Introduction

The shape of the neuronal dendritic arbor is an important determinant for neuronal function. The size, density, and geometry of the dendritic branches determine the types and number of synaptic connections formed with afferent axons. Dendritic arbor patterns also influence neuronal computation by affecting the diffusion of signal molecules and propagation of electrical activity. During dendritic development, the rules governing the manner in which dendrites grow impacts final morphology. Neurons show highly diverse morphology dependent on cell type, and such diversity contributes to the functional variety of neuronal circuits.

Cerebellar Purkinje cells have several unique morphological features (Fig. 1). The somata of Purkinje cells are located in a thin, single-celled layer by maturation (Fig. 1a). Each soma extends one or sometimes two primary dendrites into the molecular layer, which are arborized into a highly branched, fan-shaped structure in a single parasagittal (translobular) plane (Fig. 1a, c). Purkinje cell dendrites form a typical space-filling and non- (or minimal-) overlapping branch pattern, which covers the field in their receptive territory with dense, evenly-spaced branches (Fig. 1c). These flat dendritic arbors are aligned along the sagittal axis of the cerebellum. The axons of cerebellar granule cells, called parallel fibers, pass perpendicularly through the Purkinje cell dendritic arbors, forming a three-dimensional grid-like structure (Fig. 1a). The planar, space-filling, and non-overlapping arrangement of the dendritic arbor allows each Purkinje cell to form approximately 105 non-redundant synapses with parallel fibers in rodents; conversely, this arrangement results in a single parallel fiber making only one or two synapses with each Purkinje cell it passes through [1].

Fig. 1
figure 1

a Schematic showing the organization of granule cells and Purkinje cells in the cerebellar cortex. b Calbindin staining of a parasagittal section of murine cerebellum at postnatal day 24. Scale bar: 20 μm. c Representative image of sagittal (c-1) and coronal (c-2) views of a cerebellar Purkinje cell at postnatal day 14, visualized by EGFP introduced by adeno-associated viral vector [65]. The dendrites show the typical fan-shape and expand in a single parasagittal plane. c-3 A magnified view of the boxed region in c-1. The dendritic branches fill their receptive field with minimal overlap. Scale bars: 20 μm in c-1), 10 μm in c-3

Full size image

While the morphology of the Purkinje cell is unique, the individual features of its development, such as its space-filling, non-overlapping, and flat characteristics, are conserved in many neuronal types. It therefore provides a good model to assess the mechanisms and principles of dendrite morphogenesis to achieve a functional neural circuit. In this review, we will discuss how Purkinje cells develop their characteristic dendritic structures, and focus particularly on the formation of the space-filling pattern.

Expansion of the Fan-Shaped Dendrites of Purkinje Cells

A fan-shaped arbor is one of the distinguishing features of Purkinje dendrites. Computational simulation of different growth patterns demonstrated that stochastic addition of branches to dendritic terminals resulted in an arbor morphology reminiscent of Purkinje cells [2]. Consistent with this notion, we confirmed by live-imaging of cultured Purkinje cells that their dendritic arbors were formed by constant elongation of dendrites and random dichotomous branching of the dendrite termini [3]. In contrast, hippocampal pyramidal cells are known to sprout collateral branches from the shaft of the dendrite [4, 5]. Thus, the symmetric formation of dendritic termini may not be a general growth feature of dendrites and may instead be a major contributor to the fan-shaped formation of the Purkinje cell dendritic arbor.

The cell-type-specific dynamics of dendritic outgrowth are likely regulated by a combination of intrinsic programming and environmental factors. One of the key regulators of the intrinsic programming for Purkinje cell dendrite formation is the transcriptional factor retinoic acid-related orphan receptor α (RORα). This molecule controls multiple processes of dendrite differentiation in Purkinje cells, with its loss resulting in abnormal arborization of dendrites [6]. Environmental cues that have been shown to modulate the dendritic outgrowth of Purkinje cells include thyroid hormone (T3) [7,8,9,10], neurotrophins such as NT-3 [11] and BDNF [12], and neuronal activity [13,14,15,16]. T3 stimulates Purkinje cell dendrite formation via the thyroid hormone receptors TRα and TRβ1/2 [17, 18]. Application of T3 was shown to induce the expression of PGC1α, a master regulator gene of mitochondrial biogenesis, and increased ATP production and dendrite growth in Purkinje cells [10]. Indeed, mitochondrial delivery and ATP production at the growing dendritic tips are essential for stable dendrite growth of Purkinje cells [19, 20]. T3 was also shown to regulate RORα expression during dendrite differentiation [21]; thus, T3 functions at multiple levels of dendrite development, from dendrite differentiation to regulation of cellular metabolism.

NT-3 is secreted from cerebellar granule cells and potentiates Purkinje dendrite growth via the TrkC receptor expressed in Purkinje cells. NT3-TrkC signaling modulates dendrite formation through competition among neighboring Purkinje cells [11]. When TrkC was knocked out in a sparse number of Purkinje cells, dendritic arbors developed to be smaller than the surrounding wild-type cells. This phenotype was recovered by the global deletion of NT-3 expression in granule cells, suggesting that Purkinje cell dendrites can grow in the absence of NT-3 and TrkC, but the NT-3/TrkC signaling activity relative to neighboring Purkinje cells modulates dendrite arborization. Although the detailed mechanism is not yet understood, such an example of competition among Purkinje cells utilizing extrinsic signals is distinct from the mechanisms mediated by direct dendro-dendritic contact, as discussed in the following sections.

Self-Avoidance and Dendritic Structure

The space-filling and non-overlapping branching pattern is one of the most important morphological dendrite characteristics conserved across species [22, 23]. This configuration is well-suited for collecting the maximum amount of information from afferent connections and minimizing connection redundancy in a given receptive field. Generally, “self-avoidance” plays a key role in the formation of non-overlapping arbors. Self-avoidance, also called isoneuronal avoidance, refers to the repellence of dendrites emanating from the same neuron, also known as sibling dendrites, to prevent their crossing and bundling. Live-imaging of cultured Purkinje cells showed that growing dendrites underwent at least two types of self-avoidance behavior upon contact with adjacent sibling dendrites: retraction and stalling (Fig. 2a) [3]. After a collision with a sibling dendrite, the dendritic terminus stopped growing and was either retracted by a single or a few distal-most dendritic segment(s), or was stabilized and maintained at the site of contact. Branch retraction and stalling have different impacts on the topological pattern of the dendritic arbor. Contact-dependent retraction is dominant at the earlier stages of dendritic arbor growth, reducing overcrowding of branches around the cell soma and contributing to the branch length and density distribution along the proximal-distal axis. On the other hand, contact-dependent stalling, which is more evident at the later stages of arbor development, contributes more to the formation of a non-overlapping arbor pattern.

Fig. 2
figure 2

a Schematic showing the two types of self-avoidance behaviors identified in growing Purkinje cell dendrites. Dendritic branches are eliminated (contact-dependent retraction) or stall (contact-dependent stall) after collision with a neighboring branch. b The filopodia function as sensors of dendritic contact. c Physical interactions between Purkinje cell dendrites are recognized by adhesion molecules (protocadherins) and signaling molecules (Slit glycoproteins and their ROBO receptors)

Full size image

In the early stages of development in vivo, up until the first postnatal week, Purkinje cells have multiple primary dendrites stemming from the soma. Later, one or two dendrites are selected to grow and expand their branches in the molecular layer, while the other primary dendrites are eliminated. The excess primary dendrites may be eliminated by contact-dependent retraction in mouse Purkinje cells, as it has been observed that some of the primary dendrites are removed after dendro-dendritic contact in vitro [3]. In contrast, other studies using zebrafish Purkinje cells have implicated a cell polarization process mediated by atypical PKC in specification of the primary dendrites [24], suggesting the involvement of a cell intrinsic program in addition to contact-dependent retraction. After entering the molecular layer, Purkinje cell dendrites form numerous secondary and higher-order branches. At this stage, the self-avoidance behavior of the dendrites gradually switches from “retract” to “stall”. Contact-dependent stalling results in dendrites being arranged in a non-overlapping pattern, while maintaining a dense and highly-branched structure. Thus, the two types of self-avoidance behaviors, contact-dependent retraction and stalling, play important and different roles in the establishment of the characteristic morphology of Purkinje cells.

Heteroneuronal Avoidance and Tiling

In addition to self-avoidance, some types of neurons undergo heteroneuronal avoidance, in which dendrites of neurons of the same subtype or functional group repel one another. The heteroneuronal avoidance among these neurons on a two-dimensional (2D) substrate (e.g., Drosophila DA sensory neurons) results in a “tiling”, or non-overlapping, of dendritic receptive fields.

Previously, we showed by live-cell imaging that Purkinje cells grown on a 2D glass substrate in culture also exhibited contact-dependent avoidance between dendrites from adjacent Purkinje cells, in addition to those from the same neuron. This tiling arrangement suggests the involvement of heteroneuronal avoidance, at least in culture [3]. However, unlike Drosophila sensory neurons or Purkinje cells in dissociated culture, the flat dendritic arbors of Purkinje cells in vivo fill a three-dimensional space by stacking in parallel neighboring neurons in different planes (Fig. 1). This results in an apparent overlap of dendritic arbors in different planes, without crossing between dendrites in the same plane. Although it differs from the canonical definition of tiling, contact-dependent heteroneuronal avoidance may thus play some roles in the alignment of dendritic planes in vivo. As we discussed above, Purkinje cells undergo another type of competition mediated by NT3/TrkC signaling. Contact-dependent heteroneuronal avoidance and competition for afferent input such as NT3/TrkC signaling might determine the boundary of dendritic territory as well as dendritic size via distinct pathways.

Recognition of Contact by Dendritic Filopodia

In Purkinje cells, dendro-dendritic contact is mediated by dendritic filopodia: small protrusions emanating from the dendritic shaft (Fig. 2b) [3, 25]. Dendritic filopodia are morphologically dynamic structures in neurons and have been reported to act as fine sensors of the environment surrounding a dendrite [26, 27]. While canonical filopodia, including those from cell lines or on axonal growth cones, are composed of bundled filamentous actin (F-actin) oriented in one direction, dendritic filopodia in contrast contain both bundled F-actin and branched F-actin, reminiscent of the neck region of dendritic spines [28]. In dendritic spines, branched F-actin formation triggers the expansion of the spine head region, while linear F-actin formation elongates the structure [29]. Thus, dendritic filopodia morphology is likely regulated by the balance of linear and branched F-actin assembly. These are formed through distinct pathways by different actin nucleators, the formins and ARP2/3 complex, respectively. We recently found that MTSS1, an I-BAR-domain containing protein, controls dendritic filopodia length by regulating formin and ARP2/3 activities [30]. MTSS1 knockout Purkinje cells showed extended filopodia, induced by over-activation of the formin DAAM1. Extended filopodia led to increased contact-dependent dendrite retraction, which led to the loss of proximal branches. These results suggest involvement of the regulation of filopodial length to control sensitivity of Purkinje cells to dendritic contact, by influencing the probability of dendro-dendritic contact.

Two-photon observation of dye-loaded Purkinje cell dendrites in acute cerebellar slices revealed the formation of “filopodial bridges” by a small number of filopodia connecting two adjacent dendrites from the same cell. Interestingly, these filopodial bridges were only temporarily observed around postnatal days 10–12, when dendrites are actively growing [25]. The number of filopodial bridges declined over time and were never found in adult Purkinje cells. During dendrite development, the dendrite may utilize filopodia for sensing whether the surrounding dendrites originate from the same cell, other cells of the same neuronal subtype, or from other distinct cell types. This is consistent with the idea that dendritic filopodia function as not only dendritic spine precursors but also environmental sensors for correct dendritic growth in young neurons.

Molecular Mechanisms of Self-Avoidance in Purkinje Cells

It is likely that dendro-dendritic contact is recognized by some molecular interaction on the surface of the dendritic membrane and/or filopodia. DSCAM [31], Semaphorin 6A [32], clustered Protocadherins [33, 34], and Slit2/Robo2 [35] are cell-surface molecules that have recently been identified to partake in dendritic self-avoidance of mammalian space-filling neurons. Of these, the clustered Protocadherins and Slit/Robo have been shown to be required for self-avoidance in Purkinje cells (Fig. 2c).

Clustered Protocadherins

The clustered Protocadherin (Pcdh) family is the largest subfamily of the cadherin superfamily [23, 36], which is involved in many biological events, including neuronal survival, axon pathfinding, dendrite arborization, synapse formation, and self-avoidance of neurites [37]. In mammals, more than 50 Pcdh proteins are encoded by the Pcdh α, β, and γ gene clusters that are tandemly arranged on a single chromosome [38]. Stochastic and selective expression of Pcdh isoforms in a single neuron is controlled by alternative promoter choice [39, 40]. Pcdhs are homophilic adhesion molecules that form homophilic interactions in trans [41,42,43,44]. Thus, interaction between Pcdhs on neighboring dendrites from the same cell may evoke signals culminating in dendritic self-avoidance behavior, although the detailed signaling mechanisms remain elusive (see below).

In vitro cell aggregation assays have revealed the high specificity of homophilic association of Pcdh isoforms; cells expressing the same 5 Pcdh isoforms were able to form aggregates, whereas a single mismatch (i.e., 1 out of 5 isoforms) resulted in segregation of the cells [42, 44]. Such a highly specific homophilic association is thought to endow self/non-self recognition within the same subtype of neurons, although it is unclear to what extent the specificity of this homophilic interaction can be adapted for the avoidance behavior observed in dendritic branch formation.

Retinal starburst amacrine cells (SACs) provide an excellent model for demonstrating the function of Pcdhs in self and non-self recognition of dendrites. The SAC soma extends space-filling dendrites in 2D in the sublaminae of the inner plexiform layer of the retina, which overlap with neighboring SAC dendrites but not with its own. Conditional mutant animals lacking the Pcdh γ cluster in SACs showed a significant increase in crossings and fasciculation of sibling dendrites, indicating defects in self-avoidance due to the loss of Pcdh γ. In contrast, ectopic expression of the same single isoform of Pcdh γ in all SACs decreased the dendritic overlap among neighboring SACs, reminiscent of the heteroneuronal avoidance property, suggesting that self/non-self recognition in SACs requires multiple Pcdh γ isoforms [34].

Involvement of Pcdh in the non-overlapping arrangement of Purkinje dendrites has also been demonstrated by direct and indirect genetic alterations. Conditional deletion of the Pcdh γ cluster in Purkinje cells showed significant dendrite crossing due to self-avoidance defects. Likewise, indirect alteration of Pcdh expression in Dnmt3b knockout Purkinje cells exhibited an increase in overpassing dendritic branches, presumably due to defects in self-avoidance [45]. Dnmt3b is a DNA methyltransferase that has been shown to control the methylation state of the Pcdh variable exon promoters. Dnmt3b knockout Purkinje cells showed reduced methylation levels at Pcdh promoter sites and expressed a larger number of Pcdh isoforms than normal. With Dnmt3b deficiency, Toyoda et al. hypothesized that the increased number of Pcdh isoforms in a single cell may have led to a decrease in the probability of homophilic association of the recognition units on sibling dendrites, resulting in a phenotype similar to that of Pcdh γ loss [45]. Recently, it has been demonstrated that not only γ but also α isoforms engage in the self-repulsion of developing dendrites, including Purkinje cells [33]. In contrast to Pcdh α mutant SACs that showed no disruption in self-avoidance, Pcdh α mutant Purkinje dendrites showed errors in self-avoidance to the same extent as that of Pcdh γ mutants. In Pcdh α γ double mutant animals, both Purkinje cells and SACs showed more severe self-avoidance errors compared to the Pcdh γ single mutant, suggesting a differential contribution of α isoforms in distinct cell types to the dendritic self-avoidance behavior. Thus, it remains of great interest to understand how the α, β, and γ Pcdh isoforms behave during self-repulsion events during dendritic growth.

Slit-Robo2

Slit glycoproteins and their receptors Roundabout (ROBO) are known to regulate axon repulsion during neural circuit formation. Recently, Slit-Robo signaling was shown to mediate dendritic self-avoidance in Purkinje cells [35]. In situ hybridization analyses revealed that Slit2 and Robo2 are highly expressed in Purkinje cells, and cell-specific deletion of either disrupts self-avoidance of Purkinje dendrites. Slit proteins are secreted molecules but are also known to interact with extracellular matrix (ECM) components such as collagen and the transmembrane ECM receptor dystroglycan [46, 47]. Interestingly, the self-avoidance phenotype caused by Slit2 deficiency was not rescued by the expression of the diffusive truncate form slit2D2, but was rescued by the membrane-bound mutant slit2D2GPI, suggesting that Slit molecules bound to the dendritic membrane (or nearby) via the ECM/ECM receptor mediate contact-dependent repulsion. A question remains regarding how neurons discriminate between Slit-Robo interactions in cis and trans. Although both are likely localized to the dendritic arbor in the same cell, neurons appear unresponsive to cis-interactions. Future studies should address how dendrites may differentiate between possible cis and trans interactions.

Furthermore, it was shown that Slit-Robo signaling and Pcdhs seem to regulate self-avoidance of Purkinje cell dendrites via independent mechanisms due to observation of (1) an inability of slit2D2GPI to rescue the Pcdh γ deletion phenotype and (2) Pcdh γ and Robo2 double mutants showing a more severe disruption in self-avoidance [35]. The use of two different and independent signaling mechanisms in Purkinje cells is an intriguing observation regarding the complexity of self-avoidance behavior and highlights the further possibility of other membrane protein involvement in this phenomenon.

Possible Intracellular Signals Controlling Self-Avoidance in Dendrites

Although at least two recognition molecule groups have been shown to be required for self-avoidance in Purkinje cell dendrites, the downstream signaling that regulates the dynamic behavior of dendrites during repulsion and retraction remain largely unknown.

Previously, it has been shown that Pcdh α and γ interact with and suppress two non-receptor tyrosine kinases involved in the regulation of cell adhesion and cytoskeletal molecules: protein tyrosine kinase 2 (Pyk2) and focal adhesion kinase (FAK) [48]. Consistently, in the Pcdh γ knockout brain, FAK, Pyk2, and protein kinase C (PKC) have been shown to be activated [49]. In cortical pyramidal neurons, Pcdhs positively regulate dendrite growth and branching via a unique mechanism seemingly not associated with self-avoidance [50]. In these cells, intercellular homophilic Pcdh signaling regulates cytoskeletal arrangement via the FAK/PKC/MARCKS [49] and Pyk2/Rac [51] pathways. Although at present, there has been no direct evidence showing whether the same signaling cascades mediate self-avoidance, we previously showed that PKC/PKD (protein kinase D) inhibition suppressed contact-dependent retraction in Purkinje cell dendrites [3], providing indirect evidence. Further, Purkinje cell-specific knockout of mTORC2, which regulates PKC as a downstream target, showed increased dendrite self-crossing in Purkinje cells, presumably due to self-avoidance error [52]. Thus, these signaling molecules are possible candidates that may control the cellular response during retraction and demand further attention in the study of this phenomenon.

During axonal guidance, Slit-Robo regulates actin and microtubules via regulatory proteins including Slit-Robo GTPase activating proteins (srGAPs) and cytoplasmic kinase Abelson (Abl) [53]. It would be of great interest to clarify whether these pathways are also implicated in dendritic self-avoidance. Undoubtably, identification of the precise molecular cascades leading to dendrite repulsion during self-avoidance is of great importance for our understanding of dendritic arbor development.

Planarity and Self-Avoidance

Planarity is closely related to the space-filling/no-overlapping organization of neurites, although self-avoidance or heteroneuronal avoidance may play some roles in the three dimensional patterning of arborized dendrites [54, 55]. Indeed, neurons showing space-filling dendrites usually have planar dendrites (e.g., retinal neurons, Drosophila DA neurons, Leech sensory neurons, and cerebellar Purkinje cells) [56,57,58]. Restriction of dendritic arborization to a two-dimensional plane may play a supportive role in forming proper space-filling dendrites by contact-dependent repulsion, because such spatial restriction may increase the local branch density and enhance the possibility of collisions. In Drosophila, DA neurons form typical space-filling dendrites in contact with the extracellular matrix (ECM) on the basal surface of epidermal cells [59,60,61]. Loss of integrin function in DA neurons disrupted the planar dendrite pattern, with dendritic branches detached from the substrate and growing three-dimensionally into the epidermal cell layer.

How do Purkinje cells form their flat dendrite structure within the molecular layer of the cerebellum? It is unlikely that the molecular layer has repetitive structures aligned along the parasagittal axis serving as a support for the planar Purkinje dendrites. Several lines of evidence have indicated that the parallel fibers, the axons of granule cells, are the primary candidates for guiding the planar dendritic growth of Purkinje cells [62, 63]. It has been demonstrated that Purkinje cells extend dendrites perpendicularly to the granule cell axons, which are aligned into parallel bundles in culture [63]. Considering that parallel fibers form bundles and run along the coronal axis of the cerebellum, Purkinje cells may use this scaffolding for perpendicular restriction of dendritic branch growth to form a single parasagittal plane. Consistently, parallel fiber disorientation caused by X-ray irradiation of the postnatal cerebellum led to re-alignment of Purkinje cell dendrites at a right angle to the disoriented parallel fibers [62]. However, knockout of contactin, an adhesion molecule expressed in cerebellar granule cells, resulted in no change in Purkinje cell morphology [64], despite misorientation of parallel fibers in the upper region of the molecular layer. Although the discrepancy between these studies has not been clarified, it is possible that the rather minor fraction of disoriented parallel fibers in the contactin-deficient animal may fail to affect Purkinje dendrite orientation, in contrast to the more severe disorganization observed in the X-ray irradiated animal. Although the molecular mechanisms determining dendrite growth orientation with respect to axonal bundles have not been identified yet, it is likely that synaptic or non-synaptic cell-cell adhesions between axons and dendrites contribute to the orientation of dendrite growth with respect to parallel fibers, highlighting a future avenue of investigation.

During development, some Purkinje cells have minor branches which extend from the main sagittal plane and form distinct sagittal planes parallel to the main arbor, which are then remodeled into mono-planar arbors, dependent on afferent fiber activity [65]. It is thus suggested that dendrites occasionally grow in the wrong direction, deviating from the normal plane, at least in younger stages. Interestingly, Pcdh α γ double mutant Purkinje cell dendrites exhibited multiplanar dendrites in addition to dendrite crossing errors [33]. These results suggest that Pcdhs might restrict dendrite growth within a single parasagittal plane. How do Pcdhs control monoplanar dendrite arborization? One possibility may be that Pcdhs contribute to the recognition of contacts not only among sibling dendrites but also among dendrites arborizing in adjacent planes from neighboring Purkinje cells, which suppress the invasion of dendrites into different planes. However, Pcdhs may not be able to mediate such heteroneuronal avoidance if the homophilic interaction of Pcdhs is highly specific, as observed in the in vitro re-aggregation experiments described above. Another possibility is that Pcdh α γ double mutants may experience more frequent dendrite crossings due to a more severe defect in self-avoidance behavior compared with single mutants; dendrites crossing over one another may change the dendritic growth direction away from the main plane, leading to formation of multiplanar dendrites. Although the detailed mechanisms are unknown, these results suggest that dendritic avoidance mechanisms might play a secondary role in constructing (mono) planar dendrites.

Compared to self-avoidance mechanisms, less is known about the molecular mechanisms underlying the formation of flat dendritic arbors in Purkinje cells [66, 67]. Thus, clarification of the molecular and cellular mechanisms of how Purkinje cells form mono-planar dendrites and how granule cell axons orient Purkinje cell dendrites remains of great interest.