The total number of AII amacrine cells was estimated in B6/J and A/J strain mice (n = 7 each), in their F1 progeny B6AF1 (n = 9) and AB6F1 (n = 5), in mice of the chromosome substitution strains B6.A<9> (n = 3) and B6.A<19> (n = 3), and in 26 strains of mice from the AXB/BXA recombinant inbred strain-set (n = 101), all obtained from The Jackson Laboratories (Bar Harbor, ME, USA). Mice were between 4 and 8 weeks of age, and the number per individual strain is indicated in the bar histograms in the respective figures. Upon their arrival, mice were given a lethal injection of sodium pentobarbital (120 mg/kg, i.p.; Euthasol; Virbac), and once deeply anesthetized, they were intracardially perfused with 2–3 ml 0.9% saline followed by ∼50 ml of 4% paraformaldehyde in 0.1 M sodium phosphate buffer (PB; pH 7.2 at 20°C). Eyes were dissected from the orbits, and post-fixed for 15 min before being transferred to PB. Whole retinas were dissected from the eyes, taking care to ensure that the entirety of the retina was intact, for subsequent immunofluorescence labeling of the AII amacrine cells.

To identify AII amacrine cells we labeled retinas using an affinity-purified rabbit polyclonal antibody to PROX1 (dilutions, catalog numbers and RRIDs for all antibodies are listed in Table 1), followed by a donkey polyclonal antibody to rabbit IgG (H+L) conjugated to Cy3, using the following immunofluorescence protocol: (1) 3 h in a protein block of 5% normal donkey serum, (2) three rinses, 10 min each, with phosphate buffered saline (PBS), (3) incubation in primary antibodies for three nights, (4) three rinses, 10 min each, with PBS, (5) incubation in secondary antibodies for one night, and finally, (6) three rinses, 10 min each, with PBS. All steps were performed with agitation at 4°C, and all solutions were made using PBS with 1% TritonX-100.

After immunofluorescent staining, whole retinas were mounted under a coverslip and examined on an Olympus BHS fluorescent microscope equipped with a Sony video camera and linked to a computer running Bioquant Nova Prime software (R&M Biometrics), which was used to record the position of each cell across a sample field. Four fields were sampled per retina, using a 40× objective, one in each quadrant, centered at a mid-eccentric location (0.75–1.25 mm eccentric to the optic nerve head), each sampled field being 0.032 mm² in area. PROX1+ AII amacrine cells were identified in each field as having brightly labeled large nuclei residing in the innermost region of the inner nuclear layer (INL), termed the amacrine cell layer (ACL); counts from the four fields were then averaged to define a mean density of AII amacrine cells, which was then multiplied by total retinal area to generate an estimated total number. Only one retina was analyzed for AII amacrine cells in each mouse, and the same investigator was responsible for identifying AII amacrine cells in each of the four sampled fields in every retina analyzed from these 32 strains, being blind to strain identity. Strains were routinely sampled in groups of three, with individual mice of the different strains coded and then randomly intermingled (Keeley et al., 2017). The AII amacrine cell number data, for all but the chromosome substitution strain mice, have been permanently deposited on GeneNetwork¹ under the phenotype accession identifier number 10179 (AII amacrine cell number) in the mouse AXB/BXA published phenotypes database. These data, for all but the chromosome substitution strain mice, were originally reported in an on-line supplementary appendix of another study describing the covariation between 12 different cell types across this same RI strain-set of mice (Keeley et al., 2014), and subsequently in a review focusing upon the covariation of cell number in the rod pathway (Keeley et al., 2022).

Because the genome of each of the RI strains contains a mix of the parental haplotypes, QTL mapping takes advantage of the nearly random recombination events over the course of inbreeding each RI strain to estimate the covariance between AII amacrine cell number and the presence of the A versus B haplotype throughout the genome. GeneNetwork implements standard methods for simple and composite interval mapping to identify QTLs, while also estimating the genome-wide p value of a false-positive error by randomly permuting the strain data. 2,000 random permutations were used to determine the suggestive (p < 0.67) and significant (p < 0.05) thresholds for the logarithm of the odds (LOD) score, being an index of the strength of the linkage between the variation in AII amacrine cell number and genomic locus. The options selected in the mapping module in GeneNetwork excluded the parental strains because they do not contain recombinant chromosomes and therefore cannot contribute any mapping precision, but included a weighting function to take into consideration the variability observed in the individual retinas within each RI strain. Composite interval mapping was performed in GeneNetwork to control for the variation observed at each primary QTL to determine if any secondary QTL were unveiled, using the genetic background markers at the peak of each QTL: Upk2 (Chr 9), D11Mit74 (Chr 11), and D19Mit127 (Chr 19).

The boundaries of the QTLs on Chrs 9 and 19 were defined using the composite interval maps. First, the peak LOD score was determined at each locus, and then the boundaries for the QTL were defined as the locations where the LOD trace was higher than the peak LOD score minus 1.5 LOD. On Chr 9, there was a recombination event in one strain that caused a large drop in the LOD score (from ∼37 to 42 Mb), creating two distinct loci. Given their proximity to one another, we analyzed both regions together as one QTL. Once each QTL was defined, a comprehensive list of genes within these regions was constructed. We then identified every sequence variant present for each of these genes, including all single nucleotide polymorphisms (SNPs), insertions/deletions (INDELs), and structural variants (SVs) discriminating the B6/J and A/J genomes, using Release-1505 (GRCm38) of the Mouse Genomes Project from the Wellcome Sanger Institute² and classified their consequence using the query tool (Keane et al., 2011; Doran et al., 2016). These sequence variants were then sorted into either high-priority functional or regulatory variants as described below.

Dixdc1-knockout (KO) mice, carrying the B6.129-Dixdc1^TM1Bnrc/J allele (Kivimae et al., 2011), were obtained from The Jackson Laboratory (#029504; RRID:IMSR_JAX:029504) and bred as heterozygotes at UCSB to obtain KO and littermate wildtype (WT) mice. At 6–8 weeks of age, mice were perfused, eyes were harvested, and retinas were immunolabeled as described above. The numbers of AII amacrine cells and six additional neuronal cell types were determined from whole retinas (cholinergic amacrine cells, VGluT3 amacrine cells, dopaminergic amacrine cells, rod bipolar cells, Type 2 cone bipolar cells and horizontal cells), while changes to the retinal architecture were examined from 200 μm thick retinal sections; all tissue was labeled using the immunofluorescence protocol described above. Primary and secondary antibodies used to label Dixdc1 KO and WT retinal tissue are listed in Table 1. In addition, either NeuroTrace 530 (1:250, #N21482; Invitrogen) was used to label neuronal cell bodies or Hoechst 33342 was used to label nuclei; when used, these were added to the cocktail of primary antibodies.

A one-way ANOVA was used to examine for differences in AII amacrine cell number between mice of the following strains: parental A/J and B6/J strains, their reciprocal B6AF1 and AB6F1 offspring, and consomic B6.A<9> and B6.A<19> strains. Comparisons using post-hoc Tukey tests assessed whether the parental B6/J and A/J strains and their reciprocal F1 strains each differed from one another, and whether either of the two consomic strains differed from the parental B6/J strain. Because these results are considered at different locations in the Results section, the B6/J strain data are reproduced in the respective histograms, but the statistical analysis is based on a single ANOVA and associated Tukey tests. One-way ANOVA and post-hoc Tukey tests were also used to examine differences in AII amacrine cell number between all of the different RI strains, and between the different groups of RI strains sorted by haplotype at Chrs 9 and 19.

Student’s two-tailed t-tests were used to assess statistical significance for all comparisons between Dixdc1 KO and littermate WT mice, and between mice given Dtx4-encoding and control plasmids. The levels of Dixdc1 transcript expression determined from qPCR experiments were assessed for statistical significance between the B6/J and A/J strains of mice across the three developmental ages, and for any interaction between strain and age, using two-way ANOVAs followed by post-hoc Tukey tests. Adult parental strain Dixdc1 expression levels, run separately, were compared using Student’s two-tailed t-tests. Differences in β-catenin signaling across conditions in the in vitro luciferase assay were assessed using a one-way ANOVA followed by planned comparisons using post-hoc Tukey tests. An alpha threshold of 0.05 was used for determining statistical significance in all cases, indicated by an asterisk. Individual p values are indicated in the text.

AII amacrine cells in the mouse retina, which can be reliably labeled with antibodies to PROX1 (Perez de Sevilla Müller et al., 2017; Keeley and Reese, 2018), are positioned along the inner margin of the INL (Figure 1A, arrow), extending their processes across the entire depth of the IPL. The planar distribution of their somata lends itself to imaging the population in wholemount preparations, where the labeling is robust in all AII cells (Figure 1B). Estimates of the total number of AII amacrine cells show an average total of 69,223 cells in the B6/J retina and 64,777 cells in the A/J retina (Figure 1C), a non-significant difference (p = 0.241) of about 4,500 cells representing about 7% of the number present in each strain. Curiously, while the two F1 generation strains, B6AF1 and AB6F1, produced by crossing the parental inbred strains have very similar average numbers, being 82,050 and 80,882 cells (p = 0.992), respectively, each has a significantly larger population of AII amacrine cells (∼21% more) than either of the parental strains (p < 0.001). These results indicate the presence of countervailing genetic determinants in each parental strain that are unmasked in the heterozygous condition.

Given such large variation between the F1 strains versus the parental strains, we examined AII amacrine cell number in the AXB/BXA recombinant inbred strain-set derived from these same parental strains, with the expectation that variation across the RI strains would map to discrete genomic loci containing the source(s) of this variation. Indeed, these RI strains showed an even larger variation in average total number (Figure 1D), from a low of 57,141 cells, in the BXA26 strain, to a high of 86,557 cells, in the BXA2 strain, a difference of ∼29,400 cells, or a 51% increase from the lowest to the highest strain. A one-way ANOVA and post-hoc Tukey tests confirmed statistically significant differences amongst the various strain comparisons (Supplementary Figure 1). Note that the variation within each strain was meager (Figure 1D), the coefficient of variation averaging 0.05 across all strains, relative to the conspicuous variation across the strains. As previously reported, the heritability (h²) of this trait was calculated to be 0.74, indicating nearly three-quarters of the variation in AII amacrine cell number across these 101 RI mice can be ascribed to an effect of genotype (Keeley et al., 2014). Note as well that the variation across the strains is gradual (Figure 1D), although there are a couple of pronounced step-like transitions. The former result must indicate that AII amacrine cell number is a polygenic trait arising from multiple genomic loci, while the latter suggests that a few of these genomic loci exert outsized effects relative to the others.

By associating the variation in AII amacrine cell number to the variation in parental strain haplotype across the genome of these RI strains, we mapped this variation to three prospective genomic loci, on Chrs 9, 11, and 19 (Figure 2A). Permutation mapping indicated that none of the LOD scores associated with these three loci reached the significant threshold (pink broken line, p < 0.05), though two of them crossed the suggestive threshold (gray broken line, p < 0.67), and the third nearly reached this level. The estimated additive effects associated with the presence of two A alleles at the Chr 9 locus (LOD = 2.96; 46.49 Mb) equaled 11,139 cells, and at the Chr 19 locus (LOD = 4.44; 12.40 Mb) equaled 13,585 cells, while the presence of two B alleles at the Chr 11 locus (LOD = 3.27; 11.07 Mb) equaled 13,252 cells. Thus, the countervailing effects of the two haplotypes at these three loci, plus those at other locations that make smaller contributions, must be operating additively to render the parental strains hardly different.

Composite interval mapping can be used to map the remaining variation in cell number after accounting for the variation mapped to a primary locus. We used composite interval mapping to determine whether we might better reveal any undetected genomic loci by controlling for the variation due to these primary loci. When controlling for the variation associated to the locus on Chr 19, we found the locus on Chr 9 to surpass the significant threshold, validating this QTL, while additional suggestive loci on Chrs 1 and 16 were revealed, each with A alleles being associated with an increase in trait values (Figure 2B). Likewise, by controlling for the variation associated to the locus on Chr 9, the locus on Chr 19 far surpassed the significant threshold, while an additional suggestive locus on Chr 14 was revealed (Figure 2C), with the latter also indicating an effect of A alleles increasing trait values. Controlling for the variation associated to the locus on Chr 11 revealed again the slightly suggestive QTL on Chr 1 (not shown), but no other novel loci were observed. Overall, this exercise in composite interval mapping, while not revealing prominent loci at other locations, confirmed the presence of two large-effect QTL from the three originally identified, on Chr 9 and 19, where the presence of A alleles at each locus increased trait values. While the third of the three originally identified loci, the locus on Chr 11, is still detected when controlling for the other two, its magnitude is diminished (Figures 2B, C), and so we have focused our attention on the two significant QTLs on Chrs 9 and 19.

That these two loci work in an additive fashion is made apparent by considering their combined effects upon AII amacrine cell number: the four strains containing the A haplotype at both loci averaged about 84,000 AII amacrine cells, while the seven with the B haplotype at both loci averaged ∼ 62,000 cells, or a difference of ∼22,000 cells (p < 0.001; Figure 3A). This is of a magnitude roughly comparable to the additive effects ascribed to the two loci themselves, accounting for ∼75% of the total variation observed across the entire RI strain-set. The remaining strains, having the A haplotype at one locus and the B haplotype at the other locus, contain intermediate numbers of AII amacrine cells that are not significantly different (p = 0.563), while each was significantly different from the groups containing the same haplotype at both loci (p < 0.005), albeit with considerable scatter amongst the strains containing the A haplotype on Chr 9 and B on Chr 19, likely indicative of other participating smaller-effect loci.

In order to validate independently the presence of gene variants for which A alleles increase trait values on these chromosomes, we examined chromosome substitution strain mice. These are inbred strains available from The Jackson Laboratory in which a single chromosome from the donor A/J strain replaces the corresponding chromosome in the host B6/J strain, produced through standard genetic principles, including at least ten sequential backcrosses while genotyping progeny at each generation, followed by intercrossing in order to homozygose the substituted chromosome (Nadeau et al., 2012). Specifically, AII amacrine cell numbers were quantified in mice in which the entire A haplotype on Chr 9 (B6.A<9>), or on Chr 19 (B6.A<19>), had been introgressed onto the B6/J genetic background. B6.A<9> mice had an average total of 77,951 cells, while B6.A<19> mice had an average total of 79,085 cells (Figure 3B), these totals being roughly 8,700 and 9,900 more cells than observed in the B6/J strain (p = 0.02 and 0.007, respectively). These values are, of course, revealing the net effects of all variant genes on each chromosome, not just the effects associated with those respective genomic loci, and so should not be expected to match precisely those QTL effects. But the magnitude of their effects (each about 30% of the overall variation present across the RI strain-set) reinforces the mapped identification of two QTLs on these chromosomes that each exert outsized effects on the total number of AII amacrine cells. That each of these chromosome substitution strains has greater numbers of AII amacrine cells than the A/J strain, despite their all carrying the A haplotype throughout Chrs 9 and 19, must substantiate the presence of other genomic loci (including the Chr 11 locus) where the presence of the B haplotype elevates cell number.

The QTL on Chr 9 encompasses 157 genes (Figure 4A), while the QTL on Chr 19 contains 182 genes (Figure 4B). For each of these genes, sequence variants were analyzed, retinal expression was assessed, and known functions were identified, as described in the Methods. The analysis for all 339 genes is summarized in Supplementary Tables 1, 2. Nineteen genes at the Chr 9 locus (Figure 4C) and 13 genes at the Chr 19 locus (Figure 4D) were identified as top candidate genes, as they met all three of the following criteria (shown in Supplementary Tables 1, 2): (1) having at least one high-priority sequence variant between the A/J and B6/J genomes, as described in the Methods, (2) exhibiting expression in the retina during development or in adulthood, and (3) having a function that could influence gene regulation, cell proliferation, or apoptosis (Supplementary Table 3). As can be surmised from Supplementary Tables 1, 2, some of these genes in Supplementary Table 3 were regarded as more promising avenues for further exploration, because they had been found to be expressed not only in the adult retina but also during development; others had both potential regulatory as well as functional variants; and some of these genes were known to be associated with genetic pathways previously shown to affect retinal development. For instance, Cell adhesion molecule 1 (Cadm1) is a synaptic cell adhesion molecule expressed by developing cells in the retina, particularly photoreceptors, and participates in the formation of ribbon synapses in the outer plexiform layer (Ribic et al., 2014). It contains several splice region variants as well as several variants directly upstream of the transcriptional start site. Cell adhesion associated, oncogene regulated (Cdon), is a ROBO-related cell surface protein associated with holoprosencephaly and is a receptor in the Hedgehog signaling pathway involved in early eye patterning (Cardozo et al., 2014). It too contains several splice site region variants; additionally, two missense variants discriminate the parental strains. Zw10 kinetochore protein (Zw10) is a mitotic checkpoint protein that is expressed in the developing retina, and several regulatory variants exist. Family with sequence similarity 111, member A (Fam111a) is a serine protease that is important for ensuring proper DNA replication and acts as a cell cycle checkpoint protein. Single amino acid mutations in this gene are associated with Kenny-Caffey syndrome in humans, of which microphthalmia is one manifestation (Lang et al., 2021); between the A/J and B6/J parental strains, there are 46 missense variants. Syntaxin 3 (Stx3) is part of the SNARE complex and participates in vesicle exocytosis and protein trafficking. It is expressed in developing and mature AII amacrine cells, among other retinal neurons, and removal of Stx3 from photoreceptors leads to rapid degeneration (Kakakhel et al., 2020).

We began by probing these and other interesting candidates by validating their expression patterns, correlating expression levels with AII amacrine cell number, or evaluating the effects of knockout models (e.g., in situ hybridization showed elevated Cdon expression in the adult retina within the INL, but analysis of Cdon-KO mice revealed no change in the number of AII amacrine cells). As we proceeded, two genes became of particular interest, Dixdc1 and Dtx4, to which we directed increasing attention, and are discussed in detail below.

Of those higher priority candidate genes on Chr 19, Deltex E3 ubiquitin ligase 4 (Dtx4), was considered of particular interest, because of its modulatory role upon Notch signaling (Revici et al., 2022), the latter recognized to participate in the balance between progenitor cell proliferation versus cell cycle exit and differentiation (Kageyama et al., 2009), including within the retina (Rapaport and Dorsky, 1998; Mills and Goldman, 2017). It has been shown from single cell transcriptomic data to be expressed in retinal progenitor cells in embryonic and early postnatal retina (Clark et al., 2019; Balasubramanian et al., 2021), as well as in AII amacrine cells in maturity, along with several other amacrine cell types (Yan et al., 2020). Finally, there exist 309 sequence variants discriminating the B6/J and A/J genomes (Figure 5A), two variants of which were noted as high priority in the initial screen: a structural variant in the fourth intron and a SNP in the 3′UTR, both lie within evolutionarily conserved regions.

We confirmed Dtx4 expression in the mature B6/J retina using in situ hybridization, noting heightened expression in the innermost portion of the INL where amacrine cells are positioned (Figure 5B). Immunolabeling of adult B6/J sections with an antibody to DTX4 corroborated the expression pattern seen with in situ hybridization, revealing a punctate intracellular pattern of staining that was most obvious in the ACL (Figure 5C). All of the AXB/BXA RI strains have previously been profiled transcriptionally via microarray analysis of ocular mRNA (Whitney et al., 2011), which can be used to assess both variation in gene expression across the strains as well as to detect correlations between gene expression and other traits, such as cell number. Two microarray probes measured Dtx4 transcript levels across the RI strains (IDs ILMN_2922334 and ILMN_2651706), both mapping to the distal 3′UTR of the mRNA (Figure 5A); the output of both probes exhibited a high positive correlation between them (r = 0.72). These probes revealed that Dtx4 expression was variable across the RI strains, and, notably, this variation was positively correlated with the variation in AII amacrine cell number (Figure 5D). Furthermore, the variation in Dtx4 expression (measured from probe ID LIMN_2651706) mapped a suggestive expression (e)QTL to the location of Dtx4 on Chr 19, that is, a cis-eQTL, revealing the A haplotype at this locus to be associated with an increase in Dtx4 expression (Figure 5E). Given the difference in expression in Dtx4 across the strains, all 309 variants, including those found in intronic regions, were scrutinized in-depth by cross-referencing their locations with candidate cis-regulatory elements (cCREs) identified using the GENCODE database. Several INDELs were found in the promoter region; additionally multiple SNPs and INDELs were identified in putative enhancer sites (Figure 5A). Variation in the expression of Dtx4, therefore, is likely caused by one or more of these putative regulatory variants, which in turn may influence the final number of AII amacrine cells in the retina, and thus underlie the QTL on Chr 19.

Since Dtx4 is present in retinal progenitors as early as E13.5 and continues to be expressed in progenitors through the perinatal period (Clark et al., 2019; Balasubramanian et al., 2021), we sought to assess a functional role for Dtx4 during development using an overexpression strategy via electroporation on the day after birth. Previous studies in our lab (Kautzman et al., 2018) confirm earlier demonstrations that electroporation at this age yields transfection only of cell types that are generated postnatally, including rod photoreceptors, bipolar cells, Müller glia, and some amacrine cells (Matsuda and Cepko, 2004, 2007). Plasmids encoding gfp or both Dtx4 and gfp were used to transfect progenitor cells at this time-point; when mice developed to 3 weeks of age, we identified the position of all GFP-positive cells and determined the proportion of them positioned in the ACL. Twelve retinas showed successful GFP expression, including seven control retinas and five Dtx4-overexpressing retinas, each group averaging ∼850 transfected cells per retina, the vast majority of cells (75–80%) being positioned in the ONL. Because most AII amacrine cells are generated before birth (Voinescu et al., 2009), their expected frequency within the GFP+ cohort should be small, and so rather than immunolabeling them to distinguish this rare contingent amongst the GFP+ cells, we simply counted all GFP+ cells situated within the ACL (Figure 5F). When those GFP+ cells (specifically, those in the inner half of the INL) were considered as a proportion of that total GFP population in each of these retinas (Figure 5G), a significant reduction was found in the Dtx4-overexpressing retinas (p = 0.047). Overexpression of this gene during development, therefore, would appear to reduce the production of amacrine cells, likely including the AII amacrine cell because it is one of the later-generated amacrine cell types (Voinescu et al., 2009).

Dix domain containing 1 (Dixdc1) is a positive regulator of the WNT-signaling pathway (Shiomi et al., 2003; Liu et al., 2011). Dixdc1 expression during murine development was previously characterized via RT-PCR and shown to be expressed as early as E9.5 (Shiomi et al., 2005). Specific tissues were profiled via in situ hybridization, revealing expression in the CNS as early as E9.5 (Soma et al., 2006). Within the developing neocortex, Dixdc1 has been shown to play a role in neuronal proliferation and migration (Singh et al., 2010), and dendritic development and synapse function (Guo et al., 2018; Martin et al., 2018), and multiple sequence-disrupting single nucleotide variants in DIXDC1 have been associated with mental illness and autism (Martin et al., 2018). Studies in zebrafish demonstrated a reduction in eye size following overexpression of Dixdc1 during development (Shiomi et al., 2003), while loss of Dixdc1 in the mouse has been shown to delay retinal angiogenesis (Kim et al., 2022). In situ hybridization in the mouse retina revealed expression confined to the inner retina as early as E13.5 (Shiomi et al., 2005), and continuing there through the prenatal period (Soma et al., 2006), when amacrine cells are being generated (Voinescu et al., 2009). As a Dixdc1 knockout mouse is available, we profiled the retinas of these mice to assess a potential role for Dixdc1 in the regulation of AII amacrine cell number.

AII cell densities were determined in Dixdc1 KO and littermate WT mice, to estimate total cell number. There was a significant increase in the size of the AII amacrine cell population in Dixdc1 KO retinas, by nearly 6,000 cells (p = 0.024) (Figures 6A–C). We also assessed the breadth of the effect of loss of Dixdc1 function by determining the size of other neuronal populations in Dixdc1 KO and littermate WT mice, including cholinergic amacrine cells, VGluT3 amacrine cells, dopaminergic amacrine cells, rod bipolar cells, one type of cone bipolar cell (the Type 2 cone bipolar cell), and horizontal cells (Figures 6D–I). Only the dopaminergic amacrine cell population showed a significant change (p = 0.002), being an increase of comparable magnitude to that observed for AII cells, by about 11% (Figure 6F). No change in the overall size of the retina was observed (p = 0.413) (Figure 6J), and nor were there any observable abnormalities in the cellular or synaptic architecture of the retina (Figures 6K, L). Finally, we validated this effect upon AII cells by sampling additional Dixdc1 KO and littermate control (WT) retinas, now at P10, counting the AII amacrine cell population along with one other cell type that did not change in the adult, the horizontal cell population. As in maturity, we found a comparable increase in AII amacrine cells at P10, by 12.5% (p < 0.001) (Figures 6M–O), while the horizontal cell population was unchanged (p = 0.921) (Figure 6P).

The RefSeq database lists three annotated mRNA transcripts, each coding for the full-length DIXDC1 protein (Figure 7A), that is, isoforms which contain all of the known functional domains, including a calponin homology domain, an actin binding domain, two coiled coil domains, and a DIX domain (Shiomi et al., 2005; Wang et al., 2006). Other isoforms are known to arise from the use of multiple promoters and alternative splicing, coding for proteins that lack various lengths of the N-terminal region (Shiomi et al., 2005), and thus altering DIXDC1 function. Amongst the sequence variants in Dixdc1 discriminating the B6/J from A/J genomes, two particularly interesting variants were scrutinized, including one splice-region disrupting INDEL (Figure 7C) and one missense variant (Figure 7D). The missense variant results in a change from a nonpolar isoleucine to a polar threonine within the actin-binding domain of Dixdc1 (Figure 7D). The splice region sequence variant, being a three nucleotide INDEL, was of interest as it lies in the polypyrimidine tract at the 3′ end of the second intron (Figure 7C); the location of this INDEL close to the donor splice site suggests that it may influence the functioning of the splicing machinery during mRNA processing. For both sequence variants, the B haplotype contained the evolutionarily conserved sequence (Figures 7C, D).

As indicated above, all the RI strains have been profiled transcriptionally via microarray analysis (Whitney et al., 2011). The AXB/BXA mouse whole eye mRNA expression database includes two probes for Dixdc1 (Figure 7A), one of which recognizes the 3′UTR common to each of the annotated transcripts (ILMN_2665301); the other is complementary to a sequence within the second intron of these transcripts (ILMN_1214173). Yet both probes exhibited expression levels that were higher than background (Figure 7E), indicating that the latter probe must be recognizing a novel transcript or transcripts that contain all or part of this intron. Searching various databases revealed that, indeed, two protein coding Dixdc1 transcripts are expected to terminate within this very intron (Figure 7B). These “truncated” transcripts contain a STOP codon in this intronic sequence, so translation of these transcripts would be expected to yield a truncated protein product lacking most of the C-terminal domains of DIXDC1. Both probes exhibit expression levels that vary across the parental and RI strains of mice but show minimal covariation with one another (Figure 7E). Furthermore, while the variation in the expression of full-length transcripts across the strains does not map to any genomic locus, variation in the expression of truncated transcripts does map a significant cis-eQTL on Chr 9 at the Dixdc1 locus (Figure 7F), implicating a sequence variant in those strains carrying the A haplotype at this locus in lowering the level of truncated transcripts, increasing the potential candidacy of the splice region INDEL. Unlike Dtx4, adult ocular expression of Dixdc1 was not correlated with AII amacrine cell number, whether measuring the expression of full-length transcripts (r = −0.15) or truncated transcripts (r = 0.03).

To compare expression of the different Dixdc1 transcripts across development, we designed two sets of primers to quantify levels of full-length transcripts versus those that are truncated in the second intron, using qPCR (Figure 7A). We first compared adult retinal mRNA, finding that both strains expressed similar amounts of the full-length (p = 0.27) as well as truncated (p = 0.42) transcripts (Figure 7G), yielding comparable ratios of truncated to full-length transcripts (p = 0.92), with the truncated transcripts being more prevalent.

We subsequently examined mRNA levels during postnatal development, at P1, P5, and P10, during the period when AII amacrine cells are concluding their neurogenesis and commencing their differentiation (Voinescu et al., 2009; Gamlin et al., 2020). There, we observed a progressive increase in both sets of Dixdc1 transcripts during the first 10 postnatal days, with a two-way ANOVA detecting a significant main effect of age (Figure 7H; full-length, p = 0.020; truncated, p < 0.001). A significant main effect of strain was also detected for the truncated transcripts (p = 0.027), with the B6/J strain exhibiting higher expression than the A/J strain; although no significant effect of strain was observed for the full-length transcripts (p = 0.062), the A/J strain appeared to have a higher expression than the B6/J strain at the later postnatal ages (Figure 7H). These opposing effects seen between the strains resulted in a ratio of truncated to full-length transcripts that also differed, yielding a significant main effect of strain (p = 0.005) with B6/J having a higher relative abundance of truncated transcripts. In sum, the efficiency of this splicing event may be altered transiently after birth due to the presence of the splice region INDEL; as development proceeds beyond P10, the abundance of full-length transcripts decreases and the truncated transcripts become the dominant form, and differences between the strains in the ratio of these transcripts are no longer present (Figure 7G).

The shortened Dixdc1 transcripts have a STOP codon within the retained section of the second intron, yielding a truncated protein product lacking most of the C-terminus. We subsequently designed a luciferase assay to determine if a similarly truncated protein product would have any effect on β-catenin signaling; additionally, we examined any potential functional consequence of the missense variant mentioned above. HEK cells were co-transfected with a β-catenin luciferase reporter plasmid and expression plasmids that either did not encode a protein product (control) or expressed one of three different Dixdc1 transcripts coding for different DIXDC1 isoforms (Figure 7I): a full-length DIXDC1 (full-length), a full-length DIXDC1 that contains the amino acid substitution found in the A sequence (missense), and a DIXDC1 isoform lacking most of the C-terminus (truncated). Additionally, half the wells were co-transfected with a plasmid encoding Disheveled2 (Dvl2), an effective activator of WNT signaling. For each DIXDC1 condition (control, full-length, truncated, and missense), the luciferase expression in the presence of DVL2 was normalized to the expression observed absent DVL2 (i.e., with no activation of WNT signaling), to compare the fold change in β-catenin signaling upon WNT activation in the presence of each DIXDC1 isoform.

A one-way ANOVA confirmed significant differences across the four conditions (p = 0.008). Tukey’s post-hoc testing confirmed that the full-length DIXDC1 isoform significantly increased DVL2-induced β-catenin signaling relative to controls lacking any Dixdc1 transcript (p = 0.04), as expected given its role as a positive regulator of WNT signaling (Figure 7J). The substitution of the A missense mutation in that full-length DIXDC1 isoform yielded comparable DVL2-induced β-catenin signaling (p = 0.99), suggesting that its presence does not alter DIXDC1 function (Figure 7J). Substituting the truncated DIXDC1 isoform in place of the full-length isoform yielded levels of DVL2-induced β-catenin signaling that were significantly lower (p = 0.04), being comparable to those achieved in the control condition lacking any Dixdc1 transcript (p = 1.00) (Figure 7J). We conclude, consequently, that any truncated protein that is produced in vivo should be ineffective in either promoting or reducing WNT-β-catenin signaling, but whether it might possess any other functional significance, such as acting as a dominant negative isoform, remains to be determined.