Conservation of retinal direction-selective circuits across vertebrate species

Abstract

Retinal circuits extract visual features such as color, contrast, and motion. Direction-selective (DS) circuits provide a well-studied model for understanding how the brain builds feature-selective circuits. Recent work highlights remarkable evolutionary conservation of DS circuit components. Across species, these circuits share common transcriptional programs that diversify DS cell types and molecular cues that help establish precise synaptic organization. Directional tuning emerges during development without visual input through gene regulatory networks, extracellular signaling, and activity-dependent mechanisms. Disruption of these processes can impair motion processing and contribute to visual and oculomotor deficits. Therefore, DS circuits provide a powerful system for dissecting the molecular programs involved in neuronal subtype diversification, neurite guidance, and synaptic development. In this review, we summarize recent advances in our understanding of DS circuits, their development, cell type composition, and tuning across species.

“Life moves pretty fast. If you don’t stop and look around once in a while, you could miss it.”

-Ferris Bueller

Introduction to Direction-Selective Circuits

Understanding how the retina encodes feature information such as directional motion has been a central question in visual neuroscience. Direction-selective (DS) circuits in the retina depend upon precisely orchestrated interactions between multiple neuronal classes. Motion detection begins with cone photoreceptors, which pass visual information to select ON and OFF subtypes of bipolar cells (BCs) that serve as key input channels to direction-selective circuits [1–3]. These BCs provide excitatory inputs to both starburst amacrine cells (SACs) and direction-selective retinal ganglion cells (DSGCs), the latter of which relay processed motion signals to the brain [4,5]. SACs, themselves divided into ON and OFF subtypes stratifying in distinct retinal layers, are critical sculptors of directional tuning. SACs asymmetrically inhibit DSGCs, suppressing responses to motion in the null direction while permitting strong responses to preferred-direction motion [6,7].

DS circuits encode directional information through two types of circuits: ON-OFF and ON (Fig 1). These circuits are defined by their distinct cellular compositions and response properties. ON-OFF and ON DS circuits contain defining ganglion cell types, ON-OFF direction-selective ganglion cells (ooDSGCs) and ON direction-selective ganglion cells (oDSGCs), respectively [7,8]. ON-OFF DS circuits encode fast motion and respond to both increases and decreases in luminance. ON DS circuits, part of the accessory optic system (AOS), encode slow motion and mediate the optokinetic reflex (OKR). This evolutionarily conserved reflex is essential for image stabilization [9,10]. Together, these parallel ON-OFF and ON DS circuits enable the retina to encode a broad range of motion features, from detection of fast object motion to slow global image motion required for gaze stabilization.

Fig: 1. ON-OFF and ON Direction-Selective Retinal Circuits.

Diagram illustrating the synaptic organization of the direction-selective (DS) circuits in the mammalian retina, shown in cross-section from outer nuclear layer (ONL) to ganglion cell layer (GCL). Left: The ON-OFF DS circuit, which detects fast motion, comprises cone photoreceptors, horizontal cells (HCs), ON and OFF bipolar cells (BCs), and bistratified ON and OFF starburst amacrine cells (SACs) that provide asymmetric inhibitory input onto ON-OFF direction-selective ganglion cells (ooDSGCs), whose dendrites stratify in both the ON and OFF sublaminae of the inner plexiform layer (IPL). Right: The ON DS circuit, which detects slow motion and drives the optokinetic reflex (OKR), is comprised of cone, HCs, ON BCs, and ON SACs restricted to the ON sublamina of the IPL, which provide input onto ON direction-selective ganglion cells (oDSGCs). OPL, outer plexiform layer; INL, inner nuclear layer.

Comparisons of direction selective circuits across vertebrates

Since DSGCs were first described in the rabbit retina in 1963 [8], DS circuits have been primarily studied in rodents (reviewed in [11]). Transcriptomic profiling using single cell RNA sequencing (scRNA-seq) is providing an evolutionary perspective on these circuits across species [12–15]. Whereas invertebrate species have well-studied DS circuits [16], they have major differences in cellular composition, activity, and circuit structure compared to vertebrates. We focus on vertebrate DS circuits in this review.

SACs:

SACs appear to be evolutionary ancient neurons. As the only cholinergic neurons in the retina, SACs have been identified in sea lamprey, indicating that DS circuit cells emerged over 500 million years ago, early in vertebrate evolution [13,14]. Molecular segregation of ON and OFF SAC subtypes occurs across species [14]. Complementing transcriptomics, anatomical approaches remain valuable where molecular tools are limited. For example, although primates show a clear transcriptional separation of ON and OFF SACs [14], these populations remain anatomically segregated into distinct ON and OFF laminae. In humans, SACs are generated embryonically and their cell bodies assume laminar positions in the INL and GCL, similar to ON and OFF SACs in rodents [17]. Querying human fetal retinal transcriptomic data [18] for established ON and OFF SAC markers reveals segregated expression within the SAC cluster, suggesting that a transient transcriptional distinction may diversify SAC subtypes, warranting further investigation.

Zebrafish display surprising differences. Whereas asymmetric inhibition from strongly-tuned SACs is vital for retinal DS in mammalian retinas, ON and OFF SACs in zebrafish do not appear to be strongly motion selective [19], suggesting that different mechanisms establish DS circuits in zebrafish and posing important questions about how DS computation is implemented across species.

DSGCs:

DSGC-associated gene expression programs are broadly conserved, and experimentally-validated DS responses in RGCs have been reported across a wide range of vertebrates including mouse [11], rabbit [8], zebrafish [20], cat [21], bird [22,23], turtle [24], and frog [25], and most recently, primates [26–28].

Early efforts to identify DSGCs in non-human primates were largely unsuccessful, likely due to the significant differences between the relative proportions of RGC subtypes in rodents and primates [29–31,26]. Morphological and transcriptomic studies suggest 12–19 distinct subtypes of RGCs in the primate retina [12,29,32], differing dramatically from the >40 types in the mouse retina [33]. In the primate retina, ON and OFF midget RGCs (MGCs) account for ~80% of RGCs, while ON and OFF parasols make up ~10%, leaving less than 10% of RGCs to constitute remaining subtypes. Moreover, it was thought that rodents lack clear counterparts for the midget ganglion cells that make up ~80% of RGCs in the primate retina. This led to the hypothesis that primates may rely less on feature detectors such as DSGCs and instead utilize MGCs to project unprocessed visual information to the cortex for more complex processing [26,30,31]. This assumption has now been challenged by single-cell transcriptomic studies across 17 vertebrate species that identified non-primate MGC orthologs [15]. In mice, these orthologs together comprise less than 2% of all RGCs [15]. This suggests an expansion of this MGC population in the primate retina as opposed to a primate-specific evolutionary innovation. What then is the selection pressure that drove the expansion of MGCs, and did it come at a cost of the retinal DS circuits?

Transcriptomic profiling, EM reconstruction, and electrophysiological analyses revealed that primates have retinal DS circuits [27,26,28]. The primate DS circuit begins with the photoreceptors and horizontal cells that provide center-surround information to midget and diffuse types of BCs [26,27]. BCs transmit the center-surround via excitatory signaling onto SACs and DSGCs [26]. Directionally tuned SACs provide asymmetric inhibition to DSGCs, contributing to the generation of the preferred and null directions. ooDSGCs are also coupled via gap junctions to a DS polyaxonal A1 amacrine cell [26]. This A1 cell is directionally tuned and has multiple axons extending several millimeters long, allowing for long range directional information inhibitory signaling [26]. DSGCs have not been fully characterized in the human retina, however they may express the putative markers BNC2 and ROBO1 [28,34].

The conservation of SACs and DSGCs, spanning over 500 million years of vertebrate evolution, argues that direction selectivity represents a fundamental and ancient computation of the vertebrate retina. Yet, striking differences remain in the proportion of DS cells, in the strength of SAC tuning, and in the possible existence of species-specific circuit elaborations. Whether the DS computation itself is conserved, or whether convergent circuit architectures have independently arrived at similar outputs through distinct mechanisms, is a critical open question for comparative studies.

Transcriptional programs differentiate DS cell subtypes

A growing body of work has identified the developmental mechanisms that regulate cell fate specification of DS neurons, revealing both cell-type-specific programs and varying degrees of conservation across species. Understanding which regulators are shared across vertebrates and which have been rewired is essential for interpreting the functional persistence of DS circuits across species.

SACs:

During development, SACs are born in the neuroblastic layer and migrate to either the inner nuclear layer (INL), adopting the OFF subtype fate, or the ganglion cell layer (GCL), adopting the ON fate. In mice, expression of the transcription factor Fezf1 specifies the ON SAC fate [35]. Importantly, in zebrafish, Fezf1 is specifically expressed in Sox2⁺ SACs, and genetic disruption of Fezf1 results in the repositioning of Sox2⁺ SAC somata from the GCL to the INL, directly paralleling its role in mice and supporting its functional conservation [19]. Fezf genes are also expressed in a subset of SACs in several species including the lamprey, chicken, amphibians, and marmosets [35,13,14], suggesting phylogenetic conservation, but their functional roles in these species have yet to be examined.

Beyond the Fezf1-dependent ON/OFF fate switch, the transcription factor Sox2 has also emerged as a regulator of SAC development, though with intriguing species-specific differences. In the mouse retina, Sox2 deletion results in altered SAC soma distribution and produces bistratified dendritic morphology, pointing to a role in subtype positioning and dendritic organization [36,37]. In zebrafish, however, sox2 disruption results in a decrease in the overall numbers of Sox2+ SACs, indicating a role in SAC specification [19]. Thus, while Sox2 is a conserved molecular feature of SACs across vertebrates, its function within the SAC developmental program has diverged.

DSGCs:

oDSGC specification shares some molecular bases between mice and primates. BNC2 marks oDSGCs in both primates and mice [28,38]. In the human retina, BNC2⁺ RGCs show a pronounced perivascular distribution closely mirroring the perivascular pattern of mouse oDSGCs [28,32,39]. Thus, both the transcriptional identity and spatial relationship of oDSGCs to the vasculature appear to be conserved from mouse to human [39].

DSGC soma position adjacent to vasculature may be neuroprotective. Perivascular ooDSGCs express the mechanosensitive channel Piezo2. Perturbing Piezo2 alters vascular patterning, suggesting reciprocal interactions between DSGCs and the vasculature [40]. In a model of ocular hypertension, perivascular oDSGCs survived at higher rates than non-perivascular ooDSGCs, indicating that vascular proximity can be neuroprotective for some DSGCs [39]. Given the vasculature’s key roles in shaping the extracellular environment and supporting neuronal survival, further studies should explore how interactions between developing DSGCs and retinal blood vessels influence DS circuit maturation and function.

In contrast to BNC2, other molecular determinants do not appear to be shared between vertebrates. Though the transcription factor Satb2 is expressed in ooDSGCs in mice and rabbits, SATB2 cells are not direction-selective in primates [41].

Although not yet studied in primates, the transcription factor Tbx5 is expressed in upwardly-tuned oDSGCs and is required for their development as well as vertical OKR in rodents [38]. Complementing this, the transcription factor Gfi1 is expressed in downwardly-tuned oDSGCs, and is essential for their generation, dendritic elaboration, and function [42]. Together, Tbx5 and Gfi1 partition the two vertical-axis oDSGC subtypes. Whether they also play comparable roles in oDSGC specification in other species is an important open question.

Together, transcription factor-controlled gene regulatory programs specify SAC and DSGC subtype identity, with both conserved and species-specific regulators contributing to the diversification of direction-selective circuits. For many regulators, conservation of gene expression is present across species, however functional conservation has only been rigorously demonstrated for a select few (Fig 2). Notably, several molecular markers of DS cells have not been functionally validated, such as Bnc2 [28], FSTL/Spig1 [43], Fibcd1 [38], Ptprk [38], Pcdh9 [44], CART [45], Col25a1 [45], Cdh6 [45], Satb2 [41]. The challenge now lies in determining whether this conservation of gene expression reflects conservation of developmental mechanisms or simple evolutionary recycling of transcription factors into new circuit contexts. Beyond fate specification, the assembly of functional DS circuits requires precise spatial organization of neurons and their processes, which is achieved through extracellular mechanisms.

Fig 2. Validated Regulators of Direction-Selective Circuit Development.

Table includes molecules with loss- or gain-of-function evidence demonstrating a role in the development of direction-selective retinal circuits, encompassing transcriptional regulators, cell adhesion molecules, axon guidance cues, and synaptic organizers. Entries are organized by cell type and developmental process. Genes and proteins identified solely as molecular markers of DS circuit cell types without functional validation are not included here but are discussed in the text.

Cell adhesion molecules, guidance cues, and their receptors organize DS circuits

Extracellular factors including cell adhesion molecules (CAMs), extracellular matrix composition, as well as guidance cues and their receptors play central roles in assembling retinal DS circuits. Precise temporal and spatial coordination of these molecules establish laminar organization within the inner plexiform layer (IPL) and synaptic specificity between BCs, SACs, and DSGCs. A central emerging theme from cross-species retinal profiling is that many molecules used to build DS circuits are conserved across vertebrates. However, because most mechanistic work has been done in mice (Fig 2), whether the same functional roles are conserved across species remains a critical open question.

The segregation of neurites into ON and OFF layers of the IPL is a conserved feature across vertebrates and is coordinated by both cell adhesion molecules and axon guidance cues. In mice, the repulsive cue Semaphorin 6A (Sema6A) and its receptor Plexin A2 (PlexA2) regulate SAC morphology and ensure proper lamination of SAC dendrites within the IPL, thus delineating the ON and OFF circuits [46,47]. RGCs are the primary source of Sema6A to drive laminar segregation of SACs [47]. In contrast, SAC-autonomous Sema6A determines distal dendrite self-avoidance revealing a cell-type-specific division of labor within the same molecular pathway [47]. Sema6A/PlexA2 signaling also regulates retinal patterning in the zebrafish retina [48], supporting a conserved role for this pathway in visual circuit assembly. However, it is unclear whether SAC lamination specifically is regulated in a similar manner as in mice.

Megf10 and Megf11 are cell surface proteins selectively expressed by SACs that are critical for establishing mosaic spacing of SACs [49,50]. Megf10 acts as both ligand and receptor on the SAC membrane to mediate homotypic repulsion [49,50]. SAC dendritic arbors physically contact and repel neighboring homotypic somata during a defined developmental window (E16-P0) [51].

Adhesive interactions mediated by cadherins further refine laminar targeting and synaptic specificity. In mice, the Type II cadherins Cdh8 and Cdh9 are expressed in OFF BC2s and ON BC5s, respectively. Loss of Cdh8 disrupts the laminar targeting of BC2 axons to OFF layers of the IPL, which reduces directional tuning of ooDSGCs to OFF stimuli. Similarly, loss of Cdh9 impairs laminar targeting of BC5 axons to IPL ON layers, leading to reduced directional tuning of ooDSGCs to ON stimuli [1]. Beyond BC inputs, cadherins also govern the dendritic organization of the DSGCs themselves. Cdh6, Cdh9, and Cdh10 pattern the dendrites of ventral motion-preferring ooDSGCs, and loss of these cadherins abolishes their directional tuning [1], while Cdh7 and Cdh18 regulate lamination of nasal motion-preferring ooDSGCs.

In addition to cues that recruit DS neuron processes to their correct sublayers, a complementary repulsive system prevents DSGC arbors from straying into adjacent non-DS territory. In mice, FLRT2 is expressed selectively by DS circuit neurons and binds to UNC5C/D on non-DS neurons in adjacent sublayers, eliminating ectopic dendritic branches and preventing synapse formation between inappropriate partners [52].

Following laminar targeting events, synapse elaboration occurs. In mice, the guidance cue Slit2 and its Robo1 receptor refine ON SAC-oDSGC connections. Slit2 expression is enriched in ON SACs but not OFF SACs, and the transmembrane Slit2 receptor Robo1 is expressed predominantly in oDSGCs. Robo1 or Slit2 loss-of-function results in an increased OKR gain and alterations in the excitatory/inhibitory balance involving oDSGCs [34]. In the astray/robo2 zebrafish mutant, DSGCs maintain normal directional tuning despite mistargeted tectal projections [53], suggesting that Robo2 does not share Robo1’s role in shaping the intrinsic directional tuning properties of DSGCs. ROBO1 is enriched in primate and murine RGCs that are likely oDSGCs, further supporting a role for Robo1 signaling in DS circuit organization [34,18,15]. In humans, ROBO1 missense mutations cause a form of congenital nystagmus that affects vertical eye tracking [54], although a direct link to retinal DS circuit dysfunction remains to be established. Further work is required to determine the extent to which these mechanisms are conserved in primates.

Collectively, extracellular cues such as cell adhesion molecules, axon guidance cues, and their receptors are essential for proper laminar organization of DS connections and synapse elaboration (Fig 2). Much of this insight comes from the mouse retina, and a priority is to extend mechanistic loss-of-function analysis beyond the mouse, especially into zebrafish, where genetic tools are powerful, and to primates, where clinical relevance is highest. While these molecular signals establish the structural framework of DS circuits, directional tuning ultimately emerges from how synaptic inputs are organized and functionally integrated to compute motion.

Directional tuning during development

Directional tuning in DSGCs arise during development as a result of careful coordination between BCs, SACs, and DSGCs. Asymmetric wiring of SACs onto DSGCs is essential for the generation of preferred directional tuning for DSGCs [55,56]. In rodents, SAC/DSGC wiring emerges by ~postnatal day 6 (P6) and becomes asymmetric by ~P8 [55,57,58]. Determining what regulates asymmetrical SAC inhibition onto DSGCs is an outstanding goal in the field.

Spontaneous retinal activity prior to eye opening plays a selective and conserved role in refining horizontal directional tuning in mammals. Disrupting cholinergic waves from SACs by knocking out β2-containing receptors selectively diminishes asymmetric inhibition of horizontally-preferring ooDSGCs and reduces horizontal OKR responses, while vertical tuning remains unaffected [59,60]. The selectivity of this effect for the horizontal axis, across both the ooDSGC and oDSGC pathways, points to a common activity-dependent mechanism that differentially shapes cardinal axis tuning during development.

Identification of asymmetric wiring regulators between SACs and DSGCs has remained elusive. Although cerebellin-4 (Cbln4) was proposed as a candidate, it is not essential for establishing this asymmetry [58]. Transcriptomic studies show that SAC and DSGC profiles are largely similar before and after directional tuning [38,59], suggesting that major transcriptional differences do not drive wiring. Instead, asymmetric connectivity may rely on differential expression of a few genes or on post-transcriptional and post-translational mechanisms that control protein localization, modification, or function.

DS circuit function requires FRMD7, a FERM domain–containing cytosolic protein involved in cytoskeletal organization and synaptic connectivity, that is expressed in SACs. In humans, FRMD7 mutations cause a rare congenital X-linked nystagmus, characterized by involuntary eye movements, that selectively impairs horizontal OKR [61]. In mice, Frmd7 mutation leads to the loss of horizontal direction selectivity, phenocopying the selective horizontal OKR deficit seen in FRMD7 human patients [62].

The contribution of BCs to directional motion detection has largely been underappreciated relative to SACs, but cross-species analyses highlight a more prominent role. In rodents, BC terminals exhibit directional biases, forming four distinct bouton populations whose preferred directions align with the preferred directions of their ooDSGC partners [4]. These BC terminals interact closely with SAC dendrites and receive cholinergic input. Disrupting this cholinergic signaling to BCs reduces directional GABA release from SACs onto ooDSGCs, and ultimately reduces direction-selective responses in ooDSGCs, indicating that SAC modulation of BC output is an active contributor to directional tuning [63].

A key question is whether BC-intrinsic mechanisms or SAC-SAC lateral inhibition constitute the primary cause for SAC directional tuning. In rodents, SAC-SAC lateral GABAergic inhibition has been considered a major amplifier of centrifugal directional preference [64], though blocking GABA receptors fails to fully abolish SAC directionality in ON SACs, implying the existence of additional intrinsic mechanisms. In the macaque, center-surround antagonism from BCs is the primary contributor to centrifugal DS establishment in SACs [26], a departure from the predominant rodent model. Interestingly, a similar function for BC surround antagonism in rodent SAC DS [65], and SAC-SAC lateral inhibition is present in the primate retina and anatomically positioned to compute directionally relevant motion signals [27]. Together, both mechanisms co-exist, but their relative weights diverge between species.

Together, DS emerges from coordinated developmental programs that are shaped by multiple processes, including: retinal neuronal process targeting, select synapse elaboration, and cholinergic waves, resulting in asymmetric SAC inhibition and directionally tuned oDSGC responses. Disruption of these finely tuned mechanisms likely impairs retinal motion processing and may underlie visual and oculomotor deficits observed in human visual disorders.

Discussion

Novel studies across phylogeny, from lamprey to primates, reveal a striking conservation of core DS circuit components across species. Recent transcriptomic atlases suggest that the molecular identities of SACs and DSGCs were established at the base of the vertebrate lineage and have been maintained with remarkable fidelity across more than 500 million years of evolution.

During development, transcriptional programs diversify DS cell subtypes and refine their connectivity, enabling precise functional tuning required for direction selectivity. Cell adhesion molecules and guidance cues further direct dendritic stratification and synaptic partner matching to ensure proper circuit assembly. Disruption at any of these levels, from gene regulation to circuit assembly and maturation, can impair DS processing and contribute to human eye movement disorders.

DS circuit dysfunction underlies several human visual disorders, providing a direct window into the clinical relevance of circuits whose core architecture has been conserved since evolution of the earliest vertebrates. Nystagmus is characterized by involuntary eye movements, often accompanied by deficits in the OKR. A retinal origin for nystagmus suggests that congenital nystagmus arises from instability in retinal direction-selective circuits, rather than only a central oculomotor disorder as has been assumed [66]. In this model, abnormal retinal gain can generate oscillatory eye movements that are then propagated through downstream visual pathways, reframing congenital nystagmus as a primary retinal circuit disorder. However, future work will address these causal links using cell type-specific manipulations of defined retinal DS populations. Nevertheless, these results provide a framework for understanding pathology in inherited congenital nystagmus and will suggest further conservation of DS circuit function across vertebrates.

The DS circuit is also vulnerable to neurodegeneration. Impaired vertical optokinetic nystagmus, particularly for downward-moving stimuli, is associated with Parkinson’s Disease (PD) [67]. PD is characterized by the loss of dopaminergic neurons in the brain and dopaminergic ACs in the retina [68]. SACs form synaptic connections with dopaminergic ACs, providing a pathway by which dopaminergic loss in PD may alter SAC function and disrupt DS signaling in DSGCs. Indeed, SACs display reduced densities in PD patient retinas, and they receive fewer synaptic inputs from dopaminergic (TH-positive) ACs [61]. These changes in retinal cholinergic and dopaminergic circuitry likely impair DS signaling and optokinetic responses in PD patients. Notably, a very recent study demonstrates strong links between altered dopaminergic signaling and oscillatory activity in AII ACs implicated in driving the infantile nystagmus seen in Congenital Stationary Night Blindness (CSNB), providing a direct connection between dopamine and nystagmus in CNSB [69]. Future studies are needed to determine how these circuit alterations contribute causally to the optokinetic deficits observed in patients.

Classification of human DS circuits has been elusive. Nonhuman primate DSGCs only account for ~3.5% of all RGCs, making them a rare population of cells in the retina [26,27]. MERFISH and other spatial transcriptomics techniques may identify rare cells across the retina [70]. Additionally, we are lacking robust in vitro models of DSGCs. Human retinal organoids recapitulate the development of the fetal retina, but it is unclear if RGC subtypes, including DSGCs, functionally arise in these models [71]. Future research may need to employ novel approaches to identify DSGCs, such as human retina explant cultures and improved retinal organoid protocols. These advances will be critical to bridge cell-type specification with circuit function and disease relevance in the human visual system. Ultimately, the phylogenetic depth of the DS circuit is its greatest clinical asset, and every molecule and developmental program reviewed here and elsewhere is a candidate entry point for understanding why this circuit fails in oculomotor disorders.

In conclusion, recent studies reveal a striking conservation of DS circuit anatomy and expression of molecular hallmarks across vertebrate species. Yet the mechanistic studies that define DS circuit assembly have been conducted almost exclusively in mice, as illustrated in Fig 2. In primates, the confirmation of DS activity marks only a beginning. Expanding studies across vertebrate species will be essential to reveal whether the regulation of DS circuit assembly is as conserved as the circuits themselves.