Reduced Neuronal Self‐Avoidance in Mouse Starburst Amacrine Cells With Only One Pcdhg Isoform

ABSTRACT

The clustered protocadherins (cPcdhs) are a family of ∼60 homophilic cell adhesion molecules expressed across three gene clusters (Pcdha, Pcdhb, and Pcdhg) with a variety of essential roles in the developing nervous system. Some of these roles rely on specific isoforms, whereas others are more consistent with a model of isoform redundancy or a requirement for diversity. The γ‐Pcdhs (expressed from the Pcdhg gene cluster) are particularly important for neuronal self‐avoidance in starburst amacrine cells in the mouse retina. Here, we used mouse mutants to test two of the C‐type isoforms—γC4 and γC5—and found that neither was required for normal self‐avoidance. Conversely, when we analyzed a mutant with only γC4 intact, we found significant failures in self‐avoidance that could not be completely rescued by overexpression of this isoform from a transgene. We have recently found that this isoform is essential for normal neuronal survival during development, and our new findings here support the hypothesis that γC4 is specialized for the survival function at the expense of a significant role in self‐avoidance.

1. Introduction

Self‐avoidance is an essential mechanism in neuronal morphogenesis. As developing dendrites and axons branch and elaborate, self‐avoidance promotes neurites from the same cell (“self”) or same cell type (“homotypic”) to move away from each other and sample potential “non‐self” interactors rather than becoming entangled with each other. It has been described in both vertebrate and invertebrate systems, with a growing list of cell‐surface molecules implicated in mediating self‐contacts leading to avoidance (Zipursky and Grueber 2013; Dong et al. 2023). Closely related to self‐avoidance is the concept of “self/non‐self‐discrimination,” in which individual neurons have a unique recognition signal differentiating “self” processes from any other “non‐self” protrusion. This type of individual barcoding requires an extremely diverse signal exemplified by two families of cell adhesion molecules (CAMs): the 38,000 isoforms generated via alternative splicing of the dscam1 gene in Drosophila and the ∼60 isoforms generated by the clustered protocadherin (cPcdh) locus in vertebrates (Zipursky and Sanes 2010).

In mice, the cPcdhs comprise three gene clusters organized across nearly 1 Mb on chromosome 18: Pcdha encodes for 14 α‐Pcdhs, Pcdhb for 22 β‐Pcdhs, and Pcdhg for 22 γ‐Pcdhs (Wu and Maniatis 1999; Wu et al. 2001). Pcdha and Pcdhg have a similar structure with individual variable exons encoding the entire extracellular domain, transmembrane domain, and variable cytoplasmic domain (VCD) of individual isoforms, whereas three cluster‐specific constant exons encode a common C‐terminal cytoplasmic domain. β‐Pcdhs are encoded by single exons for each isoform with no constant domain. Isoforms are further subdivided on the basis of sequence homology: There are five C‐type isoforms—αC1 and αC2 encoded in the Pcdha cluster and γC3, γC4, and γC5 by the Pcdhg cluster. The other γ‐Pcdhs isoforms are categorized as γA or γB subtypes. All isoforms are CAMs in the cadherin superfamily with homophilic binding specificity in trans (Schreiner and Weiner 2010; Thu et al. 2014). They are not strictly homophilic when dimerizing in cis, resulting in lattices of dimers between membranes to confer recognition specificity in a combinatorial manner (Rubinstein et al. 2015; Goodman et al. 2017, 2016; Rubinstein et al. 2017; Brasch et al. 2019). Thus, when expressed in stochastic isoform combinations, they can generate recognition signals with extreme diversity.

cPcdhs are not only engaged in self‐avoidance, but they also contribute to neural developmental functions, including dendrite arborization (Garrett et al. 2012; Molumby et al. 2016), neuronal survival (Wang et al. 2002; Lefebvre et al. 2008; Prasad et al. 2008), and synaptic development (Weiner et al. 2005; Garrett and Weiner 2009; Meltzer et al. 2023). At least some of these functions rely on interactions common to specific clusters. For example, dendrite arborization is regulated by protein interactions with the constant domain of γ‐Pcdhs (Keeler et al. 2015; Hanes et al. 2024). In recent years, it has emerged that some functions require specific isoforms, particularly the C‐type isoforms, rather than cluster diversity or redundancy. For example, γC3 is crucial for astrocyte morphogenesis and dendrite arborization in the cerebral cortex (Mah et al. 2016; Steffen et al. 2023; Lee et al. 2025). γC4 is the only isoform essential for normal neuronal survival, as mice mutant for Pcdhgc4 undergo excessive apoptosis during the wave of developmental cell death (Garrett et al. 2019; Mancia Leon et al. 2024; McLeod, Hanes, et al. 2025). Even in the case of self‐avoidance, single isoforms can be dominant, as in the projections of serotonergic axons, which rely on αC2 for avoidance leading to dispersion (Chen et al. 2017; Katori et al. 2017). γ‐Pcdhs are particularly important for self‐avoidance in starburst amacrine cells (SACs) in the retina and Purkinje neurons in the cerebellum (Lefebvre et al. 2012). In SACs, loss of expression from Pcdhg but not Pcdha resulted in loss of normal self‐avoidance, whereas disruption of both clusters caused a more severe phenotype than loss of Pcdhg alone (Ing‐Esteves et al. 2018). However, the loss of the three γC‐type isoforms in an allele compound heterozygous with a conditional allele disrupting the whole cluster did not result in self‐avoidance failures, and expression of a single isoform from a transgene (γA1 or γC3) was able to rescue loss of the whole cluster (Lefebvre et al. 2012; Kostadinov and Sanes 2015). Together, this suggests a model of redundancy for the γ‐Pcdh isoforms in self‐avoidance per se. Consistent with this, we previously studied a mutant for Pcdhgc3 and did not find indication of self‐avoidance failures in SACs (Steffen et al. 2023). Here, we investigated Pcdhgc4 and Pcdhgc5 mutants and found that neither was essential for SAC self‐avoidance. Conversely, when we analyzed mutants with only Pcdhgc4 left intact but lacking the other 21 isoforms, we found significant reductions in self‐avoidance that could not be completely rescued by overexpression of transgenic Pcdhgc4.

2. Materials and Methods

2.1. Animals

All experiments included a mix of male and female animals. Previously described mouse strains include Pcdhg^1R1 (Garrett et al. 2019), Pcdhg^C4KO (Garrett et al. 2019), Pcdhg^C5KO (Hanes et al. 2025), Pcdhg^fcon3 (Jax strain number 012644 (Lefebvre et al. 2008; Prasad et al. 2008)), Chat^Cre (Jax strain number 031661 (Rossi et al. 2011)), and C4‐GFP (McLeod, Hanes, et al. 2025).

2.2. Viral Vectors

Brainbow viral particles were purchased from Addgene (Catalog numbers 45185‐AAV9 and 45186‐AAV9). For Brainbow with shRNA knockdown of Pcdha, we used the following targeting sequence: AACAGTATCCAGTGCAACACC (Fan et al. 2018). We first cloned a U6 promoter into a NotI site with NEBuilder HiFi DNA assembly using a synthesized fragment with the following sequence: CTCCATCACTAGGGGTTCCTGCGGCCGCGAATTCATGATGATGGGATCCCGCGTCCTTTCCACAAGATATATAAACCCAAGAAATCGAAATACTTTCAAGTTACGGTAAGCATATGATAGTCCATTTTAAAACATAATTTTAAAACTGCAAACTACCCAAGAAATTATTACTTTCTACGTCACGTATTTTGTACTAATATCTTTGTGTTTACAGTCAAATTAATTCTAATTATCTCTCTAACAGCCTTGTATCGTATATGCAAATATGAAGGAATCATGGGAAATAGGCCCTCGCGGCCGCACGCGTAAGCTTTGCAAAGATGGA. We then annealed two oligos (AATTTTTCCAAAAAAAACAGTATCCAGTGCAACACCTCTCTTGAAGGTGTTGCACTGGATACTGTTG and GATCCAACAGTATCCAGTGCAACACCTTCAAGAGAGGTGTTGCACTGGATACTGTTTTTTTTGGAAA) and ligated them into a site made by EcoRI/BamHI double digest. AAV particles were generated using the AAV‐DJ Helper Free system (Cell Biolabs) as previously described (Chifor et al. 2022). Briefly, Brainbow–shPcdha constructs were co‐transfected with pAAV‐DJ and pHelper into 293FT cells (Invitrogen). Transfected cells were lysed 3 days after transfection, and AAVs were purified with 1‐mL HiTrap heparin columns (GE Healthcare).

2.3. Intravitreal Injection

For Brainbow labeling, adult mice were anesthetized by isoflurane. A sharp 27G needle was used to make a hole in the sclera below the limbus, and 10⁹ vg AAV particles in 1 µL were injected into the vitreous using a blunt 33G Hamilton needle. AAV vectors for shRNA knockdown were introduced at P1. Pups were anesthetized on ice; a sharp needle was used to open the eyelid and puncture the sclera, then a blunt Hamilton needle was used to introduce 0.5 µL AAV into the vitreous. At both ages, topical proparacaine was used for analgesia after injection. Retinas were collected for analysis 2–3 weeks after injection.

2.4. RNAscope

Retinas from adult wild‐type mice were fixed and processed for RNAscope according to the manufacturer's instructions with the RNAscope Multiplex Fluorescent Assay v2 for fixed‐frozen tissue. Probes included Pcdghc5 (Advanced Cell Diagnostics catalog number 850581); Chat (ACD catalog number 408731‐C3); Pcdha (ACD catalog number 572201‐C2); and Pcdhg (ACD catalog number 548661).

2.5. Immunohistochemistry

Eye cups were fixed by immersion in 4% PFA for 1–3 h with the cornea and lens removed, then rinsed in PBS. Retinas were dissected in PBS. Primary antibody staining was performed for 48 h at 4°C in blocking solution containing 2.5% bovine serum albumin (BSA) and 0.25% Triton‐x‐100 in PBS, and secondary antibody staining was performed in the same solution overnight. The following antibodies were used at the indicated concentrations: goat anti‐ChAT, 1:500 (Sigma‐Aldrich Cat# AB144P, RRID:AB_2079751); chicken anti‐GFP, 1:2000 (Aves Labs Cat# GFP‐1010, RRID:AB_2307313); rabbit anti‐mCherry, 1:1000 (Millipore Cat# AB356482, RRID:AB_2889995); Alexa Fluor 647 conjugated alpaca anti‐TagFP, 1:500 (NanoTag Biotechnologies Cat# N0502‐AF647‐L, RRID:AB_3075936); guinea pig anti‐Vglut3, 1:10,000 (Millipore Cat# AB5421, RRID:AB_2187832); mouse anti‐Brn3a, 1:500 (Millipore Cat# MAB1585, RRID:AB_94166); and rabbit anti‐melanopsin, 1:2000 (Advanced Targeting Systems Cat# UF006, RRID:AB_2314781). Secondary antibodies were Alexa Fluor conjugated and used at 1:500.

2.6. Synaptosome Preparation and Western Blotting

Cerebral cortical tissue was homogenized in Syn‐PER synaptic protein extraction buffer (ThermoScientific) at 10 mL of buffer per gram of tissue. All centrifugation occurred at 4°C. Lysate was first cleared by centrifuging at 1200 × g for 10 min. The pellet was discarded, and the resulting supernatant (total homogenate) was transferred to a new tube. This supernatant of ∼200 µL was saved; the remaining supernatant was centrifuged at 15,000 × g for 20 min. The resulting supernatant (cytosolic fraction) was transferred to a new tube and saved. The synaptic pellet was resuspended in ∼400 µL of Syn‐PER buffer. Protein samples of equal amounts were loaded into TGX precast gels (Bio‐Rad), separated via SDS/PAGE, and transferred onto nitrocellulose membranes using a TransBlot Turbo System (Bio‐Rad). After protein transfer, membranes were blocked in 5% nonfat milk in TBS with 0.1% Tween 20 (TBST) for 1 h. Membranes were then washed 3× for 5 min in TBST. Primary antibodies were diluted in 2.5% BSA in TBST, and membranes were incubated overnight at 4°C. Blots were washed 3× in TBST. The membrane was then incubated for 1 h in horseradish peroxidase (HRP)‐conjugated secondary antibodies (1:1000–1:5000, Jackson ImmunoResearch) diluted in 2.5% BSA in TBST. Blots were washed 3× in TBST with a final wash in deionized water. Signals were detected using SuperSignal West Pico Enhanced Chemiluminescent Substrates (Thermo Fisher Scientific) on a LI‐COR Odyssey FC imager. Primary antibodies were used at the following concentrations: Pan‐γ‐Pcdh N159/5, 1:500 (UC Davis/NIH NeuroMab Facility Cat# N159/5, RRID:AB_2877195 (Lobas et al. 2012)), γ‐Pcdh B2 N148/30, 1:500 (UC Davis/NIH NeuroMab Facility Cat# N148/30, RRID:AB_2877457 (Lobas et al. 2012)), γ‐Pcdh C4, 1:500 (rabbit polyclonal antibody produced at Affinity BioReagents against the peptide sequence VAGEVNQRHFRVDLD (Garrett et al. 2019)), PSD95, 1:1000 (Cell Signaling Technology Cat# 36233, RRID:AB_2721262), GAPDH, 1:1000 (Abcam Cat# ab8245, RRID:AB_2107448).

2.7. Image Analysis and Statistical Comparison

Whole mount retinas were imaged en face with a Leica Sp8 confocal microscope. For individual SACs, z‐projections were compiled in Fiji, and their radial symmetry was measured using the Azimuthal Average plugin as described previously (Ing‐Esteves et al. 2018). This method measures the variance in fluorescent signal between eight radial bins centered on the SAC soma. Values across genotypes were compared by one‐way ANOVA with Tukey's post hoc comparisons.

For Elo analysis, we used the ImageEchelon tool available here: https://github.com/TheJacksonLaboratory/ImageEchelon and described in detail previously (Garrett et al. 2018). This tool simply selects random image pairs from a dataset and presents them to a trained observer who chooses which of the two is the “winner” according to whatever parameter is set by the experiment. In this case, the SAC with the more intact self‐avoidance was deemed the “winner.” After the observer selects the winner, ImageEchelon updates image scores based on an Elo algorithm such that the “winner” gains points equal to the value lost by the “loser” image. More points are exchanged if the “winner” entered the matchup with the lower starting score (i.e., as the underdog). We have found that this method can efficiently rank images without performing every pairwise comparison. We continued comparisons until the average matchup number was 25. For individual SACs, each neuron was considered a biological replicate. All conditions were compared in a single set of images. As Elo scores are ordinal, we compared between genotypes using a Kruskal–Wallis test with Dunn's multiple comparisons. Summary statistics for individual SACs with all pairwise comparisons are in Supporting Information File 1. For Elo comparisons of ON SAC plexus images, observers were instructed to choose images with more diffuse dendrite coverage rather than those with fascicles and gaps. For these, two to four images were scored per retina (technical replicates). Scores were averaged to yield a single score per retina (biological replicates). Retinas were compared across genotypes using Kruskal–Wallis with Dunn's post hoc test for multiple comparisons. Summary statistics with all pairwise comparisons are in Supporting Information File 1.

Maximum interdendritic area was measured in Fiji by sampling multiple candidate gaps within an image to identify the largest. Values from two to four images per retina were averaged to yield a single number per retina. These values were compared between all genotypes by one‐way ANOVA with Tukey's post hoc test.

To analyze the distribution of γC4‐GFP in SACs, we stained for GFP in P14 retinas expressing the transgene in cholinergic cells. We collected confocal stacks spanning from the RGL through the IPL. Each stack was normalized identically using the “Subtract Background” command in Fiji with a rolling ball radius of 50 pixels. We made sub‐stacks with 10 steps through the somas and then made a maximum projection through the ON plexus. The mean intensity through the entire field of the plexus projection was calculated. In the soma sub‐stack, we used a circular ROI with a 15.9 µm diameter to measure the mean fluorescence intensity of all 10 steps centered on a soma. These values were summed to approximate the fluorescence intensity through a single soma. Ten somas per image were measured, and their values were averaged to yield a single number per image. The ratio of these values (soma/dendrite) was used to approximate the relative distribution of the protein and presented as a “Distribution score.” Values were compared between genotypes with a Mann–Whitney test.

Cell spacing was measured by first defining x,y coordinates for each cell body using the cell counter plugin in Fiji. An R script was used to quantify spacing from these coordinates by three methods: The sjedrp calculated Density Recovery Profile (DRP) based on previous methods (Rodieck 1991); Voronoi domains were calculated using the deldir package according to previous methods (Lee and Schachter 1980), then the variance of the area of the domains was calculated as the Voronoi score; the nearest neighbor regularity index (NNRI) was calculated using the SpatialEco package. Both Voronoi and NNRI values were compared between wild type and Pcdhg^1R1/1R1 using a t‐test.

The statistical analyses above were performed in GraphPad Prism.

3. Results

3.1. Individual C‐Type Isoforms Are Not Required for ON SAC Self‐Avoidance

To test if γC4 or γC5 are essential contributors to ON SAC self‐avoidance, we analyzed retinas in two mouse mutants. Pcdhg^C5KO is a recently described allele harboring a 7 base‐pair deletion in the Pcdhgc5 exon (Hanes et al. 2025). These mice are overtly normal, have a normal lifespan, and can be analyzed as homozygous Pcdhg^C5KO/C5KO mutants. Due to the essential role that γC4 serves in promoting neuronal survival, Pcdhg^C4KO/C4KO homozygous mutants only survive for a few hours after birth (Garrett et al. 2019; McLeod, Hanes, et al. 2025). Therefore, to analyze later developing morphologies like those of SACs, we made these compound heterozygotes with the conditional allele Pcdhg^fcon3 to make Pcdhg^C4KO/fcon3 (Lefebvre et al. 2008; Prasad et al. 2008). SACs are the only cholinergic neurons in the retina, so to limit recombination of the conditional allele to SACs, we crossed these mice with Chat^Cre , which harbors an IRES‐Cre cassette knocked into the endogenous Chat locus (Rossi et al. 2011).

Chat^Cre was also useful for labeling SACs in conjunction with Brainbow AAV vectors. These are fluorescent Cre reporters that express a stochastically chosen color after recombination (Cai et al. 2013). We injected a 1:1 mixture of the two available vectors to label in four colors (two per vector) into the vitreous of Chat^Cre eyes. After 2 weeks, we isolated the retinas, immunostained for the antigenically distinct fluorescent proteins, and imaged SACs in whole‐mount retinal preparations. This multicolor approach allowed us to achieve sparse labeling to image individual SACs in isolation. We analyzed SACs for self‐avoidance defects with two methods: a unitless radial asymmetry index and a qualitative Elo scoring method, which leverages the power of the human eye to capture differences in morphology that are difficult to measure, such as more discrete fasciculation. Briefly, the radial asymmetry index is a measure of how SAC dendrites are arranged in 360° around the soma, where a number close to zero indicates there is low variance/high radial symmetry of the SAC, and higher numbers indicate a higher variability and asymmetry (Ing‐Esteves et al. 2018; Ing‐Esteves and Lefebvre 2024). The Elo score was obtained by having observers, blind to genotype, choose between two randomly presented images by judging which is less fasciculated. On the basis of “wins” and “losses” in iterative pairwise matchups, the images received scores using an Elo algorithm. Scores above zero (more “wins”) indicate that a SAC has normal self‐avoidance, whereas a score below zero (more “losses”) indicates failure in self‐avoidance compared with other SACs (Garrett et al. 2016). We compared morphologies of ON SACs across three genotypes: Chat^Cre::Pcdhg^+/+ , Chat^Cre::Pcdhg^C4KO/fcon3 , and Chat^Cre::Pcdhg^C5K0/C5KO . SACs from the two mutant genotypes were indistinguishable from Pcdhg^+/+ controls (Figure 1).

FIGURE 1.

Individual γC‐type isoforms are not required for ON SAC self‐avoidance. (A) A schematic of the Pcdhg locus with γA‐ and γB‐type variable exons at the 5′ end and γC‐type isoforms at the 3′ end adjacent to the three constant exons. The Pcdhg^C4KO allele harbors a frameshifting 13‐bp deletion disrupting the γC4 isoform. To circumvent the neonatal lethality of homozygous mutants, Pcdhg^C4KO is made compound heterozygous with Pcdhg^fcon3 and crossed with Chat^Cre to limit the loss of function to starburst amacrine cells (SACs). Pcdhg^C5KO includes a 7‐bp frameshifting deletion in the Pcdhgc5 exon. (B) RNAscope in adult retina verified that, along with many other retinal cell types, SACs expressed Pcdhgc5. Projections of ON SACs labeled with Brainbow in Chat^Cre from (C–E) Pcdhg^+/+ , (F–H) Pcdhg^C4KO/fcon3 , and (I–K) Pcdhg^C5KO/C5KO retinas show a range of normal, radially symmetrical morphologies. Individual SAC morphologies were quantified by (L) radial asymmetry and (M) Elo score. The number of individual SACs is indicated below each column with the number of labeled retinas in parentheses. Scale bar is 50 µm in (B) and 100 µm in (C–K).

3.2. Disrupted Self‐Avoidance in Pcdhg^1R1/1R1 Mutant Retinas

These results are consistent with the “redundancy” hypothesis that no isoform is essential for SAC self‐avoidance and any isoform is sufficient for this process (Lefebvre et al. 2012). However, γC4 has several features that make it unique. It alone among γ‐Pcdhs cannot form homodimers in cis (Goodman et al. 2017), and it is the sole isoform responsible for promoting neuronal survival, which likely involves distinct protein interactions (Garrett et al. 2019; Mancia Leon et al. 2024; McLeod, Lanthier, et al. 2025). To test if γC4 is also capable of promoting self‐avoidance, we analyzed SACs from Pcdhg^1R1/1R1 mutants. This allele includes a large deletion disrupting Pcdhga1 to Pcdhgc3 but sparing Pcdhgc4, then a frameshifting indel in Pcdhgc5; γC4 is the only γ‐Pcdh produced in these mutants (Figure 2A) (Garrett et al. 2019). We crossed these mice with Chat^Cre animals and analyzed individual SACs as with the mutants described above. Unlike in Pcdhg^C5K0/C5KO or Pcdhg^C4KO/fcon3 mutants, we found significant failures of self‐avoidance in Pcdhg^1R1/1R1 SACs as measured both by radial asymmetry and Elo score (Figure 2J–N). Similarly to that reported for whole cluster mutants (Lefebvre et al. 2012), there was variability in the severity of these defects. To assess these defects at the population level, we stained whole mount retinas for Chat, imaged them by confocal microscopy, and made maximum projections through the ON plexus but excluded cell bodies. Here, failure in self‐avoidance could be observed as interdendritic gaps vacated by SAC neurites as they became entangled with themselves and with each other. We quantified this effect by measuring the maximum interdendritic gap in each image and by using the pairwise comparison approach to generate Elo scores. By both measurements, Pcdhg^1R1/1R1 SACs were significantly more fasciculated than controls (Figure 2B–I).

FIGURE 2.

Disrupted self‐avoidance in Pcdhg^1R1/1R1 mutant retinas. (A) A schematic representation of the Pcdhg^1R1 allele indicates a large deletion from Pcdhga1 to Pcdhgc3 and a frameshifting mutation in Pcdhgc5, leaving only Pcdhgc4 intact. Projections through the ON SAC plexus stained for Chat in (B–D) Pcdhg^+/+ and (E–G) Pcdhg^1R1/1R1 retinas. Plexus in Pcdhg^1R1/1R1 is characterized by large gaps between dendritic fascicles, as quantified in (H) and (I). (J–L) Individual Pcdhg^1R1/1R1 SACs labeled with Brainbow in Chat^Cre show self‐avoidance failures with a variety of severity, quantified in (M) and (N). The number of individual SACs is indicated below each column with the number of labeled retinas in parentheses. Scale bar is 100 µm. ****p < 0.0001, **p < 0.01 by Tukey's (H and M) or Dunn's (I and N) multiple comparison test.

This suggests a hypothesis that γC4 alone is not sufficient to promote self‐avoidance. However, one caveat is that the total level of γ‐Pcdh constant domain is reduced in these mice, and even γC4 protein levels are lower than in wild‐type animals (Garrett et al. 2019). Thus, even if γC4 could promote self‐avoidance, there may not be enough present to form the complexes required for the dynamic process of self‐avoidance during development (Ing‐Esteves and Lefebvre 2024). To test this alternate hypothesis, we used a recently developed transgenic line, C4‐GFP, that expresses γC4 with a C‐terminal GFP fusion from a CAG promoter after Cre‐mediated excision of a stop cassette (Figure 3A) (McLeod, Hanes, et al. 2025). We found that this transgene produces approximately 10‐fold expression at the mRNA level and, when expressed in neurons, was able to rescue neonatal lethality in Pcdhg^C4K0/C4KO mice (McLeod, Hanes, et al. 2025). Of particular importance here, the design of this allele used the same promoter and insertion site (at the Rosa locus) as the C3‐mCherry and A1‐mCherry transgenic lines previously found to rescue self‐avoidance after disruption of the entire Pcdhg locus (Lefebvre et al. 2012; Kostadinov and Sanes 2015). To express the transgene in SACs, we crossed mice to generate Chat^Cre::Pcdhg^1R1/1R1::C4‐GFP animals and analyzed SAC morphology at the individual neuron level (Figure 3G–K) and at the population level (Figure 3B–F). We found partial rescue by some measurements. By the asymmetry index, SACs expressing the transgene were highly variable, so although they were not significantly better than those from Pcdhg^1R1/1R1 mutants without the transgene, they were also not statistically significantly worse than controls (Figure 3J). By Elo score, however, they were clearly distinct from Pcdhg^+/+ but did not separate from transgene‐negative Pcdhg^1R1/1R1 SACs (Figure 3K). Similarly, Chat^Cre::Pcdhg^1R1/1R1::C4‐GFP plexus images were not separable from Pcdhg^1R1/1R1 by Elo score but did have intermediate values for interdendritic areas (Figure 3E,F).

FIGURE 3.

Transgene expression from C4‐GFP partially rescues self‐avoidance in Pcdhg^1R1/1R1 . (A) Schematic representation of the C4‐GFP allele knocked in at the Rosa locus. Expression from the transgene occurs after Cre‐mediated recombination of a stop cassette. (B–D) ON SAC plexus images from Pcdhg^1R1/1R1::C4‐GFP retinas demonstrate interdendritic gaps and dendritic fasciculation intermediate between wild‐type controls and Pcdhg^1R1/1R1 mutants, as quantified in (E) and (F). (G–I) Individual SACs labeled with Brainbow in Chat^Cre also showed incomplete rescue after transgene expression. Quantified in (J) and (K). The number of individual SACs is indicated below each column with the number of labeled retinas in parentheses. Scale bar is 100 µm. ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05, ns is not significant by Tukey's (E and J) or Dunn's (F and K) multiple comparison test.

3.3. Normal Cell Spacing in Pcdhg^1R1/1R1 Mutant Retinas

The self‐avoidance failures in Pcdhg^1R1/1R1 SACs suggest this line could be a useful tool to answer other questions about the role of γ‐Pcdhs in directing how neurons arrange themselves during development. Self‐avoidance analyses in the retina have focused on SACs because robust morphological phenotypes have not been described in other retinal neurons in Pcdhg mutants. In the retina, neuronal apoptosis and mosaic cell spacing are closely linked, where an increase or decrease in neuronal density may drive errors in spacing and the random distribution of cells throughout the retina. For example, increasing cell density by inhibiting cell death drives some homotypic cells to cluster with each other (Keeley et al. 2021, 2012). Disruption of the entire Pcdhg cluster or Pcdhgc4 alone results in massive apoptosis of amacrine and RGC types via a Bax‐mediated mechanism (Lefebvre et al. 2008; McLeod, Lanthier, et al. 2025). This cell death has made it difficult to test if mosaic spacing relies on γ‐Pcdhs as it does on other self‐avoidance molecules like the Dscams (Fuerst et al. 2009, 2008), as the reduced density may mask self‐avoidance failures in non‐SAC cell types. As the Pcdhg^1R1/1R1 mutants have normal cell density (McLeod, Lanthier, et al. 2025) but disrupted SAC self‐avoidance, we sought to test if previously unappreciated self‐avoidance errors were more visible in this mutant.

First, we analyzed the dendrites of vGlut3+ ACs with confocal imaging. These cells densely project to the sublamina between the ON and OFF SACs and undergo excessive apoptosis in whole cluster Pcdhg mutants (McLeod, Lanthier, et al. 2025). Investigating either confocal projections through the entire plexus or individual confocal planes (Figure S1A), we did not observe any difference between control and Pcdhg^1R1/1R1 retinas. Second, to test if mosaic spacing relied on γ‐Pcdhs, we analyzed the distributions of dopaminergic ACs, SACs, Vglut3+ ACs, ipRGCs, and Brn3a+ RGCs in control and Pcdhg^1R1/1R1 mutants. We used three metrics to assess spacing and mosaicism for each neuronal type: nearest neighbor index (NNI), variance of Voronoi domains, and the DRP, as we have previously used to analyze spacing defects in mutants in the Dscam family (Garrett et al. 2016). For all cell types we did not find any evidence of clustering or loss of mosaic spacing in the absence of 21 γ‐Pcdh isoforms (Figure S1). In fact, two cell types—ON SACs and Brn3a+ RGCs—became slightly more regularly spaced by the Voronoi measure.

3.4. C4‐GFP Transgene Expression Cannot Rescue Self‐Avoidance in Retinas With Both Pcdha and Pcdhg Clusters Disrupted

Simultaneous disruption of the Pcdha and Pcdhg clusters caused more severe retinal disorganization than what was observed in mutation of either cluster alone (Ing‐Esteves et al. 2018). We hypothesized that targeting α‐Pcdhs in our Pcdhg^1R1/1R1 mutants would decrease the variability of the self‐avoidance defects by making them consistently more severe. To knock down Pcdha, we used shRNA sequence previously validated for efficacy and specificity (Fan et al. 2018). To verify specificity in our hands, we made a new vector by inserting this shRNA sequence driven by the human U6 promoter as a cassette into a vector encoding eYFP with expression driven by an EF1α promoter and produced AAV particles. We performed RNAscope for Pcdha and Pcdhg constant exons on retinas transduced with this vector and with a scramble shRNA control. Pcdha was consistently knocked down in YFP‐positive cells (29% of control, p < 0.0001 by Student's t‐test), whereas Pcdhg was modestly affected (75% of control, p = 0.019 by Student's t‐test) (Figure S2). To specifically analyze SACs, we subcloned this U6‐shRNA cassette into Brainbow vectors in the antisense direction to the expression frame to avoid interference (Figure 4A). With this organization, the shRNA was expressed in any cell infected by the virus, but the fluorescent reporter was only expressed after Cre‐mediated recombination. We injected this vector into the vitreous of P1 pups with three different genotypes: Chat^Cre::Pcdhg^+/+ , Chat^Cre::Pcdhg^1R1/1R1 , and Chat^Cre::Pcdhg^1R1/1R1::C4‐GFP. Knockdown of Pcdha had no discernable effect on SAC morphology in Chat^Cre mice but resulted in severe self‐avoidance defects in Chat^Cre::Pcdhg^1R1/1R1 retinas (Figure 4B–G). The addition of the C4‐GFP transgene did not improve morphologies (Figure 4H–L).

FIGURE 4.

Transgene expression from C4‐GFP is unable to rescue self‐avoidance in SACs with both Pcdhg and Pcdha disrupted. (A) Brainbow vectors were modified to include a cassette for shRNA knockdown of Pcdha by introducing a human U6 promoter and hairpin RNA sequence antisense to the EF‐1α promoter regulating fluorescent protein expression. shα::Brainbow AAV vectors were injected into (B–D) Pcdhg ^+/+, (E–G) Pcdhg^1R1/1R1 , and (H–J) Pcdhg^1R1/1R1::C4‐GFP retinas. Self‐avoidance defects quantified in (K) and (L) were severe, and the presence of the C4‐GFP transgene was unable to provide significant rescue. The number of individual SACs is indicated below each column with the number of labeled retinas in parentheses. Scale bar is 100 µm. ****p < 0.0001, ***p < 0.001, ns is not significant by Tukey's (K) or Dunn's (L) multiple comparison test.

3.5. γC4‐GFP Localizes to Dendrites and Cell–Cell Junctions in Chat^Cre::Pcdhg^1R1/1R1::C4‐GFP SACs

One possible reason for incomplete rescue in Chat^Cre::Pcdhg^1R1/1R1::C4‐GFP SACs could result from the fact that γC4 does not form homodimers in cis but requires a carrier isoform among the other γ‐ or β‐Pcdhs to efficiently localize to the cell surface (Thu et al. 2014; Goodman et al. 2017). In Pcdhg^1R1/1R1 with the 21 other γ‐Pcdh isoforms absent, there may not be enough carrier isoform available to get γC4 to the surface. To test this hypothesis, we analyzed the localization of γC4‐GFP in Chat^Cre::Pcdhg^1R1/+::C4‐GFP and Chat^Cre::Pcdhg^1R1/1R1::C4‐GFP SACs. We reasoned that we could see more GFP‐tagged protein localized to the soma rather than to the dendrites in homozygous mutants if other γ‐Pcdh isoforms were required for proper γC4 localization. We analyzed retinas at P14, as self‐avoidance is still actively occurring at this time point (Ing‐Esteves and Lefebvre 2024). We did not discern any differences in localization (Figure 5A–E). Expression was high in these transgenic animals, and localization to dendrites may not necessarily mean the protein was efficiently inserted into the membrane. However, even in Pcdhg^1R1/1R1 retinas we saw pairs of SAC somas adjacent to each other with increased GFP signal at the junctions, consistent with surface localization (Figure 5F). These results suggest that γC4‐GFP localization is not grossly different in Pcdhg^1R1/1R1 mutants.

FIGURE 5.

Localization of γC4‐GFP does not depend on other γ‐Pcdh isoforms. The distribution of GFP‐tagged protein in ON SACs was compared between Pcdhg^1R1/+ (with one copy of the other 21 Pcdhg isoforms, A and B) and Pcdhg^1R1/1R1 mutants (without any of the other 21 Pcdhg isoforms, C, D, F). The ratio of fluorescent signal between somas and dendrites was measured in (E), with no significant difference between the two genotypes. The number of labeled retinas is indicated below each column in parentheses. When SAC somas were adjacent to each other (F), there was increased GFP labeling at cell–cell junctions even in Pcdhg^1R1/1R1 , consistent with γC4 successfully trafficking to the cell membrane and engaging in trans interactions in the absence of other γ‐Pcdh isoforms. (G–J) Lysates from P21 wild‐type cerebral cortex were subjected to synaptosome fractionation. (G) PSD95 was used as a positive control to ensure proper synaptic enrichment and was detected in the synaptic (“S”) but not cytosolic (“C”) fraction. Localization of γ‐Pcdh isoforms detected using (H) a pan‐γ‐Pcdh antibody as well as an isoform‐specific antibody against (J) γB2 revealed enrichment in the synaptic fraction. Conversely, (I) γC4 was enriched in the cytosolic fraction. GAPDH, which localized to both fractions, was used as a loading control.

To ask if neuronal γC4 normally acquires a unique subcellular localization as has been described in other contexts (Hanson et al. 2010), we performed synaptosome fractionation using protein lysates from wild‐type cerebral cortex at P21. Using an antibody directed against the constant domain, we found that γ‐Pcdh isoforms in general were enriched in the synaptic fraction (Figure 5H). With isoform‐specific antibodies, we found that this was also the case for γB2 (Figure 5J). Conversely, γC4 was enriched in the cytosolic fraction (Figure 5I). This suggests that γC4 takes on a distinct subcellular localization consistent with its distinct function in neuronal survival.

4. Discussion

Here we report that the γC4 isoform of the Pcdhg cluster is insufficient to promote neuronal self‐avoidance in ON SACs. Previous studies had found that no single isoform was required for SAC self‐avoidance (Lefebvre et al. 2012; Kostadinov and Sanes 2015). Consistent with that, we saw no defects in SAC morphology in mutants lacking γC4 or γC5. The most parsimonious explanation for this redundancy is that self‐avoidance is regulated by a signaling response common to all isoforms mediated by the γ‐Pcdh constant domain. Conversely, we found that mice with only γC4 (Pcdhg^1R1/1R1 ) were unable to achieve normal SAC self‐avoidance, even with overexpression of γC4 from a transgene. One might have expected similar results to what was found when γC3 or γA1 was overexpressed from a similarly organized transgene: normal individual self‐avoidance with reduced non‐self‐contacts (Lefebvre et al. 2012) and reduced synaptic connectivity between neighboring SACs (Kostadinov and Sanes 2015). Rather, our results were more reminiscent of Pcdhg whole cluster loss‐of‐function mutants.

We previously found that γC4 is necessary and sufficient to promote neuronal survival (Garrett et al. 2019; McLeod, Hanes, et al. 2025). Further, we found that the VCD was able to rescue survival in an in vitro system (McLeod, Lanthier, et al. 2025). Our finding here that γC4 cannot fully promote self‐avoidance supports the hypothesis that it has become specialized for that survival function independent of extracellular interactions. γC4 is unique among γ‐Pcdhs in that it cannot engage in cis homodimers. Goodman et al. found that cPcdh cis dimers formed asymmetrically, with EC6 from one molecule interfacing with EC5/EC6 of the dimer pair. γC4 could only participate as the EC5/EC6 side of the cis interaction (Goodman et al. 2017). This could still allow γC4 to participate in recognition multimers required for self‐avoidance. In Pcdhg^1R1/1R1 mutants with all other γ‐Pcdh isoforms absent, γC4 would primarily depend on β‐Pcdhs for heterodimer partners. β‐Pcdhs are intact in these mutants, and we previously reported a compensatory increase in Pcdhb isoform expression at the mRNA level in Pcdhg^1R1/1R1 animals (Garrett et al. 2019). Here, we were able to observe γC4‐GFP localization consistent with the cell membrane and trans interactions between SAC cell bodies (Figure 5).

Isoform localization is not strictly dependent on cis interactions. It was recently demonstrated that EC6 can regulate surface localization independent of dimer formation (May and Gaudet 2025). Another study used γB4 as an example to find that the intracellular domain, including the variable region, can regulate the structure of transmembrane lattices (Zhang et al. 2025). This builds on previous findings that posttranslational modifications of VCD regions can regulate trafficking (Ptashnik et al. 2023). Indeed, by electron microscopy, γC4 was found to traffic to different classes of intracellular membranes than other γ‐Pcdhs: transfected γA3 and γB2 localized to membrane tubules, whereas γC4 was found in sheets attributed to ER origin (Hanson et al. 2010). Consistent with this, we found that γC4 was more enriched in the cytosolic fraction than either γB2 or γ‐Pcdh constant domain (Figure 5). This suggests several possibilities for why γC4 does not efficiently promote self‐avoidance, including potential structural differences in trans dimers, isoform‐specific protein interactions, and distinct subcellular localization.

One interesting aspect of our findings is the high variability between individual SACs in the severity of self‐avoidance defects. Because of the neonatal lethality in whole cluster mutants, previous studies relied on Chat^Cre to both disrupt Pcdhg constant exons with Pcdhg^fcon3 and achieve individual SAC labeling with Brainbow (Lefebvre et al. 2012; Kostadinov and Sanes 2015). Cre expression from Chat^Cre is sporadic in neonatal SACs before uniformly labeling SACs by ∼P10 (Xu et al. 2016; Ray et al. 2018), but SAC self‐avoidance begins earlier than this, and significant differences between wild‐type and Pcdhg mutants have been observed by P6 (Ing‐Esteves and Lefebvre 2024). This raised the possibility that variability in SAC self‐avoidance could be secondary to variability in when Cre‐mediated recombination occurred in individual cells. As disruption in Pcdhg^1R1/1R1 mutants was not dependent on Cre expression, our findings suggest that a substantial amount of this variability resulted from biological causes other than the timing of Cre expression. This does raise an important caveat when comparing the transgene results presented here with those from previous studies. Our “loss of function” mutants were homozygous Pcdhg^1R1/1R1 animals, whereas previous studies used Cre to both induce loss of 22 isoforms and introduce single isoforms. This means that, unlike those previous studies, we had mutant cells that never saw transgene expression (i.e., non‐cholinergic cells) and mutant SACs that expressed the transgene sporadically before P10.

If γC4 is specialized for neuronal survival to the exclusion of other functions, our Pcdhg^1R1/1R1 mutants could be useful for identifying other cell types that require cPcdhs for self‐avoidance or other functions for the γ‐Pcdhs that were previously masked by the widespread neuronal apoptosis in whole cluster mutants. We began by looking here at cell spacing in several retinal cell types, since spacing is sensitive to changes in cell number (Keeley et al. 2012). We did not find any major disruption in spacing such as cell clustering as is observed in mutant retinas for other self‐avoidance molecules like Dscam (Figure S1). We also investigated dendrites of Vglut3+ amacrine cells, as their coverage is also influenced by cell number (Keeley et al. 2021), but we did not observe any gross changes. We were also able to increase the severity of self‐avoidance failures by knocking down Pcdha in Pcdhg^1R1/1R1 mutants, which could be a fruitful strategy for affecting multi‐cluster disruption in future studies. ON SACs are particularly prone to self‐avoidance defects, but future analyses may reveal additional cell types dependent on γ‐Pcdhs for self‐avoidance or other functions.

Author Contributions

All authors contributed to the conception and execution of the studies. Material preparation, data collection, and analyses were performed by Cathy M. McLeod, Seoyoung Son, Muhammad Nazmul Haque, Camille M. Hanes, and Andrew M. Garrett. The manuscript was drafted by Cathy M. McLeod and Andrew M. Garrett with final approval by all authors.

Ethics Statement

All procedures using animals were performed in accordance with The Guide for the Care and Use of Laboratory Animals and were reviewed and approved by the Institutional Animal Care and Use Committee of Wayne State University.

Consent

This study did not involve human subjects.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supplementary Materials: dneu70005‐sup‐0001‐SuppMat.xlsx

Supplementary Figures: dneu70005‐sup‐0002‐Figures.pdf

McLeod, C. M. , Son S., Haque M. N., Hanes C. M., and Garrett A. M.. 2026. “Reduced Neuronal Self‐Avoidance in Mouse Starburst Amacrine Cells With Only One Pcdhg Isoform.” Developmental Neurobiology 86, no. 1: e70005. 10.1002/dneu.70005

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.