Pias3 is necessary for dorso-ventral patterning and visual response of retinal cones but is not required for rod photoreceptor differentiation

The vertebrate retina is designed to maximize the capture, integration, and transmission of visual information and consists of a stratified architecture with three cellular layers that include six neuronal cell types (Lamb, 2013). The response to light is initiated by rod and cone photoreceptors, which are distinguished by the type of opsin visual pigment they possess. Rod photoreceptors contain rhodopsin and can respond to even a single photon, thereby mediating dim light vision. In contrast, cone subtypes are defined by opsin pigments of distinct spectral properties and mediate daylight and color vision. Recent studies indicate that evolution of rod-dominance provided adaptive advantage to early mammals (Kim et al., 2016a). The spatial distribution of cone subtypes varies by species (Viets et al., 2016). Only about 3% of photoreceptors in the mouse retina are cones, with two subtypes that express opsins maximally sensitive to medium (M-opsin) or short (S-opsin) wavelengths of light (Nikonov et al., 2006). M- and S-opsin expression exhibits a dorsoventral gradient in the mouse retina, with most cones expressing varying amounts of both visual pigments (Applebury et al., 2000; Bumsted and Hendrickson, 1999).

The differentiation of photoreceptors from multipotent retinal progenitor cells is orchestrated by the combinatorial and synergistic or antagonistic action of a small number of transcription factors (Brzezinski and Reh, 2015; Cepko, 2014; Swaroop et al., 2010). Retinal progenitors expressing Otx2 have the potential to differentiate into bipolar or photoreceptor cells, whose fates are further restricted by Vsx2 and Prdm1, respectively (Brzezinski and Reh, 2015). The post-mitotic precursors expressing downstream factors, such as Crx, Rorβ, Nrl, and Nr2e3, differentiate as rod photoreceptors, whereas those expressing Crx and Rorβ develop as S-cones by default unless they are redirected to an M-cone fate by Trβ2 and Rxrγ (Ng et al., 2001, 2011; Roberts et al., 2005; Swaroop et al., 2010; Oh et al., 2008). Differential patterning of opsins is not induced until several days after the expression of these regulatory factors, suggesting that additional downstream mechanisms are needed to establish the dorsoventral expression gradient of M- and S-opsin pigments (Ng et al., 2001; Onishi et al., 2010; Roberts et al., 2005).

Post-translational modifications, such as phosphorylation and SUMOylation, can modulate the activity of transcription factors (Seet et al., 2006), including the two key photoreceptor-specific transcription factors Nr2e3 and Nrl (Roger et al., 2010; Swain et al., 2001). SUMOylation is a reversible modification involving the conjugation of SUMO protein to lysine residues and is associated with changes in the localization and/or function of target proteins (Chymkowitch et al., 2015; Geiss-Friedlander and Melchior, 2007; Lyst and Stancheva, 2007). Several SUMO pathway genes are expressed in the retina and implicated in photoreceptor development and disease (Abad-Morales et al., 2015). E3 SUMO ligase protein inhibitor of activated STAT 3 (Pias3) is reported to regulate both rod and cone subtype differentiation (Onishi et al., 2009, 2010). Pias3 can augment Trβ-dependent activation of M-opsin promoter, with concurrent repression of Rorα-mediated S-opsin promoter (Onishi et al., 2010). Pias3 can also SUMOylate Nr2e3 to maximally repress cone-specific gene transcription (Onishi et al., 2009). Together, these results suggested a dual role of Pias3 in rod and cone photoreceptor development through modulation of distinct targets in each cell type.

We generated a null allele in mice by targeting the Pias3 gene (Pias3^−/−) to elucidate Pias3 function during development. Despite a demonstrated role of Pias3 in multiple cellular pathways (Sundvall et al., 2012; Wu and Zou, 2016; Yagil et al., 2010), Pias3^−/− mice exhibited no gross abnormality and lack of Pias3 did not have a dramatic impact on retinal development and photoreceptor differentiation. Nonetheless, Pias3^−/− mice exhibited altered dorsoventral gradient of S-opsin, reduced M-cone-mediated visual response, and misregulation of a subset of vision-related genes, highlighting a specific role of Pias3 in establishing dorsoventral patterning and visual response of cone photoreceptors in the mouse retina.

A targeting vector with LoxP sites spanning exon 2 to 5 of the Pias3 gene and neomycin selection marker flanked by FRT sites was used to establish a germline knockout mouse line on C57BL/6J background (Pias3^−/−) (see Fig. 1 and Materials and Methods for details). Pias3^−/− mice were viable and fertile despite complete loss of Pias3 protein in all tissues examined. Histological analysis of the Pias3^−/− retina using hematoxylin and eosin (H&E) staining of methacrylate sections revealed proper lamination and thickness of retinal cell layers, including photoreceptors, in both young [postnatal day (P)21] and old (18 month) mice (Fig. 2A). Given that Pias3 was reported to control photoreceptor development (Onishi et al., 2009, 2010), we assessed visual function in Pias3^−/− mice by electroretinography (ERG). Scotopic (rod-mediated) and UV (S-cone mediated) responses of Pias3^−/− retina did not differ significantly from the wild-type at P21 and begin to decline only by 7 months (Fig. 2B,C and Fig. 3). However, the maximum response of P21 Pias3^−/− mice to green stimuli (M-cone mediated) was impaired [203.6±6.1 (mean±s.e.m.) versus 167.4±15.3 μv, P=0.0158] and remained so at least until 12 months (Figs 2D and 3). These results suggest an early and predominantly M-cone defect, with gradual decline of rod and S-cone function at older ages, in the absence of Pias3.

Immunohistochemical (IHC) staining of P21 Pias3^−/− retina showed altered dorsoventral gradient of S-opsin, with more S-opsin-positive cones identified in the dorsal region of both flat-mounted (Fig. 4A) and sectioned (Fig. 4B) retina. To determine whether these opsins were expressed in cones rather than ectopically in rod photoreceptors, further staining was performed with peanut agglutinin (PNA) to specifically examine cone outer matrix sheaths. All S- and M-opsin staining was consistently localized within cone outer segments in Pias3^−/− retina (data not shown). A series of IHC stainings did not show any significant difference between wild-type (Pias3^+/+) and Pias3^−/− retina for markers of rods (Rhodopsin, Rho), cones (Cone arrestin, Arr3), Müller glia (Glutamine synthetase, Glul), activated Müller glia (Glial fibrillary acidic protein, Gfap), bipolar cells (Protein kinase C alpha, Prkca), ganglion cells (POU class 4 homeobox 1, Pou4f1 or Brn3a), or amacrine and horizontal cells (Calbindin 1, Calb1) (Fig. 5A). Since reduced ERG b-wave responses could result from a synaptic transmission defect, we performed immunostaining with synaptic markers – Ctbp2 (C-terminal binding protein 2, ribeye)+Prkca and Arr3+Gnao1 (G protein subunit alpha O1) (Fig. 5B). No apparent abnormality in photoreceptor-bipolar synapses suggested that reduced photoreceptor response in Pias3^−/− retina is caused by subtle changes within neurons and not by defects in cell fate specification or morphogenesis. A similar series of IHC was performed in 18-month-old Pias3^−/− retina and did not reveal any differences with advanced age (Fig. S1).

Next, we performed whole transcriptome analysis (RNA-Seq) of young (P21) and aged (18 month) wild-type and Pias3^−/− retina (Figs 6 and 7). Using a fold change (FC) cutoff of 1.5, FPKM (fragments per kilobase of exon per million reads mapped) of ≥2, and a false discovery rate (FDR)≤0.05, we identified 195 differentially expressed genes (DEGs) at P21 (Fig. 6B; Table S1). We selected 28 uncharacterized or retina-related DEGs at P21 with FPKM >30 for qPCR validation and 24 of these were confirmed as being up- or down-regulated in both RNA-Seq and qPCR analyses (Fig. 6C). Opn1mw (M-opsin) and Opn1sw (S-opsin) were not identified as DEGs by our criteria though small differences were detected by qPCR (Fig. 6C). Using Nrl^−/− and flow-sorted rod photoreceptor RNA-Seq data (Kim et al., 2016b), we observed several rod- and cone-enriched genes (38% and 16%, respectively) among the P21 Pias3^−/− DEGs (Fig. 6D). Further analysis using STRING highlighted Egr1 as a central node interacting with nine other genes (Fig. 6E). Pathway analysis of the 195 DEGs revealed an enrichment of genes related to retinal homeostasis, retina development, visual perception, and cilium (Fig. 6F). Circular visualization of genes in these biological processes indicated that most were downregulated in Pias3^−/− retina (Fig. 6G) and some belonged to multiple pathways as shown by Chord plot representation (Fig. 6H). To determine whether changes in gene expression resulted from alterations in SUMOylation status of upstream transcription factors, we assessed global SUMOylation status of proteins in both Pias3^+/+ and Pias3^−/− retina. No major difference was evident between SUMOylated proteins in the Pias3^−/− retina by immunoblot analysis (Fig. S2). Using the same filtering criteria, whole transcriptome analysis of 18 month old Pias3^−/− retina identified only 69 DEGs (Fig. 7B;Table S2), including fewer rod- and slightly more cone-enriched genes compared to P21 analysis (9% and 22%, respectively) (Fig. 7C). Furthermore, STRING analysis highlighted Kdm6a and Kdm5c, two histone demethylases, in relation with transcription/translation factors Ddx3x and Eif2s3x (Fig. 7D). The number of DEGs was too small to perform pathway analysis. Overall, the low number of identified DEGs in aging retina compared to P21 seems to reflect redundant mechanisms that may compensate the absence of Pias3.

Pias proteins function as E3 SUMO ligases that facilitate, but are not required for, the covalent linkage of SUMO groups to lysine residues of target proteins (Schmidt and Muller, 2003). SUMOylation plays a critical role in modifying target protein function by altering subcellular localization, protein-protein interaction, and transcriptional activity (Geiss-Friedlander and Melchior, 2007); for example, SUMOylation state can affect transcriptional activation of several retinal proteins such as Rorα, Trβ1, Trβ2, Nrl, and Nr2e3 (Hwang et al., 2009; Liu et al., 2012; Onishi et al., 2009, 2010; Roger et al., 2010) as well as protein interactions and stability of Gtf2ird1 and Phosducin (Klenk et al., 2006; Widagdo et al., 2012).

To our surprise, the consequence of Pias3 deletion was less pronounced in the knockout mouse retina compared to previous reports where in vivo electroporation was used to knockdown Pias3 in developing photoreceptors (Onishi et al., 2009, 2010). We suggest that a mild phenotype in Pias3^−/− retina is likely due to redundancy in the SUMOylation machinery and functional overlap with other Pias proteins, which have the potential to catalyze the addition of SUMO groups to the same targets (Rytinki et al., 2009). In Pias3^−/− retina, the expression of Pias genes and other E3 SUMO ligases was unchanged, indicating that compensation is not due to increased expression at the transcriptional level but rather through functional redundancy (Table S3). In addition, analysis of published RNA-Seq data from flow-sorted photoreceptors of wild-type and cone-only retina showed that multiple E3 SUMO ligases are expressed in both rod and cone photoreceptors (Table S3) (Kim et al., 2016b). Such functional redundancy has been described previously. Indeed, Rorα can be SUMOylated by Pias3 as well as Pias2 and Pias4 (Hwang et al., 2009), and Trβ can also be SUMOylated by Pias1 (Liu et al., 2012). Compensatory mechanisms are especially feasible in a germline knockout mouse model where cells can overcome imbalances during development. Relatively mild phenotypes have also been observed for Pias1^−/− (Liu et al., 2004), Pias2^−/− (Santti et al., 2005), and Pias4^−/− (Roth et al., 2004; Wong et al., 2004) mice, whereas embryonic lethality is observed in double Pias1^−/−/Pias4^−/− (Tahk et al., 2007) or SUMO-conjugating enzyme Ubc9^−/− mice (Nacerddine et al., 2005). Breeding Pias3^−/− mice with other available Pias^−/− lines would likely result in more dramatic phenotypes by preventing compensatory mechanisms; however, tissue-specific or conditional deletions might be necessary to prevent early lethality.

The observation of an early M-cone defect in Pias3^−/− retina is intriguing. Pias3 is expressed in M-cones (Onishi et al., 2010) and might serve as the primary E3 SUMO ligase in these cells to fine-tune the expression of a subset of genes. It would thus appear that rods and S-cone photoreceptors are better equipped to maintain appropriate gene expression pattern even in the absence of Pias3 by using alternative SUMO-conjugating enzymes (Table S3). This explanation could also account for higher S-opsin expression in the dorsal Pias3^−/− retina, resulting from incomplete suppression of S-opsin in cells that are fated to be M-cones. Gene expression changes in M-cone population are likely masked because of their small numbers in the whole retina, and transcriptome profiling of isolated M- and S-cones from the Pias3^−/− retina would be necessary to test this hypothesis.

The mild phenotype and seemingly unperturbed global SUMOylation status of Pias3^−/− retina suggest that Pias3 reinforces or augments gene expression changes rather than controls broader cell fate or developmental decisions. We propose that the addition of SUMO groups to target proteins, such as Nrl and Nr2e3, by Pias3 is a transient and dynamic process that serves to modulate (enhance or suppress) their activity in a context-dependent manner and that additional SUMO-conjugating enzymes are able to compensate its function in Pias3^−/− retina. Fine adjustments in gene expression patterns and redundancy in control mechanisms are necessary to establish and optimize spatial and functional organization of photoreceptors in the mammalian retina.

A gene-targeting vector was constructed to add LoxP sites flanking exons 2 to 5 of the Pias3 gene followed by neo cassette enclosed by FRT sites. Correctly targeted embryonic stem cell lines were identified by PCR at both ends designed to detect homologous recombination and used to generate chimeric mice. Original founders were mated twice with Actin-Flp recombinase to remove the neo cassette and were backcrossed to C57BL/6J mice. Specific deletion in the retina and the forebrain was achieved by crossing Pias3^f/f mice with Rx-Cre-positive males (Swindell et al., 2006). However, leaky Cre expression resulted in recombination within the oocyte in a few lines tested, leading to complete deletion of Pias3 exon 2-5 in all tissues and permitting us to establish a germline knockout mouse line (Pias3^−/−). Wild-type and knockout mice of either sex were used for all experiments. All experiments were conducted according to protocols approved by a local Institutional Animal Care and Use Committee and adhered to the Association for Research in Vision and Ophthalmology Statement for animal use in ophthalmic and vision research.

Genomic DNA was extracted from tail clippings and both wild-type and floxed alleles were detected via amplification using primers flanking either side of the 5′ LoxP site yielding products of 166 bp and 200 bp, respectively (5′ LoxP Insertion, Table S5). Alleles with Pias3 exon 2-5 deletion were detected in the same manner but using a reverse primer within exon 6 of the gene to amplify a 1 kb product generated by the knockout allele (Pias3-KO Allele, Table S5). Rx-Cre was detected by amplification of a 350 bp fragment (Rx-Cre, Table S5).

Retina and spleen tissues were isolated from mice at P21 and lysed by sonication in ice-cold radioimmunoprecipitation buffer supplemented with protease inhibitors (Roche Applied Science) and 20 mM N-ethylmaleimide (NEM) (Sigma-Aldrich). Supernatants were quantified by BCA protein assay according to manufacturer's protocol (Pierce) and solubilized in 4× Laemmli buffer+β-mercaptoethanol. After denaturation, 20 µg of the total protein extract was separated by SDS-PAGE and transferred to PVDF membrane (Bio-Rad). Standard immunoblot procedure was followed, as previously described (Roger et al., 2010).

Eyeballs were enucleated and fixed in 4% PFA for 30 min prior to dissection. For cryosections, the eyecup was fixed for an additional 30 min and cryoprotected in 30% sucrose before being frozen in OCT embedding medium (Sakura Finetek USA, Torrance CA, USA) then cut 12 µm thick. For whole mounts, the retina was fixed for an additional 30 min. Both cryosections and whole mounts were blocked for 1 h at room temperature in blocking solution (5% normal donkey serum, 0.3% Triton X-100 in PBS), then incubated overnight at 4°C with primary antibodies (Table S4) diluted in blocking solution. The next day, they were washed 4× with 0.2% Triton X-100 in PBS then incubated at room temperature for 1 h with appropriate secondary antibodies (Table S4) diluted in blocking solution. After four subsequent washings, sections were stained for 5 min with DAPI in 1× PBS and washed again, mounted with Fluoromount-G (Southern Biotech, Birmingham AL, USA) and imaged using a Zeiss LSM 700 confocal microscope.

Eyeballs were enucleated and fixed in 4% glutaraldehyde in PBS for 30 min, then transferred to 4% PFA in PBS until being embedded in glycol methacrylate. Five-micron-thick sections were cut, stained with H&E, and imaged using a Zeiss Axio Imager Z1 brightfield microscope.

Total RNA was extracted by Trizol LS reagent (Thermo Fisher), according to the manufacturer's protocol, and its quality was assessed using an Agilent 2100 bioanalyzer and RNA 6000 Nano chip. Only RNA samples with RNA integrity number (RIN) of ≥7.9 were subjected to first-strand cDNA synthesis using SuperScript II Reverse Transcriptase (Life Technologies). cDNA samples then served as templates for quantitative PCR, per Fast SYBR Green Master Mix manufacturer's protocol (Applied Biosystems) performed in biological and technical triplicates. Primer sequences are provided in Table S5 (Hprt used as internal control).

Whole transcriptome analysis was performed using three independent biological replicates from retina of Pias3^+/+ and Pias3^−/− mice at P21 and at 18 months of age. Total RNA was extracted by Trizol LS reagent (Thermo Fisher), according to the manufacturer's protocol, and its quality was assessed using an Agilent 2100 bioanalyzer and RNA 6000 Nano chip. Only RNA samples with RIN≥7.9 were used. Stranded RNA-seq libraries were constructed from 100 ng of total RNA using a modified TruSeq RNA Sample preparation kit protocol (Roger et al., 2014). Paired-end sequencing of 100 bases length was performed on HiSeq 2500 system (Illumina). Pass-filtered reads were mapped using TopHat v2.1.1 (Trapnell et al., 2009) and aligned to UCSC mouse reference genome mm10. Count table of the gene features was obtained using FeatureCounts (Liao et al., 2014). Normalization, differential expression analysis and FPKM values were computed using EdgeR (Robinson et al., 2010). An FPKM filtering cutoff of 2 in at least one of the 6 samples was applied. FDR≤0.05 was considered significant and a cutoff of fold change of 1.5 was applied to identify differentially expressed isoforms. R packages and JMP Software (SAS) were used for data mining. GO annotation and pathways enrichment analysis was performed using Panther Classification System (http://pantherdb.org/).

ERG responses to light stimulation were recorded simultaneously from both eyes on an Espion Electrophysiology System (Diagnosys LLC, Lowell, MA, USA) as previously described for both photopic and scotopic conditions (Ng et al., 2011). A minimum of six responses from each age/genotype to each light stimulus were recorded (80% power to detect an effect size of 53 μv assuming 30 μv standard deviation). Statistical analysis (two-tailed unpaired Student's t-test) was performed using GraphPad Prism 7 software.

We are grateful to Douglas Forrest for comments. We thank Jacob Nellissery for lab management, Linn Gieser, Milton English and Matthew Brooks for RNA-seq, Megan Kopera and Yide Mi for maintaining the mouse colony, Haohua Qian and Yichao Li for ERG training, and Robert Fariss and Chun Gao for assistance with microscopy. This research utilized the high-performance computational capabilities of the Biowulf Linux cluster at the NIH (https://hpc.nih.gov).