Shedding Light on Chandelier Cell Development, Connectivity and Contribution to Neural Disorders

Abstract

Chandelier cells (ChCs) are a unique type of GABAergic interneuron that selectively innervate the axon initial segment (AIS) of excitatory pyramidal neurons, the subcellular domain where action potentials are initiated. The proper genesis and maturation of ChCs is critical for regulating neural ensemble firing in the neocortex throughout development and adulthood. Recently, genetic and molecular studies have shed new light on the complex innerworkings of ChCs in health and disease. This review presents an overview of recent studies on the developmental origins, migratory properties, and morphology of ChCs. In addition, attention is given to newly identified molecules regulating ChC morphogenesis and connectivity as well as recent work linking ChC dysfunction to neural disorders, including schizophrenia, epilepsy, and autism spectrum disorder (ASD).

Chandelier Cells: A Distinctive Type of GABAergic Interneuron

Since their initial discovery in the 1970s, GABAergic chandelier cells (ChCs), also known as axo-axonic cells, have intrigued neuroscientists due to their unique axonal morphology and selective subcellular innervation of excitatory pyramidal neurons (PyNs) [1–4]. Appropriately named given their resemblance to a chandelier/candelabrum, ChCs possess a distinct, highly-branched axonal arbor that terminates in parallel arrays of short vertical sets of presynaptic boutons, referred to as cartridges [2, 4]. Initially believed to synapse at the apical dendrites of PyNs, ChC cartridges were instead found to selectively innervate the axon initial segment (AIS) (see Glossary) of PyNs in the neocortex, hippocampus (CA3, CA1, and dentate gyrus), piriform cortex, and amygdala [4–9]. Given that the AIS is the site of action potential initiation, ChCs are strategically poised to exert powerful yet precise control over PyN firing and population output, ultimately contributing to the maintenance of proper excitatory/inhibitory (E/I) balance in health as well as perturbed E/I homeostasis in disease [10–14]. Relatedly, recent work has demonstrated a key role for ChCs in the synchronization of firing patterns of large PyN ensembles in different functional states in addition to altered ChC-mediated neurotransmission in conditions including epilepsy and schizophrenia [14–19]. For further review of ChC-mediated regulation of neural firing and ChC electrophysiological properties, we direct the reader to recent excellent reviews [11–13].

Despite the field’s early excitement, studies investigating the molecular and cellular innerworkings of ChCs were, until recently, few and far between, in large part due to the scarcity of ChCs and lack of unique neurochemical ChC markers. However, aided by the development of novel tools and technology, significant discoveries have been made over recent years, revealing intriguing new aspects of ChC biology. For instance, transplantation studies, in utero viral injections, and generation of new knock-in Cre-driver mice have shed light on the developmental origins of ChCs [11, 20–23]. Moreover, recent ChC single-cell RNA sequencing (scRNA-seq), high-resolution imaging and morphometric analyses, and in utero electroporation (IUE)- and viral injection-driven molecular studies in mice have significantly advanced our understanding of ChC postnatal development and connectivity [20, 21, 24–29]. Such progress, in conjunction with ongoing research on postmortem human tissue and new animal models, continues to strengthen the association between ChC dysfunction and neural disorders [24, 30–32]. In particular, recent work studying ErbB4, DOCK7, and FGF13, molecules linked to epilepsy and/or schizophrenia, have demonstrated key roles for these factors in proper ChC morphogenesis and neurotransmission, further reinforcing the link between such neurological disorders and ChC biology [21, 24, 26, 28, 33–37]. In this review, we summarize and discuss recent work elucidating the developmental origins, migratory properties, and morphology of ChCs. In addition, special attention is given to recent studies investigating the molecular mechanisms governing ChC axonal morphogenesis and connectivity in the neocortex. Finally, we review the increasing number of human and animal model studies linking ChC dysfunction to schizophrenia, epilepsy, and ASD.

Developmental Origin of Cortical ChCs

Cortical interneurons comprise a highly heterogeneous group of neurons with diverse morphologies, biochemistry, synaptic targets, and firing properties [38–41]. These properties are largely acquired during development through the implementation of distinct transcriptional programs that are either intrinsically encoded or driven by interactions with the local extracellular environment [41–44]. Among the diverse cortical interneuron types, ChCs are arguably the most distinctive, owing to their unique axonal morphology and highly selective innervation pattern [10–13]. Despite these unique features, the spatial and temporal origin of ChCs remained elusive for decades and was only discerned recently through novel fate mapping studies.

To start with, fate mapping using transplantation and/or genetic mouse models revealed that the embryonic ventral telencephalon (or subpallium), which can be subdivided into anatomically distinct regions such as the medial and caudal ganglionic eminences (MGE and CGE, respectively) and the preoptic area (POA), is the sole source of cortical interneurons (see Box 1) [40, 41, 44, 45]. Among these, the MGE and adjoining POA, which are marked by expression of the homeodomain transcription factor NKX2.1, were found to give rise to many major subclasses of interneurons, such as parvalbumin (PV)- and somatostatin (SST)-expressing interneurons [46]. The first hint toward the origin of ChCs came from ensuing fate mapping studies using Nkx2.1-Cre transgenic mice [47] and in utero viral injections in the MGE [21]. Both strategies resulted in the labeling of ChCs among other interneurons, thus revealing that ChCs originate from the MGE/POA. Further work involving MGE transplantations into the mouse neonatal neocortex [48] and temporal/spatial fate mapping using Nkx2.1-CreER;Ai9 knock-in mice, which enables tamoxifen (TMX)-dependent ChC red fluorescent protein labeling via Nkx2.1-driven CreER expression [20], indicated that ChCs are primarily generated in the most ventral region of the MGE and at the latest stages of corticogenesis. In fact, by performing BrdU-based birth dating of ChCs labeled in Nkx2.1-CreER;Ai9 animals that received TMX between embryonic day 15.5 (E15.5) and postnatal day 1 (P1), it was determined that the peak of ChC genesis occurs between E16–17 [20]. This relatively late birth time initially came somewhat as a surprise, as the generation of SST+ interneurons ceases by E14.5 and only a small fraction of PV+ interneurons are produced during this time window [43] (Figure 1A). It should be noted though that besides being generated at this late embryonic stage, recent results suggest that ChCs are also produced as early as E12 (see below; Figure 1A–C). It is also important to point out that while work using Nkx2.1-CreER;Ai9 mice reported that only a subset of ChCs tested PV+ by immunostaining [20], more recent scRNA-seq data showed that the majority of ChCs express PV transcript albeit at almost half the levels observed in PV+ basket cells (BCs) [25]. A possible explanation for this discordance is that PVALB is an activity regulated gene and currently it is unknown whether PV mRNAs are translated in all ChCs. On a more global level, among all Pvalb mRNA-expressing cell types, only ~1.5% in the mouse anterior lateral motor cortex and primary visual cortex [49] and ~5% in human middle temporal gyrus were transcriptomically identified as ChCs [50]. These numbers, though, are quite likely variable between species and brain areas and may even change with age. Additional large-scale transcriptomic studies will undoubtedly shed further light on this.

Box 1: MGE Neurogenesis.

Neocortical interneurons derive from progenitors located in the proliferative zones of the embryonic ventral telencephalon (subpallium), specifically the medial ganglionic eminence (MGE) and caudal ganglionic eminence (CGE). In addition, a small fraction of interneurons is also produced in the preoptic area (POA), whereas striatal medium spiny GABAergic projection neurons and olfactory bulb interneurons are derived from progenitors in the lateral ganglionic eminence (LGE) (see Figure I). The ganglionic eminences are temporary embryonic structures that change dramatically in size and shape between embryonic day 14.5 (E14.5) and E17.5 in mice (as shown in Figure I), with NKX2.1 expression (red shading) persisting till late embryonic stages. In parallel, the MGE differentiates into numerous subregions delineated by combinatorial transcription factor expression, giving rise to multiple dorsal-ventral subdomains and an increasingly thick subventricular zone (SVZ) layer adjacent to the ventricular zone (VZ) [102–104]. MGE neurogenesis starts soon after ~E9.5, peaks between E12–15, and continues until ~E17.5 in mice [22, 23]. During this period, the hardly noticeable VZ expands greatly along the rostral-caudal, dorsal-ventral, and apical-basal axes.

Upon its initial formation, the MGE (and the adjacent POA) is marked by NKX2.1 expression within a small region of the subpallial VZ which is comprised of a thin layer of neuroepithelial cells (NEs). NEs subsequently proliferate and differentiate into radial glial progenitor cells (RGCs), bipolar shaped cells which actively divide at the luminal surface of the VZ. After initially undergoing symmetric proliferative divisions to amplify the progenitor pool, these cells predominantly partake in asymmetric neurogenic divisions to self-renew and simultaneously produce neurons either directly or indirectly via intermediate progenitors (IPs) in the SVZ. The IPs that divide along the ventricular surface are termed apical IPs (aIPs) whereas those that divide in the SVZ are called basal IPs (bIPs) (Figure 1B) [105–109]. As the brain develops and neurogenesis proceeds, bIPs expand in numbers and bIP-derived interneurons, in addition to aIP-derived interneurons, migrate to their final positions (dashed grey lines), laminating the neocortex in an inside-out-inside (i.e. inside to outside and to inside again) pattern.

Figure 1. Origin and Development of Cortical ChCs.

(A) Temporal generation of cortical SST+ (green), PV+ (blue), and ChC (red) interneurons. Dotted red line represents early-born ChCs. (B) Top: Schematic of cortical laminar distribution of wave 1 (early-born) and wave 2 (late-born) MGE-derived SST+ (green), PV+ (blue), and ChC (red) interneurons. Bottom: Model of progenitor and lineage mechanisms for the generation of SST+, PV+, and ChC interneurons. Wave 1 ChCs are generated from apical intermediate progenitors (aIPs) and wave 2 ChCs from basal IPs (bIPs), which are derived from early-activated and later-activated radial glial cell (RGC) progenitors (RGC^E and RGC^L, respectively), as ascertained from cell-fate mapping experiments [23], retroviral-labeling, and RGC manipulation using Pard3 conditional knockout (cKO) animals [22]. Progressive depletion (shaded blue bar) and complete loss (white cross in dark blue circle) of RGCs via conditional Pard3 knockout causes a concomitant reduction of ChCs (ChCs with white crosses in dark blue circles). It is unknown whether Pard3-cKO also causes a reduction in non-ChC PV+ interneuron number (white crosses in dark blue circles with question marks). (C) Left: NKX2.1 is expressed (red shading) in the middle-caudal regions of the MGE. Arrows denote migratory paths taken by MGE-derived cells toward the dorsal and rostral cortex. Right: Schematic depicting the perinatal and early postnatal spatial and temporal migratory dynamics of superficial and deeper layer ChCs (red cells at layer ½ (L½) border with axons extending into L2/3 and in L5/6, respectively) produced by NKX2.1+ progenitors (red dots) in the MGE. Panel B adapted from [23] and panel C adapted from [20]. SST, somatostatin; PV, parvalbumin; ChC, chandelier cell; WM, white matter; MZ, marginal zone; CP, cortical plate; SVZ, subventricular zone; VZ, ventricular zone; NE, neuroepithelial cell; RGC^E, early radial glial progenitor cell; RGC^L, late radial glial progenitor cell.

While the above studies shed light on the developmental origin of ChCs, it remained unclear whether ChCs originate from a common pool of multipotent progenitors that has the potential to generate all MGE-derived GABAergic interneurons or alternatively from a specified (i.e. fate-restricted) pool of progenitors in the MGE. To address this, mouse genetics and in utero retroviral injections were combined to selectively label dividing NKX2.1-positive radial glial progenitor cells (RGCs) in the MGE and adjoining POA at time points spanning E12–17 and systematically examine their interneuron output in the neocortex at P21 [22]. Using this approach, it was found that different neocortical interneuron types are progressively generated in an inside-out-inside order, with ChCs being produced at a steady rate during the late embryonic stage by continuously dividing MGE RGCs that had previously produced non-ChC interneurons. Accordingly, disruption of RGC asymmetric division by ablation of the cell polarity protein partitioning-defective 3 (PARD3) leads to a progressive loss of RGCs in the MGE and, importantly, a consequential loss of both superficial and deeper layer ChCs (Figure 1B) [22]. These findings indicate that late-born ChCs arise from a multipotent pool of MGE RGCs that undergoes consecutive asymmetric cell divisions.

In an independent study, Huang and colleagues combined multiple intersectional mouse driver lines to systematically fate map MGE RGCs as well as intermediate progenitors (IPs) starting from the very onset (i.e. at ~E10) and throughout the course of cortical neurogenesis (see Box 1 for MGE neurogenesis) [23]. By combining this approach with thymidine-analogue neuronal birth dating, the authors discovered that the MGE is comprised of two separate pools of multipotent RGCs, each derived from neuroepithelial cells (NE) that mediate, via distinct and temporally-defined IP cohorts, two consecutive waves of neurogenesis, each sequentially generating different sets of interneuron types that laminate the neocortex in an inside-out-inside manner (Figure 1B). With regard to ChCs, a small and early-activated (at ~E10) pool of RGCs (designated RGC^E), which is restricted to the caudal MGE, generates a small set of L5/6 and L2/3 ChCs via apical IPs (aIPs). Additionally, a more numerous RGC pool (RGC^L), which is activated later (starting at ~E12) and occupies the entire rostral-caudal regions of the MGE, gives rise to a much larger set of ChCs across L2-L6 via sequential temporal cohorts of fate-limited basal IPs (bIPs) [23]. Importantly, in line with the aforementioned study [22], this report found that ChCs are primarily generated toward the end of each wave and laminate the cortex in an outside-in manner. Of note, the first study most likely captured the RGC^L pool, but not the RGC^E pool, as fate mapping was performed between E12–17 and did not include progenitors as early as E10. Furthermore, recent single-cell transcriptomics analyzing progenitor pool diversity within germinal regions of the dorsal and ventral MGE (dMGE and vMGE, respectively) and CGE identified the transcriptomic signatures of direct and indirect neurogenic pools and noted that progenitors derived from the subventricular zone (SVZ) are more transcriptomically diverse than the ventricular zone (VZ) [51, 52]. These studies [51, 52] indirectly favor the possibility that more than one IP pool exists, each generating a separate wave of ChCs, which aligns with the findings by Huang and colleagues [23]. Lastly, a recent study looking outside of the neocortex found that the genesis of hippocampal ChCs occurs between E11–14 with peaks at E12 and E13 [53].

Migration of Developing ChCs

A process imperative for the proper allocation/lamination of interneurons in the neocortex is migration. Irrespective of their identity, interneurons generated in the MGE undertake complex migratory routes to reach their final destination in the neocortex while avoiding the striatum and other regions of the subpallium. Three major routes/phases of migration have been observed for such neurons during corticogenesis: the first is along well-defined tangential paths, either a superficial path within the marginal zone (MZ) or a deep path along the SVZ/lower intermediate zone (IZ), from the MGE toward the corticostriatal junction and into the cortical wall, the second encompasses multidirectional migration within these migratory paths upon reaching the neocortex, and the third involves a switch to a radial trajectory to enter the cortical plate (CP) [41, 43, 54].

To determine whether ChCs undergo a similar scheme of migration, early work took advantage of the Nkx2.1-CreER;Ai9 mouse line to label precursors of late-born ChCs along the middle-caudal regions of the vMGE and follow their trajectory to their final neocortical positions. It was found that late-born ChC-fated cells migrate along stereotyped routes with a defined schedule to settle at the L½ border (with axons extending into L2/3) and in L5/6 [20] (Figure 1C). More specifically, the MGE-derived neurons first migrate tangentially along the lateral wall of the ventricle, reaching the cortex at approximately E18-P0. This migrating cohort then splits into medial and lateral streams to populate the cortical SVZ by P1. In addition to this route, a smaller cohort migrates anteriorly along the ventricle wall and/or within the VZ to populate rostral cortical areas. After entering the cortex, the neurons turn radially to cross the CP, reaching L1 around P2–3, where they continue to spread for several days before descending back into the cortex, forming a dense band of cells at the L½ border. Shortly afterwards, between P3–7, a significant fraction of neurons migrates to deeper cortical layers but the details regarding their exact migratory route(s) and temporal schedule, in addition to the subsequent regulation of ChC number via developmental cell death (see Box 2), require further investigation [20] (Figure 1C). As of yet, no studies (to our knowledge) have ascertained the migratory trajectories of early-born ChCs generated in the caudal MGE.

Box 2: Developmental Cell Death of ChCs.

Previous work found that a large number of ChCs present at P7 along the L½ border were no longer detectable at P28 [20]. While such a reduction is perhaps not surprising given reports of interneuron overproduction and subsequent early developmental cell death, either through cell-autonomous or competitive mechanisms [110], new data suggests a unique and somewhat counter-intuitive cell death process occurs in ChCs, at least in the visual cortex. Specifically, between P7–14, Huang and colleagues observed dramatic ChC death at the V1/V2L border region of the binocular zone in mice, reporting ChC density to be about half that of neighboring regions [111]. This study interestingly found that ChC elimination at this border region is dependent on excitatory input from callosal projections from the contralateral cortex [111]. In an effort to manipulate ChC death, the group found that reduced callosal axon innervation results in an increase in ChC density at the V1/V2L border, while an induced broadening of callosal inputs to the border region via P0 monocular enucleation causes a further reduction in ChC density, ultimately suggesting that callosal axon invasion promotes ChC elimination. Moreover, to explore the role of retinal activity in this process, intravitreal injections of tetrodotoxin (TTX) were performed between P7–14 to attenuate spontaneous retinal activity present normally in mice during this time window prior to eye opening. This strategy left callosal projections unaffected but significantly compromised ChC elimination at the V1/V2L border, indicating that spontaneous retinal activity plays a major role in regulating ChC density at the V1/V2L border [111]. To investigate the functional impact of such cell death, neuronal response properties in V1 binocular neurons were studied in mice with an induced increase in ChC density at the V1/V2L border. The group found that the ocular dominance index (ODI) of mice with excess ChCs was shifted toward the contralateral eye and that their contralateral/ipsilateral response ratios (C/I ratios) were significantly increased relative to control animals. Of note, monocular response properties, including orientation and direction selectivity, were normal in animals with excess ChCs. Use of a visual cliff test to test stereopsis (depth perception) found ChC-excess mice had difficulties detecting the visual cliff, resulting in a significant increase in the number of crosses into the cliff region relative to controls [111]. In sum, these data suggest that such targeted developmental elimination of ChCs helps to shape neural circuitry essential for binocular depth perception.

Development of ChC Morphology and Connectivity

ChCs are perhaps the most readily recognizable neuronal cell type in the mammalian brain. They display a very distinctive, ‘chandelier-like’ axonal geometry with multiple arrays of vertically oriented cartridges, each harboring a string of synaptic boutons [2, 4]. Unlike other interneurons that typically form somatodendritic synapses, ChC cartridges selectively innervate individual PyNs at their AIS, the site of action potential initiation [10–13] (Figure 2). This raises the important question as to how ChCs establish this idiosyncratic/unique axonal organization and synaptic connectivity during development. Recent work taking advantage of genetic labeling of neocortical ChCs in mice has found that after reaching the border of L½ in the somatosensory cortex by approximately P7, ChCs begin to elaborate their axon with a marked increase in branching/complexity first observed at P11/P12. This increase in ChC axonal complexity, branch points, and axon length continued through P14 before plateauing at approximately P28 [29]. Importantly, this developmental program also relies on ChC axonal pruning between P16–21 as a means to eliminate excess axon branches and nonsynaptic axonal terminals/boutons; a key requirement for proper establishment of ChC/PyN connectivity in adolescent/adult mice [29, 55]. By selectively forming synapses at PyN AISs and pruning non-synaptic axonal arbors and off-target (non-AIS) boutons, it has been suggested that ChCs establish axonal geometry through remodeling, but develop subcellular synapse specificity via genetically predetermined mechanisms [55]. Of note though, work in non-human primates showed that pruning of ChC boutons/synapses also occurs at the AIS in what appears to be a PyN-specific manner [56]. Such a developmental strategy may also help explain the seemingly contrasting programs employed by ChCs and BCs to organize their axonal arbors in mice. Namely, the high density of proximal dendrites and cell bodies in the cortical neuropil presents BCs with an abundance of postsynaptic target sites to form synaptic varicosities, excluding the need to prune non-synaptic (off-target) axonal branches and boutons. These findings are further supported in non-human primates where BC synapses are not pruned [56]. In contrast, given the relatively sparse distribution of AISs, a large number of ChC axonal branches are likely not able to form synapses on their subcellular targets, ultimately leading to their pruning [55]. Later inspection of ChC morphology in the murine somatosensory cortex noted that the axonal arborizations established by the end of the first postnatal month are quite fixed/stable with only a slight expansion in the lateral axis, and no change in the vertical axis, observed at P90 [57].

Figure 2. Molecular Mechanisms Governing ChC Cartridge/Bouton Morphogenesis and Innervation of the AIS of Excitatory PyNs in the Neocortex.

GABAergic ChCs possess a unique and highly-branched axonal arborization characterized by terminal strings of synaptic boutons, known as cartridges, that selectively innervate the AIS of neocortical PyNs [1–4]. ChC cartridge/bouton and synapse development is dependent upon proper ChC ErbB4-DOCK7- and FGF13-mediated signaling, as depletion of any of these molecules causes a significant reduction in ChC bouton/synapse formation selectively at the PyN AIS, the proximal subcellular domain of the axon necessary for action potential (AP) initiation/generation [21, 24, 26, 28, 33]. The AIS is a highly complex structure comprised of numerous molecular components, including a high density of voltage-gated sodium and potassium ion channels (NaV, Kv7, Kv1) required for AP initiation [76, 78]. In addition, the AIS has ɑ2-containing GABA_A receptors (GABA_AR-ɑ2) bound to gephyrin, which are necessary for ChC-mediated GABAergic neurotransmission [71], as well as a highly complex cytoskeleton composed of ankyrin-G (AnkG), ɑ2- and β4-spectrin, actin rings, and microtubules [76–78]. Among the many cell adhesion molecules (CAMs) present at the AIS of PyNs, AnkG-bound L1CAM is critical for ChC/PyN AIS innervation in the neocortex [29]. The identity of the presynaptic partner(s) of L1CAM present on ChC cartridges and/or in the extracellular milieu remains to be determined. For visualization purposes, innervation between a single ChC and one neighboring PyN is depicted. Of note though, in vivo, single neocortical ChCs are able to simultaneously innervate hundreds of neighboring PyNs with each PyN AIS typically being innervated by three to four cartridges originating from separate ChCs [57].

With regards to the timing of ChC synapse formation, murine ChC/PyN synaptic connectivity profiling noted minimal PyN AIS innervation between P8–11, when ChC axons are still relatively underdeveloped, but a sharp increase in synaptic contact at P12, aligning with the initial spike in ChC axonal complexity/branching. This increase in neocortical ChC/PyN AIS innervation continues into the third/fourth postnatal week before stabilizing at approximately P28 when, on average, each cartridge is approximately 16–28 µm in length and possesses four to nine boutons [28, 29, 57]. At this time, most PyN AISs in L2/3 of the murine somatosensory cortex are innervated by three to four cartridges typically originating from separate ChCs [57]. Moreover, given the complex arborization of their axons, single ChCs are able to simultaneously innervate hundreds of neighboring PyNs, allowing them to exert decisive control over the spiking of excitatory neocortical ensembles and directly regulate E/I balance [13–18]. Interestingly, such innervation was recently found to be quite plastic and dependent on the polarity of axo-axonic synapses in the mouse somatosensory cortex [58]. More specifically, increases in the activity of either PyNs or ChCs between P12–18, when ChC mediated GABAergic transmission at the AIS is depolarizing, results in a reversible decrease in axoaxonic connections [58]. Similar manipulations of PyN network activity in older mice (P40–46), when axo-axonic synapses are inhibitory, instead, caused an increase in ChC/PyN AIS synapses [58]. These findings suggest that the direction of ChC synaptic plasticity follows homeostatic rules dictated by the polarity of their synapses [58].

In regard to the spatial distribution of ChC/PyN AIS innervation, analysis of innervation at P18, P30, and P90 in L2/3 of the mouse somatosensory cortex found that the majority (73%) of cartridges are located within 60–150 µm of the ChC soma and that 86% of these cartridges innervate an AIS, further demonstrating the selective targeting of such contacts [57]. While it is possible that a fraction of the remaining 14% of observed cartridges could be non-synaptic (due to e.g. a defect in developmental pruning), the reported percentage of innervation is most likely an underestimation for several reasons, including technical limitations and the investigators’ definition of ChC-AIS contact, which required at least two adjacent boutons to be present [57]. Of note, analysis of innervation patterns across the mouse cortex did reveal areas of high and low densities of ChC/PyN AIS innervation [57, 59], which is even more apparent in primates, where the number of synapses per AIS can range from 0–50 [56]. Such differences could be due to regional differences in ChC morphology or distribution, especially given that ChCs are not distributed/positioned uniformly in certain cortical areas/regions [4, 6, 32, 57, 59–61]. It cannot be excluded though that subsets of ChCs may preferentially target certain neuronal populations in the cortex. In line with this, previous work found that ChCs preferentially contact PyNs with intracortical projections in the auditory and visual cortices and centrifugal cells in the piriform cortex [4, 6, 62–64]. Moreover, recent tracing studies combined with electrophysiological recordings found that L2 ChCs in the murine prelimbic (PL) cortex preferentially innervate PyNs projecting to the basolateral amygdala (_BLAPC) compared to those projecting to the contralateral cortex (_CCPC) [14]. In terms of profiling synaptic inputs onto ChCs, very little is known. The abovementioned report did find that L2 PL ChCs receive preferential input from local and _CCPCs instead of _BLAPCs and BLA neurons [14]. Another report also found that ascending dendrites of ChCs in L1 of the piriform cortex receive short collaterals and boutons en passant from olfactory axons therein [9]. Additional high-resolution ChC tracings and connectivity profiling is needed though to fully elucidate the wiring rules of ChCs in different cortical regions throughout development and adulthood.

While extensive developmental profiling of axonal organization and synaptic connectivity has so far been performed primarily for L2 ChCs in the murine somatosensory cortex, morphological analyses of ChCs in the hippocampus, piriform cortex, and deeper L5/6 of the neocortex at young adult age have been conducted as well [6, 7, 20, 53, 64–67]. Such work revealed the presence of distinct ChC morphological variants in the various cortical structures. For instance, in the piriform cortex, ChC cartridge number and AIS innervation were found to be higher in comparison to somatosensory and motor cortices, which could be attributed to differences in ChC axonal branching and/or density [7, 64]. Furthermore, analysis of murine hippocampal ChCs found that these cells cover twice as much area and innervate twice as many PyNs relative to neocortical ChCs [53]. These observations raised the question of whether the “cardinal” ChC type consists of multiple subtypes. To address this, Huang and colleagues combined genetic cell labeling with dual-color fluorescence micro-optical sectioning tomography (dfMOST) to achieve brain-wide imaging with high axon resolution and spatial registration of genetically-targeted single ChCs [27]. Such work unveiled that cortical ChCs in mice consist of multiple morphological subtypes that differ in laminar position as well as dendritic and axonal arborization patterns [27]. The group noted that in the neocortex, 55% of ChCs were located at the L1/2 border, 5% in L3, 22% in L5, and 18% in L6 and also identified eight different clusters of ChCs grouped according to the laminar distribution of ChC soma position and layer-specific neurite arborization. The observed differences in the laminar arrangement of ChC dendrites likely reflects differential recruitment by input streams whereas the local geometry and laminar distribution of their axons facilitates differential innervation of PyN ensembles. Ultimately, further identification/characterization of these new morphological subtypes should pave the way for future in-depth orthogonal studies of ChCs using connectomics, physiology, and molecular data.

Molecular Mechanisms Governing ChC Morphological Development and ChC/PyN Connectivity

Given the unique morphology and synaptic specificity of ChC axonal terminals, major effort has been devoted to identifying the molecular mechanisms governing ChC cartridge/bouton development and subcellular synapse targeting/formation, a multistep process involving pre- and postsynaptic initial contact, maturation/stabilization, and synaptic maintenance (Figure 2). Early work in mice identified ErbB4, a receptor tyrosine kinase expressed by PV+ interneurons including ChCs, as a key regulator of cortical ChC bouton morphogenesis [21]. Using conditional Erbb4 mice, this study found that loss of ErbB4 in ChCs causes a decrease in ChC bouton density without perturbing gross ChC morphology nor average ChC cartridge density and length. A follow-up report noted a similar decrease in ChC bouton density in Lhx6-Cre;Erbb4^fl/fl mice, which was paralleled by a significant reduction in the number of α2-containing GABA_A receptor (GABA_AR) clusters at the AIS of PyNs in the hippocampus and lateral entorhinal cortex [24]. This impairment in GABAergic synapse formation was associated with a significant decrease in miniature inhibitory postsynaptic current (mIPSC) frequency (but not amplitude) and no changes in pair-pulse ratios [24]. It was also noted that loss of ErbB4 from fast-spiking interneurons perturbed hippocampal rhythms/gamma oscillations and long-range synchrony between the prefrontal cortex (PFC) and hippocampus (see below) [24]. Interestingly, another report, published concurrently, linked ErbB4 function to the Rac/Cdc42 guanine nucleotide exchange factor (GEF) DOCK7 in ChC morphogenesis [28]. By leveraging a newly developed IUE-based strategy to target ChCs in mice, this study demonstrated that DOCK7 acts as a cytoplasmic activator of ErbB4 and, most importantly, promotes ChC cartridge/bouton development by enhancing ErbB4 activation independent of its GEF activity [28]. As seen for ErbB4 depletion in this study, DOCK7 knockdown in cortical ChCs caused a dramatic reduction in the density and size of ChC boutons and a disorganization of the network of ChC cartridges. Conversely, ectopic DOCK7 expression in ChCs significantly increased bouton size and density above control levels [28]. Finally, a decrease in ChC cartridge bouton number and mIPSC frequency was more recently also reported in Nkx2.1-CreER;Erbb4^fl/fl mice [33]. Given the reported link between ChCs and neurological disorders, it is of note that changes in the number/size of ChC boutons is one of the most salient features of individuals suffering from such conditions (see below) [32, 68–70].

In line with the reported changes in α2-containing GABA_AR clusters at the AIS of PyNs in Erbb4 mutant mice, recent work demonstrated a key role for proper GABA_AR clustering/distribution in AIS synapse formation [71]. By leveraging the Gabra2–1 transgenic mouse line, in which a mutation was made in the collybistin-binding region of the α2 subunit of GABA_ARs, this study found a significant reduction in VGAT-positive cluster number at the AIS of cortical PyNs in both homozygous and heterozygous Gabra2–1 mice. No change in the number of presynaptic VGAT-positive clusters apposing PyN somas was observed under such conditions, suggesting a critical and selective role for collybistin-α2 subunit binding in proper ChC/PyN AIS synapse formation [71].

With the surge in scRNA-seq and cell-type classification studies, it is of little surprise that such efforts have begun to identify molecular factors regulating the establishment of distinct inhibitory synapses. Of particular interest is the recent identification of the non-secretory growth factor FGF13 as a key regulator of ChC presynaptic morphogenesis [26]. More specifically, by performing RNA-sequencing and whole-transcriptome analyses of murine interneurons during peak synaptogenesis, Rico and colleagues found that different classes of GABAergic interneurons acquire unique molecular signatures upon first establishing synaptic contacts and identified FGF13 as a candidate molecule for regulating ChC synapse development. By knocking down ChC-expressed FGF13 in mice starting at P2, the group observed a significant decrease in the density of ChC presynaptic boutons as well as ChC axonal disorganization. Notably, the decrease in bouton density was not merely the result of defects in axonal development, as FGF13 knockdown after P14 still decreased the density of ChC presynaptic boutons without perturbing axonal organization, suggesting two different/independent functions of FGF13 in ChCs [26].

Despite the identification of ErbB4, DOCK7, and FGF13 as regulators of neocortical ChC cartridge bouton/synapse development, it is important to note that ChCs depleted for these molecules still make contact with PyN AISs, suggesting that another/other molecule(s) must govern this selective form of axo-axonic innervation in the neocortex. To identify the molecular players required for neocortical ChC/PyN AIS transsynaptic contact, our group performed an extensive RNA interference (RNAi) screen targeting murine PyN-expressed cell adhesion molecules (CAMs) [29], as such molecules have previously been implicated in axonal subcellular targeting and/or synaptic innervation [72, 73]. Of particular interest was the L1 immunoglobulin (Ig) CAM family member neurofascin-186 (NF186), which is not only enriched at the AIS but also regulates GABAergic innervation of cerebellar Purkinje cell AISs and clustering of inhibitory postsynaptic proteins at the AIS of hippocampal granule cells (GCs) in mice [74, 75]. By leveraging the Nkx2.1-CreER;Ai9 mouse line and IUE-based RNAi, our group screened over a dozen PyN-expressed CAMs, Ephs, and ephrins for their potential role(s) in mediating neocortical ChC/PyN AIS innervation [29]. Such screening surprisingly unveiled a critical role for the PyN pan-axonally-expressed IgCAM L1CAM, and not the AIS-localized CAMs NF186 and NrCAM, as a key regulator of ChC/PyN AIS innervation. More specifically, embryonic depletion of PyN L1CAM caused a significant reduction in the percentage of L1CAM-depleted PyN AISs innervated by neocortical ChCs at P28 [29]. As expected, such a reduction was mirrored by a decrease in VGAT- and gephyrin-positive puncta number at the AIS of L1CAM-depleted PyNs as well as reductions in both the frequency and amplitude of L1CAM-depleted PyN mIPSCs. It is important to note that this function of L1CAM is selective for ChC/PyN AIS innervation as the number of inhibitory synapses was unchanged along the somatodendritic compartment of L1CAM-depleted PyNs [29]. Additional work unveiled a dual role for PyN-expressed L1CAM; namely, L1CAM was found to be essential for both the establishment of neocortical ChC/PyN AIS innervation at P11/P12, which aligns with the timepoint when L1CAM first becomes enriched at the AIS, and for the maintenance of such contacts into adulthood [29]. Finally, noting that L1CAM selectively facilitates synaptic innervation at the AIS despite its pan-axonal localization, it was hypothesized that the pools of L1CAM localized at the AIS and distal axon are different, possibly via interactions with AIS- or distal axon-specific proteins. Of particular interest is ankyrin-G (AnkG), the master regulator of the AIS, which is an AIS-localized scaffolding protein that links AIS transmembrane proteins to the actin cytoskeleton via βIV-spectrin [76–78]. Moreover, previous in vitro work has shown that L1CAM interacts with AnkG via its cytoplasmic domain and that such binding promotes L1CAM oligomerization and limits its diffusibility selectively at the AIS [79–82]. To test the importance of this interaction, we used an AnkG-binding-deficient L1CAM mutant (Y1229H), as well as βIV-spectrin RNAi, and found that L1CAM-Y1229H failed to rescue the L1CAM-induced innervation phenotype and that PyN βIV-spectrin knockdown caused a significant decrease in neocortical ChC/PyN AIS innervation [29]. These results support a model in which the AnkG-βIV-spectrin cytoskeletal complex anchors and clusters L1CAM at the AIS to promote high-affinity cell adhesion between ChC cartridges and PyN AISs, thereby facilitating axoaxonic synapse formation/stabilization. Together, these findings underscore the importance of postsynaptic AnkG-associated L1CAM in both early postnatal and adult murine neocortical ChC/PyN AIS synaptic innervation and raise the intriguing question as to the identity of the presynaptic binding partner(s) of L1CAM present either on ChCs or in the extracellular milieu necessary for this subcellular targeting.

ChCs and Brain Disorders

As reported for many other subpopulations of interneurons, disturbances in ChC biology, especially alterations in ChC cartridge/bouton number and their GABA-mediated signaling, have been associated with several brain disorders linked to improper E/I balance, including schizophrenia, epilepsy, and ASD.

Regarding schizophrenia, early work on postmortem tissue of subjects with schizophrenia noted a decrease in the number/density of GAT-1 immunoreactive ChC cartridges in L2–4 of the dorsolateral PFC [68–70, 83, 84]. However, recent studies using more advanced techniques have begun to question these original findings, reporting that neither VGAT, GAD67, or calbindin (CB) protein levels per ChC bouton nor the number of boutons per cartridge are altered in the PFC of schizophrenic subjects. These observations led the authors to suggest that the originally reported decrease in cartridges is likely due to reduced expression of GAT-1 [32, 85]. Surprisingly, they instead found a 2.7-fold increase in the mean density of vGAT+/CB+, but not vGAT+/CB−, cartridges exclusively in L2 in individuals with schizophrenia. These findings are layer specific, as neither vGAT+/CB+ nor vGAT+/CB− cartridge density was altered across L-36. The authors speculate that this observed increase in vGAT+/CB+ cartridge density in L2 likely results from a defect in the postnatal pruning of ChCs in this cortical layer, which notably matures much later than deeper cortical layers [32]. Importantly, such a layer- and cell type-specific increase in cartridge density may underlie some of the pathophysiological/behavioral changes associated with schizophrenia, most likely via the targeted increase in inhibition of L2 PyNs. It is noteworthy that such findings also help to explain the reported increase in the density of GABA_AR-alpha2 subunit-labeled AISs in the PFC of individuals with schizophrenia [83]. While additional studies are clearly needed to fully profile ChC changes in schizophrenia, it is apparent that alterations in ChC development/maturation and, as a result, E/I balance are intimately associated with this disorder.

Besides a potential role for ChCs in the manifestation of schizophrenia, evidence from early work in human and non-human primates found a reduction in PyN AIS synapses in subjects with epilepsy, a debilitating condition marked by perturbed E/I homeostasis [86–89]. This was particularly the case for temporal lobe epilepsy, the most common form of epilepsy with focal seizures that originate in the temporal lobe of the brain [87, 89]. For instance, a recent report noted a significant reduction in PV+ ChC boutons in the dentate granule cell layer of human subjects with hippocampal sclerosis [89]. Surprisingly though, electron microscopy studies in human tissue from temporal lobe epilepsy subjects found no change in AIS synapse number on surviving CA1 PyNs and even an increase in synapse number on the AISs of dentate granule cells [90, 91]. These contrasting findings may be explained, at least in part, by the methods used to visualize ChC cartridges or, alternatively, the brain region being analyzed, especially given that a separate study found brain region-specific neurochemical and morphological reorganizations of ChC terminals in the sclerotic hippocampus of epileptic human individuals [92]. Importantly, with regards to the pathogenesis of epilepsy, it has been suggested that a reduction in ChC number could result in the formation of microcircuits that facilitate runaway excitation and the propagation of seizure activity [18].

Another group of brain disorders thought to involve, at least in some cases, dysregulated excitability is ASD. Hence, it is perhaps of little surprise that recent work has linked ChCs to this neurodevelopmental disorder. Using a combination of Vicia villosa lectin and PV immunostaining to discern between ChCs and BCs in postmortem human tissue, recent work found a significant decrease in ChC number in the PFC of individuals with ASD [30].

Interestingly, such a reduction was not consistent across all interneuron types, as the number of BCs was not as severely impacted [30]. Preliminary immunohistochemical analysis of cortical Brodmann area 46 in autistic individuals revealed a decrease in GAT1-positive ChC cartridge number and GABA_AR subunit ɑ2 levels at the AIS of PyNs (V. Martinez-Cerdeño, personal communication). Surprisingly though, the average number of boutons per ChC cartridge remained unchanged (V. Martinez-Cerdeño, personal communication). Despite the need for additional work, such initial findings point to a possible key role for altered ChC neurotransmission in the pathophysiology of ASD.

In addition to the abovementioned human studies, emerging evidence from animal research lends further support for an association between ChC abnormalities and neural disorders. Behavioral analyses of Lhx6-Cre;Erbb4^fl/fl mice, in which the schizophrenia- and epilepsy-linked gene Erbb4 was conditionally deleted in postmitotic, MGE-derived interneurons, found these mice to have a decrease in ChC cartridge bouton density and to display behavioral phenotypes associated with schizophrenia, including increased locomotor activity, decreased prepulse inhibition (PPI), impaired sociability, working memory dysfunction, deficiency in nest building, and perturbed synchrony in gamma-range oscillations between the frontal cortex and hippocampus [24]. Subsequent work leveraging Nkx2.1-CreER;Ai9;Erbb4^fl/fl mice observed similar behavioral phenotypes; including decreased PPI, impaired working memory and social cognition, and hyper-locomotor activity [33]. This study also found that acute systemic administration or intra-mPFC bilateral preinfusion of L-838417, a partial agonist of ɑ2-containing GABA_ARs, successfully reduced locomotor activity, enhanced PPI, and normalized social novelty recognition and working memory in mutant animals [33]. In a similar vein, genetic mutations in GABA_AR subunits have also been linked to genetic epilepsies, with the ɑ2 subunit (GABRA2) being identified via a recent genome-wide mega-analysis as one of the most likely biological epilepsy genes [34, 71, 93]. In addition, studies using the Gabra2–1 knock-in line, in which mice have altered clustering of ɑ2 subunit-containing GABA_ARs, found such animals suffer from developmental seizures and early mortality. Besides being implicated in the pathophysiology of schizophrenia, haploinsufficiency of ErbB4 has also been linked to epileptic encephalopathy [35]. Interestingly, in this regard, mutations in the DOCK7 gene, whose product has been shown to act as an activator of ErbB4 in the regulation of ChC cartridge/bouton development in mice, have been causally linked to a syndromic form of epileptic encephalopathy [36].

Given the critical roles of L1CAM and FGF13 in ChC/PyN AIS innervation and synapse development, it is of clinical relevance to note that mutations in both genes have been linked to neurological conditions. L1CAM mutations have been reported in individuals with a broad spectrum of neurological abnormalities and intellectual disability (ID), together known as L1 syndrome, and have also been linked to schizophrenia and epilepsy [94–97]. As noted above, knockdown of L1CAM in neocortical PyNs in mice causes a significant reduction in ChC/PyN AIS innervation, which is reflected by a decrease in both the frequency and amplitude of L1CAM-depleted PyN mIPSCs [29]. Such perturbations in humans could contribute to some of the neurological aspects of L1 syndrome. Moreover, FGF13 mutations are linked to Genetic Epilepsy and Febrile Seizures Plus (GEFS+) as well as Börjeson-Forssman-Lehmann syndrome, an X-linked disorder characterized by seizures, ID, obesity, hypogonadism, and distinctive facial features [37, 98]. This is of particular interest with regards to ChCs, since FGF13 knockdown in these cells causes a significant decrease in the density of ChC presynaptic boutons in addition to ChC axonal disorganization in mice [26].

Combined, the above studies support a connection between aberrant ChC morphogenesis/function and brain disorders. It should be noted though that other classes of interneurons very likely contribute to the onset and/or progression of these disorders in conjunction with ChCs. Finally, as with all animal studies, some caution should be taken with regards to their translatability to humans, nevertheless these studies combined with new investigations of ChCs at the systems level will undoubtedly shed further light on the function of ChCs in circuitry in health and disease states.

Concluding Remarks and Future Perspectives

Significant progress has been made in recent years in our understanding of ChC biology, particularly in regard to their origin and migration, differentiation/maturation, and synaptic targeting mechanisms. There has also been growing recognition of the association between ChC dysfunction and neural disorders in multiple domains, including neurodevelopmental (e.g., ASD), neurological (e.g., epilepsy), and neuropsychiatric (e.g., schizophrenia) disorders. Despite the substantial progress in these areas, multiple questions remain (see Outstanding Questions). Emerging evidence suggests imbalanced E/I homeostasis and interneuron dysfunction to be shared pathophysiological mechanisms of the abovementioned neural disorders [99–101]. In this regard, recent work has implicated ChCs in the regulation of PyN firing and orchestration of dynamic PyN ensembles. Future research that addresses both the role of ChCs in the regulation of proper network/circuit activity and how ChC developmental and connectivity defects impact network/circuit function in disease is paramount. Knowledge gained from such studies may help to unveil key factors underlying these afflictions, thereby raising hopes for new preventive and therapeutic strategies.

Figure I. Interneuron Neurogenesis in the Embryonic Brain.

Coronal sections through the embryonic brain at E14.5 and E17.5 highlighting the brain regions necessary for interneuron neurogenesis, namely the MGE, LGE, CGE, and POA. Expression of the homeodomain transcription factor NKX2.1 in the MGE and POA is denoted by red shading. Interneuron migratory routes in the embryonic brain are denoted by dashed arrows.

Outstanding Questions.

• Do differences in PV protein levels among ChCs correlate with other properties such as morphology, physiology, connectivity, and/or function?

• What is/are the principal function(s) of ChCs in the neocortex and hippocampus? Does their function change throughout development and adulthood?

• Are ChCs in upper and deeper cortical layers functionally distinct?

• Do ChC morphological subtypes have distinct connectome motifs? What is the relationship between early- and late-born ChCs and their morphological subtypes?

• What is/are the presynaptic partner(s) of L1CAM on ChC cartridges and/or in the extracellular milieu necessary for proper neocortical ChC/PyN AIS innervation?

• Apart from the neocortex, ChCs are also found in the hippocampus, piriform cortex, and amygdala. Is postsynaptic L1CAM required for ChC/PyN AIS innervation in these brain regions outside the neocortex?

• What molecular factors regulate the formation and/or stabilization of excitatory and inhibitory synaptic inputs onto ChCs?

• Do other cell types (e.g., astrocytes, oligodendrocytes, microglia) regulate ChC subcellular synaptic targeting of PyN AISs? Do trophic and/or other secreted factors regulate ChC/PyN AIS synapse formation and, if so, what are their identities?

• Will manipulating ChC activity/function prove to be a useful therapeutic strategy for treating ChC-linked disorders such as epilepsy, schizophrenia, and ASD? Are there other neural diseases that are associated with ChCs?

Highlights.

• Interneurons, including GABAergic chandelier cells (ChCs), are critical for regulating circuit function in the central nervous system (CNS). Perturbations in excitatory/inhibitory balance or homeostasis via deficits in interneuron-mediated neurotransmission have been linked to debilitating neurological conditions.

• ChCs undergo a unique developmental program, which differs from that of other types of interneurons, to ensure their proper genesis, positioning, morphology, and connectivity in the neocortex.

• Recent studies using RNA interference and animal models have begun to shed light on the molecules governing ChC axonal morphogenesis and synapse formation.

• Studying ChC biology via genetic, molecular, and cellular strategies is a promising approach to elucidate the pathophysiological mechanisms involved in the development of ChC-linked disorders, including schizophrenia, epilepsy, and autism spectrum disorder.

Glossary

Axon Initial Segment (AIS)

Subcellular domain of a neuron located at the proximal end of its axon, responsible for generating and shaping action potentials (APs). The AIS is composed of a high density of voltage-gated ion channels necessary for AP generation in addition to a unique ankyrin-G-based cytoskeleton and associated cell adhesion molecules, including neurofascin-186 (NF186) and NrCAM. In the neocortex, the AIS of excitatory pyramidal neurons selectively receives GABAergic innervation from chandelier cells

Cell Adhesion Molecules (CAMs)

A group of transmembrane cell adhesion proteins present on the cell surface that are required for proper nervous system development, including neurite formation and axon pathfinding. CAMs mediate cell-to-cell and/or cell-to-extracellular matrix (ECM) interactions which can induce intracellular responses affecting cytoskeletal organization, molecular signaling, and/or gene expression. Major superfamilies of CAMs in the CNS include cadherins, integrins, and the immunoglobulin superfamily of CAMs (IgCAMs)

Excitatory/Inhibitory Balance (E/I Balance)

The balance of excitatory and inhibitory synaptic membrane currents received by a neuron. The interplay between such excitatory and inhibitory inputs, in combination with complex feedforward, gain control, and feedback inhibition mechanisms, underlies a neuron’s spontaneous firing and/or response to sensory inputs. Maintaining proper E/I balance is critical for homeostatic cortical and subcortical circuit function and, when perturbed, has been shown to contribute to neurological disorders, including schizophrenia, epilepsy, and autism spectrum disorder

Ganglionic Eminence (GE)

A transient structure present in the brain during embryonic and fetal stages of neural development. The GE is located between the thalamus and caudate nucleus and is divided into three subdivisions; namely, the medial, lateral, and caudal ganglionic eminences (MGE, LGE, and CGE, respectively). The GE is a major source of GABAergic interneuron progenitors which give rise to distinct GABAergic interneuron types that sequentially undergo tangential and radial migration prior to reaching their final position in the cortex

In Utero Electroporation (IUE)

A technique utilizing electric pulses for acute gene delivery into live mouse embryos developing in utero. IUE is an invaluable technique for investigating the molecular and cellular mechanisms that guide neural stem cell proliferation, differentiation, migration, and maturation during development

NKX2.1

Homeodomain transcription factor that is expressed in the subpallium, including the MGE and preoptic area. NKX2.1 regulates the formation and patterning of the MGE and is critical for the generation of major GABAergic interneuron subclasses such as parvalbumin- and somatostatin-expressing interneurons

Ventricular Zone (VZ) & Subventricular Zone (SVZ)

The VZ is a transient embryonic layer of tissue containing neural stem cells, predominantly radial glial progenitor cells, that neighbors the lateral ventricles. Lying adjacent to the VZ is the SVZ, a secondary proliferative zone that contains intermediate neuronal progenitors which continue to divide into post-mitotic neurons. During embryonic and fetal development, neurogenesis, or the generation of new neurons, depletes the parent neural stem cell pool in the VZ, leading to its disappearance, whereas the SVZ is maintained through adulthood.