Rorβ regulates selective axon-target innervation in the mammalian midbrain

The development of long-range neuronal connections employs multiple mechanisms. Although axon pathfinding, topographic mapping and laminar-specific connections have been extensively studied, some related phenomena, including target selection, remain poorly understood (Dickson, 2002; Sanes and Yamagata, 2009). Axon-target selection is the process of axons choosing their targets and avoiding the adjacent ones. Multiple studies using invertebrates and periphery-to-brain connections have uncovered the basic mechanisms of axon-target selection (Dickson, 2002). However, how such selection is executed within the mammalian brain remains unclear.

The superior colliculus (SC) is a midbrain center that regulates orienting behaviors and defense responses to threat (Basso and May, 2017; Cang et al., 2018). Previous studies revealed that the SC projects axons to several subcortical areas (Ahmadlou et al., 2018; Bickford et al., 2015; May, 2006; Shang et al., 2018; Wei et al., 2015). Projections to the dorsal lateral geniculate nucleus (dLGN) are uniquely localized within the nucleus and possibly regulate direction selectivity. Projections to the lateral posterior nucleus (LP) and the parabigeminal nucleus (PBGN) control visual cue-triggered behaviors, such as freezing and escaping.

Traditionally, developmental studies have focused on the arrangement of sensory inputs to the SC, including topographic map formation (Constantine-Paton et al., 1990; Feldheim and O'Leary, 2010). However, little is known about the mechanisms regulating the development and output pathways of SC neurons.

To examine such mechanisms, we previously sought molecules labeling specific types of SC neurons (Byun et al., 2016), focusing on the superficial layer of the SC (sSC), which receives visual inputs from the retina and cortex and projects axons to subcortical areas. We discovered that a transcriptional factor, retinoid-related orphan receptor β (Rorβ), is highly expressed within the sSC. As sSC neurons from specific sublayers selectively innervate different thalamic nuclei (Harting et al., 1991; May, 2006), we hypothesized that sSC Rorβ⁺ neurons may project axons to distinct midbrain areas.

Several studies have demonstrated the importance of Rorβ in the nervous system (Abraira et al., 2017; Jabaudon et al., 2012). Global Rorb mutants show impaired circadian behavior and neuronal differentiation (André et al., 1998; Liu et al., 2013; Oishi et al., 2016), and disruption of the human RORB locus has been reported in neurodevelopmental disorders (Baglietto et al., 2014; Rudolf et al., 2016). However, little is known about the role of Rorβ in the establishment of the long-range axonal projections (Moreno-Juan et al., 2017).

We assessed the role of Rorβ in sSC neuronal development using global and midbrain-specific Rorb mutants. A novel genetic technique that allows targeted GFP expression was used to visualize Rorβ⁺ neurons. We found that they project axons to distinct subcortical areas: dorsal LGN (dLGN), ventral LGN (vLGN), pretectum (PT), PBGN and LP. Rorb loss led to a severe decrease in dLGN projections and an increase in LP projections; no obvious changes were detected in other areas. Rorb overexpression increased projections to the dLGN and decreased projections to the LP. Altogether, our study identifies a molecular mechanism regulating sSC target selection and sSC-thalamic connections.

To assess the role of Rorβ in sSC development, we performed basic characterization of Rorβ⁺ neurons. We previously reported that Rorβ⁺ somas are localized at the top layer of the sSC (Byun et al., 2016). Here, we confirmed this by analyzing axonal projections of retinal ganglion cells (RGCs) using cholera toxin β subunit (CTB)-conjugated fluorescent dyes (Fig. 1A,C). Strong CTB labeling marks the axonal arborization layer of RGCs (stratum griseum superficiale, SGS), and weak labeling marks the axonal bundle layer (stratum opticum, SO). To identify Rorβ⁺ neurons, we used Rorb^1g/+ mice, in which the reading frame of GFP replaces the Rorb1-specific exon leading to complete loss of Rorb1 mRNA (Liu et al., 2013; Fig. 1B). Here, we refer to Rorb^1g/+ as Rorb^+/− mice, as Rorb1 is a predominant isoform in the sSC (Byun et al., 2016). CTB labeling showed that Rorβ⁺ neurons occupy the upper layer of the sSC (SGS; Fig. 1C).

We then examined expression of Rorb in the sSC during development (Fig. S1A,B). At embryonic day (E)13, Rorb is expressed by a few cells in the sSC, an intermediate zone (IZ) (Edwards et al., 1986). We used a Ki67 antibody, which labels proliferating cells, to visualize the ventricular and subventricular zones (VZ/SVZ) of the SC. Rorb signal was barely detectable in the VZ/SVZ, suggesting that expression occurs mostly post-mitotically. At E16, Rorb expression was increased in the top layer, close to the pia, and remained there at postnatal day (P)1-P6.

Layer-restricted expression of Rorb suggested that sSC Rorβ⁺ neurons may innervate distinct subcortical targets. As GFP-based tracing was challenging because of strong widespread GFP signal in Rorb^+/− brains, we utilized a novel strategy exploiting a GFP-dependent recombinase, FLP-DOG (Tang et al., 2017) and FLP-dependent mCherry expression (fDIO-mCherry). FLP-DOG contains destabilized GFP-binding proteins (dGBP1) and codon-optimized FLP (Flpo). Binding to GFP protects dGBP1 from degradation resulting in active Flpo expression (Fig. 1D). fDIO-mCherry contains two pairs of incompatible FLP recognition (FRT) sequences flanking an inverted mCherry. Interaction of stabilized Flpo with FRT places mCherry into the forward orientation, allowing expression. We generated adeno-associated viruses (AAVs) that express FLP-DOG and fDIO-mCherry and co-injected them into the sSC of Rorb^+/− mice at P0-P1. Immunostaining with anti-GFP and anti-mCherry at ∼P20 revealed that ∼98% of mCherry⁺ cells expressed GFP (n=745 cells, 3 mice), suggesting high specificity of the FLP-DOG strategy. We found that Rorβ⁺ neurons innervate several subcortical areas (Fig. 1E-I): dLGN, vLGN, PT and PBGN. Axonal projections to the vLGN and PT were less abundant. Projections to the LP were detectable as long as medial sSC neurons were mCherry labeled.

To assess the percentage of Rorβ⁺ neurons among the total sSC population projecting to each target, we employed retrograde labeling. We chose to focus on the dLGN and LP – the target areas showing the most and least abundant projections, respectively. Neurons were labeled by injecting CTB-555 into the dLGN or LP of adult mice (Fig. S1C-H). We found that most dLGN-projecting sSC neurons were located at the SGS, and ∼62% of the labeled cells were GFP⁺ (Rorb expressing). Most LP-projecting sSC neurons were located at the SO and only ∼13% of them were GFP⁺. Our results are consistent with previous findings that neurons in the upper layers of the sSC project mainly to the dLGN but barely to the LP (Harting et al., 1991; May, 2006).

Based on temporal and spatial expression of Rorb, we hypothesized that Rorβ might control neuronal positioning or postmitotic differentiation, including axonal projections. To test these hypotheses, we conducted loss-of-function experiments.

No obvious differences were observed in the cytoarchitecture of the SC in control and Rorb mutants at P1 (Fig. S2A-D). Similar to the control, Rorβ⁺ neurons were enriched in the upper layer of the sSC of the mutants. However, the Rorβ⁺ neuronal layer was visibly narrower in the mutants at P12 (Fig. S2E-G). Examination of neurofilament (SMI312) staining demarcating the layer below the SGS (SO) and quantification of Rorβ⁺ neurons confirmed a decrease of neuronal distribution and number. These results suggest that Rorb loss might affect late developmental aspects, such as axonal projections.

To examine whether Rorb is required for sSC axon development, we injected AAV-FLP-DOG and AAV-fDIO-mCherry into the SC of E15.5 embryos and examined the brains at P2. Because of no prior reports on sSC output pathways, time points were chosen based on the Rorb expression and cortical projection studies (Arlotta et al., 2005; Galazo et al., 2016). Very few, if any, labeled cells were found in the sSC (not shown). As this could be due to weak embryonic GFP expression insufficient to produce enough stabilized FLP-DOG (Fig. 1D), we injected AAVs at P0-P1 and examined brain tissues at P10-P11. Owing to obvious changes in Rorβ⁺ neuronal number and distribution at P12, analysis was not conducted beyond this stage. We discovered a strong decrease in projections to the dLGN in the mutants (Fig. 2A-D,I), which was most pronounced in the rostral dLGN. In contrast, projections to the LP were increased. Quantification of mCherry fluorescence (Materials and Methods) revealed ∼87% decrease in projections to the dLGN and ∼5-fold increase in projections to the LP in the mutants. Axonal innervation in other areas, including the PBGN, showed no obvious changes (Fig. 2E-I). Projections to the vLGN and PT were not quantified because the boundaries between axon bundles and terminals were less clear. Examination of the other areas failed to detect any mistargeting or random stalling of axons. Altogether, our data indicate that Rorb deletion strongly affects sSC-thalamic projections.

Changes in the sSC Rorβ⁺ neuronal layer and alterations of axonal projections in the Rorb mutants might be directly related to each other. However, such changes could be caused by general retinal disorganization (André et al., 1998; Sakurai and Okada, 1992; Smith and Bedi, 1997). To rule this out, we employed a conditional knockout (cKO) strategy.

We utilized a Rorb mutant, in which an exon encoding a part of the DNA-binding domain is flanked by loxP sites (Rorb^flox/flox; Fig. S3A). Engrailed1-Cre (En1-Cre) mice were used as colliculus-specific Cre line (Dhande et al., 2012; Kimmel et al., 2000). To validate the En1-Cre line, we crossed it to the Ai14 line expressing Cre-dependent td-Tomato (Madisen et al., 2010) and confirmed the restricted Cre expression (Fig. S3B). To demonstrate ablation efficiency, we crossed Rorb^flox/flox to Rorb^+/−; En1-Cre (Rorb cKO) and confirmed Rorb deletion by RT-PCR (Fig. S3C).

We examined whether conditional deletion of Rorb affects the development of sSC neurons. Given that Rorb expression is detected at E13, we examined early developmental aspects, including cell survival and laminar positioning. Quantification of cleaved-caspase 3⁺ cells at E16 and P1, and NeuN (Rbfox3)⁺ cells at P12 showed no difference between genotypes, suggesting that Rorb is not required for neuronal survival (Fig. S3D-F). Quantification of Rorβ⁺ neurons throughout development (Fig. S3G,H) demonstrated that in both controls and mutants, the number of Rorβ⁺ neurons decreased from embryonic to postnatal stages. We found no differences in layer distributions between genotypes (Fig. S3I). Together, these results demonstrate that conditional deletion of Rorb does not disrupt SC development and that Rorb is not required for cell survival and migration in the sSC.

Next, we examined alterations of sSC neuronal projections in conditional mutants. Using the FLP-DOG strategy, we discovered that, similar to the global knockout, Rorb loss led to a decrease in axonal innervation in the dLGN and an increase in axonal innervation in the LP of cKO mice at ∼P20 (Fig. 3A-C). Quantification of mCherry fluorescence revealed ∼66% decrease of projections to the dLGN and ∼2.4-fold increase to the LP. No obvious changes were detected in the PBGN, vLGN and PT (Fig. 3D-H). These findings suggest that alterations of axonal projections in the Rorb mutants are caused by Rorβ ablation in SC.

Finally, we injected AAV-FLP-DOG and AAV-fDIO-mCherry into the sSC at P0-P1 and investigated axonal innervation in the dLGN and LP at P4/P5, P7/P8 and P10/P11 (Fig. 3I-O). In controls, axonal projections to the dLGN were detectable at P4, became abundant at P7 and were complete by P10. The cKO mutants showed no obvious projections to the dLGN at P4 and relatively few at P7 and P10. Remarkably, analysis of projections to the LP revealed the opposite phenotype. In controls, axonal targeting (not just passing) was undetectable at P4 but became rather obvious at P7 and P10. However, the mutants showed clear axonal branching at P7. Quantification demonstrated decreased projections to the dLGN (∼85% at P4, ∼89% at P7 and ∼57% at P10) and increased projections to the LP (∼4.2-fold at P4, ∼3.9-fold at P7 and ∼3.5-fold at P10).

To investigate whether mis-expression of Rorb is sufficient to redirect SC neuronal projections to thalamus, we delivered Rorb by electroporation into wild-type mice at E13 and AAV injections at E13 and E15.5. Delivery at E13 produced no interpretable results because of very little labeling in the sSC (electroporation) or excessive labeling in thalamus (AAV). Therefore, we analyzed the results obtained by delivering AAV-Rorβ and AAV-YFP into the SC at E15.5 and characterizing axonal projections at ∼P18 (Fig. 4A). The fidelity of co-expression was validated by injecting AAV-YFP and AAV-mCherry (∼90% of YFP⁺ cells expressed mCherry; n=325 cells; Fig. 4B). Rorb overexpression was confirmed by RT-qPCR (Fig. 4C). Quantification of YFP fluorescence revealed that Rorb overexpression increased projections to the dLGN by ∼1.6 fold and decreased projections to the LP by ∼69% (Fig. 4D-K). This result strongly suggests that Rorb is sufficient to bias axon-target selection of sSC neurons.

Here, we examined how long-range neuronal connections are established during development of the mouse midbrain. Using a novel FLP-DOG genetic strategy, we identified Rorβ as a key molecule regulating sSC-thalamic projections. We cannot completely rule out Rorβ involvement in specifying neuronal identity because of the lack of the markers for other sSC neuronal types. However, the normal layer distribution of Rorβ⁺ neurons in the mutants suggests that Rorβ has no obvious effects on the identity acquisition of neurons projecting to the dLGN/LP.

Based on axonal phenotypes at different time points, we speculate that Rorβ might regulate the developmental timing of sSC projections: Rorb loss seems to delay projections to the dLGN but accelerate them to the LP. Projections to the dLGN, although never reaching the control level by P20, increase over time (∼15% of control at P4 and ∼34% at P20), suggesting a delay in targeting. Reduced projections to the dLGN could be also explained by failed axonal arborization. Although we saw no obvious projections to the dLGN in P4 mutants, supporting the former possibility, single cell analysis or advanced imaging may provide more definitive answers.

Specific recognition molecules crucial for establishing neuronal connections (Maness and Schachner, 2007; Riccomagno and Kolodkin, 2015) could acts as downstream effectors of Rorb. We previously reported that several adhesion molecules are differentially expressed in sSC Rorβ⁺ neurons (Byun et al., 2016). It will be interesting to examine whether their altered expression could phenocopy the projection changes observed in Rorb mutants. As sSC projections to the dLGN have been implicated in direction selectivity (Bickford et al., 2015), it may be important to assess whether Rorb deletion alters this function in sSC or dLGN neurons.

All animal procedures were approved by the Institutional Animal Care and Use Committee at Yale University and were in compliance with federal guidelines. We used both female and male animals in the study. The age of the animals was specified in each experiment. For timed pregnancies, the date of the vaginal plug detection was considered as E0.5.

Generation of the Rorb^1g/+ mouse, in which a reporter construct was knocked into the retinoid-related orphan nuclear receptor β (Rorb) gene, has been described previously (Liu et al., 2013). Briefly, the coding sequence of green fluorescent protein (GFP) replaced the Rorb1-specific exon, leading to complete loss of Rorb1 mRNA in the Rorb^1g/1g mouse. Given that Rorb1 is a predominantly expressed isoform in the sSC (Byun et al., 2016), we referred to Rorb^1g/+ mouse as Rorb^+/− mouse throughout the text. A conditional Rorb mutant, in which an exon encoding the first zinc finger of the DNA-binding domain is flanked by loxP sites (Rorb^flox/flox), was derived by targeted homologous recombination in embryonic stem cells. The Neo selection cassette was deleted by FLP recombinase, then FLP was removed by out-crossing (Ozgene, Australia). For genotyping wild-type and flox alleles, the following primers were used: Rorb-F2, 5′-TTTAGCGGAAGCCCAGGAAGG-3′; Rorb-R5, 5′-TCAAATGGAGTCAGTGTTGC-3′, which yielded a 520 bp band for the wild type, a 720 bp band for the floxed allele and a 210 bp band for deletion of the exon between the loxP sites. For deletion of Rorb in the colliculus, we used engrailed1-Cre (En1-Cre) mice. Cre expression in En1-Cre mice is limited to the midbrain and hindbrain and turns on around E9.5 (Kimmel et al., 2000; Jackson Laboratory, #007916). To validate the properties of the En1-Cre line, we crossed it to the Ai14 line, which expresses Cre-dependent tdTomato (Madisen et al., 2010; Jackson Laboratory, #007914) and confirmed restricted Cre expression in the midbrain and hindbrain. Conditional deletion of Rorb in the Rorb^flox/−; En1-Cre mouse was confirmed by the standard reverse transcription PCR. The primers used for amplification were: forward 5′-TCATCACGTGTGAAGGCTGCAAG-3′ and reverse 5′-ATCCTCCCGAACTTTACAGCATC-3.

Mice were anesthetized by intraperitoneal injection of a combination of 100 mg ketamine plus 10 mg xylazine/kg of bodyweight and perfused transcardially with 4% paraformaldehyde (PFA)/PBS. For embryonic studies, the heads of the embryos were separated from the bodies, post-fixed for 3-4 h at 4°C, then incubated with 15% sucrose/PBS for ∼6 h at 4°C followed by 30% sucrose/PBS overnight at 4°C and sectioned using a cryostat (16-20 µm sections). For postnatal studies, whole brains were dissected, post-fixed overnight at 4°C and prepared for sectioning. For cryosectioning, tissue was incubated with 15% sucrose/PBS followed by 30% sucrose/PBS overnight at 4°C and sectioned at 20-35 µm. For free-floating sectioning, tissue was post-fixed with 4% PFA/PBS overnight at 4°C, washed with PBS and sectioned at 35-50 µm using a vibratome. For quantification of axonal projections to the subcortical areas, 35 µm-thick sections were chosen. For immunostaining, the sections were washed twice for 5 min with PBS, blocked with 3-5% donkey serum/0.1% Triton X-100/PBS for 30 min at room temperature, and incubated with the primary antibodies for 2-3 days at 4°C. Then, the sections were incubated with appropriate secondary antibodies for 2 h at room temperature.

Primary antibodies used were: rabbit anti-GFP (1:1000, Millipore, AB3080P), chicken anti-GFP (1:1000, Aves Laboratories, GFP-1020), rabbit anti-DsRed (1:1000, Clontech, 632496), goat anti-ChAT (1:500, Millipore, AB1440), mouse anti-NeuN (1:1000, Millipore, MAB377), mouse anti-calretinin (1:2000, Milipore, MAB1568), mouse anti-SMI312 (1:1000, BioLegend, 837904), rabbit anti-cleaved caspase 3 (1:1000, Cell Signaling Technology, 9661), rabbit anti-Ki67 (1:500, Thermo Scientific, RM9106). Secondary antibodies were conjugated to Alexa Fluor-488, Cy3 or Cy5 (all from Jackson ImmunoResearch Laboratories; 703-545-155, 711-545-152, 711-165-152, 715-165-151, 715-175-151, 705-175-147) and diluted 1:500.

To generate the plasmid that carries FLP-dependent mCherry, we amplified the mCherry sequence by PCR, created AscI and NheI sites at each end and replaced PCR products with the DNA fragment in the AAV vector (Addgene #74291). The final product contained mCherry between two nested pairs of incompatible FRT sequences (pAAV-CAG-fDIO-mCherry-WPRE-SV40pA). The AAV plasmid carrying FLP-DOG was obtained from Addgene (#75469).

For overexpression experiments, full-length Rorb was excised with EcoRI from the previously generated plasmid Rorb-pCR8/GW/TOPO TA vector (Invitrogen; Byun et al., 2016), treated with Klenow fragment and sub-cloned into the AAV vector (Addgene #18917) blunted after BamHI and EcoRI digestion. Generation of AAV-YFP construct or AAV-mCherry has been previously described (Fink et al., 2017).

Virus production was based on a triple-transfection, helper-free method, and virus was purified as described previously (Park et al., 2015). Briefly, HEK293 cells in exponential growth phase were transfected with the DNAs using polyethylenimine. DNA mixtures contained a plasmid carrying AAV capsid 2/1 genes (UPenn Vector Core), delta F6 plasmid (UPenn Vector Core) and a plasmid carrying the gene of interest. Cells were harvested 72 h after transfection. Viral vectors were purified using a step gradient of iodixanol by ultracentrifugation, buffer-exchanged to PBS, and concentrated using Ultracel (Millipore). The titer was determined by quantitative PCR using primers that recognize WPRE or human growth hormone poly A sequences; the concentrated titers were >10¹³ viral genome particles/ml in all preparations. Viral stocks were stored at −80°C.

To visualize axonal distribution of retinal ganglion cells in the sSC, Rorb^+/− mice were anesthetized by intraperitoneal injection of a combination of 100 mg ketamine plus 10 mg xylazine/kg of bodyweight. A small hole was made in the eye with an insect pin to release intraocular pressure. Cholera toxin B subunit conjugated to Alexa Fluor-555 (CTB-555; 1-2 µl of 1 mg/ml, Invitrogen) was injected through the same hole using a Hamilton syringe. Intraperitoneal injections of the analgesic buprenorphine (0.05-0.1 mg/kg of bodyweight) were given before and after the surgery.

To analyze axonal projections of sSC Rorβ⁺ neurons, we injected AAV-FLP-DOG and AAV-fDIO-mCherry (ratio of 1:5 or 5:1) into the sSC. We kept the 5:1 ratio between FLP-DOG and fDIO-mCherry in most experiments except when labeling the brains that were examined at P4 and at P7. We changed the ratio of FLP-DOG to fDIO-mCherry to 1:5 to increase the level of mCherry expression. Quantification revealed that ∼98% of mCherry⁺ cells express GFP (n=745 cells, 3 mice) at P20 and ∼95% of mCherry⁺ cells express GFP (n=364 mCherry⁺ cells, 3 mice) at P7. These results indicate that the FLP-DOG to fDIO-mCherry ratio did not affect the fidelity of the FLP-DOG strategy. For injections, P0-P1 pups were placed on ice to induce hypothermia, and a small incision was made on the skin over the SC. The skull and brain tissue were simultaneously penetrated with disposable glass pipettes containing AAVs. Then ∼200 nl of AAVs were administrated by a pressure injector. After injection, pipettes were gently withdrawn to prevent backflow and the skin was sealed using cyanoacrylate glue. Oral injections of nonsteroidal anti-inflammatory analgesics (meloxicam, 0.3 mg/kg of bodyweight) were given before and after the surgery for 2 days. Animals were sacrificed at P4/P5, P7/P8, P10/P11 and P20/P21. The brains were dissected, sectioned and immunostained with antibodies to GFP (to amplify Rorβ⁺ signal) and DsRed (to amplify mCherry signal). Axonal projections of the labeled neurons were examined in sections of the entire brain extending from the suprachiasmatic nucleus to the cerebellum.

To retrogradely label sSC neurons that project to the dLGN or LP, we used a fluorescent dye (CTB-555). For stereotaxic injections, the mice were anesthetized by intraperitoneal injection of a combination of 100 mg ketamine plus 10 mg xylazine/kg of bodyweight and a small craniotomy was made over the dLGN or LP. Coordinates used for dLGN injection were bregma −2.3 mm, lateral ±2.5 mm, dura −2.5 mm. Coordinates used for LP injection were bregma −2.0 mm, lateral ±1.8 mm, dura −2.4 mm. Fluorescent dye (∼200 nl) was injected with a glass pipette at a rate of ∼15 nl/min and the pipette was left in place for 5 min after injection. Intraperitoneal administration of the analgesic buprenorphine (0.05-0.1 mg/kg of bodyweight) was given before and after the surgery. Two days later, we dissected the brain and performed immunostaining with an antibody to GFP to detect Rorβ localization. Labeled sSC neurons (CTB⁺ or CTB⁺ and GFP⁺) were quantified in each animal using three coronal sections, 200 µm apart. Neurons were counted in areas of equal size within the colliculus.

In utero injections were conducted following an in utero electroporation procedure previously described (Saito, 2006) without electric shock. For overexpression experiments, pregnant CD-1 females (E15.5) were anesthetized by intraperitoneal injection of a combination of 100 mg ketamine plus 10 mg xylazine/kg of bodyweight. The embryos were exposed after laparotomy and a mixture of AAV-Rorβ and AAV-YFP (1:1 ratio) or AAV-YFP alone (1:1 with PBS) were administered into the SC of each embryo by a pressure injector. Intraperitoneal injections of the analgesic buprenorphine (0.05-0.1 mg/kg of bodyweight) were given before and after the surgery. Embryos were allowed to develop and pups were sacrificed at P18. The brains were dissected, sectioned and immunostained with antibodies to GFP (to amplify YFP signal) and calretinin (to delineate dLGN and LP). For quantification, the number of labeled neurons in the sSC and fluorescence intensity of axonal terminals in the thalamic nuclei were analyzed.

To examine co-transfection efficiency, AAV-YFP and AAV-mCherry (1:1 ratio) were injected. For quantification, sectioned brains were immunostained with antibodies to GFP and DsRed.

To assess overexpression of Rorb after in utero injections, the brains were dissected at P2 and the SCs were isolated. Total RNA was prepared using the RNeasy Mini Kit (Qiagen) and cDNA was synthesized using either the Superscript III First-Stand system (Thermo Fisher) or the iScript gDNA clear cDNA synthesis kit (Bio-Rad). RT-qPCR was performed in duplicate using iQ SYBR Green Supermix (Bio-Rad) on the CFX96 real-time system (Bio-Rad). The Ct values of samples were normalized to that of Gapdh and overexpression levels of Rorb in relation to controls were calculated by the ΔΔCt method. Primers used for qPCR were as follows: Gapdh, 5′-GTGGAGTCATACTGGAACATGTAG-3′ and 5′-AATGGTGAAGGTCGGTGTG-3′; Rorb, 5′-TCATCACGTGTGAAGGCTGCAAG-3′ and 5′-ATCCTCCCGAACTTTACAGCATC-3′.

Images were acquired using a Zeiss Imager M2 fluorescence microscope and a Zeiss LSM 800 confocal microscope. z-stacks were obtained with 1 µm steps using a 20× objective (NA 0.8) and 0.5 µm steps using a 40× objective (NA 1.4). ImageJ software (National Institutes of Health) was used for data analysis.

To quantify axonal projections to subcortical areas, images were taken using 35 µm-thick coronal sections, ∼200 µm apart, extending from rostral to caudal areas of the brain. Sections containing both the dLGN and LP were selected, and the regions of the dLGN and LP covered by axon terminals were demarcated. Fluorescence intensity of each demarcated area was measured using ImageJ as the sum of pixel values after subtracting background signal; the total intensity from two or three sections (depending on the animal's age) was calculated for each nucleus. To account for injection variability, the summed fluorescence intensities from the dLGN and LP sections were divided by the total number of labeled sSC cells that were counted in the sSC sections, ∼200 µm apart, covering the entire infected sSC area. Cell counting for overexpression experiments was conducted in the sSC areas spanning 350 µm from the pia. Data obtained for dLGN and LP in mutants were normalized to controls and presented as percentages.

To quantify the total number and distribution of Rorβ⁺ neurons, anatomically matched sections were selected. For embryonic analysis, images were taken using sections that spanned ∼120 µm from the medial edge to the lateral edge of the ventricular zone. For postnatal analysis, images were taken using sections that spanned ∼200 µm from the medial edge to ∼200 µm from the lateral edge. For cell counting, the images were cropped to 350 µm×450 µm (width×height). Such cropping covers an entire area from the pia to the ventricular zone in embryonic tissues and visual input areas from the pia in postnatal tissues. To analyze the distribution of neurons, images were horizontally divided from the pia and cells were grouped into 25 µm bins.

To quantify retrogradely labeled sSC neurons in adult brains, we adapted counting strategies employed in the analyses of cortical projection neurons (Arlotta et al., 2005; Galazo et al., 2016). Briefly, images were taken using three coronal sections, 200 µm apart. For cell counting, images were cropped at ∼200 µm from the midline to avoid the most medial part where labeling of axons and neurons was mixed. The cropped sizes were either 450 µm×450 µm or 450 µm×600 µm (width×height).

Axonal projections were quantified using two or three different sections for measurement of fluorescent intensity in thalamic nuclei, three to seven different sections for counting the labeled sSC cell number per animal and summed for each of five to seven animals per condition. Total number or layer distribution of Rorβ⁺ or NeuN⁺ neurons were analyzed using three to six different sections per animal and averaged for each of three or four animals per condition. Animals of either sex were collected from two to five different litters for each time point.

All the data are reported as mean±s.e.m. and analyzed using GraphPad Prism software (GraphPad Software). Means between two groups (control and either global or conditional mutants or control and overexpression) were compared using an unpaired two-tailed Student's t-test. The value of n represents the number of animals used per condition. The following significance levels were used: *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

We thank Joshua Sanes (Harvard University) for helpful discussions. This work was supported by a NIH grant (EY026878: a Core Grant for Vision Research for Yale University) and by an unrestricted grant from Research to Prevent Blindness to the Department of Ophthalmology and Visual Science at Yale University.