Cerebellar Modules and Their Role as Operational Cerebellar Processing Units

It is difficult to give a consensus of informed opinion because, although there is much informed opinion, there is rather little consensus. David Colquhoun (1971) Lectures on Biostatistics. Oxford, UK: Clarendon Press.

The cerebellum has long been considered as a uniform structure with well-organized in- and output relations that ultimately serves a particular adaptive control function that is mainly, if not completely, used for coordinating, modifying, adapting, and learning motor functions [1, 2]. By now, we have learned that the idea of an operational uniform cerebellar cortex needs to be revised [3]. In addition, the functional extent of cerebellar influence extends to cognitive, affective, and autonomic domains [4, 5]. Yet, although not completely resolved, one consideration that is generally accepted is that the basic operational unit is the cerebellar module. Each cerebellar module includes a longitudinal, i.e., (para-)sagittally organized, zone of Purkinje cells (PCs) in the cerebellar cortex that receives common climbing fiber input from a particular region of the inferior olive, and in turn, the same PCs target a discrete part of the cerebellar nuclei. This part of the nuclei is also targeted by collaterals of the same olivocerebellar axons that provide the climbing fibers to the zone of PCs, and harbors a population of small GABAergic neurons that project back to the same part of the inferior olive. This precise olivo-cortico-nuclear circuitry forms the core of individual cerebellar modules (Fig. 1). The basic cerebellar modules, A, B, C, and D, as defined by Voogd [7] have now each been subdivided into several smaller entities and in some cases, based on similar peripheral receptive fields, these have been shown to comprise yet smaller units, termed microzones, which are the cortical component of micromodules [6, 8‐12].

Several decades ago, it became clear that the apparent uniformity of the cerebellar cortex masked underlying differences in the expression of a multitude of genetic markers in a broad transverse and finer parasagittally organized patterns, which are commonly referred to as stripes [8]. Much work has been devoted to describe the organization of the anatomically defined zones in relation to these biochemically defined stripes [13]. This interest has gained new impetus given the additional finding that differences in physiological properties can be related to this biochemical heterogeneity [14‐17]. Such a finding raises the important possibility that individual cerebellar modules may not be uniform in their operation [3]. The current paper brings together up-to-date views on cerebellar modules. The general approach is at a systems level in order to understand the neural circuit basis of cerebellar modules and to establish to what extent they are functional entities and can fulfill functions that are independent of other modules.

Jan Voogd, who first used the term “cerebellar module” to describe the basic operational unit of the cerebellum, provides an historical synopsis. Izumi Sugihara subsequently reviews his work on the relation between modules and several biochemical markers. His detailed scheme of the relation of olivocerebellar organization and the aldolase C (zebrin II) pattern is now widely used, but he also points to the shortcomings of the aldolase C pattern and the great potential that additional markers may have in studying both the development and the adult organization of cerebellar modules. Doug Wylie uses the vestibulocerebellum system in the pigeon to examine sagittally organized zones of PCs and how they modulate their activity in response to optic flow. Although these zones are present in lobule IXcd and in lobule X, their relation to the zebrin pattern of stripes differs, as there is no distinctive pattern in lobule X, whereas the same functional zones cover adjacent stripes of zebrin II-positive (ZII+) and zebrin II-negative (ZII-) PCs in lobule IX. This raises the important issue that zebrin alone is insufficient as a marker to describe the functional heterogeneity of PCs. Richard Hawkes subsequently explores the extent to which cerebellar modules can be divisible into their smallest processing units, leading to the idea of the “cerebellar quantum.” As such, the cerebellar cortex may be made up of short strips or microzones (i.e., positioned within an anatomically defined zone or biochemically defined stripe) or, maybe, elongated patches, which, together, may comprise several thousands of individual processing units. Parallel processing power, positional coding, improving signal-to-noise ratios, and functional processing diversity are potential advantages of such modular processing. The question of what constitutes the basic functional unit of the cerebellum is also asked by Fredrik Bengtsson and Henrik Jörntell. However, they address this important question from a systems level physiological perspective and propose that the fundamental unit of the cerebellar cortex is a population of PCs located within a given microzone, working together as a “super PC.” In pinpointing the cerebellar quantum (Hawkes) or the super PC (Bengtsson and Jörntell), both sections touch upon the role of mossy fiber afferents that show a more prominent transverse orientation but also adhere to modular organizational principles. This aspect is further discussed by Roy Sillitoe and colleagues who explore the relation between the organization of the mossy fiber systems, granule cells, and cortical interneurons.

These initial sections mostly deal with the anatomical foundations of the cerebellar modular functionality and are followed by sections that concentrate on their physiological properties. Martijn Schonewille reviews differences in several physiological properties of PCs with different molecular signatures. This significant recent development in cerebellar physiology is also highlighted by Gang Chen and Tim Ebner, who further explore the physiological and functional differences of modules based on ZII+ and ZII− stripes. Philippe Isope, Ludovic Spaeth, and Antoine Valera, on the other hand, return to the effect of mossy fiber input on plasticity within modular circuits and propose that modular identity may not be rigid but adaptable.

Exploring the fate of cerebellar modular output, Sho Aoki and Tom Ruigrok survey how this output is distributed and used by other areas—does the output from individual modules remain separated or can the outputs of different modules converge to be jointly processed in common receiving areas? Finally, Richard Apps and colleagues review recent developments on cerebellar involvement in emotional behavior. In line with the ideas developed in the previous section, they call attention to a body of evidence that the various modular constituents of the vermal A zone are connected to widespread brainstem and diencephalic (limbic) areas. They suggest that different components of the A module (possibly relating to micromodules) may carry out different, but orchestrated, aspects of an integrated emotional response.

The term “modules” was first used for Purkinje cell zones defined by their cerebellar and vestibular target nuclei and their climbing fiber afferents by Voogd and Bigaré [18] in a paper read at a meeting in Montreal. Our paper was based on the work of Groenewegen et al. [19] and Bigaré [20]. Cerebellar modules, however, were recognized before this term was used by us. In Brodal’s [21] study of the olivocerebellar projection in the cat and Jansen and Brodal’s [22, 23] studies of the corticonuclear projection, the lobules were the units or modules in their description. As a byproduct, they described an intermediate zone, located in the anterior lobe hemisphere, lateral to the vermis, that, like the vermis, received an olivocerebellar projection from the accessory olives but projected to the interposed nucleus. This was the first definition of a longitudinal Purkinje cell zone as we know it today. Attempts to extrapolate the intermediate zone to more posterior parts of the cerebellum failed, because the authors did not recognize the loops in the folial chains in the posterior cerebellum (Fig. 2(a1)).

My contribution to the distinction of longitudinal Purkinje cell zones was based on the following considerations [26, 28]. Bolk’s [25] description of the cerebellar vermis and hemisphere as folial chains with ansiform and (para-) floccular loops defined the topography of the Purkinje cell zones (Fig. 2(a2)). The distinction of anterior and posterior subdivisions in Brunner’s [29] interposed nucleus and of dorsal and ventral subdivisions of the lateral cerebellar nucleus as target nuclei of the zones was based on the localization of the relatively small myelinated fibers from the posterior interposed nucleus in the medial one-third and of the larger fibers from the anterior interposed and the dorsal part of the lateral cerebellar nucleus in the lateral two-thirds of the brachium conjunctivum [30] (Fig. 2(b)). Finally, the observation of compartments in the white matter that channeled the Purkinje cell axons to their target nuclei provided an intrinsic coordinate system for the zones. The innervation of Purkinje cell zones by specific subdivisions of the inferior olive followed from the localization of their olivocerebellar fibers in the corresponding white matter compartments [7]. Their termination as longitudinal zones of climbing fibers was first shown by Courville et al. [31], the organizer of the Montreal meeting. As a consequence, seven zones were distinguished (Fig. 2(a3)). Two were located in the vermis. The medial A zone projecting to the fastigial nucleus, the lateral B zone to Deiters’ nucleus. In the hemisphere, the C1 and C3 zones that connect with the anterior interposed nucleus and C2 that projects to the posterior interposed nucleus replaced Brodal and Jansen’s intermediate zone. The hemisphere was found to be composed of the two D zones that project to different parts of the dentate nucleus.

This simple zonal pattern was found to be inadequate after Hawkes and Leclerc’s [32] discovery of the “stripy”’ distribution of ZII+ and ZII− PCs. Apart from the identification of the B, C1, and C3 Purkinje cell zones as being positioned within ZII− stripes and the C2, D1, and D2 zones within ZII+ stripes, a number of narrow, ZII+ “satellite bands” were found to be present. These narrow bands, like their broad counterparts, are characterized by their climbing fiber afferents and, presumably, also by their corticonuclear projection [10, 33, 34]. The reconstruction of this more complicated map now serves as the standard reference for the description of zonal organization of the cerebellum [13].

Where the history of the Purkinje cell zones goes back to the early twentieth century [35], microzones made their appearance much later. They were first identified in the B zone of the cerebellum of the cat by Andersson and Oscarsson [27]. They consist of 50-mm-long and at least 200-μm-wide strips of PCs sharing the same climbing fiber receptive fields. The five microzones distinguished in the B zone differ in their input from forelimb or hindlimb nerves or a mixture of these nerves and the short or long latency of the response (Fig. 2(c)). The somatotopical localization in the B zone with the forelimb medially and the hindlimb laterally earlier was described by Oscarsson and Uddenberg [36]. Evoked potentials from the dorsal spino-olivary climbing fiber system [37] and the exteroceptive component of the cuneocerebellar mossy fiber system [38] are distributed in a similar, but more detailed microzonal pattern in the anterior lobe C3 zone of the cerebellum of the cat [39]. Overall, mossy fibers innervating these microzones had receptive fields resembling the climbing fiber receptive field defining that microzone [40].

What is the morphological basis for the microzones? The termination of mossy fibers in narrow longitudinal aggregates of rosettes in the granular layer was already described by Scheibel [41]. A similar, microzone-like distribution of individual climbing fibers was reported by Sugihara et al. [42]. The significance of the termination of mossy fibers in multiple longitudinal strips of mossy fiber terminals is difficult to understand, because this pattern would be erased by the parallel fibers [43]. Microzones, defined by their cutaneous receptive field of olivary mediated complex spike responses, thus far, only have been identified in the C1 and C3 zones of the anterior lobe. The microzone-like terminations of single or small groups of climbing and mossy fibers are present in the entire cerebellum. It would be interesting to know what these thousands or even millions of microzones in other parts of the cerebellum represent.

Self-motion of an organism through a world cluttered with visual stimuli results in “optic flow” across the entire retina [53]. This visual information is analyzed by retinal-recipient nuclei in the pretectum [54] and accessory optic system (AOS) [55], and reaches the vestibulocerebellum (VbC) via particular subnuclei in the inferior olive [56]. The VbC includes the flocculus, nodulus, and uvula, and is a site of visual-vestibular integration important for the generation of compensatory eye movements and the analysis of self-motion [57‐59].

In birds, where the cerebellum essentially appears as a vermis without hemispheres [60], the VbC includes folia IXcd and X [61]. The optic flow information to the VbC originates in the pretectal nucleus lentiformis mesencephali (LM) and the nucleus of the basal optic root (nBOR) of the AOS [62‐65]. The pigeon VbC shows many aspects of the classic modular organization of the cerebellum [18] as shown in Fig. 4a. The complex spike activity (CSA) of Purkinje cells (PCs) in the pigeon VbC responds best to particular patterns of optic flow resulting from self-translation or self-rotation through space, and these PCs are organized into sagittal zones across folia IXcd and X. As in mammals, CSA in the flocculus is modulated by rotational optic flow about either the vertical axis (VA neurons) or an horizontal axis oriented 45° to the midline (HA neurons) [70‐72]. In pigeon, there are two VA zones interdigitated with two HA zones [73]. In the uvula/nodulus, the CSA responds best to optic flow resulting from self-translation [66]. There are four response types organized into three sagittal zones. In the most medial zone, CSA responds best to optic flow resulting from translation backwards along an horizontal axis 45° to the midline such that there is a focus of contraction at 45° contralateral azimuth. Medial to this is a zone where the CSA responds best to optic flow resulting from either (i) forward translation along an horizontal axis 45° to the midline such that there is a focus of expansion at 45° ipsilateral azimuth, or (ii) upward translation along the vertical axis. Lateral to this is a zone where the CSA responds to the optic flow resulting from downward translation along the vertical axis [66]. A sagittal organization is also apparent with respect to the projection of PCs in the VbC: PCs in each of the optic flow zones project to particular regions in the vestibular and cerebellar nuclei [74‐76]. Also, each of the optic flow zones receives climbing fiber (CF) input from particular regions of the medial column of the inferior olive (mcIO) [77, 78] (see also Fig. 4c).

A sagittal organization in IXcd is apparent with respect to the expression of Zebrin II (ZII; a.k.a. aldolase C [79]. As in mammals [44], ZII is heterogeneously expressed such that there are sagittal stripes of PCs exhibiting high ZII expression (ZII+) alternating with sagittal stripes of PCs that show little or no ZII expression (ZII−) [80]. In the VbC, there are seven stripe pairs (Fig. 4a). The most medial ZII− stripe, P1−, is bisected by a thin ZII+ stipe, such that P1− is divided into medial and lateral region (P1−med, P1−lat) (Fig. 4b). Similarly, the P2+ stripe is bisected by a notch that contains no PCs, effectively dividing the stripe in two halves (P2+med, P2+lat) (Fig. 4b). Using electrophysiological recordings combined with immunochemistry, we showed that the optic flow zones spans a ZII+/− stripe pair (Fig. 4a) [66, 67]. For example, the contraction zone spans P1+ and P1−med. As such, we consider that a ZII+/− pair represents a functional unit in the VbC, but what are the differences between the ZII+ and ZII− stripes within the unit? We have shown that they receive CF input from separate, but adjacent areas of the mcIO (Fig. 4c) [81, 82], and there is some suggestion that the ZII+ and ZII− PCs have differential projections [76]. We have some evidence that the CSA of ZII+ PCs shows a greater depth of modulation to optic flow stimuli, compared to the ZII− PCs within the same functional unit [83]. This applies if one compares ZII− and ZII+ PCs in IXcd, and if one compares the ZII− PCs in IXcd with the PCs in X (all ZII+). The depth of modulation of ZII+ PCs in IXcd is not different to that of PCs in X [83]. Moreover, the ZII+ and ZII− stripes likely receive different mossy fiber (MF) inputs. Both nBOR and LM project directly to IXcd as MFs [62, 63], and the majority (~ 85%) of these terminate adjacent to the ZII+ stripes [68] (Fig. 4a). It is not known if other MF afferents target the ZII− stripes.

Note that the optic flow zones span folia IXcd and X, but the ZII stripes do not. Rather, all the PCs in X are uniformly ZII+ [80]. Folia IXcd and X also differ with respect to MF inputs. The optic flow MFs from nBOR and LM mentioned above innervate IXcd, but not X. In contrast, there is a primary vestibular projection to folium X, but not IXcd [69] (see Fig. 4a).

In summary, the pigeon VbC contains optic flow modules that are sagittally oriented and span folia IXcd and X. The classic sagittal zonal organization is apparent with respect to PC response properties, CF inputs, and PC projections. However, there is clearly a transverse component to the modules as well, since IXcd and X receive discrete MF inputs carrying optic flow and vestibular information, respectively. Finally, the modules clearly contain subregions defined by neurochemistry, as each module encompasses a ZII+/− stripe pair. Whether this type of modular organization applies to other parts of the cerebellum, or the VbC in other vertebrate classes, remains unknown.

The modular nature of the cerebellar cortex suggests that it represents a map or family of maps, although what exactly is being “mapped” is less evident. The afferent topography is perhaps the simplest answer, in which case the map is fundamentally discontinuous in the sense that neighboring representations of body regions are neither anatomically nor physiologically continuous. What is the cerebellar “quantum”? In this context, the central idea is topographical equivalence: all cells in the “quantum” share a common chemistry, receive statistically identical inputs, project to the same target field(s), and have equivalent interneuron connectivity. Such a quantum would represent the smallest unitary processing unit.

Cerebellar modular architecture arose early in vertebrate evolution as the ground plan across birds and mammals is generally conserved. The largest cerebellar cortical compartments are the transverse zones (note that these are distinct from the sagittally oriented zones defined by olivocorticonuclear connectivity). In the mammalian vermis, four transverse zones are found in all species studied—the anterior zone (AZ), central zone (CZ), posterior zone (PZ), and the nodular zone (NZ) [84, 85] (in mouse a subdivision of the CZ has been identified—[86]: in birds, the ground plan has an additional transverse zone—the LZ [80]). Transverse zones evolve independently in response to different lifestyles (mosaic evolution). For example, in bats the echolocation centers in lobules VI/VII are accommodated by an expansion of the CZ—[87], and in the blind star-nosed mole, the CZ and NZ (visual receiving areas) are reduced and the trigeminal (star)-receiving areas (NZ and crus I/II) are expanded [88]. In sum, the cerebellar cortex comprises of the order ~ 10¹ transverse zones: in a mouse each of ~ 10⁴ Purkinje cells (PCs).

Transverse zones are further divided into parasagittal stripes. How these stripes relate to the microzones identified by Oscarsson and his group [for review, see 88] is not certain: a tentative common framework is provided by the group of Voogd and Sugihara [10, 11, 13]. Stripes are discontinuous across transverse zone boundaries [85], suggesting that the earliest parcellation of the cerebellum during development is into transverse zones and subsequently these further subdivide into stripes. As is the case for zones, the number and variety of PC stripes is also not properly understood. The problem of how many stripes are present is exacerbated because many stripes revealed by ZII expression are, in fact, composite (e.g., heat shock protein-HSP25+/− subtypes within the ZII+ population [84]; PLCβ4+ sub-stripes within the ZII− population [89] etc.). As a consequence, the absolute number of stripes remains uncertain. Secondly, when molecular markers and mutant phenotypes are used in combination, some 10 PC subtypes can reliably be identified: this is likely an underestimate. By way of estimate, 5 transverse zones, each duplicated on either side of the midline, and 20 stripes per zone (based on connectivity plus chemistry) yields ~ 200 stripes per cerebellum, each comprising < 10³ PCs in the mouse. This is almost certainly an underestimate.

Stripes are further subdivided into strings of patches. For example, tactile receptive field mapping of trigeminal representations reveals an elaborate mosaic of somatosensory patches (so-called fractured somatotopy: [90‐92], which in some cases have been shown to align with ZII+/− stripe boundaries [93, 94]. A complementary heterogeneity was also revealed by Garwicz et al. [43], further dividing microzones in the anterior paravermis (C3) of the cat into multiple rostrocaudal patches. Possible anatomical correlates of patches—blebs (e.g., [95] and expression markers, such as NOS [96] and dystrophin [97]—confirm an elaborate parcellation of the granular layer. The upshot is the dicing of stripes into several thousand functional patches, each comprising ~ 10² PCs [98].

The cerebellar cortex is close to a pure feed-forward structure with little or no cross talk between neighboring stripes, so their proximity would seem irrelevant. However, this simplistic view may be wrong. Functional aggregates—limb inputs to the AZ, eye inputs to the flocculonodular lobes, trigeminal inputs to crus II, etc.—are found throughout the cerebellar cortex: indeed, this is the reality beneath the long-outdated idea of cerebellar homunculi. Such “neighborhoods” may be functionally critical due to MF data sharing via parallel fiber innervation.

So why does the cerebellum need a modular structure? We can suggest three reasons. First is the requirement for parallel processing. It is mandatory for the motor system to respond in a timely fashion and where there are so many degrees of freedom to control in an integrated manner serial processing is a non-starter. Hence, a highly parallel modular architecture has evolved to serve real-time motor control.

Thirdly, topographically equivalent quanta are a means to manage cerebellar signal-to-noise problems by exploiting the internal redundancy afforded by multiple, statistically identical PCs as a filter to generate a smoothed, more reliable output. The number of PCs needed—and hence the minimum quantum size—depends on how noisy each input is and how reliable the output needs to be.

In conclusion, the speculations above suggest that the cerebellar quantum is either a stripe (several hundred per cerebellum, each < 10³ PCs in mouse) or a patch (several thousand per cerebellum, each < 10² PCs). This is not to imply that multiple quanta do not work in tandem to generate specific behaviors. First, perhaps cerebellar neighborhoods reflect a higher functional order—functionally related stripes/patches arrayed mediolaterally within a transverse zone and innervated by a common set of parallel fibers: stripes in the AZ processing forelimb signals also having access to hind limb information; vibrissal patches in crus I/II receiving contextual data about the lips and teeth, etc. Secondly, stripes may work as pairs—for example, ZII+/− stripe pairs in the pigeon NZ respond in concert to optic flow [79; and above]. Finally, multiple stripes may cooperate. Support for this view comes from data showing that networks of patches are linked by common MF inputs (see section by Spaeth et al.) and evidence that multiple stripes cooperate to control single muscles [99].

Based on anatomical and physiological mapping studies, there are some indications to support this view, but also some caveats that prevent us from drawing a definite conclusion.

First of all, one needs to define the terms used to describe functional units of the cerebellum. The terms modules and micromodules have historically been used in a confusing non-conformative way and here we try to disentangle the terminology. The relationship between a module and a micromodule is that a module is a sagittal zone of cerebellar cortex, the parts of the inferior olive (IO) that supplies that zone with climbing fibers (CFs), and the subdivision of the cerebellar nuclei (CN) that the sagittal zone sends its Purkinje cell (PC) axons to. A micromodule, or what members of our lab originally referred to as a microcomplex, consists of a microzone within the sagittal zone (each sagittal zone may contain several 10’s of microzones [100] and its associated subdivisions of the IO and CN [12, 40, 101]. The PCs of each microzone predominantly contact a small group of neurons in a specific CN subdivision, and here we refer to this set of neurons as a “micro-group.” Similarly, the PCs of each microzone receive CFs from a small part of a specific subdivision of the IO, and we refer to this set of IO neuron as a “micro-part” [101].

To date, there is no evidence to support that different PCs of the microzone control specific CN cells within the micro-group. Rather, individual PCs diverge extensively in their projection to the CN and each CN cell receives a wide convergence of PC inputs [102]. The lack of differential CN cell control within the micro-group is the rationale for assuming that it is acting as one unit, which consequently has one functional contribution. Caveat to this assumption is if separate PCs within the microzone are eventually shown to have differential control of these CNs, or if the mossy fibers that drive the CN cells [103] split this group into smaller functional units. Notably, there is a specific relationship between the receptive fields of the mossy fiber input and of the PC-mediated CF input to the individual CN cell [103], which suggests that the mossy fiber input to the CN cell is defined by learning and can therefore be expected to be homogenous for CN cells within the same micro-group. However, in the adult animal, the mossy fiber to CN plasticity do not seem highly active or easily induced [104], which of course does not contradict the possibility that it exists or that it might be highly active under development.

Although not included in the original concept of a micromodule, recent findings suggest that the inhibitory nucleo-olivary pathway should be included [105]. As the name suggests, the pathway originates in the CN and is under control of the PC output. A decreased PC firing will result in a disinhibition of the IO, thus forming a closed inhibitory feedback loop between the IO and the cerebellar circuit. The pathway seems to be zonally specific [106]. The spontaneous activity in the PCs is controlled by the level of IO input [107, 108]. Given that the assumption of a uniform micro-group of CN cells above applies, the total level of nucleo-olivary inhibition within a micromodule would be expected to be uniform and most of its PCs would have the same set point for their spontaneous firing activity. Different micromodules, however, may well have different levels of total nucleo-olivary inhibition and hence different levels of spontaneous PC activity. This scenario could work as an explanatory model for multiple reports that there are overall differences in the PC and CF activity between zebrin stripes [16, 17], as these appear to have a large degree of congruence with the functionally defined microzones [8].

The general idea that the modules of the cerebellum are functionally specific is supported by inactivation of specific areas of the IO, which results in functionally specific deficits in motor control [109]. The functional effects of the olivary inactivation can readily be explained as different modules predominantly project to different motor systems, i.e., vestibulospinal, tectospinal, reticulospinal, or rubrospinal systems as well as the corticospinal system [110]. For each micromodule, each CN micro-group can be expected to activate specific aspects of the function of the specific motor system for the module, which would be the cause of functional differences between micromodules. On the output side, each micro-group is divergent and contact strongly divergent upper motor neurons that in turn contact divergent spinal interneurons [111]. Yet, some center of gravity for which combinations of muscles each micromodule controls exists [99].

As every microzone has a specific function, assuming that it is the control of a specific set of muscles, for example, the PCs of the microzone will learn or potentiate specific mossy-/parallel fiber input that relates (sensory, motor, or sensorimotor) to the activation of that particular set of muscles. Depending on the specifics of a particular movement, different parallel fiber inputs will be active to a different degree and perhaps with a different temporal relationship to the CN output of the micromodule. Depending on the degree of correlation with the output effect of the CN group, subsets of parallel fiber inputs to PCs within a given microzone will be either potentiated or depressed. If the micromodule indeed is the minimal functional unit of the cerebellar circuitry, then the consequence is that the population of PCs in the microzone effectively is combined into one “super PC,” which operates with the same micro-group of CN neurons. The advantage of a super PC would be that it provides the possibility to sample a much higher total number of mossy fibers, from which the mossy fibers with the highest possible correlations with the micromodule activity functions can be selected, to the control function of the micromodule than a single PC alone would be capable of.

Sagittal zones originate during early cerebellar development, and nearly all major cell types in the cerebellum respect the boundaries of zones [8, 112]. The zonal patterns of developing and adult Purkinje cells (PCs) have been extensively studied, but we are far from fully understanding how mossy fibers and the various types of interneurons are restricted within the zonal framework. This is an intriguing problem to consider from a circuit perspective because mossy fibers form mono- and di-synaptic connections to each class of interneurons in the cerebellar cortex.

Mossy fibers project from over two-dozen brainstem and spinal cord nuclei. Functionally similar mossy fibers terminate on granule cells within the same transverse domains in the cerebellar cortex. Within these transverse domains, mossy fiber terminal fields organize into parasagittal zones that have a reproducible anatomical relationship with olivo-cortico-nuclear modules. In contrast to climbing fibers, which terminate on just one or two contralateral zones of PCs, mossy fibers branch to terminate in multiple bilateral zones [113]. Furthermore, sensory information from different mossy fiber sources can converge onto single granule cells [114]. Cues derived from Purkinje cell clusters are thought to provide the organizational scaffold for the zonal distribution of both climbing fibers and mossy fibers. Purkinje cell clusters initially express transient parasagittal molecular markers as early as E14 in mice. Although Purkinje cell and climbing fiber patterning starts early, mossy fiber arrival in the cerebellum spans mid-embryonic and postnatal development [115]. This suggests that a protracted relationship might exist for module patterning to occur. Indeed, mossy fibers directly contact PCs through the second postnatal week in mice [116]. This idea is consistent with data showing that mossy fibers do not exhibit clear-cut zones until after birth [117]. Despite the clear heterogeneity of mossy fiber terminal field domains, their zones are generally broader and not as sharply defined as those of climbing fiber projections or the PCs [6]. Adding to this complexity is that mossy fiber receptive fields, mapped by recording granule cell responses to tactile stimuli, reveal multiple sensory representations of body parts in mosaic patches that form a “fractured somatotopy” [91, 93]. This complexity is mirrored in the organization of the mossy fiber targets, the granule cells. Granule cells are also restricted to transverse and parasagittal patterns of gene expression and these patterns are reflected by abnormalities detected after various experimental manipulations [96, 118‐120]. Granule cell progenitors arise from the rhombic lip and proliferate in the external granular layer (EGL). Despite potential molecular differences in the progenitor populations, it is not clear how lineage influences the final patterning of granule cells. However, it seems that temporal mechanisms may distinguish broad transverse domains such that specific granule cells are fated to specific lobules [121]. It is also possible that interactions between the EGL progenitors and/or recently differentiated granule cells and Purkinje cell signals may direct parasagittal granule cell patterning. Between E11 and E14 in mice, cells arising from the rhombic lip travel to the EGL where, under the control of Purkinje cell signals, the EGL expands through progenitor proliferation. Granule cells must then traverse past the Purkinje cell dendrites and somata in order to reach what will become the granule cell layer [112]. During this time, Purkinje cell parasagittal zones could influence granule cell molecular phenotypes. It has also been suggested that mossy fibers might play an active role in patterning granule cell zones [120]. Interestingly, granule cell parallel fiber projections are also patterned relative to the Purkinje cell map (see section by Isope, Spaeth, and Valera).

Similar to granule cells, the excitatory unipolar brush cells also exhibit transverse and parasagittal zonal restriction. After they are born, unipolar brush cells migrate through the white matter en route to lobules IX and X, and by adult they localize to the granule cell layer [122]. Differential molecular expression distinguishes them into three subtypes, calretinin+, mGlrR1α+, and PLCβ4+, and mGlrR1α− and PLCβ4+, which all respect the parasagittal Purkinje cell zones [123]. There is compelling evidence to suggest that PCs have a large impact on the distribution of unipolar brush cells. For instance, unipolar brush cells lose their restriction to lobules IX and X when normal Purkinje cell patterning is disrupted by genetic lesions (e.g., via the deletion of Ebf2: [124]).

Much less is known about the zonal patterning of the inhibitory interneurons. Golgi cells, for example, exhibit molecular restriction in the anterior-posterior axis with some degree of morphological restriction to parasagittal zones. There are multiple molecular subtypes of Golgi cells, but so far, only the subtype expressing ZAC1 is known to be restricted to the posterior zone [125]. Golgi cell apical dendrites, which ascend into the molecular layer and contact parallel fibers, respect the borders of Purkinje cell parasagittal zones. Fewer than 3% of Golgi cell dendrites cross the borders of Purkinje cell zones and, though mechanisms have been suggested for this restriction, it is not clear how this relationship develops or is maintained [126].

Least is known about the patterning of basket and stellate cells in the molecular layer. Like Golgi cells, basket and stellate cells could exhibit a morphological restriction to zones wherein, particularly for basket cells and less so for stellate cells, their axons extend in the parasagittal plane. This may result in restriction of the inhibitory influence of the basket or stellate cells to specific zones [127, 128].

To achieve this restriction, it is possible that the parasagittal orientation of basket and stellate cell axons could have followed the spreading of Purkinje cell clusters into zones during cerebellar development. This argument is supported by the idea that modules might have their origins in the earliest stages of cerebellar development and therefore cells that are born later in cerebellar development, such as interneurons, develop within a circuit that is already committed to a zonal map. The outcome of these multicellular rearrangements plus the targeting of mossy fibers to the cerebellar input layer is thought to be modulation of Purkinje cell simple spikes via parallel fiber projections [129]. Both the frequency and regularity of simple spikes are dynamic during postnatal development and consistent with the maturation of parallel fiber synapses and establishment of mature Purkinje cell zonal expression patterns [129]. The maturation process of zones is mediated by spontaneous activity and sensory experience, which may intersect with genetic programs to integrate or sculpt mossy fibers into modules [112]. Ultimately, however, the formation and function of an operational module may depend on several factors including regional variations in Purkinje cell morphology, Purkinje cell packing density, granule cell packing density, neuronal soma size, intrinsic Purkinje cell firing properties, synaptic plasticity, the positions of mossy and climbing fiber synapses within their target layers, the distributions of the various cerebellar interneurons, and perhaps even glia [3].

A wealth of anatomical and immunohistochemical data has revealed the modular organization of the cerebellum and its chemical landmarks, as described above. The efforts to understand the physiological and functional features of this organization have thus far not matched that. This section will discuss the progress made so far in analyzing the differences at the physiological level between modules in relation to the differential gene expression patterns.

As detailed in the contributions to this consensus paper by Voogd, Sugihara, and Hawkes, a dominant feature of the cerebellum is its longitudinal architecture as defined by the parasagittal organization of its afferent and efferent projections and by the molecular compartmentalization of these parasagittal zones (see Figs. 2 and 3). Although highlighted by the expression of zebrin II/aldolase C (ZII), the parasagittal organization involves a host of other molecules, expressed on PCs in either a ZII+ or ZII− banding pattern of stripes [8, 158] [8, 158]. Importantly, many of these molecules control neuronal excitability, for example EAAT4 and mGluR1 subtypes.

The contribution by Schonewille describes recent studies on the differential firing characteristics of PCs in ZII+/− stripes, with the key observations that the spontaneous simple spike and complex spike firing rates are higher in ZII− than in ZII+ stripes (see Fig. 5 and Table 2) [16, 17]. Several of the firing differences are intrinsic to PCs as they persist when synaptic inputs are blocked, either pharmacologically or genetically [17]. The mGluR1 signaling pathway associated with ZII− PCs plays a role. However, neither EAAT4 nor aldolase C contributes to the intrinsic differences in firing rates, both of which are expressed in a ZII+ pattern. Building on the Schonewille review, this section focuses on two additional aspects of the physiological properties of ZII+/− stripes: responses to afferent inputs and synaptic plasticity.

In this section, we will review how the interplay between the mossy-fiber (MF)/granule cell (GC)/Purkinje cell (PC) pathway and the olivo-cerebellar system determines a functional modular organization.

The cerebellum is typically recognized for its role in movement coordination and motor learning, but increasing evidence suggests it may also be involved in higher-order functions, including emotional behavior [4, 242, 243]. As described in the previous section, anatomical pathway tracing studies in animals have found that the cerebellum projects to an extensive list of brainstem and diencephalon nuclei, including a number of limbic structures [244‐247]. In humans, structural and functional abnormalities can sometimes lead to impaired mood regulation and anxiety disorders (the cerebellar cognitive affect syndrome) [248‐252]. In addition, neuroimaging studies have found changes in BOLD signal in the human cerebellum during fear learning paradigms [for review see 253]. In many cases, such changes are associated with the midline cerebellar vermis [249, 254, 255], and experimental studies in animals have found that lesions and other interventions of this cerebellar compartment have effects on defensive behaviors evoked by emotionally salient events [256‐264].

The cerebellar vermis primarily consists of the “A” module. Individual modules are defined by their olivo-cortico-nuclear projections. In the case of the A module, the cortical parasagittal zonal component receives olivocerebellar (climbing fiber) input from the caudal medial accessory olive, and the Purkinje cells (PCs) located within this region of cortex have a corticonuclear output to the medial (fastigial) cerebellar nucleus [8, 45, 265]. This medial nucleus has widespread connections to midbrain and cerebral cortical regions including the amygdala, hypothalamus, periaqueductal gray (PAG), striatum, prefrontal cortex, parietal cortex, and hippocampus [218, 245‐247]. Other cerebellar modules (notably the lateral vermal B and paravermal C3 modules) have been shown to be subdivided into smaller units, the cortical component of which are termed microzones [9, 27, 100, 101]. As discussed in previous sections, microzones and their micromodular connections are thought to represent the basic functional units of the cerebellum. Given that a finer grain olivocerebellar topography is present within the broader A module [266], this suggests a micromodular organization may also be present in this region of the cerebellum (see also section by Sugihara). Since emotional behaviors involve an integration of cognitive, somatomotor, and autonomic activity, this raises the possibility that different parts of the A module (possibly corresponding to micromodules) may be associated with different aspects of a coordinated emotional response.

To date, most studies have not attempted to explore cerebellar contributions to emotional behaviors at the modular (or micromodular) level of resolution. However, large cerebellar lesions involving vermal lobules III–VIII have shown various behavioral changes in relation to fearful or predator-prey interactions in rats. These include (i) fewer signs of fear when animals were placed in a brightly lit arena versus a dimly lit arena; (ii) decreases in freezing behavior and other signs of fear in the presence of a cat; (iii) faster recovery time than controls to the neophobic response to a novel taste test; and (v) attenuated spontaneous predation of mice [261, 264].

Some evidence for a lobular organization has also been found. For example, lesions of the anterior cerebellar vermis (lobules III–VI), but not the hemispheres resulted in impaired acquisition and retention of fear-conditioned bradycardia [256, 257]. Moreover, a subpopulation of PCs was found to respond to the conditioned stimulus, and in some cases this activity was correlated with the magnitude of the conditioned bradycardia response [267]. However, Supple et al. [267] did not investigate the possibility of correlated activity with other aspects of defensive behaviors, which might be expected if a finer grain localization of function was present.

In relation to human studies, a meta-analysis of fMRI mapping of the cerebellum supports a role for the human anterior lobe (vermal lobules IV and V) in fear learning and affective state [see review by 251]. And Sacchetti and colleagues found that reversible inactivation of a similar region in rats impairs the consolidation of fear memories [259, 263, 268]. By comparison, regions of the posterior lobe vermis appear to be involved in different aspects of behavior. In particular, it has long been known that vermal lobules VI and VII are important in the control of saccadic eye and head movements in a range of species including humans (the oculomotor vermis) [239, 269‐273], while lesions of vermal lobule VIII in rats results in deficits in innate and conditioned fear induced freezing behavior but no detectable changes in general motor activity [258, 274]. By contrast, vermal lobules IX and X are related to autonomic functions including regulation of blood pressure, heart rate, respiration, and the baroreceptor reflex [275, 276].

Thus, different cerebellar vermal lobules (that may relate to different components of the A module) appear to be associated with different aspects of an integrated array of behaviors. From rostral to caudal: lobules IV–VI with fear memory and affective state; lobule VI–VII with orientation of gaze; lobule VIII with fear-induced freezing behavior; and lobule IX and X with cardiorespiratory control (Fig. 8). Different parts of the A module, possibly relating to individual micromodules, could therefore regulate and integrate the cognitive, motor, and autonomic aspects of fear-related behavior.

The present collection of views on the anatomical and functional organization of the cerebellum has resulted in the realization that as a first requirement for a general consensus would be to agree to a general terminology. In this review, the authors agreed to use the following definitions of terms.

There is broad consensus that this battery of terms reflects different ways of describing and identifying the same complex modular architecture. First, there is the olivocerebellar projection pattern—separated into the cortical climbing fibers and their nuclear collaterals—which matches the corticonuclear projections of the targeted PCs. Matching nucleo-olivary projections close the modular loop, while the bulbar projections of the other nuclear neurons form the executive branch of such a module (Sugihara, Voogd: Fig. 1). Beyond this, cerebellar architecture at higher resolution is the topic of intense debate.

Although it is evident that there is generally a good correspondence between the chemical signature of Purkinje cells and their anatomical connections, the level of chemical complexity surpasses that of our current anatomical knowledge. To what extent are modules subdivided into smaller anatomical entities and can these really be seen as micromodules (Apps, Wylie)? What is the smallest operational unit in which the modular connectivity pattern can be recognized (Hawkes, Jörntell) and could this be different for different cerebellar modules (Hawkes, Apps)? Would such a cerebellar quantum or super PC require an anatomically fully closed integrated circuit between the incorporated inferior olivary cells, Purkinje cells, and nuclear cells? Another aspect that is still not fully understood is the anatomical incorporation of the nucleo-cortical connections into the modular circuitry [224, 277‐279].

Questions concerning the anatomical equivalency of the relation between olivocorticonuclear connections and chemical identity of Purkinje cells also relate to the much more difficult question concerning the functional interpretation of cerebellar modules, as it is far from settled to what extent the olivocorticonuclear circuitry (as modules or micromodules) represent functional entities. This will necessitate a better understanding of how the organization of mossy fibers fits in the modular scheme (Hawkes, Sillitoe, Isope)? Questions of how the distributed mossy fiber-parallel fiber system functionally and adaptively interacts with individual (micro-)modules are far from answered (Ebner, Isope)? Likewise, it remains to be determined to what extent information processing in modules with different chemical signatures fundamentally differs (Schonewille, Ebner) and how this subsequently may be used in the same functional setting (Wylie, Ruigrok)? What, really, is the function of individual (micro) modules? Do modules cooperate both within and outside of the cerebellum and in what way? Are the different modules with the same basic connectivity (e.g., C1 and C3) signs of redundancy (Hawkes, Isope)? To what extent do cerebellar modules serve multiple but integrated functions (Apps)?

Although, these and other questions cannot yet be readily answered, hypotheses have been formulated and a host of new and innovative techniques are at hand to begin to explore them all. For now, it should be clear that the cerebellum cannot be seen as a single operational machine, but that its basic modular organization has the potential to serve a great many functions in both individual and integrated ways.