Development of the Vertebrate Retinal Direction-selective Circuit

Abstract

The vertebrate retina contains an array of neural circuits that detect distinct features in visual space. Direction-selective (DS) circuits are an evolutionarily conserved retinal circuit motif – from zebrafish to rodents to primates – specialized for motion detection. During retinal development, neuronal subtypes that wire DS circuits form exquisitely precise connections with each other to shape the output of retinal ganglion cells tuned for specific speeds and directions of motion. In this review, we follow the chronology of DS circuit development in the vertebrate retina, including the cellular, molecular, and activity-dependent mechanisms that regulate the formation of DS circuits, from cell birth and migration to synapse formation and refinement. We highlight recent findings that identify genetic programs critical for specifying neuronal subtypes within DS circuits and molecular interactions essential for responses along the cardinal axes of motion. Finally, we discuss the roles of DS circuits in visual behavior and in certain human visual disease conditions. As one of the best-characterized circuits in the vertebrate retina, DS circuits represent an ideal model system for studying the development of neural connectivity at the level of individual genes, cells, and behavior.

INTRODUCTION

The retina includes a remarkable array of neural circuits that detect, extract, and process key features in the visual world. Like many other CNS structures, the retina has a laminar architecture, separating the cell bodies and synapses of the six main neuronal types into distinct layers (Fig. 1). Vision begins in rod and cone photoreceptors, which detect photons of light and reside at the back of the retina in the outer nuclear layer (ONL). Bipolar cells (BCs), located in the inner nuclear layer (INL), transduce signals from photoreceptors to the projection neurons of the retina, the retinal ganglion cells (RGCs), which reside in the ganglion cell layer (GCL) and target visual brain areas. Horizontal and amacrine cell interneurons, which also reside in the INL, filter the synaptic inputs received by BCs and RGCs, respectively. There are two synaptic layers: the outer plexiform layer (OPL) that separates the ONL from the INL, and the inner plexiform layer (IPL) that separates the INL from the GCL. The processing power of the retina follows from this precise organization of cell bodies and synapses, allowing for interpretation of color, contrast, and motion even before visual information reaches the brain.

Fig. 1. Cell types of the retina.

Cell bodies of the retina are organized into three layers: the outer nuclear layer (ONL), containing rod and cone photoreceptors; the inner nuclear layer (INL), containing horizontal cells (HCs), bipolar cells (BCs), which are divided into ON and OFF subtypes, and amacrine cells (ACs); and the ganglion cell layer (GCL), containing retinal ganglion cells (RGCs) and displaced ACs. Synapses between OPL-residing and INL-residing cells form within the outer plexiform layer (OPL), while synapses between INL-residing and GCL-residing cells form within the inner plexiform layer (IPL). RGCs depicted include direction-selective ganglion cells (DSGCs) which are categorized as responsive to light onset (ON DSGCs) or both light onset and offset (ON-OFF DSGCs). ACs depicted are starburst amacrine cells (SACs), which provide asymmetric inhibition onto DSGCs; SACs responding to light onset (ON SACs) reside in the GCL, while SACs responding to light offset (OFF SACs) reside in the INL. Specific subtypes of OFF and ON BCs wire in direction-selective circuits, including OFF BC2s/BC3as and ON BC5s/BC7s.

One of the best-characterized retinal circuits, the direction-selective (DS) circuit, specializes in motion detection. The DS circuit motif is evolutionarily conserved in flies, zebrafish, rodents, rabbits, and primates (Giolli et al., 2006; Simpson, 1984; Masseck & Hoffmann, 2009, Fredericks et al., 1988). In many vertebrates it consists of two separate components – one for fast and one for slow motion – that contribute to visual behavior. DS circuits include three specialized types of neurons: direction-selective ganglion cells (DSGCs), BC subtypes, and starburst amacrine cells (SACs) that together respond to increases (ON) or decreases (OFF) in illumination. BCs and SACs wire precisely onto DSGCs such that a given DSGC fires robust action potentials in response to a single preferred direction (PD) but does not fire in response to motion in the opposite, or null, direction (ND).

DSGCs were first discovered by Barlow and Hill in the rabbit retina (Barlow & Hill, 1963). Subsequent studies established that DSGCs fall into two broad classes, ON-OFF and ON, based on whether or not they respond to light decrements in addition to light increments. ON-OFF DSGCs detect motion in four PDs spaced 90 degrees apart – dorsal, ventral, nasal, and temporal – and are tuned to fast motion; they project to brain areas involved in pattern vision. ON DSGCs were originally described as having three PDs spaced about 120 degrees apart – upward (up-ON DSGCs), downward (down-ON DSGCs), and forward (forward-ON DSGCs). ON DSGCs respond to slower motion. They project to central targets that belong to the accessory optic system (AOS), and they generate the optokinetic reflex (OKR) that compensates for retinal slip during image motion (Simpson, 1984). The three canonical PDs of ON DSGCs have classically been aligned with the axes of the vestibular system; however, a recent study has identified a fourth nasal-preferring ON DSGC and describes ON DSGC PDs spread 90 degrees apart and aligned with those of ON-OFF DSGCs but not vestibular coordinates (Sabbah et al., 2017). Nevertheless, this newer view of ON DSGC tuning only adds to our understanding of AOS-projecting ON DSGCs accrued over the last several decades. In the early 2000s, the study of DSGCs expanded into the mouse retina (Yoshida et al., 2001), enabling the use of the vast genetic tools available in mouse models.

In this review, we trace the development of DS circuits over time (Fig. 2) – from early events that govern the birth and differentiation of transcriptionally distinct retinal neuron subtypes to the mechanisms that specify sophisticated patterns of synaptic connectivity – and we explore their function in normal visual behavior and certain disease conditions. With its well-characterized catalog of cell types and subtypes, its genetic accessibility, and its amenability to manipulation and behavioral testing, the retinal DS circuit is an ideal model for studying the molecular mechanisms that generate neuronal subtype diversity, that instruct patterns of neurite morphology and lamination, and that guide synapse formation and maintenance both in the retina and throughout the CNS.

Fig. 2. Developmental time course of ON/ON-OFF DS circuit components and effects of key LOF mutants.

(A) Cross-section view of the developing retina, E11-P21. E11-E16.5: After birth, SACs migrate to their appropriate laminar positions, with OFF SACs ultimately settling in the outer neuroblastic layer (ONBL) while ON SACs migrate into the inner neuroblastic layer (INBL). E17.5-P3: SACs refine their dendritc arbors to organize the ON and OFF DS sublaminae of the IPL. P4-P21: DSGCs organize dendrites into discrete ON and OFF laminae; SACs form asymmetric inhibitory connections onto DSGCs. (B) En face view of developing SACs, P3-P12; SAC dendrites develop self-avoidance over time. (C) Pcdhγ LOF impairs SAC dendritic self-avoidance. (D) Fezf1 LOF causes all SACs to settle in the INL and become OFF SACs. (E) Megf10 LOF reduces the regularity of the mosaic spacing of SAC somas. (F) LOF of Sema6a and/or Plxna2 causes the dendrites of OFF and ON SACs to form crossovers across DS sublaminae. (G) Cdh8 LOF induces OFF BC2s to terminate in the ON sublamina, while Cdh9 induces ON BC5s to terminate in the OFF sublamina. Dashed lines in (G) denote original laminar positions.

EARLY EVENTS IN DS CIRCUIT DEVELOPMENT

BIRTH, FATE SPECIFICATION, AND DIFFERENTIATION OF DIRECTION-SELECTIVE RETINAL NEURON SUBTYPES

The assembly of DS circuits begins with the birth and differentiation of each of their primary cellular components: RGCs, SACs, and BCs. Cell genesis in the nascent retina begins at embryonic day (E) 8, giving rise during the rest of embryogenesis and early postnatal development to the seven retinal cell types in defined temporal and spatial (central-to-peripheral) orders as multipotent progenitors pass through competence states. Classic birthdating experiments revealed that RGCs are the firstborn neurons in the embryonic retina, followed by horizontal cells, cones, and amacrine cells; bipolar cells, rods, and Müller glia arise later in embryonic and postnatal development (Rapaport et al., 2004). RGC genesis occurs between E8 and E17 (with birth peaking at E11), while ACs are born between E8 and P5 (peaking at E16), and BCs are born between E17 and P6 (peaking at P2; Voinescu et al., 2009). Immunolabeling with cell type-specific markers following birthdating with bromodeoxyuridine (BrdU) revealed that subtypes of amacrine cells, differing in neurotransmitter expression and soma position, are born in overlapping but distinct windows. Specifically, the peak timing of GABAergic amacrine cell birth (E14) precedes that of glycinergic amacrine cells (P0), and within the former group SACs are among the firstborn, with 20-30% postmitotic by E10 and 90% postmitotic by E14 (Voinescu et al., 2009). These birthdating experiments also demonstrated a correlation between SAC birthdate and soma position: ON SACs, which are mostly postmitotic after E14 and reside in the GCL, are born earlier than OFF SACs, which are mostly postmitotic by E17 and located in the proximal portion of the INL (Voinescu et al., 2009) (Fig. 2).

After exiting the cell cycle, myriad transcription factors mediate both the specification of retinal cell types and the differentiation of diverse subtypes within each class. The LIM-homeodomain transcription factor Islet-1 (Isl1) is required for the differentiation of ON bipolar cells (BCs) and also SACs (Elshatory et al., 2007), and conditional deletion of Isl1 in the retina results in a severe reduction in the numbers of these cells. The forkhead transcription factor Foxn4 makes multipotent retinal progenitors competent to produce amacrine and horizontal cells and acts upstream of the bHLH transcription factor Ptf1a; deletion of Ptf1a results in a marked depletion of horizontal and amacrine cells and an increase in RGCs (Fujitani et al., 2006). Further, the homeodomain transcription factor Chx10 (Burmeister et al., 1996) and the basic helix-loop-helix (bHLH) transcription factors Mash1 and Math3 (Hatakeyama et al., 2001) mediate the early specification of BCs, while the paired-like homeodomain protein Vsx1 (Ohtoshi et al., 2004) and the Olig family bHLH protein Bhlhb4 (Bramblett et al., 2004; Chow et al., 2004) are required specifically for the differentiation of OFF cone BCs and rod BCs, respectively.

Though many molecules have been implicated in the differentiation of broad retinal cell classes, it is less clear how closely related subtypes within a class acquire different characteristics and also at what point in development these molecular differences are specified. Single-cell transcriptomic profiling of SACs at E16 revealed two molecularly distinct populations, indicating that genetic differences between ON and OFF SAC subtypes arise early in development (Peng et al., 2020). At E14, expression of the transcriptional regulator Fezf1 initiates in a subset of SACs at the boundary between the outer and inner neuroblastic layers. This promotes the expression of ON genes and represses the expression of OFF genes, thus determining the fate of ON SACs, which subsequently migrate into the forming GCL. Fezf1 loss-of-function results in loss of the ON SAC population in the GCL, with ON SACs undergoing a conversion into OFF SACs (Peng et al., 2020) (Fig. 2d).

In addition, single cell transcriptomic profiling of BCs in the juvenile (P17) mouse retina revealed fine-scale molecular differences among BC subtypes, distinguishing BCs with distinct response properties and lamination patterns. Previously identified morphological subtypes correspond to 15 molecularly distinct subtypes, and the molecular relatedness among BC subtypes correlates with IPL lamination depth, with more closely related subtypes laminating at similar depths within the IPL (Shekhar et al., 2016). Identified BC subtypes that wire DS circuits include the ON BC5s and BC7s, and the OFF BC2s and BC3as (Duan et al., 2014; Kim et al., 2014; Greene et al., 2016).

Single-cell transcriptomic profiling of mouse RGCs revealed roughly 40 transcriptionally distinct subtypes at P5 (Rheaume et al., 2018). A variety of molecular markers and transgenic mouse lines have been identified that label ON and ON-OFF DSGC subtypes (Morrie and Feller, 2016; Dhande et al., 2015; Sanes and Masland, 2015; Kay et al., 2011). These markers and mouse lines are invaluable for guiding targeted electrophysiological recordings, visualizing cells for morphological analysis, and establishing DSGC identity during immunohistochemistry and in situ hybridization experiments. However, no markers for a single subtype have yet been tied to a role in DS circuit development, reflecting a limited understanding of the genetic mechanisms that drive DSGC subtype specification. Many of the genes targeted in commonly used BAC transgenic lines are not actually expressed by labeled DSGCs (Kay et al., 2011; Duan et al., 2018; Laboulaye et al., 2018); in these lines, GFP expression is due to positional effects of transgene integration, thus limiting their ability to shed light on the genetic programs giving rise to directional preference. However, among genes used as markers and for genetic analysis in mouse lines, Cadherin 6 (Cdh6) is a marker for vertical-preferring ON-OFF DSGCs (Kay et al., 2011) and has the best characterized functional role in DSGC development. Cdh6 is associated with a bias towards vertical-preferring ON-OFF DSGC fate as early as the E10 progenitor stage; although progenitors labeled in Cdh6-CreER retinas at this stage give rise to cells in all three retinal cell body layers, around 80% of the RGCs they produce are ON-OFF DSGCs, with under 10% of labeled RGCs expressing the nasal-preference marker Mmp17 (De La Huerta et al., 2012).

Taken together, results of these and other transcriptomic profiling experiments suggest that retinal neuron subtypes, including those specialized for motion detection, are transcriptionally distinct populations shortly after birth, and that these molecular differences arise independent of visual experience.

MIGRATION, CELL BODY POSITIONING, AND MOSAIC SPACING OF STARBURST AMACRINE CELLS

Following their birth in the outer neuroblastic layer (ONBL), retinal neurons must migrate varying distances to settle in their appropriate locations within the retina. Cellular mechanisms that direct migration and positioning of SACs have recently been described and may represent a general strategy employed by other DS cell types that reside in the inner retina. ON and OFF SACs must migrate differentially such that ON SACs reside in the GCL, while OFF SACs reside in the INL (Fig. 1 and Fig. 2a). Just after birth, SACs migrate basally with bipolar morphology and by E15.5 accumulate at the boundary between the ONBL and the inner neuroblastic layer (INBL). There, ON SACs retract their trailing process, leaving a prominent leading process directed toward the INBL. However, in the absence of the Fezf1 transcription factor, “ON” SACs instead have a prominent process directed toward the ONBL, much like OFF SACs (Fig. 2d). This change in ON SAC migratory pattern is consistent with transcriptomic profiling data indicating that Fezf1 loss-of-function (LOF) results in a switch from an ON-gene expression program to an OFF one (Peng et al., 2020). Fezf1 represses the expression of the Rho GTPase Rnd3, which is differentially expressed in OFF SACs at least as early as E16. Interestingly, Rnd3 has well-characterized roles in regulating excitatory neuron migration in the cerebral cortex, in axon elongation and guidance, and in neurite outgrowth (Kaur et al., 2020; Azzarelli et al., 2014; Peris et al., 2012). Knockdown of Rnd3 causes SACs to reside in the GCL, indicating that Rnd3 might directly regulate cytoskeletal dynamics during OFF SAC radial migration, or that Rnd3 promotes the outgrowth of OFF SAC dendrites, stabilizing the position of these cells in the INL. Taken together, these results show that early transcriptional differences between ON and OFF SACs mediate their differential migration into separate cell body layers in the inner retina (Peng et al., 2020).

The transcription factor sex determining region Y box 2 (Sox2) also regulates SAC migration and cell body positioning. Though conditional deletion of Sox2 does not alter total SAC number, significantly fewer SACs reside in the GCL, suggesting perturbed migration of ON SACs. Of the remaining ON SACs in the GCL in the absence of Sox2, some exhibit stratification within both the OFF and ON strata of the IPL (Whitney et al., 2014) and consequently display electrophysiological responses to both increases and decreases in illumination (Stinic et al., 2018). Although Sox2 is expressed in both OFF and ON SACs, Sox2 might act downstream of Fezf1 to regulate the expression of genes that specifically promote ON SAC migration and also later events such as laminar dendritic stratification.

Upon reaching the appropriate cell body layer, SAC somas tile the retina in a roughly evenly spaced, or mosaic, arrangement. Mosaic spacing creates a minimum exclusion zone between neighboring SAC somas, ensures even coverage of SACs across the retina and ultimately facilitates parallel processing of visual stimuli that span multiple receptive fields. Several factors contribute to the mosaic spacing of SAC cell bodies, including appropriate radial and tangential migration, apoptosis, and SAC-SAC homotypic interactions. During migration, SAC somas are in a random spatial arrangement, indicating that SACs must migrate tangentially upon reaching their appropriate laminar destination to achieve mosaic spacing. At late embryonic ages, even as additional SACs continue to enter the nascent INL and GCL, SAC somas form a mosaic similar to what is observed in the mature adult retina, suggesting that the formation of mosaics is a dynamic process that depends on homotypic SAC-SAC interactions (Galli-Resta et al., 1997).

Multiple EGF Like Domains 10 (Megf10) was identified in a P6 microarray profiling study as a gene with selective expression in early postnatal SACs. Megf10 encodes a single-pass transmembrane protein that contains an extracellular EMI-domain and 17 EGF-like repeats, and an intracellular domain containing putative phosphorylation targets at PTB-domain and SH2-domain binding sites. Megf10 LOF disrupts the mosaic spacing of SAC somas in the INL and GCL without affecting characteristic SAC dendritic morphology, including the self-avoidance of dendritic processes observed in individual SACs (Kay et al., 2012). Gain-of-function (GOF) experiments performed by electroporating later-born retinal neurons with a fluorescent Megf10 fusion protein revealed that SACs avoid areas of high Megf10 expression, showing that Megf10 likely acts as a repulsive cue. Because Megf10 overexpression failed to induce SAC repulsion in a Megf10^−/− background, Megf10 likely participates in a homophilic signaling complex as both a receptor and a ligand (Kay et al., 2012). The RNA binding protein cytoplasmic polyadenylation element binding protein 3 (CPEB3) negatively regulates the translation of Megf10, and CPEB3 LOF alters the mosaic distribution of ON, but not OFF, SACs (Chen et al., 2016). The appropriate positioning and spacing of SAC cell bodies sets the stage for SAC dendrites to laminate the IPL, which later serves as a scaffold to recruit neural processes of DS circuit partners.

LAMINAR STRATIFICATION AND ELABORATION OF DENDRITES AND AXONS IN DIRECTION-SELECTIVE CIRCUITS

The laminar stratification and morphogenesis of axons and dendrites in select sublayers of the IPL brings synaptic partners in close apposition and mediates proper DS circuit connectivity. A diverse array of molecules – including Megf10, axon guidance cues, cell adhesion molecules such as cadherins, protocadherins, and immunoglobulin superfamily members, and also several transcription factors – regulates these processes during DS circuit development. Among the neurons that comprise DS circuits, SACs are the first to laminate the nascent IPL, with distinct OFF and ON laminae apparent between P0 and P2 (Stacy & Wong, 2003). Prior to IPL lamination, SACs form homotypic soma-layer contacts with each other that persist until P2–P3, after which SAC processes stratify solely within the IPL. In addition to regulating the mosaic spacing of SAC somas (Kay et al., 2010), Megf10 acts earlier, homophilically, between neighboring SACs to mediate proper IPL innervation (Ray et al., 2018). Megf10 LOF causes SACs to form more soma-layer contacts with each other, resulting in abnormal IPL lamination (Fig. 2e). After P5, the OFF and ON strata eventually form in Megf10 mutants, albeit with discontinuities and some mistargeting to ectopic IPL layers. Specifically, OFF SACs form an ectopic IPL substratum close to the INL, which is presumed to arise from the increased soma-layer contacts observed prior to P3/P5. Cre lines allowing for either early (Six3^cre) or late (Chat^cre) conditional deletion of Megf10 reveal that Megf10 activity is dispensable for SAC IPL lamination after P3. Ultimately, ectopic SAC IPL lamination resulting from Megf10 LOF leads to laminar targeting defects in ON-OFF DSGCs and some BCs, consequently causing a broadening and weakening of ON-OFF DSGC directional tuning responses (Ray et al., 2018). These observations show that a SAC dendritic scaffold guides the laminar stratification of DSGC dendrites and BC axons by providing either permissive or attractive cues, and also that appropriate patterns of SAC IPL lamination ensure proper DS circuit function.

The transmembrane axon guidance cue semaphorin 6A (Sema6A) and its receptor plexin A2 (PlexA2) also regulate the laminar stratification of SAC dendrites. SACs express both Sema6A and PlexA2, and LOF of Sema6A, PlexA2, or both disrupts the laminar stratification of SAC dendrites, resulting in crossovers among SAC dendritic processes between the OFF and ON sublaminae (Fig. 2f). Furthermore, ON SACs, but not OFF SACs, exhibit reduced dendritic field area and symmetry, with some portions of their arbors largely missing. In addition, ON SACs display loss of self-avoidance in their distal-most dendrites. Exogenous Sema6A repels neurites of presumptive OFF SACs in vitro, suggesting that repulsive Sema6A/PlexA2 signaling in vivo maintains the separation between ON and OFF SAC dendrites (Sun et al., 2013). Furthermore, Sema6A LOF reduces the directional tuning of the ON, but not the OFF, responses of posterior motion-preferring ON-OFF DSGCs, in line with the observed disruption in ON, but not OFF, SAC dendritic arbors in these mutants. Despite their altered dendritic morphology, SACs in Sema6A mutants still form asymmetric inhibitory inputs onto DSGCs and retain direction selectivity in individual neurites (Morrie & Feller, 2018). Whole-cell voltage-clamp recordings of Sema6A^−/− DSGCs revealed an attenuation of null-side inhibition. Computational simulations show that reducing the SAC coverage factor, defined as cell density multiplied by the field area, is sufficient to produce comparably reduced null-side inhibition in DSGCs. These results suggest that SAC dendritic morphology and coverage are important for appropriate connectivity in DS circuits. In addition, in these mutant backgrounds the processes of many OFF and ON SACs still laminate appropriately within S2 and S4, indicating that other molecules act together with Sema6A/PlexA2 to ensure the appropriate sublaminar targeting of OFF and ON SAC dendrites. Furthermore, the selective effects of Sema6A and PlexA2 LOF on the overall morphogenesis of ON, but not OFF, SAC dendritic arbors suggest that OFF and ON SACs use distinct molecular mechanisms to acquire their radially symmetric dendritic morphology.

After arriving at appropriate laminar strata in developing presumptive layers S2 and S4 of the immature IPL, SAC dendrites at P3 are highly branched and dendritic processes from individual SACs overlap extensively (Fig. 2b). However, by P12 SAC dendrites exhibit self-avoidance, giving rise to a radially-symmetric arbor (Lefebvre et al., 2012). Deletion of the 22 genes in the γ-subcluster of the Protocadherin (Pcdh) locus – which contains a tandem array of three α, β, and γ clusters – leads to a severe disruption in SAC dendritic self-avoidance (Fig. 2c). This includes tangled, bundled branches apparent as early as P3, without affecting overall dendritic arbor diameter, IPL sublaminar stratification, or the mosaic spacing of SAC somas. Much like the numerous Drosophila DSCAM protein isoforms (Zipursky & Sanes, 2010), Protocadherin γ genes (Pcdhgs) undergo alternative splicing in regions encoding the extracellular domain combined with a constant transmembrane and intracellular domain; each splicing isoform exhibits stochastic expression in SACs and acts cell-autonomously to mediate isoneuronal repulsive interactions that mediate self/non-self-discrimination (Lefebvre et al., 2012; Zipursky & Sanes, 2010). Pcdhγs are partially redundant with Pcdhαs, and combined loss-of-function of these protocadherins in SACs leads to even more profound defects in dendritic self-avoidance (Ing-Esteves et al., 2018).

SACs laminate the IPL before DSGCs, suggesting that SAC dendrites provide laminar cues for DSGC dendritic stratification. RGC dendrites begin to stratify at P4 and are initially diffuse through the depth of the IPL but later separate into distinct sublaminae during the first postnatal week (Stacy & Wong, 2003). Two classical cadherins, Cdh6 and Cdh10, pattern the dendritic lamination of dorsal- and ventral-ON-OFF DSGCs (Duan et al., 2018). Loss of Cdh6 and Cdh10 results in severe defects in dorsal- and ventral-ON-OFF DSGC dendritic lamination, with diffuse dendrites apparent as early as P7 (Duan et al., 2018). SACs, which also express Cdh6 but do not exhibit lamination defects in these mutants, likely function as a scaffold to recruit ON-OFF DSGC dendrites. Ablation of SACs by select expression of the diphtheria toxin receptor results in similar lamination defects in ventral-ON-OFF DSGCs, phenocopying the effects of Cdh6/Cdh10 LOF. Nasal-ON-OFF DSGCs labeled by the transgenic mouse line Drd4-GFP exhibited no lamination defects in a triple Cdh6-9-10 mutant background; however, ablation of SACs also disrupted the lamination of these ON-OFF DSGCs, indicating that various DSGC subtypes interact with the SAC dendritic scaffold for proper IPL lamination. Single-cell RNA-sequencing of nasal-ON-OFF DSGCs and SACs showed that nasal-ON-OFF DSGCs selectively express Cdh7, whereas SACs express its heterophilic binding partner Cdh18. RNAi-mediated knockdown of Cdh7 or ectopic expression of Cdh18 by subretinal electroporation was sufficient to disrupt the lamination of nasal-ON-OFF DSGCs but not ventral-ON-OFF DSGCs. Taken together, these results show that different classes of DSGCs use unique combinations of cell adhesion molecules to interact with the SAC dendritic scaffold during IPL innervation.

The transcription factor Satb1 is expressed in over 90% of Hb9-GFP (ventral-ON-OFF DSGCs), Drd4-GFP (nasal-ON-OFF DSGCs), and BD (bistratified-dendrite) ON-OFF DSGCs, with retinal Satb1 expression largely confined to ON-OFF DSGCs (Peng et al., 2017). Satb1 is required for the stabilization of the ON dendritic arbor of ventral-ON-OFF DSGCs. Although Satb1^−/− ventral-ON-OFF DSGCs initially stratify in both S2 and S4 at P6, the ON/S4 arbor is eventually lost. Arborization is also affected in nasal-ON-OFF DSGCs: one third of mutant cells develop a single arborization in S2, and the remaining cells were evenly split between having a single stratification within S3 and having normal S2/S4 bistratification, possibly as a result of a compensatory increase in Satb2 expression following Satb1 deletion. Satb1 drives ON-OFF DSGC expression of the immunoglobulin superfamily protein contactin 5 (Cntn5), which homophilically binds CNTN5 found in ON, but not OFF, SACs. This stabilizes the S4 arbor through its cis signaling partner Caspr4; arbor stabilization requires Cntn5 expression in both DSGCs and SACs (Peng et al., 2017). These and the studies mentioned above underscore the role for cadherin and contactin cell adhesion molecules (CAMs) in regulating the laminar organization of ON-OFF DSGC dendrites. Furthermore, the finding that distinct sets of cadherins regulate the dendritic lamination of ON-OFF DSGCs with different tuning preferences raises the possibility that cadherin interactions are important for establishing DSGC tuning preference (Duan et al., 2018).

A number of DSGC subtypes have dendritic arbors that are spatially asymmetric (Kay et al., 2011; Lilley et al., 2019). Though the majority of these subtypes do not align their dendrites with their PD, this property is found in Hb9-GFP ventral-ON-OFF DSGCs. Since this asymmetry is observed in a molecularly-labeled ON-OFF DSGC population, a genetic program is likely at play. However, recent dark-rearing experiments reveal that this dendritic arbor asymmetry is due in part to visual activity. Hb9-GFP arbors are asymmetric at eye opening but not aligned towards the ventral retina. Though normally-reared mice show alignment of these arbors toward the ventral retina in young adults, this is not observed in dark-reared mice (El-Quessny et al 2020). These results suggest that a combination of activity-dependent and activity-independent cues govern the dendritic morphology patterns of certain ON-OFF DSGC subtypes.

As the latest born cell type in DS circuits, BCs are also the last to wire during retinal development. In addition to regulating the wiring specificity of various DSGC subtypes, the classical cadherins also regulate laminar stratification of BC axons that wire in DS circuits. Cadherin 8 (Cdh8) and Cadherin 9 (Cdh9) interact heterophilically to instruct the stratification of OFF BC2 and ON BC5 axonal arbors in the OFF and ON sublayers of the IPL, respectively (Duan et al., 2014). Deletion of Cdh8 results in the mistargeting of roughly half of OFF BC2 terminals to the ON IPL sublayer, whereas deletion of Cdh9 results in the mistargeting of roughly one third of ON BC5 terminals to the OFF sublayer of the IPL (Fig. 2g). Optogenetic stimulation of mutant BCs while recording from ventral-ON-OFF DSGCs (Hb9-GFP) revealed that Cdh8 or Cdh9 deletion alters the incidence, but not the strength, of individual synaptic connections. As a consequence, Cdh8 deletion specifically reduced OFF responses of ON-OFF DSGCs, while Cdh9 deletion only affected the ON response. These results show that, in addition to acting as an adhesive code for BC axon terminal lamination, these cadherins also regulate the formation of functional synapses, though whether this is direct or indirect remains to be determined.

Taken together, these studies showcase the ability of select molecules to direct the appropriate laminar targeting, stratification, and morphogenesis of SAC, DSGC and BC processes during development of DS circuit function.

SYNAPSE FORMATION AND EARLY NEURAL ACTIVITY PRIOR TO EYE OPENING

SAC dendritic morphogenesis is closely linked to synapse formation and maintenance. Underscoring this relationship, Pcdhγs regulate synapse formation by SACs in addition to their roles in regulating process self-avoidance and self/non-self-discrimination (Kostadinov & Sanes, 2015). SAC-specific deletion of all Pcdhγ isoforms causes SACs to form autapses and abrogates the developmental pruning of synapses that form between closely spaced (< 100 μm) SACs. Replacement of the 22 Pcdhγ isoforms with a single Pcdhγ isoform restores dendrite process self-avoidance but leads to defects in self/non-self-discrimination, resulting in fewer synaptic connections between the distal dendrites of neighboring SACs but no difference in GABAergic synapses onto ventral-ON-OFF DSGCs. Interestingly, the deletion of all Pcdhγs, or the expression of a single isoform, both diminish the DS tuning of ventral-ON-OFF DSGCs. This shows that defects in SAC dendrite morphogenesis alter SAC–SAC connectivity and that the refinement of SAC-SAC connections contributes to the sharpening of DS tuning.

Prior to eye opening, SAC-SAC connectivity drives early stage (between P0 and P11) ‘retinal waves,’ spontaneous bursts of action potentials and calcium transients that propagate across the retina during the period of RGC dendritic refinement and retinorecipient targeting (Firth et al., 2005, Xu et al., 2016). Excitatory acetylcholine released by SACs activates nicotinic acetylcholine receptors (nAChRs) in RGCs, generating patterned activity in the retina, superior colliculus (SC), and visual cortex (Ackman et al., 2012). SAC-specific deletion of β2-nAChRs revealed that mutual excitation among SACs is critical for the propagation of cholinergic retinal waves (Xu et al., 2016). Abolishing cholinergic retinal waves by deletion of β2-nAChRs disrupts the retinotopic refinement of retinocollicular projections, prevents the segregation of retinogeniculate projections into eye-specific layers of the dorsal lateral geniculate nucleus (dLGN), and delays the refinement of RGC dendritic lamination into separate ON and OFF sublayers in the IPL. Initial studies of DSGC tuning responses following global deletion of β2-nAChRs suggested that cholinergic retinal waves are not required for direction-selective responses in DSGCs, since strongly-tuned DSGCs could still be identified (Elstrott et al., 2008). However, a additional analyses reveal that deletion of β2-nAChRs leads to a sharp reduction in the fraction of both ON-OFF and ON DSGCs preferring horizontal motion, and a genetically-labeled population of nasal-ON-OFF DSGCs (TRHR-GFP) was observed to respond to moving stimuli regardless of direction. This reveals a role for cholinergic retinal waves in the development of horizontal DSGC tuning (Tiriac et al 2021), which may be related to the observation that cholinergic retinal waves are biased to propagate in the nasal-temporal axis (Stafford et al., 2009).

In addition to cholinergic waves, DSGCs receive input from cone BCs that drives glutamatergic retinal waves between P11 and P14 (Akrouh & Kerschensteiner, 2013). Loss of Cacna1h, which encodes the T-type voltage-gated calcium channel CaV3.2 found in BC synaptic terminals, disrupts the normal propagation of these retinal waves. It is also sufficient for impairing the segregation of eye-specific inputs within the dLGN, indicating this manipulation can disrupt wave-dependent developmental events critical for normal retinorecipient connectivity patterns in the dLGN. However, Drd4-GFP and TRHR-GFP ON-OFF DSGCs, and also ON-OFF DSGCs identified functionally by multi-electrode array (MEA) recording, showed normal direction selectivity following loss of CaV3.2 (Hamby et al., 2015). Therefore, although glutamatergic retinal waves also show a bias in the nasal-temporal axis and could theoretically provide directional information to developing DSGCs (Elstrott and Feller, 2010), they are not required for proper DS circuit wiring.

RETINORECIPIENT TARGETING OF DIRECTION-SELECTIVE GANGLION CELL AXONS

As connectivity develops in the retina, DSGC axons target distinct retinorecipient nuclei in the brain in order to convey directional motion information to CNS visual centers. RGC axons make several choices upon exiting the optic nerve in order to project to their correct central targets (Zhang et al., 2017). In brief, axons in the optic nerve must choose whether to cross the optic chiasm, which the vast majority do, and are then presented serially with many potential exit points from the optic nerve. ON and ON-OFF DSGCs project to largely distinct sets of targets. ON-OFF DSGCs have two primary targets: the dLGN, the thalamic nucleus responsible for relaying visual information to primary visual cortex (V1; Chalupa and Werner, 2003), and the SC, a midbrain nucleus responsible for multisensory integration that is crucial for mediating saccades and stimulus-guided behaviors, including those guided by vision (Cang et al., 2018). While many ON-OFF DSGC subtypes project to both structures, their projection patterns within the dLGN and SC differ. The projections of nasal-ON-OFF DSGCs (Drd4-GFP and TRHR-GFP) occupy a thin, superficial, lamina within the dLGN, distinct from the deeper lamina occupied by axons of transient OFF-alpha RGCs. Within the SC, ON-OFF DSGCs elaborate their axonal arbors within the most superficial layer, with Drd4-GFP axonal projections evenly distributed while TRHR-GFP and BD-RGCs form more clustered projections (Rivlin-Etzion et al., 2011). In addition to these targets, TRHR-GFP cells send projections into ventral LGN (vLGN) and the zona incerta, a region immediately adjacent to the optic tract and ventral to the LGN. These differences suggest that Drd4-GFP and TRHR-GFP cells represent distinct DSGC subtypes despite sharing a preference for nasal/posterior motion, a notion supported by TRHR-GFP cells showing a broader tuning curve and longer ON response duration (Rivlin-Etzion et al., 2011; Kay et al., 2011). The molecular cues that guide these select DSGC axon innervation patterns in the dLGN and the SC remain to be identified.

The main retinorecipient targets for ON DSGCs comprise the AOS and include the medial terminal nucleus (MTN), the nucleus of the optic tract (NOT), and the dorsal terminal nucleus (DTN). Each of these AOS nuclei are easily distinguished and well-separated from each other and from ON-OFF DSGC targets. The MTN and NOT/DTN are responsible for mediating the vertical and horizontal components, respectively, of the OKR (Simpson, 1984). This division is reflected in the ON DSGC subtypes that project to each target. The MTN receives input from ON DSGCs with preferences for upward or downward motion; those preferring upward motion are labeled in the Spig1-GFP mouse line (Yonehara et al., 2009). The NOT and DTN, in contrast, receive input from forward-preferring ON DSGCs and also a class of ON-OFF DSGCs tuned to slow forward motion; these DSGCs, and also vertical-preferring ON DSGCs, are labeled in the Hoxd10-GFP line (Dhande et al., 2013), suggesting shared properties among these subtypes relating to AOS-specific functions.

Experiments exploring the molecular mechanisms driving RGC axon targeting have informed our understanding of retinorecipient targeting in the AOS. Although previous work explores how RGC axons form retinotopic maps within various targets, the following experiments are among the few that describe how DSGC axons in particular innervate and stabilize their terminals within a target region, and more generally how RGC subtypes target distinct retinorecipient regions as they extend along the optic tract. Forward-ON DSGC axons express the cell adhesion molecule contactin-4 (Cntn4), which binds to amyloid precursor protein (APP) within the NOT. This interaction is critical both for innervation and normal forward-ON DSGC axon arborization within the NOT. Proper innervation of the NOT is required in order to restrict other DSGCs from ectopically targeting the NOT. Normally, nasal-ON-OFF DSGCs (Drd4-GFP) transiently innervate the NOT at P8, but these projections are pruned by P20. However, In Cntn4 mice this innervation is retained (Osterhout et al., 2015). Targeting of ON DSGCs tuned to slow vertical motion in the MTN is mediated by the transmembrane protein semaphorin 6A (Sema6A), which is expressed by up-ON DSGCs and down-ON DSGCs. Sema6A in this context acts as a receptor in ON DSGCs and facilitates MTN axon targeting through interactions with the ectodomains of both plexin A2 (PlexA2) and PlexA4. These large transmembrane proteins are both expressed in the MTN during the time when ON DSGCs innervate it and serve redundant functions as ligands for Sema6A (Sun et al., 2015). Despite these advances in understanding ON DSGC targeting, knowledge of the mechanisms guiding retinorecipient targeting by ON-OFF DSGCs remains limited.

LATER EVENTS IN DIRECTION-SELECTIVE CIRCUIT DEVELOPMENT

ASYMMETRIC STARBURST AMACRINE CELL-DIRECTION-SELECTIVE GANGLION CELL CONNECTIVITY

Once axons and dendrites of DS circuit components align and coalesce in the correct IPL sublaminae, synapse formation and refinement among BCs, SACs, and DSGCs leads to the mature connectivity patterns that underlie DS tuning properties. Models from the Barlow and Hill era (reviewed in Borst & Euler, 2011) presented two possible circuit motifs for generating DS responses; both focus on spatiotemporal delays but differ as to whether they reside in excitation or inhibition. In the Hassenstein-Reichardt model, excitation from one set of inputs is delayed such that during PD stimuli the first set of excited inputs releases glutamate at the same time as the second, but during ND stimuli the two sets do not synchronize glutamate release. The Barlow-Levick model, in contrast, proposes that inhibitory SAC input is shifted laterally and either delayed or sustained; in this model, ND stimuli activate SACs that block activity in the DSGCs next in line to receive excitation, while PD stimuli only elicit inhibition long after the DSGC has fired. Seminal work in the rabbit retina demonstrated that DSGCs receive asymmetric inhibition, that SACs are required for retinal DS, that SACs themselves are DS, and that null-side SACs ultimately drive greater inhibition than those on the preferred side of DSGC dendritic arbors through larger releases of GABA (Fried et al., 2002). Thus, the asymmetric inhibitory wiring of SACs onto DSGCs is essential for motion preference tuning, supporting the Barlow-Levick model.

When, and how, is asymmetric SAC input onto DSGCs formed during development? Two landmark studies established that for both ON-OFF DSGCs and ON DSGCs, asymmetric inhibition from SACs arises during the second postnatal week during a relatively narrow window between ~P6 and P8 (Wei et al., 2011; Yonehara et al., 2011). This asymmetry is independent of both spontaneous activity and visual experience, since direction selectivity arises prior to eye opening (at P13) and pharmacological blockade of activity from P6-P12 does not alter the direction-selectivity of DSGCs. During the first postnatal week, the somas of SACs synaptically connected to DSGCs are equally distributed between the null and preferred side of DSGC dendritic arbors. Further, at P6 the spatial distribution and strength of inhibitory synapses onto both ON-OFF DSGCs and ON DSGCs is also symmetric. However, at P8, SACs located on the null side of these DSGCs elicit greater inhibitory currents than those on the preferred side, and at P14, GABAergic conductances are higher when elicited from depolarization of null side SACs than from the preferred (Wei et al., 2011; Yonehara et al., 2011). Direct support for the conclusions drawn from these observations comes from experiments combining two-photon calcium imaging with serial block-face EM, revealing that SACs on the null side of DSGC dendrites form a greater number of inhibitory synapses onto DSGCs than those on the preferred side (Briggman et al., 2011).

Furthermore, these experiments (Briggman et al., 2011) revealed that asymmetric connectivity is driven by the relationship between SAC dendritic orientation and DSGC PD, rather than by SAC soma location alone. Maximal connectivity occurs when SAC dendrites and the PD of a postsynaptic DSGC are antiparallel; since SAC dendrites extend radially, this gives null-side SACs the greatest level of connectivity and preferred-side SACs the least. When SAC somas are at more intermediate locations along the preferred-null axis, however, these SAC dendrites overlap with the DSGC arbor at a variety of orientations. Though a model driven by soma location predicts that all SAC dendrites form synapses equally, synapse formation is biased towards those in an antiparallel orientation, indicating that this is a key determinant of asymmetric connectivity. Since this analysis defines SAC dendritic orientation as the vector connecting the SAC soma to the synapse, the orientation used does not represent the distal dendrite immediately surrounding the synapse, but rather the proximal dendrite to which it belongs. Experiments performed in Sema6A^−/− SACs underscore the importance of the orientation of the proximal, rather than distal, SAC dendrite in asymmetric SAC-DSGC connectivity (Morrie & Feller, 2018). In these mutants, distal SAC dendrites take on disorganized orientations, while proximal dendrites mostly retain their normal orientations. Although DS responses are weakened in Sema6A mutants, asymmetric SAC-DSGC wiring remains intact, indicating that the correct orientation of proximal SAC dendrites is sufficient to preserve this asymmetric connectivity.

The importance of proximal dendritic orientation supports the idea that asymmetric connectivity between SACs and DSGCs is driven by synaptogenic cues that are selectively trafficked to a specific quadrant of the SAC arbor and that recognize a subtype-specific partner in the DSGC arbor, with these cues being distributed such that antiparallel orientations of SAC proximal dendrites and DSGC PDs are preferred. Rather than requiring detection of the local orientation of each distal dendritic segment and sending out the appropriate molecule, correct distribution of such cues could be achieved by sorting at the soma-dendrite border. The search for the molecular determinants of asymmetric connectivity, therefore, should benefit from focusing on identifying sets of proteins localized to specific quadrants of the SAC arbor and complementary cues, perhaps CAMs, with expression limited to a specific DSGC subtype.

Although work on DS circuits has largely focused on asymmetric SAC inhibition of DSGCs, ON DSGCs also receive asymmetric excitation as well. Excitatory inputs with different kinetic properties, comprising multiple BC subtypes and a glutamatergic AC, are arranged in a slow-to-fast gradient across the diameter of the ON DSGC arbor: BCs with the longest delay between stimulation and glutamate release synapse onto the distal portion of the preferred-side arbor; the slow glutamatergic AC is biased towards the preferred-side arbor; and BCs with the shortest delay synapse onto the distal portion of the null-side arbor. During PD stimulation, slow inputs are stimulated first and delay their release of glutamate until fast inputs on the other side of the arbor also receive stimulation and rapidly release glutamate, providing synchronous neurotransmitter release onto ON DSGCs for a large peak conductance. In contrast, stimuli in the ND first stimulate fast inputs, which rapidly release glutamate and thus do not synchronize neurotransmitter release with the slow inputs, leading to a smaller peak conductance. This may underlie the preference of ON DSGCs for slower stimulus speeds than ON-OFF DSGCs; a stimulus moving too quickly will trigger the fast inputs before the slow inputs begin releasing glutamate, preventing their synchronization (Matsumoto et al., 2019). This also provides support for the Hassenstein-Reichardt model describing a portion of the DS responses of ON DSGCs. Such an arrangement of excitatory inputs implies the existence of a developmental mechanism guiding different subtypes of excitatory neurons to synapse onto distinct portions of the ON DSGC arbor. One possible scenario involves each excitatory subtype expressing a specific CAM, with each CAM recognizing a specific binding partner in a restricted portion of the ON DSGC arbor. In turn, each ON DSGC subtype could arrange these binding partners across its dendritic arbor so that the different excitatory inputs are lined up in a slow-to-fast gradient along the preferred-null axis for that subtype.

DIRECTION-SELECTIVE TUNING REFINEMENT

While the PD of a given DSGC appears to be developmentally programmed, the strength of this preference is refined over time. Multi-electrode array (MEA) recordings of retinas just before eye opening revealed DS tuning in both ON-OFF DSGCs and ON DSGCs, showing that DS circuit functionality precedes visual experience. Following eye opening, DSGCs exhibit a narrowing of tuning width and, for ON-OFF DSGCs, an increase in their direction selectivity index (DSI), a measure of how well DS responses of a given DSGC are correlated with directional motion of the stimulus (Chen et al., 2014).

Initial two-photon Ca²⁺ imaging studies suggested that at ~P14 DSGCs do not cluster their PDs around the classical directions, but instead have their PDs broadly distributed in directional space. These studies also suggested that refinement of DSGC PDs requires visual experience, since dark-reared mice failed to cluster their PDs around the classical directions (Bos et al., 2016). These studies, however, were performed prior to the finding that the axes of DSGC preferences vary across the retina to align with local patterns of optic flow (Sabbah et al., 2017). More recent work that includes assessment of DSGC retinal location finds that the mature directional alignment of DSGC PDs is present at eye opening, indicating that visual experience is not required for this alignment. These experiments also reveal a modest but significant increase in DSI for vertical-ON DSGCs following eye opening (Tiriac et al., 2021). Furthermore, these studies corroborate the existence of nasal-ON DSGCs. It is interesting that the previous Ca²⁺ imaging experiments (Bos et al., 2016) found no nasal-ON DSGCs despite using the same DSI cutoff that was subsequently used to reveal their existence (Sabbah et al., 2017). Why might this be? Perhaps this arises from the fact that these two previous studies used different stimulus speeds: one at 300 μm/s (Sabbah et al., 2017) whereas the other at 500 μm/s (Bos et al., 2016); this most recent study (Tiriac et al, 2021) also uses a slower stimulus speed (250 μm/s). This may indicate that nasal-ON DSGCs possess a narrower range of preferred speeds than other ON DSGCs, in addition to being more weakly direction-selective. Examining circuitry and transcriptomic differences among nasal-ON DSGCs and other ON DSGCs may prove illuminating with respect to defining the developmental mechanisms that set the direction selectivity and speed preferences of any DSGC subtype.

CONNECTIVITY BETWEEN RETINAL DIRECTION-SELECTIVE CIRCUITRY AND HIGHER-ORDER VISUAL AREAS

How do DSGCs relay directional information to visual cortex? Retrograde tracing experiments reveal a retino-geniculate-cortical circuit that di-synaptically links ON-OFF DSGCs to the superficial, but not deep, visual neocortex in area V1. Nearly all RGCs di-synaptically connected to layer 2/3 (L2/3) V1 neurons were identified as DSGCs, whereas no RGCs connected to layer 4 (L4) were DSGCs (Cruz-Martin et al., 2014). From V1, visual information flows to a number of higher-order visual areas, including the rostrolateral (RL) area, whose L2/3 neurons are DS. When retinal direction selectivity is disrupted, the stimulus preferences of DS neurons within area RL shift; additionally, the fraction of RL-projecting L2/3 neurons in V1 that are DS is reduced (Rasmussen et al., 2020). This suggests that the tuning properties of neurons in area RL depend on input from the DSGC retino-geniculate-cortical circuit being relayed beyond V1. Together, these experiments outline a role for DSGCs in transmitting directional information to multiple regions of visual cortex.

The elaboration of these circuits requires an extensive network of developmental regulation: DSGCs must target the correct cells of the dLGN, which in turn must choose to connect with L2/3 and not with L4; L2/3 cells in V1 must also send their axons to the correct higher-order areas and choose the correct laminar targets within them. The principles of laminar organization and exquisite partner matching illustrated by the retina are thus reflected throughout the CNS, highlighting the importance of this organization and providing insight into development across the nervous system. The mechanisms guiding the wiring of these circuits are largely unknown. It is likely that these processes involve molecular recognition cues that mediate partner matching, classical axon guidance cues, and also activity-dependent regulation in a manner similar to that observed in the retina.

VISUAL BEHAVIOR AND DISEASE

BEHAVIORAL ASSAYS FOR DIRECTION-SELECTIVE CIRCUIT FUNCTION

A direct behavioral output of retinal DS circuit function in the accessory optic system is the optokinetic reflex (OKR). The OKR is a conserved visual reflex across species – from zebrafish to mice to humans – that generates compensatory eye movements to stabilize images on the retina during slow object or head motion. In head-fixed mice, the OKR can be quantified using eye tracking software to detect eye tracking movements (ETMs) as moving gratings are presented in either horizontal (nasal/temporal) or vertical (up/down) directions at a specific contrast level, spatial frequency, and angular velocity (Stahl, 2004; Cahill & Nathans, 2008; Kodama & du Lac, 2016) (Fig. 3a). Selective ablation of SACs abolishes the OKR (Yoshida et al., 2001), highlighting not only the critical role played by SACs in mediating this visual behavior but also the ability of OKR measurements to reflect defects in DS circuit function and development. DS neurons in the pretectum of zebrafish also generate an OKR, which can similarly be elicited by placing larval zebrafish in a panoramic arena with continuously rotating gratings (Wu et al., 2020). OKR measurements can be used as an assay to screen gene candidates for involvement in DS circuit function and development, including those implicated in human disorders, such as infantile nystagmus, that disrupt the OKR (Yonehara et al., 2016).

Fig. 3. Selective role for Frmd7 in development of horizontal DS circuitry.

(A) Mice with Frmd7 LOF exhibit a loss of horizontal, but not vertical, OKR. Shown are eye tracking movements (ETMs) in response to slow motion stimuli in the indicated directions (yellow, purple, green, and blue arrows). (B) ON DSGCs tuned to prefer horizontal motion respond equally strongly to all directional stimuli following Frmd7 LOF, whereas those tuned to prefer vertical motion are unaffected. (C) SAC morphology is normal in Frmd7 mutants. (D) FRMD7 protein is distributed symmetrically across the SAC dendritic arbor between P3 and P7. (E) Two Frmd7 mutations identified in congenital nystagmus patients impair the ability of FRMD7 to associate with GABRA2, a component of inhibitory neurotransmitter receptors. D = dorsal; N = nasal; V = ventral; T.= temporal.

FRMD7 AND INFANTILE NYSTAGMUS

One gene whose association with human disorders spurred further study is FERMDomain Containing 7 (Frmd7). Approximately 70% of human patients with idiopathic congenital nystagmus have mutations in Frmd7 (Tarpey et al., 2006), which encodes a protein containing an N-terminal FERM domain and a C-terminus that shows no homology with other proteins (Yonehara et al., 2016). These patients show specific deficits in the horizontal, but not vertical, OKR, suggesting a role for Frmd7 in the development of the DS circuit for detecting horizontal motion (Thomas et al., 2008). A landmark study by Yonehara and colleagues (2016) investigated the role of Frmd7 in wiring the DS circuit in the mouse. Frmd7 mutant mice show deficits in OKR selectively along the horizontal axis (Fig. 3a and 3b), matching the aberrant OKR observed in human patients and thus revealing a remarkable conserved role for Frmd7 across species. In Frmd7 mutants, the population of RGCs labeled in the Hoxd10-GFP line, which labels all AOS-projecting DSGCs, contains a normal number of vertical-ON DSGCs but also includes untuned RGCs instead of forward-ON-OFF and ON DSGCs, suggesting that DSGCs programmed to prefer horizontal motion are rendered non-DS by Frmd7 LOF.

How might Frmd7 influence development of the DS circuit? Frmd7 mutant SACs have normal morphology (Yonehara et al., 2016, Fig. 3c), suggesting that established roles in neuroblastoma cells for FRMD7 in neurite outgrowth through associations with CASK (Watkins et al., 2013) and Rac1 (Pu et al., 2013) are not relevant in the context of DS tuning. Localization of FRMD7 protein is symmetric in developing SACs (Fig. 3d), suggesting that FRMD7 is not selectively trafficked to specific dendrites involved in processing horizontal motion (Yonehara et al., 2016). However, this study did not assess FRMD7 localization past P7; this leaves open the possibility that FRMD7 could become selectively trafficked at P8, when asymmetric inhibition emerges (Yonehara et al., 2011), or at later developmental timepoints.

A recent study reveals an association between FRMD7 and the a2 subunit of GABARs, which are the neurotransmitter receptors at inhibitory synapses. This work describes three human patients with novel Frmd7 mutations. A biochemical assay for protein-protein interactions found that, when expressed in heterologous cells, WT human FRMD7 binds GABRA2 in mouse retinal tissue but the mutant forms found in patients do not (Fig. 3e). The C. elegans homologue of FRMD7, FRM-3, binds the C. elegans homologue of GABRA2, UNC-49B, at the same TM3-TM4 loop at which FRMD7 binds GABRA2. A locomotion phenotype caused by loss of frm-3 can be rescued using human Frmd7, though only in conjunction with a variant of UNC-49B that also includes the human TM3-TM4 loop added after the endogenous loop; variants based on mutations in patients could not. These variants also showed an impaired ability to anchor GABARs at fixed points within the plasma membrane. These results establish a context in which FRMD7 influences nervous system function through association with GABRA2 (Jiang et al., 2020). This line of investigation was inspired by the finding that loss of Gabra2 eliminates DS responses for all directional stimuli (Auferkorte et al 2012); therefore, FRMD7 can function as a binding partner of a protein known to be critical for retinal DS tuning. Altogether, Frmd7 illustrates how insights from human disease can inform basic research of the visual system and, in turn, how work in simpler model systems can be translated to more complex organisms with the potential for aiding human health.

CONCLUSION

The assembly of DS circuits reflects the precise coordination of gene expression, neural activity, and visual experience to generate highly specific patterns of connectivity over the course of retinal development. Recent advances in single-cell transcriptomic profiling have resolved some of the transcriptional differences among subtypes of RGCs, BCs, and SACs that constitute DS circuits. However, it remains unclear which factors are functionally linked to distinct subtypes of DSGCs. Transcriptomic profiling of genetically labeled DSGCs holds the promise of uncovering gene expression differences among subtypes with different motion preferences, and whether certain genes underlie their early differentiation, specification, and their distinct tuning properties.

Early transcriptional differences specify the fate of ON and OFF SACs, and these distinct “ON” and “OFF” gene expression programs ultimately mediate differential patterns of migration and laminar dendritic stratification in the IPL. Additional molecules likely cooperate with Sema6A/PlexA2 to target and restrict the dendrites of OFF and ON SACs to S2 and S4 of the IPL, respectively, since many SACs in these and other mutants still laminate appropriately. Further assessment of the genes differentially expressed in OFF and ON SACs might lend insight into additional cues that govern their specific patterns of lamination and dendritic elaboration. And the select expression of a variety of CAMs, including classical cadherins, in subsets of ON-OFF DSGCs may facilitate the establishment of directional tuning through regulation of inhibitory synapse distribution.

An outstanding question in DS circuit development is how asymmetric inhibitory connectivity arises between SACs and DSGCs, and why, at least in mice, it occurs during a specific postnatal developmental window. Again, transcriptomic profiling of SACs and DSGCs spanning these developmental time points may reveal genes with the appropriate temporal and spatial expression patterns to mediate this developmental switch in connectivity. An additional question is what role, if any, is played by DSGC dendritic morphology in the DS circuit and what programs are responsible for shaping it. The asymmetric morphology of some DSGC subtype dendritic arbors, even if these arbors are not aligned with the cell’s PD, suggests that dendritic asymmetry is important for circuit functionality. Yet, at least in Hb9-GFP ON-OFF DSGCs, there is no significant correlation between dendritic asymmetry and tuning strength (El-Quessny et al., 2020). Why, then, do some DSGCs orient their dendrites in this fashion, and what developmental programs give rise to this orientation?

We have presented in this review the blueprint that organizes DS circuit development, from early events that establish cell type diversity, process morphology, and lamination, to later events that govern sophisticated patterns of connectivity and behavior. Advances in genetics, molecular profiling, and physiology, have allowed for characterization of DS circuit components, showing that DS circuits are ideal model systems for understanding neural circuit function at the level of individual genes, cells, and behavior, throughout the CNS and across species. Further, insights into DS circuit function revealed through analyses of visual system behavior in humans holds great promise for understanding and ameliorating visual system disorders.

• Direction selective (DS) circuits show strong phylogenetic conservation

• Distinct transcriptional programs differentiate DS retinal neuronal subtypes

• Cell adhesion molecules and neuronal guidance cues organize the spatial distribution of retinal DS neuronal subtype dendrites and synapses

• Analysis of DS tuning responses reveals the roles of neural activity and visual experience in DS circuit assembly

• Human eye movement disorders provide important clues for understanding the molecular basis of DS tuning