguing avenue for future research.

UBCs are the only excitatory interneurons among them and are located in the GL of the cerebellar vermis. Specifically, UBCs are abundant in cerebellar regions associated with vestibular functions, such as lobules IX and X in the vermis, but are relatively rare in other lobules (Diño et al., 1999). They have a distinctive morphology, characterized by a single short dendrite with a brush-like end. UBCs receive an excitatory input from a single MF and inhibitory inputs from GoCs (Figure 4F-a; Harris et al., 1993; Floris et al., 1994; Kalinichenko and Okhotin, 2005; Mugnaini et al., 2011; McDonough et al., 2020). Their axons branch out locally and form MF terminal-like structures in the GL, which make synapses onto GCs or other UBCs (Mugnaini et al., 2011; van Dorp and De Zeeuw, 2015). UBCs have been thought to temporally amplify short-lived signals arising from a single MF by utilizing their unique circuit structures (Kinney et al., 1997; Mugnaini et al., 2011; van Dorp and De Zeeuw, 2014). However, several studies suggested that UBCs have more complex circuits and functions. Two biochemical subtypes of UBCs have been identified: mGluR1α-positive UBCs, which lack calretinin, and mGluR1α-negative UBCs, which express calretinin (Nunzi et al., 2002; Diño and Mugnaini, 2008). Functionally distinct subtypes appear to be associated with these biochemical differences: MF inputs excite mGluR1α-positive UBCs, but suppress firing in mGluR1α-negative UBCs (Russo et al., 2008; Borges-Merjane and Trussell, 2015; Zampini et al., 2016). Subtype-specific circuit structures have also been reported (Figure 4F-b). Primary sensory afferents from the vestibular organ selectively innervate mGluR1α-positive UBCs, while secondary sensory afferents from the vestibular nucleus innervate both UBC subtypes (Balmer and Trussell, 2019). Additionally, inhibitory connections from PC collateral axons specifically innervate mGluR1α-positive UBCs (Guo et al., 2021). Given these specific circuit structures, UBCs may mediate bidirectional activity patterns in distinct GC pathways by amplifying excitatory signals in one pathway while suppressing firing in another.

LuCs are a type of inhibitory interneuron in the GL, characterized by fusiform somas located beneath PC somas, and are intermediate in size between GoCs and globular cells, which are two other inhibitory interneurons in the GL. Anatomical studies demonstrated that LuCs extend poorly ramified horizontal and bipolar dendrites beneath the PC somas, where they receive inhibitory inputs from PC collateral axons, and project their axons to the ML, where they innervate MLIs (Lainé and Axelrad, 1996, 1998; Simat et al., 2007b). As mentioned above, the functional synaptic connections from PCs to LuCs were confirmed by electrophysiological recordings upon optogenetic activation of PCs (Witter et al., 2016). Another recent study presented a wider range of anatomical connections, revealing inputs from CFs, MFs, GC ascending axons, GoCs, other LuCs, and serotonergic fibers, and axonal projections to GoC dendrites (Figure 4G), but not to PC dendrites (Miyazaki et al., 2021). This raises an intriguing possibility that LuCs integrate various inputs and then coordinate two types of inhibitory circuits, though further functional studies are necessary (Miyazaki et al., 2021).

Globular cells have often been considered a subtype of LuCs (Lainé and Axelrad, 2002; Schilling et al., 2008; Prestori et al., 2019). Although globular cells exhibit different morphological properties, such as having a small and round soma with radiating dendrites, they share circuit properties with LuCs, in terms of inputs from PC collateral axons and projections to the ML (Lainé and Axelrad, 2002). One functional difference reported was that globular cells responded to both serotonin and noradrenaline, whereas small LuCs responded only to serotonin (Hirono et al., 2012). In addition, studies using single-cell transcriptomic analysis suggested three types of inhibitory interneurons around the PCL, presumably LuCs, globular cells, and candelabrum cells (Kozareva et al., 2021; Osorno et al., 2022). Although the functional relevance of distinguishing LuCs and globular cells is still unclear, the possibility is gradually emerging that they exhibit unique functions due to their distinct molecular expressions.

Candelabrum cells were identified more than two decades ago as neurons with a small soma situated in the PCL, with a few dendrites extending toward the surface of the ML and axons winding through or above the PCL (Lainé and Axelrad, 1994). They have long been the most enigmatic cell type in the cerebellum. Interestingly, a recent study established a mouse line showing fluorescent labeling in candelabrum cells, sufficiently distinguishing them from other cell types (Osorno et al., 2022). The study found that GABAergic candelabrum cells receive excitatory inputs from MFs and GCs, and strong inhibitory inputs from PC collateral axons, while they inhibited MLIs (Figure 4H). This mouse line may thus facilitate in unraveling the long-standing mystery of candelabrum cells.

Feedforward and feedback microcircuits are fundamental components of neural computations, widely distributed across various brain regions (D'Souza and Burkhalter, 2017; Park et al., 2023). Among these regions, the cerebellum stands out: GCs receive both feedforward and feedback inhibitions from GoCs, and PCs receive feedforward inhibition from MLIs (D'Angelo et al., 2013; Kim and Augustine, 2021). Furthermore, studies have revealed additional microcircuits within the cerebellar cortex, including feedback inhibitions on various neuron types from PCs and feedforward excitation on GCs mediated by UBCs (Kinney et al., 1997; Mugnaini et al., 2011; van Dorp and De Zeeuw, 2015; Guo et al., 2016; Witter et al., 2016; Guo et al., 2021; Osorno et al., 2022; Hariani et al., 2024). Although the specific arrangements of these microcircuits have been gradually elucidated (Balmer and Trussell, 2019; Guo et al., 2021; Halverson et al., 2022), a comprehensive understanding remains elusive. For example, PC collateral axons send feedback inhibitions to GCs, UBCs, LuCs, candelabrum cells, and MLIs, some of which connect each other, yet their circuit arrangements at the individual neuron level are not clear. There might be organized connections among them, similar to the connections between MLIs and PCs: MLIs appear to receive feedback inhibition from PCs that are distinct from the PCs innervated by those particular MLIs (Halverson et al., 2022). Since the resulting neuronal activity and potential regulations can differ according to the circuit arrangements (Figure 5A), further analyses utilizing advanced techniques, such as a combination of sparse labeling and anterograde transsynaptic labeling, are expected.

A large amount of effort has been made over the past decade to advance our understanding of cerebellar circuits. We assume that the additional organization of cerebellar circuits increases the options of regulating PC activity, consequently leading to pattern separations of signals provided by the cerebellum. However, it is still difficult to picture exactly how such circuit organization collaboratively controls PC activity. Superficially contradicting aspects make it even more challenging. For example, in MLI circuits, MLIs in the inner ML or MLI1s likely have strong inhibitory effects on PCs due to their high efficiency of individual inhibitory synaptic inputs (Arlt and Häusser, 2020) and strong electrical coupling (Alcami and Marty, 2013; Lackey et al., 2024). However, they are inhibited by MLIs in the outer ML or MLI2s (Rieubland et al., 2014; Lackey et al., 2024). This raises concerns that the strong inhibitory effects from the inner MLIs might be underutilized or wasted. Additionally, glutamate spillover from CFs (Coddington et al., 2013; Arlt and Häusser, 2020) and feedback signals from PCs (Witter et al., 2016) could have both excitatory and inhibitory effects on MLIs, raising the possibility that they might functionally cancel each other out. These concerns might be alleviated by obtaining more precise information about the spatiotemporal aspects of MLI circuits, such as temporal gaps or the distinct spatial distances between excitatory and inhibitory effects (Figure 5B). A study provided a clue about the first concern by recording the activity of MLI1s, MLI2s, and PCs during licking, which periodically modulates PC firing (Lackey et al., 2024). Changes in PC firing rates were inversely correlated with MLI1s and correlated with MLI2s, suggesting the organization of bidirectional control of PC firing by these two groups of MLIs. Alternatively, it would be interesting to computationally predict the tuning of PC activities by including all known components of the complex MLI circuit organizations.

While this article focuses on neurons and excitatory inputs in the cerebellar cortex, many other factors also contribute to cerebellar circuit structures as components or regulators. These include non-neuronal cells, neuromodulators, and molecules arriving from circulation (Schweighofer et al., 2004; Koibuchi and Ikeda, 2013; Araujo et al., 2019; Guedes et al., 2022; Gruol, 2023; Li et al., 2024). To comprehensively understand cerebellar circuit structures and their functions, these factors must also be taken into account. Indeed, information on their contributions has gradually been updated. For example, immunohistochemical analyses have investigated layer-or lobule-dependent distributions and orientations of neuromodulatory projection fibers (Carlson et al., 2021; Longley et al., 2021). Other studies have reported novel interactions between PCs and Bergmann glial cells through dopamine released from PCs (Li et al., 2023), and cerebellar circuit formation through cytokine-dependent regulation of microglial functions (Kana et al., 2019; Guedes et al., 2023). Regulation by these factors appears less selective than synaptic transmission-mediated regulation, as evidenced by volume transmission of neuromodulators (Özçete et al., 2024) and delivery of cytokines through circulation (Wu et al., 2023). Nevertheless, recent studies have shown their involvement in specific cerebellar functions or disfunctions (Yamamoto et al., 2019; Kim et al., 2021; Cutando et al., 2022; Snell et al., 2022; Guedes et al., 2023; Li et al., 2023; Stanley et al., 2023; Dewa et al., 2024). As the broad range of cerebellar functions appear to be achieved through computations in different functional modules/domains, as shown in human studies (Guell et al., 2018; Diedrichsen et al., 2019; King et al., 2019), it would be intriguing to determine how the factors described in this paragraph exhibit a certain extent of spatial selectivity within the cerebellum (Figure 5C).

Well-organized circuit structures of the cerebellum gradually form during postnatal development through dynamic processes regulated by both cell-autonomous mechanisms and interactions between developing neurons in the cerebellum (Leto et al., 2016; Park et al., 2021; van der Heijden and Sillitoe, 2021; Kim et al., 2023a). While extensive research has focused on these processes for traditionally known circuit structures, less is understood about the developmental mechanisms of newly identified structures. The updated structures are relatively minor or complex, complicating the investigation of their developmental processes. In general, a useful approach to understand developmental processes may be to examine the effects of molecular deficiencies during specific developmental windows. Developmental time-and cell type-dependent transcriptome analysis data could help to identify relevant molecules (Figure 5D). This approach indeed led us to discover developmental mechanisms that maintain the PC soma monolayer after its formation. Specifically, multiple epidermal growth factor-like domains protein 11, which substantially increases during the late stage of GC development (Rosenberg et al., 2018), plays a role in this maintenance (Jun et al., 2023). Furthermore, integrating transcriptomic data with known information about circuit formation could be beneficial for hypothesizing the developmental mechanisms of the updated structures. For example, neuropilin-1 (Nrp1) expressed in BCs is crucial for the formation of pinceau synapses in the AIS of PCs (Telley et al., 2016). Whereas candelabrum cells also express high levels of Nrp1, they do not innervate PCs (Osorno et al., 2022). Two potential mechanisms are proposed to explain this discrepancy (Schilling, 2024): the timing of expression of Nrp1 and its partner proteins in PCs, MLIs, or candelabrum cells might be regulated, or different splice variants of Nrp1 may contribute differently to their synaptic connections. Thus, hypotheses based on existing data would help advance our understanding of how cerebellar circuits, including newly identified structures, are coordinately organized during postnatal development.