AP-2ε Expression in Developing Retina: Contributing to the Molecular Diversity of Amacrine Cells

AP-2 transcription factors play important roles in the regulation of gene expression during development. Four of the five members of the AP-2 family (AP-2α, AP-2β, AP-2γ and AP-2δ) have previously been shown to be expressed in developing retina. Mouse knockouts have revealed roles for AP-2α, AP-2β and AP-2δ in retinal cell specification and function. Here, we show that the fifth member of the AP-2 family, AP-2ε, is also expressed in amacrine cells in developing mammalian and chicken retina. Our data indicate that there are considerably fewer AP-2ε-positive cells in the developing mouse retina compared to AP-2α, AP-2β and AP-2γ-positive cells, suggesting a specialized role for AP-2ε in a subset of amacrine cells. AP-2ε, which is restricted to the GABAergic amacrine lineage, is most commonly co-expressed with AP-2α and AP-2β, especially at early stages of retinal development. Co-expression of AP-2ε and AP-2γ increases with differentiation. Analysis of previously published Drop-seq data from single retinal cells supports co-expression of multiple AP-2s in the same cell. Since AP-2s bind to their target sequences as either homodimers or heterodimers, our work suggests spatially- and temporally-coordinated roles for combinations of AP-2 transcription factors in amacrine cells during retinal development.


AP-2ε is expressed in amacrine cells. Expression of four members of AP-2 family has previously been
documented in the developing retina, with AP-2α, AP-2β and AP-2γ all expressed in amacrine cells. We examined whether AP-2ε might also be expressed in the retina by carrying out in situ hybridization of mouse retinal tissue sections at E16.5 (mostly proliferative cells), P1 (early stage of differentiation), P7 (intermediate stage of differentiation) and P15.5 (late stage of differentiation). Only background staining was observed at E16.5, indicating that AP-2ε is not expressed in proliferating cells (Fig. 1a). By P1, AP-2ε RNA was detected in the inner part of the inner neuroblastic layer where amacrine cells are located. At P7 and P15.5, there were AP-2ε-positive cells throughout the inner part of the inner nuclear layer, along with a few positive cells in the ganglion cell layer of P7 retina. The AP-2ε distribution patterns at P1, P7 and P15.5 are consistent with expression in amacrine cells, as displaced amacrine cells are also found in the ganglion cell layer.
We then examined whether AP-2ε expression in amacrine cells is evolutionarily conserved. In situ hybridization of chick retina tissue sections was carried out at E10 which is roughly equivalent to mouse P7 retina 35,36 . Similar to mouse, AP-2ε RNA in chick retina was found in the amacrine cells located in the inner part of the inner nuclear layer (indicated by arrowheads in Fig. 1b). No signal was observed in the ganglion cell layer, likely reflecting the reduced numbers of displaced amacrine cells in the ganglion cell layer of chick retina compared to  37,38 . However, there was a layer of AP-2ε-positive cells in the outer part of the inner nuclear layer where horizontal cells are located (indicated by arrow in Fig. 1b).
To confirm our findings that AP-2ε is expressed in both mouse and chicken retina, we carried out RT-PCR analysis at different stages of development. In mouse retina, no signal was detected at E16.5, in agreement with the in situ hybridization data ( Fig. 1c and Supplementary Fig. S1). A strong signal was obtained in P1 retina, with progressively weaker signals in P14 and adult retina. These semi-quantitative data were verified by quantitative RT-PCR (Fig. 1d). In chick retina, no signal was detected in the relatively undifferentiated E5 retina, with a peak signal observed in E10 retina ( Fig. 1c and Supplementary Fig. S1).
Next, we carried out immunohistochemical analysis to examine the distribution of AP-2ε protein in retina. We first tested the specificity of our AP-2 antibodies by western blot analysis of HeLa cells transfected with different AP-2 expression constructs. Based on western blotting, the AP-2α, AP-2β, AP-2γ and AP-2ε antibodies are highly specific ( Fig. 2a and Supplementary Fig. S2). The presence of doublet bands suggests post-translational modification of AP-2 proteins. We then used the AP-2ε antibody to immunostain mouse retina. In P7 mouse retina, AP-2ε-positive cells were observed in the inner nuclear layer (arrowheads point to positive cells) (Fig. 2b). We also examined the distribution of AP-2ε in human fetal retina at 17 weeks gestation, a stage when amacrine cells are differentiated 39 . Similar to what we observed in mouse retina, AP-2ε-positive cells in human retina were mostly confined to the inner part of the inner nuclear layer where amacrine cells are located (Fig. 2c). A few AP-2ε-positive cells were also found in the ganglion cell layer, likely displaced amacrine cells.
As we observed a few AP-2ε-positive cells in the ganglion cell layer, we co-immunostained mouse tissue sections with anti-AP-2δ (ganglion cell-specific) and anti-AP-2ε antibodies. There was no co-localization of AP-2δ and anti-AP-2ε at any of the stages analyzed (P1, P7 and P14) (data not shown). These combined data demonstrate complex spatial relationships between AP-2ε and AP-2α, β, γ in the developing retina, with strongest associations observed with AP-2α and AP-2β at P7 and P14. The limited overlap in AP-2ε and AP-2γ co-expression at P1 suggests that there is little need for AP-2γ/AP-2ε-positive amacrine cells at early stages of retinal development.
AP-2ε is expressed in GABAergic amacrine cells. There are two major categories of amacrine cells based on the neurotransmitter used to transmit signals across the retina: glycinergic and GABAergic 40,41 . Previous studies have shown that AP-2α and AP-2β-positive cells can be either glycinergic or GABAergic 26 . We carried out immunofluorescence studies to determine whether AP-2ε is preferentially expressed in glycinergic or GABAergic amacrine cells. Anti-GLYT1 (glycinergic) and anti-GAD67 (a biosynthetic enzyme for GABA) antibodies were used to identify the two different categories of amacrine cells. Examination of P14 mouse retina tissue sections revealed no co-expression of AP-2ε with GLYT1 (Fig. 7a). On the other hand, virtually every AP-2ε-positive cell expressed GAD67, although AP-2ε-positive cells represented a small fraction of GAD67-positive cells (~10%) (Fig. 7b). These results indicate that the amacrine cells that express AP-2ε are GABAergic. We also examined AP-2ε/GAD67 co-immunostaining in the ganglion cell layer. The observed co-immunostaining of AP-2ε and GAD67 confirmed that the few AP-2ε-positive cells in the ganglion cell layer are indeed displaced amacrine cells ( Supplementary Fig. S3).

Co-expression analysis of AP-2 transcription factors using single cell RNA sequencing data.
Macosko et al. 42 carried out single-cell Drop-seq analysis of 44,808 cells isolated from P14 mouse retina and obtained 39 transcriptionally distinct clusters ranging in size from 50 to 29,400 cells. Ganglion cells, cones, rods, horizontal, Müller glia and astrocytic cells were each placed in a single cluster, and bipolar cells in 8 clusters. Amacrine cells were placed in 21 clusters (clusters 3 to 23): 12 GABAergic, 5 glycinergic, 1 excitatory based on glutamate transporter Slc17a8 expression and 3 with low levels of GABAergic, glycinergic and glutamatergic markers. Subgrouping of the GABAergic and glycinergic clusters was based on differential expression of known amacrine markers. Approximately 10% of sequenced single cells (~4,400 of 44,808 cells) were classified as amacrine cells.
In keeping with previously reported data on AP-2 expression in mouse retina 26 , there were clusters (e.g. cluster 6) with elevated levels of AP-2α, AP-2β and AP-2γ sequences. As co-expression of AP-2α and AP-2γ has not previously been reported in the literature, we tested this in silico finding by co-immunofluorescence analysis using anti-AP-2γ and AP-2α antibodies. Our analysis revealed only a few AP-2α/AP-2γ co-expressing cells in P1 mouse retina, with the number of AP-2α/AP-2γ-co-expressing cells increasing at P7 compared to P1 (Fig. 8). Notably, AP-2γ-positive cells located in the middle of the inner nuclear layer, presumably migrating amacrine cells, were exclusively negative for AP-2α (Fig. 8). The pattern of expression of AP-2γ compared to AP-2α suggests that AP-2γ is expressed in migrating cells committed to the amacrine lineage as well as differentiated amacrine cells, and that AP-2γ-expressing amacrine cells differentiate later than AP-2α-expressing amacrine cells. This interpretation of our results is in agreement with Drop-seq data showing amacrine clusters that were positive for AP-2γ, with no or very little AP-2α or AP-2β (clusters 17 and 18). Thus, the Drop-seq data for the different members of the AP-2 family suggest overlapping expression patterns that are aligned with co-immunofluorescence data obtained by us and others.
AP-2ε RNA is expressed in retinoblastoma cell lines. We have previously shown that AP-2α RNA is expressed in retinoblastoma cell lines 43 . However, retinoblastoma cells do not express AP-2β RNA 43 . To further investigate expression of amacrine lineage-specific AP-2s in retinoblastoma cells, we examined the expression of AP-2γ and AP-2ε RNA in 13 retinoblastoma cell lines. Semi-quantitative RT-PCR analysis revealed expression of AP-2γ and AP-2ε in subsets of retinoblastoma lines ( Fig. 9 and Supplementary Fig. S1). With the exception of RB778 cells which were positive for both AP-2γ and AP-2ε, there was a trend towards mutual exclusion for these two AP-2s. These results provide further support for a link between RB and amacrine cells and suggest that AP-2 expression patterns in retinoblastoma cell lines mimic developmentally-regulated amacrine cell differentiation patterns.

Discussion
Four members of the AP-2 family (α, β, γ and δ) have previously been shown to be expressed in subsets of cells in the retina. In this study, we demonstrate that the fifth member of the AP-2 family, AP-2ε, is also expressed in vertebrate retina. Like AP-2α, AP-2β and AP-2γ, AP-2ε is specifically expressed in a subset of amacrine cells. The AP-2ε expression pattern is similar in chicken, mouse and human, although peak expression is observed at an earlier developmental stage in mouse (P1) compared to chicken (E10).
AP-2ε has previously been shown to have a limited expression pattern during mouse embryogenesis compared to AP-2α, AP-2β and AP-2γ. AP-2ε is mainly expressed in developing brain and spinal cord in mouse, with highest levels in the olfactory bulb 11 . AP-2ε is no longer detected in the olfactory bulb by P14 11 . AP-2ε-knockout mice have an abnormal olfactory bulb architecture 21 , with defects in lamination of projection neurons and their associated axons. Surprisingly, AP-2ε−/− mice can still sense odors, suggesting that other members of the AP-2 family may be compensating for loss of AP-2ε in the olfactory bulb 21 . The expression pattern of AP-2ε in amacrine cells is more restricted than that of the other AP-2s. For example, both AP-2α and AP-2β are widely distributed in amacrine cells, with AP-2α/AP-2β co-expression observed in a high percentage of cells 27 . AP-2γ is primarily found in a subset of amacrine cells distinct from those expressing AP-2α and AP-2β (AP-2β/AP-2γ co-localization is described in Bassett et al. 26 ; AP-2α/AP-2γ co-localization is shown in Fig. 8). Although ocular abnormalities have not yet been reported for AP-2ε−/− mice, this may be due to the relatively small number of cells expressing AP-2ε in the retina. As a case in point, it was only upon detailed analysis of the retina and visual centers of the brain that an ocular phenotype was identified in AP-2δ knockout mice 20 . Furthermore, as described for AP-2α and AP-2β 26 , other members of the AP-2 family may compensate for loss of AP-2ε expression.
Structurally, AP-2 proteins have a helix-span-helix dimerization domain at the carboxy terminal region preceded by a basic region. The helix-span-helix motif along with the central basic region constitute the DNA binding domain 44,45 . The DNA binding domain of AP-2ε is evolutionary conserved and is highly similar to that of the other AP-2s suggesting that these transcription factors bind to similar AP-2 recognition elements. In agreement with this, AP-2ε binds to the consensus AP-2 recognition element GCCNNNGGC as either a homodimer or heterodimer with other members of the AP-2 family 46,47 . The N-terminus contains the activation domain which is generally less conserved between different AP-2 proteins except for a proline-rich region found in all AP-2s except AP-2δ [44][45][46][47] . These similarities and differences in the structure of AP-2 proteins may affect protein-protein and protein-DNA interactions, thereby determining AP-2 target gene specificity. For example, AP-2α, AP-2β and AP-2γ can all bind as either homodimers or heterodimers to an AP-2 recognition site in the c-erbB2 promoter; however, AP-2α and AP-2γ are four times more active than AP-2β at activating a c-erbB2-driven reporter construct 48 . Considering that: (i) there are four members of the AP-2 family expressed in amacrine cells with distinct and overlapping patterns, (ii) AP-2 can function as either homodimers or heterodimers, and (iii) there are at least 33 amacrine subtypes in mammalian retina 32,33 , AP-2s could play key roles in the determination of amacrine subtype-specific functions in the retina.
Only a few genes have been identified as AP-2ε target genes in mammals, including ITGA10 encoding integrin α10 (important for chondrocyte differentiation) 49 , COL2A1 (involved in modulation of cartilage development) 50 , DKK4 (associated with resistance to chemotherapy in colon cancer) 51 , and CDKNA1 encoding p21WAF1  52 . The Mmp13 gene has also been shown to be upregulated in adult AP-2ε−/− mice, although it's not known whether Mmp13 is a direct target of AP-2ε 53 . Similarly, Xenopus AP-2ε activates neural crest-specific genes Snail2 and SOX10 54 , and zebrafish AP-2ε activates kita expression in melanophores and helps promote melanophore differentiation 55 . Additional putative AP-2ε target genes come from cDNA microarray analysis of colorectal cancer cells that either express or don't express AP-2ε 51 . Of the top 50 genes identified in this cDNA microarray, one was previously shown to be expressed in amacrine cells in chick retina: Tenascin C (TNC) 56 . Cluster analysis of Drop-seq data, dividing amacrine cells into 3 clusters [group 1 including clusters 5 and 9 (>30% AP-2ε-positive cells), group 2 including clusters 8 and 15 (10-15% AP-2ε-positive cells), group 3 including the remaining clusters with few or no AP-2ε-positive cells], revealed no correlation with any of the putative AP-2ε target genes mentioned above. A next step would be to compare gene expression in wild-type versus AP-2ε−/− retina in an attempt to identify AP-2ε target genes.
Amacrine cells are broadly defined on the basis of the neurotransmitter that they use to transmit signal. There are two main categories of neurotransmitters in amacrine cells: GABAergic (GAT1 positive) and glycinergic (GLYT1 positive) 40,41 . GABAergic amacrine cells are further divided based on whether they express cholinergic (VAChT positive) or tachykinin (Tac1 positive) neurotransmitters. AP-2α, AP-2β and AP-2γ are found in both GABAergic and glycinergic amacrine cells. In contrast, AP-2ε is exclusively found in GABAergic amacrine cells, a result that is in agreement with the Drop-seq data 42 . Thus, AP-2ε may play a specialized role in the regulation of genes involved in fate determination of GABAergic amacrine cells.
A previous study showed that AP-2α RNA is expressed in retinoblastoma cells 43 . Transfection of either AP-2α or AP-2β expression constructs in retinoblastoma cells induced apoptosis, suggesting incompatibility with expression of AP-2 amacrine cell differentiation markers and survival in retinoblastoma cells 43 . Our results demonstrating AP-2γ and AP-2ε expression in retinoblastoma cells further support an amacrine cell lineage for retinoblastoma tumors. The trend towards mutual exclusion of AP-2ε and AP-2γ RNAs in different retinoblastoma cell lines also supports the idea that retinoblastoma tumors are derived from different amacrine subtypes, in keeping with the low level of overlap between AP-2ε-and AP-2γ-expressing amacrine cells at early stage of retinal differentiation in mouse. In conclusion, we show that AP-2ε is expressed in a subset of amacrine cells in developing vertebrate retina with peak expression at P1 in mouse and E10 in chick. AP-2ε expression overlaps with the other three AP-2s previously shown to be expressed in amacrine cells, with extensive overlap with AP-2α and AP-2β at all stages tested. Our immunostaining data are in agreement with previously reported sequencing data obtained from single    cell lines, as well as RNA preparation, have been previously described 43 . The following primers were used for RT-PCR analysis of retinoblastoma cells: human AP-2ε (forward primer: 5′-CAATGTGACGCTGCTGACTT-3′; re vers e primer : CACTGC C CACACTGCT TAG-3′ ), and human AP-2γ (for ward primer : 5′-AAAGCCGCTCATGTGACTCT-3′; reverse primer: TGGTCTCCAGGGTTCATGT-3′). Quantitative RT-PCR was carried out using the SYBR green based qPCR system (Applied Biological Materials Inc. Canada) and analyzed on an ABI 7900HT PCR system, with primers designed to amplify a 150 bp region of AP-2ε (forward primer: 5′-ATTGCAGGCGATAGATGACC-3′; reverse primer: 5′-GAGCAGAAGACCTCACTGG-3′).
In situ hybridization. Tissue sections (7-8 μm) were prepared from E16.5, P1, P7, P15.5 mouse retina, as well as E10 chick retina. Tissue sections were fixed in 4% paraformaldehyde, and incubated with DIG-labelled probes overnight at 55 °C, as previously described 58 . Tissue sections were then washed and digested with ribonuclease A. The signal was detected with anti-DIG antibody using NBT and BCIP as substrates 24 . For in situ hybridization probes, we PCR amplified a 700 bp mouse AP-2ε cDNA fragment and a 560 bp chicken AP-2ε cDNA fragment. cDNAs were cloned into the pBluescript vector and pGEM-T Easy vector, respectively. The constructs were linearized and electrophoresed in a polyacrylamide gel. Bands were cut out, electrophoresed out of the gel and the DNA extracted with phenol and chloroform. Probes were generated using T3 and T7 RNA polymerases and digoxygenin (DIG)-labeling mix, as specified by the manufacturer (Roche). The probes were quantified by comparing labeling intensity to a control probe supplied by manufacturer. Image acquisitions were made using a Zeiss Axioskop2 Plus microscope and AxioVision 4.7.1 software.
Immunostaining. Immunohistochemistry and immunofluorescence analyses were carried out as previously described 43,59 . Mouse P1 (from 2 pups), P7 (from 3 pups) and P14 (from 2 pups) retinas and human fetal retina tissue at 17 weeks gestation were fixed in formalin and paraffin-embedded. Note that mouse retina tissue at P14-16 is roughly equivalent, with most cells in the retina being fully differentiated by P11 36 . Tissue sections were deparaffinized in xylene, rehydrated and microwaved in a pressure cooker for 20 min for antigen retrieval. The rabbit anti-AP-2ε antibody (1:1,500, generated by Dr. Markus Moser, Max Plank Institute of Biochemistry) was used for immunohistochemistry. The following antibodies were used for immunofluorescence analysis: anti-AP-2α, mouse monoclonal antibody (1:400, 3B5, Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by the University of Iowa) 26 , anti-AP-2β, rabbit polyclonal antibody (1:1,000, #2509, Cell Signaling Technology) 26 , anti-AP-2γ, mouse monoclonal antibody (1:200, 6E4/4, Santa Cruz Biotechnology) 26 , and anti-AP-2ε, rabbit polyclonal antibody (1:1,500, described above). GABAergic For co-immunofluorescence of same species anti-AP-2β and anti-AP-2ε rabbit antibodies, we used the Tyramide Signal Amplification (TSA) kit (PerkinElmer). Tissue sections were first immunostained with anti-AP-2ε antibody followed by goat anti-rabbit HRP-conjugated secondary antibody (DAKO). Cy3-conjugated tyramide was then applied, resulting in a tyramide-protein-antibody complex, with HRP activating bound tyramide. The protein-antibody complex was then heat-denatured in sodium citrate, leaving the Cy3-conjugated tyramide intact. Next, tissue sections were immunostained with AP-2β primary antibody followed by Alexa 488-conjugated secondary antibody. Immunofluorescence images were captured on a Zeiss LSM710 confocal laser scanning microscope with a plan-Apochromat 20X lens using ZEN software.
To ensure AP-2 antibody specificity, HeLa cells were transfected with each of the five AP-2 expression constructs in p3xFLAG-CMV vector. Cells were harvested, and total cell lysates were prepared using RIPA buffer. Lysates were electrophoresed through a 10% SDS-PAGE gel and western blot analysis carried out using antibodies to each of the five AP-2s.

Analysis of AP-2 expression in mouse retinal cells.
We utilized the merged expression data set generated by Drop-seq of single P14 mouse retinal cells (GEO accession viewer GSE63472) 42 . Individual cells were grouped as per the 39 clusters defined by Macosko et al. 42 , and further analysis was performed on those clusters identified as amacrine cell types (clusters 3 to 23, comprising a total of 3,711 individual cells). These cells were then scored as positive or negative for expression of AP-2α, AP-2β, AP-2γ or AP-2ε. A more detailed analysis of AP-2ε was performed with cell clusters containing >30% AP-2ε-positive cells (high expression or HE, clusters 5 and 9: 343 cells), >10% AP-2ε-positive cells (mid-level expression or ME, clusters 8 and 15: 182 cells), with the remaining cell clusters labelled as low or no expression (NE, 3,186 cells).
Data availability statement. The datasets analyzed in the current study are from a published study by Macosko et al. 42 . These datasets are publicly available through GEO accession viewer repository GSE63472.