Feedback Induction of a Photoreceptor-specific Isoform of Retinoid-related Orphan Nuclear Receptor β by the Rod Transcription Factor NRL*

Background: The orphan receptor isoforms RORβ1 and RORβ2 encoded by Rorb are differentially expressed during retinal differentiation. Results: The RORβ2 isoform is photoreceptor-specific and induces the rod differentiation factor NRL, which, in turn, induces RORβ2 through positive feedback. Conclusion: NRL induces the RORβ2-specific promoter of Rorb, its own inducer gene. Significance: Positive feedback between Nrl and Rorb genes reinforces the commitment to a rod differentiation fate. Vision requires the generation of cone and rod photoreceptors that function in daylight and dim light, respectively. The neural retina leucine zipper factor (NRL) transcription factor critically controls photoreceptor fates as it stimulates rod differentiation and suppresses cone differentiation. However, the controls over NRL induction that balance rod and cone fates remain unclear. We have reported previously that the retinoid-related orphan receptor β gene (Rorb) is required for Nrl expression and other retinal functions. We show that Rorb differentially expresses two isoforms: RORβ2 in photoreceptors and RORβ1 in photoreceptors, progenitor cells, and other cell types. Deletion of RORβ2 or RORβ1 increased the cone:rod ratio ∼2-fold, whereas deletion of both isoforms in Rorb−/− mice produced almost exclusively cone-like cells at the expense of rods, suggesting that both isoforms induce Nrl. Electroporation of either RORβ isoform into retinal explants from Rorb−/− neonates reactivated Nrl and rod genes but, in Nrl−/− explants, failed to reactivate rod genes, indicating that NRL is the effector for both RORβ isoforms in rod differentiation. Unexpectedly, RORβ2 expression was lost in Nrl−/− mice. Moreover, NRL activated the RORβ2-specific promoter of Rorb, indicating that NRL activates Rorb, its own inducer gene. We suggest that feedback activation between Nrl and Rorb genes reinforces the commitment to rod differentiation.

the differentiation of amacrine and horizontal cells (31), but the function of ROR␤2 is unknown. Here we show that ROR␤2 is expressed in rod photoreceptors and that its targeted deletion increases the cone:rod ratio in the mouse retina. Further loss and gain of function analyses indicate that both ROR␤2 and ROR␤1 induce Nrl. Surprisingly, ROR␤2 itself was identified as a target of NRL, indicating that NRL promotes positive feedback over Rorb, its own inducer gene. We propose that a mutual activation loop between the Rorb and Nrl genes reinforces the commitment to a rod fate.

EXPERIMENTAL PROCEDURES
Targeted Mutagenesis-The ROR␤2-specific exon was identified by alignment of mouse genomic and cDNA sequences (accession no. NM_146095). The ROR␤2 exon was replaced with lacZ, and homology arms isolated from W9.5 ES cell DNA using Phusion polymerase (New England Biolabs) were inserted into pACN carrying a neomycin resistance gene that self-excises upon germ line transmission by male chimeric mice (32) (5Ј arm coordinates, Ϫ1965 to Ϫ1 relative to the ROR␤2 ATG start codon and 3Ј arm, ϩ103 to ϩ4153). Targeted W9.5 ES cell clones were used to generate chimeric mice at the Genetics Core Facility, Mount Sinai School of Medicine. Chimeras were crossed with C57BL/6J mice, and then Rorb ϩ/2z mice were intercrossed to generate ϩ/ϩ, ϩ/2z, and 2z/2z progeny on a mixed C57BL/6J ϫ 129/Sv background for analysis. The targeted allele structure was confirmed by Southern blot using probes external to the targeting construct. Genotypes were determined by Southern blot or PCR using the following primers: common forward, CCA CTG CTG AAT GAA GTC TGC G; wild-type reverse, GCC TCT TCT ACC CAA AGT CAC; and 2z mutant reverse, GTA AAA CGA CGG GAT CCC CG; yielding wild-type and mutant bands of 710 and 463 bp, respectively. Genotyping has been described previously for the Rorb Ϫ total deletion (27), Rorb 1gfp allele (31), and Nrl deletion (9). For the Nr2e3 rd7 allele, genotype primers were as follows: wild-type forward, CAG CCA CTG AAC TCC AAA CCT C; mutant forward, TGT GCT GAT TTT TAT ACA ATG; and common reverse, AAC ATA CTT CCA CTG TCA CTG CCC. Rorb 2z and Rorb 1g alleles were on C57BL/6J ϫ 129/svJ backgrounds, and the Nrl null allele was on a C57BL/6J background. The studies were performed in accordance with protocols approved by the Animal Care and Use Committee at the NIDDK, National Institutes of Health.
Analysis of RNA by PCR-Reverse transcription PCR analysis of ROR␤1 and ROR␤2 mRNAs from pooled mouse retinas (Ն 3 mice/pool) was performed using the following primers: ␤1 forward, AAC AAA ACC CAG GCA CCA G; ␤2 forward, CAG ACC TCA AGT GAA ACG G; and common reverse AAC AGT TTC TCT GCC TTG G; yielding ROR␤1 and ROR␤2 bands of 442 and 249 bp, respectively. Gapdh was used as a control (primers: forward, ACC TGC CGC CTG GAG AAA CC; reverse, GAC CAT GAG GTC CAC CAC CCT G), yielding a 252-bp product. Real-time quantitative PCR (qPCR) analysis of mRNA levels of Opn1mw, Opn1sw, Rho, Nrl, Crx, Nr2e3, Gnat1, Pde6b, Mef2c, Chx10, and the two Rorb isoforms were normalized to Gapdh RNA levels. The primer pairs for qPCR were as follows: Opn1mw, TCG AAA CTG CAT CTT ACA TCT C and GGA GGT AAA ACA TGG CCA AA; Opn1sw ,  TGC TGG GGA TCT GAG ATG AT and AAT GAG GTG  AGG CCA TTC TG; Rho, CCC TTC TCC AAC GTC ACA GG  and TGA GGA AGT TGA TGG GGA AGC; Nrl, TCA CCC  ACC TTC AGT GAG C and CCC GAG AAC CTC ATC CGA  C; Nr2e3, ACT TGT GCT AAA CTG GAG CCA GAG and  AAG CTC ATT CCA TGC CTC TTC C; Gnat1, GAT GCC  CGC ACT GTG AAA C and CCA GCG AAT ACC CGT CCT  G; Pde6b, GCA GCA CTT TTT GAA CTG GTG and CAT  TGC GCT GGC GGT ACA TA; Mef2c, TGC CAG TTA CCA  TCC CAG TGT C and ATC AGA CCG CCT GTG TTA CCT  G; and Chx10, GCT CCG ATT CCG AAG ATG TTT C and AAA GAT TGT CCT GTG TCG CCG C. The transcription initiation site was mapped using an Ambion 5Ј rapid amplification of cDNA ends kit (Ambion, catalog no. AM1700), followed by cloning and sequencing of product bands.
In Situ Hybridization, Immunohistochemistry, and Histology-Tissues were fixed in 4% paraformaldehyde, and 10-m-thick retinal cryosections were hybridized with digoxigenin-labeled antisense and sense riboprobes as described previously (21). ␤-Galactosidase encoded by lacZ was detected on cryosections after incubation at 37°C overnight with 1 mg/ml X-gal. For histology, tissues were fixed in 3% glutaraldehyde/2% paraformaldehyde. 3-m-thick methacrylate plastic sections were stained with hematoxylin and eosin as described previously (21). Cells were counted on 160-m lengths of ONL in the central retina on three sections each from six eyes for each group at an age of ϳ3 months. Data represent mean Ϯ S.D. Statistical analysis used Student's two tailed t test to compare ϩ/ϩ with the test genotype.
Electroporation and Retinal Explant Culture-Retinal explants from neonatal mice were immersed in DNA solution (1 g/l in PBS), and DNA plasmids were introduced by electroporation using five square pulses (30 V) of 50-ms duration with 950-ms intervals using a pulse generator (BTX Harvard apparatus) (34). Retinas were cultured at 37°C on Nucleopore polycarbonate filters (Whatman, 0.2-m pore size) with 1:1 mixed Ham's F-12 medium (Invitrogen) and DMEM (Invitrogen) containing 10% FCS (HyClone). After 8 -10 days, retinas were analyzed by immunohistochemistry or qPCR analysis. ROR␤1, ROR␤2, NRL, and tdTomato were expressed from the pUb vector (35). The Rorb2p-GFP reporter contained the ROR␤2-specific promoter (Ϫ4959 to Ϫ1, ATG start codon as ϩ1) and a 0.6-kb conserved intron region (ϩ1519 to ϩ2109) that was inserted after the poly-A signal of the GFP cassette. Cell count data represent mean Ϯ S.D. Statistical analysis used Student's two tailed t test to compare ϩ/ϩ with the test genotype.
Isolation of Rod Cells and Strand-specific RNA Sequencing-Retinas were dissected from mice carrying the Nrlp-GFP transgene (36) at P2, P4, P6, P10, P14, and P28 in Hanks' balanced salt solution (Invitrogen), and cells were dissociated at 37°C for 10 min using Accutase (Invitrogen). After incubation, cells were washed with 4 ml of PBS. Dissociated cells were collected by centrifugation at 1600 rpm for 5 min and resuspended in 1 ml of PBS. GFP ϩ cells were collected by flow sorter using FACS Ari-aII (BD Biosciences) in "precision mode" to minimize contamination from different retinal cell types. After flow sorting, samples were reanalyzed with the same settings to check purity. Samples over 97% purity were analyzed. Flow-sorted GFPϩ cells were lysed in TRIzol LS (Invitrogen) to isolate total RNA. RNA quality was evaluated by Bioanalyzer RNA 6000 Pico assay (Agilent Technologies). Using 20 ng of total RNA, mRNA sequencing libraries were generated with poly-A selection using a TruSeq RNA sample prep kit v2 (catalog no. 15025062, Illumina, San Diego, CA). Strand-specific sequencing libraries were generated from at least two biological replicates as described previously (37) and sequenced on a Genome Analyzer IIx platform (Illumina). Transcript quantitation was performed with eXpress v1.3.1 by streaming pass filter reads to Bowtie2 v2.1.0 for alignment to GRCm38.p2/Ensembl v73 annotation.

Expression of ROR␤2 and ROR␤1 in Differentiating Rod Cell
Populations-ROR␤2 and ROR␤1 isoforms possess identical DNA binding and C-terminal domains but divergent N termini encoded by unique 5Ј exons, each with its own promoter (30,31) (Fig. 1A). RNA sequencing analysis during retinal development revealed that ROR␤2 mRNA expression rose at late embryonic stages and attained a plateau in the second postnatal week (Fig. 1B), a temporal pattern that correlated with peak generation of rod photoreceptors in mice (2,3). In contrast, ROR␤1 displayed an earlier rise and broader peak of expression, partly reflecting expression in undifferentiated progenitor cells and in horizontal and amacrine precursor cells (31). These results were in accord with a previous quantitative PCR analysis of ROR␤2 and ROR␤1 mRNA levels during retinal development (31).
To investigate the role of ROR␤ isoforms specifically in rod differentiation, cells were purified by fluorescence-activated flow sorting from mice carrying a transgene that expresses GFP from the Nrl promoter during the final mitosis in rod genesis (36). Transcriptome analysis by RNA sequencing of these GFP ϩ rod populations revealed ROR␤1 expression at postnatal day 2 (P2), the earliest stage examined, with a decline to undetectable levels by P14 (Fig. 1C). In contrast, ROR␤2 continued to rise after P2 to a plateau by P14. Stages earlier than P2 could not be examined by this method because of the small, limiting num-bers of rod precursors at embryonic stages. NRL and NR2E3 mRNA levels rose as expected in these differentiating rod populations (Fig. 1C). Rod sample purity was ascertained by detection of little or no expression of amacrine and horizontal (Calb1), bipolar (Prkca), and ganglion cell (Brn3b/Pou4f2) markers (data not shown). Coexpression of ROR␤1 and ROR␤2 at neonatal stages, when Nrl expression is amplified and large numbers of rods differentiate, implicated both isoforms with a role in rod differentiation fate.
Cellular Expression of ROR␤2 Traced with a Targeted lacZ Marker-To determine ROR␤2 function in mice, targeted mutagenesis ( Fig. 2A) was used to replace the ROR␤2-specific exon with lacZ, which also facilitated detection of expression. It has not been possible previously to distinguish cellular expression of ROR␤2 and ROR␤1 because these isoforms differ by only 13 and two unique N-terminal amino acids, respectively (Fig. 1A). RT-PCR analysis showed that Rorb 2lacZ/2lacZ (2z/2z) mice lacked ROR␤2 mRNA, whereas previously generated The alternatively spliced ROR␤2-e* mRNA (302-bp band) was also deleted in 2z/2z mice. ROR␤2-e* encodes a non-functional, 16-amino acid peptide without a DNA binding domain (24). C, Western blot analysis showing loss of ROR␤2 and retention of ROR␤1 protein in 2z/2z mice at the ages indicated. In 1g/1g mice, ROR␤1 is lost, but ROR␤2 is retained. A control blot for actin is shown below. Arrowheads, ROR␤1 ϳ52-kDa and ROR␤2 ϳ53-kDa bands. D, expression of ROR␤2 in the immature photoreceptor layer (ONL) detected by histochemistry for ␤-galactosidase (turquoise) in ϩ/2z mice at P7. Control ϩ/ϩ mice gave no detectable ␤-galactosidase signal. RPE, retinal pigmented epithelium; INL, inner nuclear layer; GCL, ganglion cell layer. E, developmental expression of ROR␤2 and ROR␤1 in retinas monitored by, respectively, histochemistry for ␤-galactosidase (turquoise) in 2z/2z mice and immunohistochemistry for GFP (brown) in ϩ/1g mice. IS/OS, inner/outer segments; OPL, outer plexiform layer; IPL, inner plexiform layer. F, ROR␤1 in photoreceptor precursors detected by double fluorescence in Rorb ϩ/1gfp mice for GFP (turquoise) with TR␤2 (purple) for cones or NRL (purple) for rods in the outer neuroblast layer (ONBL) at the indicated ages. Scale bars ϭ 25 m. ROR␤1-deficient mice (Rorb 1gfp/1gfp , 1g/1g) (31) lacked ROR␤1 mRNA (Fig. 2B). PCR bands for ROR␤1, ROR␤2, and an alternatively spliced, non-functional product, ROR␤2-e* (24), were sequenced to confirm that the predicted products were deleted in 2z/2z mice. Western blot analysis of the retina at P14, when ROR␤2 expression is normally prominent and exceeds that of ROR␤1, confirmed loss of the 53-kDa ROR␤2 protein and retention of the 52-kDa ROR␤1 in 2z/2z mice (Fig. 2C). Conversely, analysis at P0, when ROR␤1 expression is normally more prominent than that of ROR␤2, confirmed loss of ROR␤1 protein but retention of ROR␤2 in 1g/1g mice. ROR␤2 expression traced by histochemistry for ␤-galactosidase activity was detected specifically in the photoreceptor layer (ONL) of the retina at postnatal stages in heterozygous ϩ/2z mice (Fig. 2D). No expression was detected in other retinal cell layers. Expression throughout the ONL was consistent with expression in the rod population, which spans the entire width of the ONL. Expression was weak in ϩ/2z mice but was more readily detected in 2z/2z mice, with a similar spatial pattern in both genotypes. Developmental analysis in 2z/2z mice revealed an initially weak expression in the neonatal retina, with a marked increase in levels in the ONL by P7 and later stages (Fig. 2E). In contrast, ROR␤1 expression traced by immunostaining for GFP in Rorb ϩ/1g mice was prominent in embryonic and neonatal neuroblast layers and in horizontal and amacrine precursors. After P3, GFP expression declined but also spread at low levels to Müller glia and bipolar cells, as reported previously (31). GFP expression was transient in the photoreceptor layer but could be colocalized with the TR␤2 cone marker at embryonic day 17.5 and the NRL rod marker at P1 (Fig. 2F). The results indicated that ROR␤2 was photoreceptor-specific, whereas ROR␤1 was more widely expressed in photoreceptors and other retinal cell types.
Retinal Morphology and Excess Cones in ROR␤2-and ROR␤1-deficient Mice-The expression of ROR␤2 in rods at early postnatal stages suggested a role in determining the rod versus cone differentiation outcome. This hypothesis was supported by detection of ϳ2-fold increased cone numbers in 2z/2z mice compared with ϩ/ϩ mice at 3 months of age (p Ͻ 0.001) (Fig. 3, A and B). Cones are distinguished histologically by their large nuclei and dispersed chromatin, whereas rods possess smaller nuclei with condensed chromatin (6). Moreover, in 2z/2z mice, ϳ30% of cone nuclei were misplaced in the inner 3 ⁄ 4 zone of the ONL (Fig. 3, B and D) compared with the typical location in the outer 1 ⁄ 4 zone of the ONL in ϩ/ϩ mice (6).
In accord with the histological evidence, in situ hybridization also revealed increases in cells expressing cone S and M opsins (Fig. 4A) and cone arrestin 3 (data not shown) in both 2z/2z and 1g/1g mice. Many S opsin ϩ and M opsin ϩ cells in 2z/2z mice were dispersed in the inner 3 ⁄ 4 zone of the ONL. In 1g/1g mice, most S opsin ϩ and M opsin ϩ cells resided in the outer 1 ⁄ 4 zone, although a few cells were misplaced in the outer plexiform layer and inner nuclear layer. Overt changes in rhodopsin expression were not detected in 2z/2z or 1g/1g mice, in contrast to the nearly complete lack of rhodopsin in Rorb Ϫ/Ϫ mice.
Photoreceptor Phenotypes following Combined Deletion of ROR␤ Isoforms-The moderate, ϳ2-fold increase in the cone: rod ratio resulting from loss of either ROR␤ isoform compared with the extreme predominance of cone-like cells caused by the absence of all ROR␤ isoforms suggested that ROR␤1 and ROR␤2 substitute for each other in rod versus cone differentiation. Therefore, we investigated whether a more pronounced phenotype would be unmasked in compound heterozygotes in which the Rorb 2z (ROR␤2-deficient) allele was combined with the Rorb Ϫ null allele (lacking both isoforms). In comparison with Rorb 2z/2z mice, the stepwise removal of additional ROR␤ isoforms of both types in Rorb 2z/Ϫ mice further increased the proportion of S opsin ϩ and M opsin ϩ cells 2.0-and 1.5-fold, respectively (each p Ͻ 0.005 versus 2z/2z) (Fig. 4A). Similarly, the combination of the Rorb 1g (ROR␤1-deficient) allele with the null allele in Rorb 1g/Ϫ compound heterozygotes also resulted in increased proportions of S opsin ϩ and M opsin ϩ cells (Fig. 4A).
These results, together with the evidence of an almost completely cone-dominated retina in Rorb Ϫ/Ϫ mice lacking all ROR␤ isoforms, indicate that ROR␤2 and ROR␤1 cooperate during retinal development and are capable of substantial substitution for each other in rod versus cone differentiation. Analysis by qPCR confirmed the increased cone opsin mRNA levels in 2z/2z and 1g/1g mice and a much exacerbated increase in Rorb Ϫ/Ϫ mice lacking all ROR␤ isoforms (Fig. 4B).
Electroretinogram Analysis of Cone and Rod Function-We examined whether 2z/2z and 1g/1g mice exhibit altered cone or rod function in accord with the increased cone:rod ratio because Nrl Ϫ/Ϫ mice display enhanced S cone function and loss of rod function (9). The scotopic electroretinogram revealed rod responses in the normal range in 2z/2z mice (Fig. 5A). The photopic electroretinogram evoked with light wavelengths optimal for stimulation of S cones (360 nm) and M cones (520 nm) in mice (33) revealed no difference for S cone responses and a slight decrease in M cone responses. Therefore, despite the ϳ2-fold increased cone numbers, no enhanced cone response was evident.
In 1g/1g mice, no scotopic or phototopic responses could be evoked, even with high intensity stimuli. No a wave or b wave was detected (Fig. 5B). The lack of an a wave in the electroretinograms suggested the contribution of a photoreceptor-intrinsic defect because the a wave reflects the response of photoreceptors rather than the indirect response of bipolar neurons. Therefore, ROR␤1 is essential for the acquisition of phototransduction function in all rod and cone photoreceptors. Rorb Ϫ/Ϫ mice, deficient for both ROR␤1 and ROR␤2, also lacked all rod and cone responses.
ROR␤ Isoforms Induce Nrl and Rod Differentiation Genes-To demonstrate, by gain of function, that both ROR␤ isoforms promote a rod versus cone outcome and do so by induction of Nrl, vectors expressing ROR␤1 and ROR␤2 were electroporated into retinal explants from neonatal Rorb Ϫ/Ϫ mice that contain almost exclusively cone-like cells instead of rods (27). ROR␤ protein has been shown to bind to the Nrl gene promoter (28,29), but a functional induction of Nrl has not been demonstrated. Electroporated cells were identified by coelectroporation of a vector for the Tdt marker. Control retinal explants from ϩ/ϩ pups electroporated with empty expression vector (Ub/empty) expressed NRL, NR2E3, and rhodopsin in many cells in the ONL, whereas explants from Rorb Ϫ/Ϫ mice lacked almost all expression of these rod markers (Fig. 6A). Electroporation of the Ub/ROR␤1 and Ub/ROR␤2 expression vectors into explants from Rorb Ϫ/Ϫ mice resulted in the detection of NRL, NR2E3, and rhodopsin in many cells in the ONL (Fig. 6, A  and C). Analysis by qPCR of explants electroporated in parallel demonstrated the induction of mRNA for many rod genes, including Rho, Nrl, Nr2e3, Pde6b, Mef2c (Fig. 6D), and Gnat1 (data not shown). As a negative control, the Chx10 bipolar cell marker showed no change in response to ROR␤1, ROR␤2, or empty control vector.
To demonstrate that both ROR␤1 and ROR␤2 promote rod differentiation through induction of NRL, electroporations were performed in retinal explants from Nrl Ϫ/Ϫ pups (Fig. 6B). Neither ROR␤ isoform induced expression of any rod gene examined by immunofluorescence or qPCR analysis (Fig. 6D). These results indicate that the activation of rod genes in conelike precursors by ROR␤1 and ROR␤2 depends upon the induction of NRL.
NRL-dependent Expression of ROR␤2-Rod differentiation requires the sequential induction of NRL and NR2E3, but the place of ROR␤2 and ROR␤1 in this hierarchy is undefined. The Rorb gene is required for Nrl expression, but Western blot analysis demonstrated, unexpectedly, that ROR␤2 was undetectable in Nrl Ϫ/Ϫ mice (Fig. 7A), suggesting that ROR␤2 was downstream of NRL. The residual ROR␤ protein band on the Western blot represented ROR␤1, as indicated by comparison with 2z/2z mice that express only ROR␤1. Analysis by qPCR confirmed the severe loss of ROR␤2 mRNA in Nrl Ϫ/Ϫ mice in postnatal development (Fig. 7B). In contrast, ROR␤1 protein and mRNA levels were largely unchanged in Nrl Ϫ/Ϫ mice at any postnatal stage (Fig. 7, A and B), consistent with ROR␤1 being upstream of NRL in photoreceptor precursor cells. Neither ROR␤1 nor ROR␤2 expression was altered in Nr2e3 rd7/rd7 mutant mice, indicating that expression of both ROR␤ isoforms was independent of NR2E3. These results suggested that ROR␤  NOVEMBER 21, 2014 • VOLUME 289 • NUMBER 47 isoforms were sequentially expressed relative to NRL in the order ROR␤1-NRL-ROR␤2 and, surprisingly, that ROR␤2 itself was a target of NRL.

Retinoid-related Orphan Receptor ␤2 in Rod Differentiation
NRL-dependent Promoter in the Rorb Gene-To determine whether ROR␤2 was induced at the transcriptional level by NRL, we first demonstrated the existence of a transcriptional start site in the putative ROR␤2-specific promoter of the Rorb gene. A major 5Ј end point of ROR␤2 mRNA was localized 74 nucleotides upstream of the ATG initiator codon of ROR␤2 by rapid amplification of 5Ј complementary DNA ends made from mouse retinal RNA (Fig. 8A). Further analysis of ChIP sequencing (ChIPseq) data of NRL-bound chromatin from mouse retinas at P21 (38) revealed two peaks located in a 5-kb region upstream and a 2-kb region downstream of the ROR␤2-specific exon of the Rorb gene. No other major NRL-bound peaks were detected over the ϳ200-kb span of the Rorb gene.
To investigate the functional consequence of NRL interaction with the ROR␤2 promoter, a reporter (Rorb2p-GFP) carrying 5.0-kb promoter and 0.6-kb intron fragments that included the two NRL-bound regions mapped by ChIPseq was electroporated into retinal explants of neonatal ϩ/ϩ mice (Fig.  8B). After 8 days in culture, GFP expression was detected in the ONL and colocalized with rhodopsin (data not shown) and NRL, indicating expression in rods (Fig. 8C). GFP ϩ cells displayed a rod-like morphology with projections extending to the outer edge of the ONL and to the outer plexiform layer. Rorb2p-GFP expression was not detected in explants from Nrl Ϫ/Ϫ mice but was recovered by coelectroporation with an NRL expression plasmid (Ub/NRL) (Fig. 8D). Most of these GFP ϩ cells also displayed a recovered expression of endogenous NR2E3, indicating that NRL induced the expression of both the ROR␤2 promoter of the Rorb gene and the endogenous Nr2e3 gene in the same rod precursor cells (Fig. 8E).

Mutual Activation of Rorb and Nrl Genes in Rod
Differentiation-We report that the Rorb gene displays a novel form of sequential expression of two orphan receptor isoforms, ROR␤1 and ROR␤2, during the process of induction of Nrl, a key gene that dictates rod differentiation (Fig. 9). We propose that, after FIGURE 6. Induction of Nrl and rod gene expression by ROR␤ isoforms. A, retinal explants from ϩ/ϩ or Rorb Ϫ/Ϫ mice at P0 electroporated with ROR␤1 or ROR␤2 expression vectors. After 8 days in vitro, electroporated cells were identified with Tdt marker (purple), and NRL, NR2E3, and rhodopsin were detected by immunostaining (turquoise; or double-positive, whitish). Rorb Ϫ/Ϫ explants severely lacked NRL ϩ , NR2E3 ϩ , and rhodopsin ϩ cells, but substantial expression of these markers was recovered by electroporation of Ub/ROR␤1 or Ub/ROR␤2. INL, inner nuclear layer. Scale bar ϭ 25 m. B, similar electroporations with ROR␤2 or ROR␤1 (data not shown) expression vectors in Nrl Ϫ/Ϫ explants failed to recover NR2E3 or rhodopsin expression. C, percentage of electroporated cells that express rod markers in explants from Rorb Ϫ/Ϫ mice counted per 160-m length of retina. p Ͻ 0.001 for Ub/ROR␤1 or Ub/ROR␤2, each versus Ub/empty. D, analysis by qPCR showing reinduction of rod genes (rhodopsin, Nrl, Nr2e3, Pde6b, and Mef2c) after electroporation of Ub/ROR␤1 or Ub/ROR␤2 in Rorb Ϫ/Ϫ explants but not in Nrl Ϫ/Ϫ explants. A control bipolar cell gene, Chx10, did not vary in response to ROR␤ isoforms in any genotype. In a control electroporation in Nrl Ϫ/Ϫ retina, Ub/NRL expression vector reinduced mRNA for rod markers (data not shown).
the initiation of NRL expression by the ROR␤1 isoform together with the homeodomain factors OTX2 and CRX (39,40), NRL exerts dual actions to stabilize rod fate. First, NRL induces ROR␤2, which, in turn, reinforces the induction of NRL through a mutual activation loop. This positive feedback might elevate NRL expression to a threshold, or sustain expression at a threshold, that biases the precursor cell to a rod fate. The second action of NRL is the induction of NR2E3 (9, 10), which promotes the activation of downstream rod genes and suppression of cone genes (11)(12)(13)(14)(15). On the basis of the developmental expression patterns of ROR␤1 and ROR␤2, it is possible that, in early differentiating rods, NRL is induced by ROR␤1 and, in the later expanding rod population, by both ROR␤ isoforms because ROR␤2 is induced by NRL in the neonatal phase.
The proposed model resembles other systems in which feedback loops bias a poised cell to commit to one of two alternative fates (41). Feedback loops have also been reported to stabilize photoreceptor cell fates in the Drosophila compound eye (42). An interesting speculation is that the self-reinforcing activation of the Rorb and Nrl genes may be part of the counting mechanism that produces a greater number of rods than cones in the mouse retina. In contrast, a simple linear induction process without such feedback may be less decisive in directing a rod outcome, a proposal that is supported by the elevated cone:rod  Location of potential NRL response element (NRE, turquoise ovals) and CRX response element (CRE, purple diamonds) in the promoter and intron around the ROR␤2-specific exon with nucleotide coordinates noted relative to the ATG (ϩ1). Location of NREs is on the basis of NRL ChIPseq peaks and CREs on a consensus TAATCC motif and location of CRX ChIPseq peaks (39). Triangle, ROR␤2 mRNA 5Ј end mapped by 5Ј rapid amplification of cDNA ends. B, ROR␤2 promoter-GFP reporter (Rorb2p-GFP) carrying promoter and intron regions with the NREs and CREs shown in A. A schematic for electroporation and culture of retinal explants is shown below. C, Rorb2p-GFP expression (turquoise) in rod precursors in electroporated retina of ϩ/ϩ mice. Almost all GFP ϩ cells were NRL ϩ (purple, or whitish for doublepositive). Scale bar ϭ 20 m. INL, inner nuclear layer. Right panel, GFP ϩ cell showing rod-like morphology. Scale bar ϭ 5 m. D, Rorb2p-GFP lacked activity in the retinas from Nrl Ϫ/Ϫ pups but was reinduced by coelectroporation with the Ub/NRL expression vector. Electroporated cells were identified by the Tdt marker (purple, or whitish for double-positive). E, expression of electroporated Rorb2p-GFP (turquoise) and endogenous NR2E3 (purple) was not detected in Nrl Ϫ/Ϫ mice, but both were recovered in the same cells (whitish) by coelectroporation of Ub/NRL. Scale bar ϭ 20 m. ratio resulting from loss of either one of the two ROR␤ isoforms.
Our model predicts that deletion of either ROR␤ isoform might give a greater increase in the cone:rod ratio than the ϳ2-fold shift observed in 1g/1g and 2z/2z mice. Why is this not the case, and why is there not a major loss of Nrl expression? A likely explanation is that transcriptional compensation masks a full-blown phenotype. For example, in ROR␤1-deficient mice, OTX2 and CRX may still be able to induce a small amount of NRL that is adequate to initiate the expression of ROR␤2, which, in turn, would more fully induce NRL. We showed that ROR␤1 and ROR␤2 individually are potent inducers of Nrl in retinal explants and can largely substitute for each other in directing the differentiation of the majority of rods in vivo. A technical limitation is that we are currently unable to quantify the fluctuations in levels of component transcription factors in an individual cell and cannot monitor the individual cellular transitions that occur in changing populations of precursors over many days in a complex tissue. Our current data concern average gene expression levels.
NRL also maintains the rod phenotype at mature stages (38,43), but, because ROR␤1 is no longer present in mature rods, how is NRL expression sustained? Potentially, CRX, which stimulates many rod and cone genes (44,45), contributes to the expression of both NRL and ROR␤2. Crx Ϫ/Ϫ mice have partly reduced ROR␤2 expression (data not shown), and the Rorb gene contains CRX binding sites near to the NRL binding sites in the ROR␤2 promoter (39) (Fig. 8A). Crx Ϫ/Ϫ mice also have reduced expression of NRL (40). It is also possible that the mutual activation of NRL and ROR␤2 eventually becomes selfsustaining so that the initial stimulus, ROR␤1, is no longer necessary for NRL expression at later stages.
Differential Functions for ROR␤ Isoforms in Retinal Development-Our evidence indicates that the functions of the Rorb gene are extended by its ability to express distinct isoforms in highly cell-specific and temporally distinct patterns. Rorb shares this property with other developmentally critical nuclear receptor genes, including thyroid hormone receptor ␤ (19) and retinoid-related orphan receptor ␥ (46) genes. A unique feature of the Rorb gene is its sequential expression of both ROR␤1 and ROR␤2 isoforms in rod photoreceptors, which is coordinated by positive feedback by the downstream target, NRL rod differentiation factor. Another form of autoinduction of a nuclear receptor isoform during eye morphogenesis has been observed for the RAR␤2-specific promoter of the retinoic acid receptor ␤ gene in response to retinoic acid and the orphan nuclear receptor TLX (homolog of the Drosophila tailless protein) (47). Our results highlight the role of nuclear receptors, along with basic helix-loop-helix, homeodomain, and other classes of transcription factors (48 -50), in retinal cell differentiation.
Although ROR␤2 is restricted to photoreceptors, ROR␤1 also promotes the differentiation of horizontal and amacrine interneurons (31). An intriguing parallel between the differentiation pathways for rod photoreceptors and interneurons is that, in both cases, ROR␤ isoforms act by induction of a key effector protein: NRL in rods and Ptf1a, a basic-helix-loop-helix factor, in horizontal and amacrine interneurons (51). The ability of ROR␤ isoforms to initiate two such different pathways is facilitated by cooperation with distinct cell-specific factors. Therefore, ROR␤1 induces the Ptf1a gene synergistically with forkhead factor FOXN4 (52), whereas it initially induces Nrl in conjunction with the homeodomain factor OTX2 (28,29,40).
ROR␤1 is also required for the development of phototransduction properties in all photoreceptors because rod or cone electroretinogram responses, including both a and b waves, were absent in ROR␤1-deficient mice. In contrast, ROR␤2-deficient mice retain rod and cone responses. However, cone responses are not enhanced in 2z/2z mice, suggesting that the excess cones lack function for reasons that remain unclear. Nr2e3 mutant mice also possess excess S cones but lack enhanced S cone function (53). In contrast, human NR2E3 mutations produce enhanced S cone responses (54), suggesting that the precise roles of factors immediately downstream of NRL may vary somewhat between species.
Another distinction between 2z/2z and 1g/1g mice is that the excess cones in 1g/1g mice align in the outer zone of the ONL but, in 2z/2z mice, are dispersed over the inner zone of the ONL, perhaps reflecting derivation from differing subpopulations of precursor cells, in accord with the temporally distinct expression patterns of ROR␤1 and ROR␤2. It has been reported that the final zonal residence of a mature photoreceptor in the ONL correlates with its time of generation from undifferentiated progenitor cells (2,3). Further study of ROR␤ isoforms may yield insights into the events that control the timing of differentiation and the migration of photoreceptor precursor cells during maturation of the mammalian neuroretina.