We utilized a murine conditional knockout model in which exon 2 of Afadin was flanked by LoxP sites (Figure 1U; Beaudoin et al., 2012). Cre-mediated recombination results in the excision of exon 2, resulting in a frameshift mutation and a premature stop codon. These conditional alleles were crossed with Six3^Cre transgenic mice to mediate gene deletion, particularly within the developing retinal neuroepithelium starting at E9 (Oliver et al., 1995; Liu and Cvekl, 2017; Diacou et al., 2018). Upon conditional knockout, Afadin mutants (hereafter referred to as AfadincKO) displayed aberrant lamination patterning at four different postnatal time points examined: P2, P7, P14, and P60 (Figure 1E–H). At P2, control retinas (Afadin^F/F) exist as a singularly laminated piece of tissue resulting from the proliferation and early differentiation of early-born retinal neurons (Figure 1A). In contrast, AfadincKO, the central retina is disrupted and contains rosette-like structures with regions devoid of cells at P2 (Figure 1E). At P7, the inner and outer plexiform layers are evident in control mice (Figure 1B) but notably disrupted in AfadincKO, with the OPL omitted and instead existing as a singular fused nuclear layer (fused INL/ONL) (Figure 1F). P7 and older mutants also show columns of neurons (Figure 1F, arrow; Figure 2F) spanning the inner plexiform layer (IPL) and rosettes in the fused INL/ONL (Figure 1F, asterisks). By P14, most, if not all, retinal neuron types and Müller glial cells have been established (Turner and Cepko, 1987; Cepko, 2014). To determine whether loss of Afadin affects retinal neuron densities or cell fate differentiation in addition to cellular layer organization, we examined mouse retinas and quantified neuronal subtypes in controls and mutants at P14 using well-established molecular markers: RBPMS to label RGCs, Chx10 to label BCs, and AP2α to label ACs. We found that the densities of the three major inner retinal neuronal types did not differ significantly between controls and mutants (Figure 1T), indicating retinal neurons in AfadincKO differentiate and proliferate via expected proportions, and Afadin is likely not involved in fate determination or cell survival regulations. When we used P14 mice to quantify the mislocalization of retinal types across three cellular layers, RGCs, ACs, and BCs were significantly scrambled across three cellular layers, compared to control (Figure 1S). Collectively, these results revealed the roles of Afadin in sorting inner retinal neurons into proper layers, likely integrating the positioning cues from the environment when specifying the layers.

In addition to scrambled neuronal locations in the wrong cellular layers, we also observed unusual outer layer ‘rosette’ structures and IPL bridging columns in AfadincKO that were evident as early as P2. To better characterize these substructures in our model, we utilized wholemount retina sections to obtain an en face view of the IPL and fused INL/ONL (Figure 2E–H). Interestingly, these rosettes retained some degree of canonical neuronal patterning (Figure 2—figure supplement 1A–F). They comprised radially arranged rod BCs projecting inwards towards ACs and RGCs (Figure 2I and J). The subtype identities of RGCs within the clusters were diverse, including Osteopontin (Spp1)⁺ αRGCs (Zhao et al., 2023; Duan et al., 2015), Melanopsin (Opn4)⁺ intrinsically photosensitive RGCs (ipRGCs) (Provencio et al., 2000; Provencio et al., 1998), and Cartpt⁺ ON-OFF direction-selective ganglion cells (ooDSGCs) (Kay et al., 2011). In addition, dendrites of Carpt⁺ ooDSGCs are projected centrally. They colocalized with starburst amacrine cell (SAC) dendrites (Figure 2J and K and Figure 2—figure supplement 1D), reminiscent of traditional synaptic pairings at the IPL among these cell types (Wei and Feller, 2011; Wei et al., 2011). We next quantified the number of ‘rosettes’, which numbered approximately 160±13 (Figure 2—figure supplement 2A and B) across four AfadincKO retinas in adults. This suggested that rosette formation was likely not a random occurrence but was likely a result of compensatory mechanisms of retinal lamination independent of Afadin and its CAM partners.

Prior work exploring the role of Afadin and N-cadherin in the mouse dorsal telencephalon revealed that loss of either led to severe defects in axonal pathfinding, in addition to neuronal mislocalization and increased progenitor cell proliferation (Yamamoto et al., 2015; Gil-Sanz et al., 2014; Rakotomamonjy et al., 2017). We asked whether loss of Afadin in the retina would affect RGC projections to the superior colliculus (SC), one of the primary retinorecipient areas in the mouse receiving input from more than 85% of RGCs (Cang et al., 2018; Matcham et al., 2024; Tsai et al., 2022). To delineate the contributions from the right and left eyes, we administered intravitreal injections with CTB-488 and CTB-555 to label projection neurons, that is, RGCs from the eyes: these dyes are subsequently transported via RGC axons to the SC (Figure 3A). Interestingly, though RGCs did project to the SC, ipsilateral/contralateral segregation was disrupted. While most (>97%*) projecting axons should cross at the midline in mice and project contralaterally, we noted an unusually high number of aberrant ipsilateral projections in AfadincKO (Figure 3A). Using an AAV(Retro)-mediated axonal projection-based retrograde labeling for RGC labeling, we delivered AAV-Retro TdTomato into the brain targets. We observed retrogradely labeled RGC distributions inside the retina (Figure 3B). To our surprise, the retrograde labeling also labeled RGCs displaced into the fused INL/ONL (Figure 3C–E). Approximately 42.2 ± 8.8% of the TdTomato labeled RGCs were mislocalized in AfadincKO (Figure 3F), while only a few RGCs were displaced into the INL in the control conditions (Figure 3C and F). Altogether, these results suggest that Afadin loss leads to axonal pathfinding deficits; on the other hand, the scrambled RGCs still grow their axons onto the central targets, including the SC.

Associated with Afadin mutants in the inner retina, we inquired about the functional changes associated with such drastic anatomical changes. We observed a progressive PR loss within the same mutants. Thus, the mutants in their current form prevented us from further inquiring about functional changes associated with the inner retina physiological functions and instead encouraged examination of how Afadin loss affected PR lamination and survival. In adult mice at P60 (Figure 4A and B), we observed a close-to-complete loss of central PRs; there was noted peripheral sparing, as Six3^Cre drives primarily central retinal neuroepithelium during development. Using Recoverin, which primarily labels rods and a small subset of BCs, we observed a significant loss of Recoverin⁺ rods in AfadincKO mice (Figure 4C–J). The mechanisms leading to the loss of PRs are currently unknown; however, the loss is likely due to the postnatal disruption of the organization and connectivity. Yet, the same mutants caused no major losses of BCs, ACs, and RGCs (Figure 1T). Using the same set of mutants at P60, we sought to obtain a comprehensive measurement of visual function in AfadincKO mice. Given the degree of disorganization and PR loss, we noted significantly reduced scotopic and photopic ERG responses (Figure 4M), with flattened a and b waves in scotopic conditions (Figure 4K and L). Altogether, these results indicate that loss of Afadin affects PR stability, leading to significant visual deficits.

Mice were maintained under regular housing conditions with standard access to food and drink in a pathogen-free facility. Male and female mice were used in roughly equal numbers; no sexual dimorphisms were observed. Animals with noticeable health problems or abnormalities were not used. All ages and numbers were documented. PCR of tail biopsy PCR determined genotypes.

The following mouse lines were used:

Six Afadin^F/F and six AfadincKO mice were dark-adapted overnight, and ERG data was collected in dim red light. Mice were first anesthetized with a combination of ketamine/xylazine/acepromazine (70/10/2 mg/kg), and proparacaine eye drops were administered as local anesthesia. Pupils were then dilated with 1% Tropicamide. The mouse was placed on a heating pad (39°C) under a dim red light provided by the overhead lamp of the Diagnosys Celeris ERG apparatus (Diagnosys LLC). The light-guide electrodes were placed onto the corneas. For scotopic conditions, we used a white stimulus with 30 cd·s/m² luminance intensity. Signals were captured for 300 ms after each step to assess scotopic a- and b-wave function. Following the dark-adapted protocol, we used a photopic intensity ramp protocol to assess function in a light-adapted state. For photopic conditions, we used a white stimulus with 10 cd·s/m² luminance intensity. Signals were captured for 300ms after each step to assess scotopic a- and b-wave function. Following the recordings, each mouse was placed in its home cage on a heating pad (39°C) to aid recovery from anesthesia.

Intravitreal injection protocol, as previously established in Zhao et al., 2023. Mice were first anesthetized with a combination of ketamine/xylazine/acepromazine (70/10/2 mg/kg). Then, CTB-488 (Invitrogen, C34775) and CTB-555 (Invitrogen, C34776) dye was injected into the vitreous chamber of the right and left eye, respectively, with a fine glass pipette (Sutter Instrument Company). The toxin was allowed to travel anterograde for 2 weeks before processing brain tissue.

Stereotaxic injection protocol as previously established by Tsai et al., 2022. Mice were anesthetized with continuous 2% isoflurane/oxygen on a stereotaxic setup (Model 940, David Kopf Instruments). Meloxicam (5 mg/kg) was administered IP before the surgery and for two consecutive days after the surgery. AAV viruses were loaded into a pulled glass pipette connected with a syringe (Hamilton, 7634-O) by a dual ferrule adaptor (Hamilton, 55750–0). Injection speed and volume were controlled by a Microinjection Syringe pump (WPI, UMP3T-1). AAV (Retrograde, RG) -Cag-tdTomato (Addgene, 59462-AAVrg) was injected into the right and left SC of P60+adult Afadin^F/F (control) and AfadincKO mice. Coordinates for superior colliculi injection: (3.9–4.2 mm posterior, 0.6–0.7 mm lateral to bregma, and 1.4–1.0 mm below the skull). Volume: 600 nl for a saturated SC injection. Three weeks later, the mice were humanely euthanized via transcardial perfusion. Brain and retinas were collected in 4% PFA for immunohistochemical analysis.

Retina section histology as previously established (Toma et al., 2024). Retina wholemount protocols were previously described in Duan et al., 2018; Duan et al., 2015. A lethal overdose of anesthesia sacrificed the mice. The eyes were dissected and post-fixed with 4% PFA on ice for 1 h and rinsed with 1× PBS. Retinas were analyzed as cryosections and whole mounts. For frozen sections, tissues were immersed in 30% sucrose for 2 h, then frozen in OCT before sectioning in a cryostat (20 μm). For immunohistochemistry, sections were incubated in PBS with 3% donkey serum and 0.3% Triton X-100 for 1 h blocking, followed by primary antibodies overnight at 4°C. For wholemount retinas, tissues were incubated with blocking buffer (5% normal donkey serum, 0.5% Triton X-100 in 1× PBS) overnight, followed by primary antibodies for 2–4 days at 4°C. Secondary antibodies were applied for 2 h at room temperature. Sections and wholemounts were washed with 1× PBS and mounted using Fluoromount-G Mounting Medium, with and without DAPI (Invitrogen). Confocal images were acquired using a Zeiss LSM900 (Carl Zeiss Microscopy).

Retina section histology was described in (N. Tsai, M.R.L., X. D, manuscript in preparation). Eyes were collected and fixed in 4% PFA on ice for 30 min, dissected to remove the cornea and lens, and then placed back into fresh 4% PFA on ice for another 30 min. The sclera was peeled off from each retina, and four radial cuts were made. Retinas were then placed in 30% sucrose/PBS and kept at 4°C until the retinas equilibrated/sank. Retinas were mounted onto a 0.45 μm membrane filter (MF-Millipore, HABG01300) and stretched until flat. The retina and filter paper then underwent two cycles of drying and rewetting and 30% sucrose/PBS before drying for 5 min. The membrane filter was then trimmed to match the retina size. The retina and filter paper were then adhered onto a stage made through mounting a block of tissue freezing medium (EMS, 72592) by mounting a block of tissue freezing medium (EMS, 72592), which was formed earlier within an embedding mold (Polysciences) onto a cryostat chuck and sectioning to form a flat stage. The retina was then quickly embedded with a thin layer of tissue-freezing medium. The chuck was placed back onto a block of dry ice to solidify the tissue freezing tissue-freezing medium. The embedded retina block with chuck was then incubated at –80°C overnight. Following equilibration within the cryostat, the retina was cut flat by maintaining the orientation of the chuck with the cryostat blade while preparing the block and subsequent sectioning. Sections of 12 um thickness were collected onto superfrost plus slides (Fisherbrand, 12-550-1).

Requests for reagents and further inquiries may be directed to the corresponding author, Xin Duan (xin.duan@ucsf.edu).