Photoreceptor Progenitors Depend Upon Coordination of gdf6a, thrβ, and tbx2b to Generate Precise Populations of Cone Photoreceptor Subtypes

Abstract

Purpose: Replacing cone photoreceptors, the units of the retina necessary for daytime vision, depends upon the successful production of a full variety of new cones from, for example, stem cells. Using genetic experiments in a model organism with high cone diversity, zebrafish, we map the intersecting effects of cone development factors gdf6a, tbx2b, and thrβ.

Methods: We investigated these genes of interest by using genetic combinations of mutants, gene knockdown, and dominant negative gene expression, and then quantified cone subtype outcomes (which normally develop in tightly regulated ratios).

Results: Gdf6a mutants have reduced blue cones and, discovered here, reduced red cones. In combined gdf6a/tbx2b disruption, the loss of gdf6a in heterozygous tbx2b mutants reduced UV cones. Intriguingly, when we disrupted thrβ in gdf6a mutants by using a thrβ morpholino, their combined early disruption revealed a lamination phenotype. Disrupting thrβ activity via expression of a dominant negative thrβ (dnthrβ) at either early or late retinal development had differential outcomes on red cones (reduced abundance), versus UV and blue cones (increased abundance). By using dnthrβ in gdf6a mutants, we revealed that disrupting thrβ activity did not change gdf6a mutant cone phenotypes.

Conclusions: Gdf6a loss directly affects blue and red cones and indirectly affects UV cones by increasing sensitivity to additional disruption, such as reduced tbx2b, resulting in fewer UV cones. The effects of thrβ change through photoreceptor development, first promoting red cones and restricting UV cones, and later restricting UV and blue cones. The effects of gdf6a on UV, blue, and red cone development overlap with, but likely supersede, those of thrβ.

Retinal degenerations are characterized by the gradual death of retinal neurons and become especially debilitating upon the loss of photoreceptors, the neurons that detect light. Cone photoreceptors, or cones, respond to high intensity light and are divided into subtypes based on maximal wavelength sensitivity; thus, they are especially important for daytime visual function and color discrimination. The human macula is populated exclusively by cones and is critical for daytime vision and visual acuity; thus, when retinal degeneration occurs in the macula, functional vision is dramatically compromised. Restoring functional vision in humans demands strategies to replace lost cones, but how cones and their subtypes are produced remains poorly understood. Creating functional cone photoreceptors for therapeutic use (e.g., from stem cells) requires better knowledge and manipulation of the regulatory networks in cone development.^1–10 

Much insight into photoreceptor development to date has come from the mouse, which has two cone types (M and S cones) typical of most mammals. Photoreceptor progenitors are characterized by Crx expression; from this pool, Nrl and its target Nr2e3 direct rod photoreceptor fate.¹¹ Thyroid hormone receptor, or Thrβ, is a transcription factor with roles in numerous developing tissues, including the retina, otic vesicle, pituitary, and jaw cartilages.^12–17 In the retina specifically, mouse Thrβ and zebrafish thrβ are required for M cones and red cones, respectively,^18,19 whereas treatment with thyroid hormone itself initiates death and regeneration of UV cones in trout^20,21 and shifts visual pigment content in zebrafish.^22,23 In mice, it is proposed that Thrβ drives the differentiation of M cones, whereas S cones are a default type.²⁴ RAR-related orphan receptor (ROR) and retinoid X receptor (RXR) nuclear receptors can either negatively (RXRγ)²⁵ or positively (RORβ, RORα)^26–28 regulate the expression of S-opsin in mice, which results in a dorso-ventral gradient pattern of S-opsin expression. Overall, the mouse retina is dominated by rods with a low density of cones and, thus, a paucity of cone-cone interactions. By contrast, the human macula is cone dominant and has three cone types, more than most mammals. How three (or more) cone subtypes can be created and what conditions are required to build a cone-rich area like the macula remain unanswered and demand a wider diversity of retina development models. 

Nonmammalian vertebrates have a greater diversity of cones and so offer insight into how multiple cone subtypes are generated. The zebrafish retina is dense with cones, with four subtypes based on maximal light spectrum sensitivity (UV, blue, green, and red), and these subtypes are produced in tightly controlled proportions in larvae and adults. Furthermore, these four subtypes are arranged in a precise row mosaic pattern in the adult retina.^29–37 The four major cone types are conserved in the vertebrate lineage, making zebrafish a highly useful model to study cone development. In zebrafish, T-box transcription factor 2b, or tbx2b, is a transcription factor required for early neuronal differentiation in the dorsal retina³⁸ and, specifically, UV cones (the homologue to mammalian S cones and human blue cones); tbx2b appears to negatively regulate rod abundance.³⁹ Growth differentiation factor 6, or GDF6, is a bone morphogenetic protein (BMP) ligand that is known to bind BMP receptors such as BMPR1 and BMPR2, thereby activating SMAD transcription complexes.^40–43 GDF6 disruption is causal of congenital photoreceptor defects in Leber congenital amaurosis (LCA17).⁴⁴ In zebrafish, the homologue to GDF6 gdf6a is required for establishing dorsal identity in the early retina,^45–47 for regulating retinal cell proliferation,⁴⁸ and for adequate blue cone abundance.³⁵ Gdf6a^s327/s327 mutants are microphthalmic and have reduced blue cone to UV cone abundance ratios. Gdf6a also enhances tbx2b for the correct production of UV cones and rods. Six7, a sine oculis homeobox transcription factor, is required for green cones.^49,50 

In sum, while the murine retina contains two cone types with one developmental transcription factor complex discerning between them,^18,24,51 zebrafish have four cone subtypes produced in exact ratios^{29–31,33,52,53} and so must use a more complex system of progenitor fate choice. Much of the knowledge of zebrafish cone subtype production is expected to be directly relevant to the development of the human macula. We sought to determine if the factors within this system interact to regulate the ratios, or relative abundances, of cone subtypes. In this study, we examined thrβ, gdf6a, and tbx2b. 

Our documentation of the cone phenotype in gdf6a mutants is expanded: gdf6a mutants have reduced blue and red cones but not UV cones, suggesting that gdf6a promotes blue and red cone development. In addition, the interaction between gdf6a and tbx2b is newly detailed by quantifying cone abundances, confirming that gdf6a loss causes progenitors to have reduced tolerance for insufficient tbx2b activity, resulting in fewer cells assuming a UV fate. We find that thrβ has developmental stage-specific roles in red cone and UV and blue cone subtype specification and interacts with gdf6a to direct proper retina lamination. 

Methods

Ethics Statement

All fish care and experiments were approved by the Animal Care and Use Committee: Biosciences at the University of Alberta under protocol AUP00000077 and were in accordance with the Canadian Council on Animal Care. Similarly, the work adhered to the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. 

Animal Care and Embryo Injections

Zebrafish were raised and maintained according to standard procedures.⁵⁴ Embryos were kept in E3 embryo medium at 28°C except where indicated in the experimental procedures. Medium with 0.003% propylthiouracil to prevent pigmentation was applied at 8 to 10 hours post-fertilization (hpf). Mutant lines were gifted as follows: gdf6a^+/s327 (Zebrafish Information network [ZFIN] identifier [ID] ZDB-ALT-050617-10)⁴⁵ from Andrew Waskiewicz (University of Alberta) and tbx2b^+/fby (ZFIN ID ZDB-ALT-070117-1)^55,56 from Josh Gamse (Vanderbilt University). Transgenic lines used were Tg(-5.5opn1sw1:EGFP)^kj9 (ZDB-ALT-080227) and Tg(-3.5opn1sw2:mCherry)^ua3011 (ZDB-TGCONSTRCT-130819-1).^34,35,57,58 For thrβ knock down experiments, embryos were injected with 10 ng of either splice-blocking thrβ morpholino (MO) (5′-TCTAGAACTTGCAATACCTTTCTTA-3′) (ZFIN ID ZDB-MRPHLNO-131114-1) (predicted to retain the intron between exons 1 and 2)^19,59 or standard control MO (5′-CCTCTTACCTCAGTTACAATTTATA-3′)⁶⁰ at the 1- to 2-cell stage and then were maintained as above. 

Generating Transgenic dnthrβ Zebrafish

Dominant negative thyroid hormone receptor beta was generated by expressing a version of the cDNA modified to lack 12 amino acids from the carboxy terminus, thereby removing the ligand binding and coactivator binding sites, as previously accomplished in Xenopus.^61,62 Primers used in construct creation were based upon dnthrβ mRNA of the National Center for Biotechnology Information Reference Sequence NM_131340.1 and are 5′-GGG GAC AAG TTT GTA CAA AAA AGC AGG CTC AGT ATG TCA GAG CAA GCA G-3′ (forward) and 5′-GGG GAC CAC TTT GTA CAA GAA AGC TGG GTG AGC TCT GTG GGA CAT TCC-3′ (reverse). Transgenic constructs were created using multisite Gateway cloning into vectors suitable for Tol2 recombination and transgenesis.⁶³ The sequence encoding dominant negative thyroid hormone receptor beta, or dnthrβ, was amplified from zebrafish cDNA and recombined with the hsp70 promoter sequence (allowing for conditional expression of the dominant negative receptor) and V2A.NLS.eGFP (for visualization of expression) to create the plasmid pDestTol2CG2.hsp70:dnthrβ.V2A.NLS.eGFP. The V2A.NLS.eGFP sequence was kindly provided by Steven Leach.⁶⁴ This construct was verified by sequencing and injected with Tol2 mRNA into zebrafish embryos at the 1- to 2-cell stage, and transient transgenic fish were raised to establish a stable transgenic line (line designation ua3113). To visualize the UV and blue cones, Tg(hsp70:dnthrb.V2A.NLS.eGFP)^ua3113 fish were crossed to Tg(-5.5opn1sw1: EGFP)^kj9; Tg(-3.5opn1sw2: mCherry)^ua3011 carriers.^34,35,57,58 

Heat shock was performed at the ages indicated by placing petri dishes (containing embryos in E3 medium) in a water bath set to 37°C for two hours. Under these conditions, the measured E3 medium temperature reached 34°C. Transgene expression was confirmed through GFP expression visualized 4 to 6 hours following heat shock. 

Genotyping for gdf6a and tbx2b Mutations

Genotyping of adult and larval fish was performed via restriction fragment length polymorphism. Genomic DNA was extracted from fin clips from adults and whole larvae as previously described by Meeker et al.,⁶⁵ and regions containing the gdf6a^s327 and tbx2b^fby loci were PCR amplified by using the primers designed by Gosse and Baier⁴⁵ for gdf6a^s327 (forward, 5′-ATGGATGCCTTGAGAGCAGTC-3′; reverse, 5′-CTACCTGCAGCCACACTGTTC-3′) and Snelson et al.⁶⁶ for tbx2b^fby (forward, 5′-TGTGACGAGCACTAATGTCTTCCTC-3′; reverse, 5′-GCAAAAAGCATCGCAGAACG-3′). PCR products underwent restriction digest with BmsI (gdf6a) and SaqAI (tbx2b) (FastDigest, FD2124 and FD2174) and were run on a 3% agarose gel. The gdf6a^s327 mutation is detected by the loss of a BmsI restriction site, such that the wildtype product is digested into 170- and 110-bp bands, and the mutant product remains a single 280-bp band. The tbx2b^fby mutation introduces a novel SaqAI restriction site; the wildtype product runs as a single 206-bp band, and the mutant product results in 169- and 37-bp products. 

Immunocytochemistry and Imaging

Larvae were fixed at 4 days post-fertilization (dpf) in 4% paraformaldehyde overnight, and whole-mount immunocytochemistry was performed as previously described.^35,36 Briefly, larvae underwent washes in 0.1 M PO₄/5% sucrose, then in 1% Tween/H₂O (pH 7.4), and then in −20°C acetone. Larvae were incubated with 10% normal goat serum/PBS³⁺ for 60 minutes to reduce nonspecific antibody binding and incubated overnight at 4°C in PBS³⁺ containing 2% normal goat serum (NGS) and antibody. Following incubation in primary antibody, larvae were washed with PBS³⁺ and incubated in secondary antibody overnight at 4°C. 

Immunohistochemistry on cryosections was performed as follows: larvae were fixed at 4 dpf in 4% paraformaldehyde overnight and dehydrated step wise in 0.1M PO₄ with increasing concentrations of sucrose. Larvae were then embedded with frozen section compound (catalog no. 95057-838; VWR, Radnor, PA, USA), and the eyes were cryosectioned (10 μm per section) and mounted on Superfrost Plus slides (catalog no. 12-550-15; Fisher Scientific, Waltham, MA, USA). Sections were stored at −80°C until used, where they were rehydrated with PBS with Tween 20 (PBSTw), blocked in 10% NGS/PBSTw for 30 to 90 minutes, and incubated in primary antibodies diluted in 2% NGS/PBSTw overnight at 4°C. Slides were subsequently washed and incubated in secondary antibodies overnight at 4°C. All immunohistochemical steps were performed in a humid chamber. 

Monoclonal primary antibodies and dilutions used were mouse 1D4 against zebrafish red cones (1:500; ZFIN ID: ZDB-ATB-110114-2; catalog no. b5417; Abcam, Cambridge, MA, USA) and rat 10C9.1 against UV opsin (1:100; ZFIN ID: ZDB-ATB-140728-2).³⁵ Secondary antibodies/stains and dilutions used were Alexafluor 647 chicken anti-mouse (1:1000; catalog no. A-21463; Invitrogen, Carlsbad, CA, USA), Alexafluor 488 chicken anti-rat (1:1000; catalog no. A-21470; Invitrogen), and 4′,6-diamidino-2-phenylindole (catalog no. D1306; Invitrogen). Retinas were then dissected and flat mounted. Imaging of flat-mount and sectioned tissues was performed using a Zeiss LSM 700 scanning confocal microscope and Zen 2010 software (Carl Zeiss, Oberkochen, Germany). 

Image Analysis and Statistical Analysis

Images of flat-mounted retinas were analyzed in ImageJ (https://imagej.nih.gov/ij/index.html, Wayne Rasband, National Institutes of Health, Bethesda, MD, USA). In each retina image, cone subtypes were counted within a 100 μm × 100-μm area (drawn as a square dorsal of the optic nerve head). For gdf6a mutants, most flat-mounted retinas were amenable to counts in a 100 μm × 100-μm area, so they were used for analysis. Raw values of cone subtype counts are available in the Supplementary Figures. Cone subtype abundance data were analyzed via Kruskal-Wallis tests with Mann-Whitney pairwise comparisons in Stata/SE 14.1 for Mac (2015, StataCorp, College Station, TX, USA). Sample sizes (n) reported in the figures represent the number of larval fish examined. 

Results

Gdf6a Influences tbx2b in UV Cone Development

We had previously investigated the interaction of gdf6a and tbx2b in photoreceptor development by using the tbx2b^lor and tbx2b^fby mutants (hypomorphic and null alleles, respectively). Tbx2b^lor/lor mutants display the “lots-of-rods” phenotype with an overabundance of rods and a small number of UV cones, and tbx2b^fby/fby mutants have lots of rods and a near-complete absence of UV cones.³⁹ Although zebrafish heterozygous for tbx2b^lor or tbx2b^fby have typical rod and UV cone abundances, in a gdf6a^s327/s327 background (often abbreviated here as gdf6a^−/−), some tbx2b^+/lor or tbx2b^+/fby embryos have an overabundance of rods.³⁵ Here, we reexamined the effects of this interaction and expanded upon it by assessing abundances of UV, blue, and red cones. 

Quantification of cone abundances confirmed phenotypes that have been documented previously, including homozygous tbx2b^fby/fby mutants that lack UV cones³⁹ and gdf6a^s327/s327 mutants that have a reduction in blue cones (39% fewer than wildtype, P = 0.003). Additionally, we found that gdf6a^s327/s327 mutant retinas have a novel phenotype of fewer red cones (29% reduction, P = 5.16 × 10⁻⁵) (Fig. 1; abundances are shown as ratios to control values, which are normalized to 1, and average values of raw counts are available in Supplementary Fig. S1). In our previous publication, only the UV and blue cones were examined and quantified,³⁵ so this red cone phenotype had not been appreciated. Cone abundances of fish heterozygous for either gdf6a^s327 or tbx2b^fby were not different from wildtypes (Fig. 1). 

Figure 1

 

gdf6a loss in tbx2b heterozygous fby carriers reduces UV cone abundance. (A) Relative abundances of UV, blue, and red cones in compound gdf6a^+/s327; tbx2b^+/fby in-crosses show near-complete loss of UV cones in tbx2b^fby/fby mutants and partial loss of UV cones in tbx2b^+/fby mutants with gdf6a^s327/s327 background (labeled gdf6a^−/− or “−/−”). Values with matching letters are not significantly different (Kruskal-Wallis test with Mann-Whitney pairwise comparisons, gray asterisks indicate *P < 0.05, **P < 0.01, ***P < 0.001 relative to wild type (WT)/WT controls). Gdf6a^s327/s327 mutants have reduced blue cones and a novel reduced red cone abundance phenotype, both of which appear either unaffected or slightly enhanced by tbx2b loss. Wild-type values shown are sibling controls. (B) Representative retinal images depicting UV (magenta), blue (cyan), and red cones (green) in wild type, gdf6a^s327/s327, tbx2b^fby/fby, and compound gdf6a^s327/s327; tbx2b^+/fby mutants.

Fish heterozygous for tbx2b^fby but lacking gdf6a (gdf6a^s327/s327; tbx2b^+/fby) had a 23% reduction in UV cone abundance compared to wild types, a partial but statistically significant reduction (P = 0.014) (Fig. 1). For gdf6a^s327/s327 mutants, blue cones were reduced further with the tbx2b^fby mutation (81% reduction, P = 8.16 × 10⁻⁵ in gdf6a^s327/s327; tbx2b^+/fby and 63% reduction, P = 6.49 × 10⁻⁶ in gdf6a^s327/s327; tbx2b^fby/fby). Red cone abundance remained low in all gdf6a^s327/s327 retinas, but interestingly, this abundance was reduced further with sequential accumulation of tbx2b mutant alleles (51% reduction, P = 8.16 × 10⁻⁵ in gdf6a^s327/s327; tbx2b^+/fby and 64% reduction, P = 2.37 × 10⁻⁶ in gdf6a^s327/s327; tbx2b^fby/fby). 

Thrβ Knock Down in gdf6a Mutants Disrupts Cone Abundances and Retinal Lamination

Based on our cone quantifications in gdf6a mutants, Gdf6a signaling positively influences red cone and blue cone development. Although the loss of gdf6a alone does not alter UV cone abundances, it appears to increase sensitivity toward UV cone loss, as seen in tbx2b heterozygous mutants. Thrβ is required for red cone specification, but it also negatively regulates UV cones, as thrβ morphants have a UV cone excess.¹⁹ Therefore, gdf6a and thrβ have similar effects on red cone populations but differing effects on UV cone populations. 

We wanted to resolve if gdf6a and thrβ act in shared or separate regulatory pathways toward UV cone fate. We generated two hypotheses: (1) gdf6a and thrβ act independently in UV cone development, where Gdf6a signaling promotes UV cones indirectly (possibly linked by tbx2b), and Thrβ suppresses UV cones separately from that; or (2) UV fate is regulated by gdf6a and thrβ in an epistatic fashion. For hypothesis 1, we predicted that reduction in both factors, and reduction of their opposing influences, would result in a wild-type abundance of UV cones. For hypothesis 2, we predicted that the reduction of both would cause either excess of UV cones (if thrβ knock down can override any sensitization to UV cone reduction as caused by gdf6a loss) or a reduced abundance of UV cones (if thrβ knock down does not rescue sensitization). To test these predictions, we knocked down thrβ in gdf6a^s327/s327 mutants with a thrβ splice-blocking morpholino, which was previously used by others.¹⁹ 

Disruption of thrβ expression with 10 ng of morpholino in gdf6a^s327/s327 mutants yielded a surprising set of results. When we examined whole-mount retinas with antibody-labeled UV and red cones, thrβ morphants showed a consistent lack of red cone labeling regardless of gdf6a genotype, assuring us of the morpholino's efficacy (91% and 97% decrease in wild type and gdf6a^s327/s327, respectively; P = 3.64 × 10⁻⁵, P = 5.53 × 10⁻⁵, respectively). In gdf6a^s327/s327 mutant retinas receiving thrβ morpholino, the low blue cone phenotype persisted and, in fact, dramatically worsened (98% decrease, P = 0.00005). In the UV cone channel, gaps or holes in the photoreceptor layer were apparent in these gdf6a^s327/s327; thrβ morphants, which affected UV cone abundance counts (Fig. 2A, 2B; average values of raw counts available in Supplementary Fig. S2). Radial sections of 4dpf gdf6a^s327/s327; thrβ morphant retinas revealed lamination defects impacting all retinal layers. Gaps in the photoreceptor layer were confirmed, and “bridges” of cells were found spanning the inner plexiform layer between the inner nuclear layer and ganglion cell layer (Fig. 2C). These lamination defects were not detectable when either thrβ or gdf6a expression were disrupted individually. This lamination phenotype suggests thrβ and gdf6a may be active in an early retinal development process, such that the loss of one or the other has no overt effect, but the loss of both negatively affects photoreceptor populations and disrupts lamination. 

Figure 2

 

Knockdown of thrβ disrupts retinal lamination, including the photoreceptor layer, in gdf6a^s327/s327 mutants. (A) Thrβ knockdown with splice-blocking morpholino in gdf6a^s327/s327 mutants (labeled gdf6a^−/− or “−/−”) fails to increase UV cone abundance, instead causing near-total loss of blue and red cones. Values with matching letters are not significantly different (Kruskal-Wallis test with Mann-Whitney pairwise comparisons, gray asterisks indicate *P < 0.05, **P < 0.01, ***P < 0.001 relative to WT + control [CTL] MO controls). UV cone abundance decreased significantly, in contrast to thrβ knockdown in wildtype fish. (B) Gaps or “holes” can be seen in the photoreceptor layer of gdf6a^s327/s327; thrβ MO-treated whole-mount retinas (two holes indicated by asterisks in UV image). These holes are not seen in CTL MO-injected gdf6a^s327s327 mutants, nor in wild-type or heterozygous morphants. (C) Radial sections of 4 dpf gdf6a^s327/s327; thrβ MO-treated retinas show disrupted retinal lamination, with gaps in the photoreceptor layer that are occupied by cells of the inner nuclear layer, and cells disrupting the inner plexiform layer between the inner nuclear layer and ganglion cell layer (n = 9 embryos examined per group).

Differential Disruption of thrβ Reveals New Roles in Cone Development

The severity of the lamination defect limited our ability to quantify cone subtypes when gdf6a and thrβ expression were both disrupted, so, to overcome this, we created a model of conditional Thrβ activity disruption via expression of a dominant negative receptor. It was determined through morpholino knock down that thrβ is required for red cone development¹⁹; however, morpholinos must be introduced at the 1- to 2-cell stage, prohibiting manipulation of thrβ expression at later developmental stages. To further investigate gdf6a and thrβ, as well as actions of thrβ at different stages of cone photoreceptor development, we created a transgenic line with a construct created with Gateway recombination⁶³ that enables the expression of a dominant negative receptor predicted to disrupt endogenous receptor activity. Modeling after the work of Ulisse et al.⁶⁷ expressing a dominant negative thrβ in Xenopus, we amplified the zebrafish thrβ sequence and removed nucleotide bases coding for the C-terminal 12 amino acids (which are necessary for ligand and coactivator binding). Without these binding sites, the receptor may dimerize with other receptors and bind to DNA target sequences but fail to recruit the rest of the activation complex and so transcription is not initiated. Expression of this dominant negative receptor (dnthrβ for short) was placed under the hsp70 promoter to enable conditional induction upon heat shock and is followed by enhanced green fluorescent protein (eGFP) (Fig. 3A). This construct was injected into 1- to 2-cell stage embryos and a stable Tg(hsp70:dnthrβ.V2A.eGFP) line was established (designation ua3113). Heat shock induction allows us to temporally manipulate dnthrβ expression, thereby disrupting the activity of endogenous Thrβ proteins at specific developmental stages. It is important to note that the presence of a dominant negative receptor stands in contrast to MO knockdown (in which overall receptor protein abundance is reduced), as it is predicted to create an incomplete complex that occupies promoter and enhancer regions to the exclusion of other complexes; therefore, it is predicted to have negative transcriptional effects. 

Figure 3

 

A transgenic zebrafish model of conditional Thrβ disruption reveals additional roles in cone development. (A) Generation of a dominant negative thyroid hormone receptor β was accomplished via an 11-amino acid C-terminal deletion in thrβ, which was then cloned into a transgene for dominant negative thrβ (dnthrβ) and eGFP under the hsp70 promoter. Endogenous Thrβ dimerizes with other factors, such as Thrβ or RXRγ, and binds thyroid hormone to activate transcription (left), whereas dnThrβ would bind endogenous receptors but, lacking ligand and coactivator binding domains, would render dimers inactive (right). (B) Transgenic embryos express the transgene, including GFP, throughout the body. Scale bar: 1 mm. (C) Thrβ MO causes dramatic reduction in red cone abundance and increased UV abundance at 10-ng dose. Our dominant negative model shows the same increase in UV cones but also a significant increase in blue cones (HS, heat shocked). Values with matching letters are not significantly different (Kruskal-Wallis test with Mann-Whitney pairwise comparisons, gray asterisks indicate *P < 0.05, **P < 0.01, ***P < 0.001 relative to CTL MO or sibling heat-shocked [HS] controls). Thrβ MO data are normalized to CTL MO data; dnthrβ HS data are normalized to sibling HS data.

Upon heat shock, we observed zebrafish embryos showing eGFP expression throughout the body (Fig. 3B). Without heat shock induction, eGFP expression is limited to the eye lens, which expresses hsp70 during normal development.⁶⁸ 

We examined the effects of dnthrβ expression on cone photoreceptor subtype determination in comparison to splice-blocking morpholino knockdown of thrβ.¹⁹ Knockdown of thrβ causes near-complete absence of red cones and an increase in UV cone abundance (by approximately 35%, P = 0.0003), whereas expression of dnthrβ via heat shock at 52 hpf leads to increased UV (by 27%, P = 0.003) and blue cone abundance (by 36%, P = 0.001) relative to heat shocked nontransgenic siblings (Fig. 3C; average values of raw counts available in Supplementary Fig. S3). Inducing dnthrβ expression at other time points, including 24 hpf, 30 hpf, and 36 hpf, did not alter cone abundances as dramatically relative to controls (<20% change, Supplemental Fig. S4). None of these interventions with dnthrβ expression caused UV, blue, or red opsin coexpression, which was verified in the orthogonal views in both Zen 2010 and ImageJ software. This revealed an effect of thrβ that is limited to later photoreceptor development: the endogenous receptor negatively regulates blue cone determination, and the dominant negative receptor would disinhibit this process, allowing for more progenitors to assume a blue fate late in development. In contrast, disrupting Thrβ activity either early (with morpholino knock down) or late leads to more UV cones. Finally, unlike the observed coexpression of red opsin with other opsins upon thrβ misexpression,¹⁹ expression of dnthrβ did not cause opsin coexpression. The methods used to label cone subtypes for analysis have limited capacity to detect coexpression of a diversity of opsin types; the transgenic expression of GFP in UV cones and mCherry in blue cones is robust, but the fluorescent proteins are restricted to the cell bodies, so visualization of the outer segments was not possible. Only two antibodies, anti-UV opsin and anti-red opsin, were available and used here, and so only coexpression of these opsins would be possible to detect. Thus, although we could not see opsin coexpression, we cannot rule it out as a possibility, especially in experiments where the total number of cones detected appears larger than normal (e.g., in dnthrβ retinas, total abundances shown in Supplementary Figures). 

Disrupting the effects of Thrβ via dnThrβ expression increased blue cone abundance (Fig. 3C, 3D), making thrβ only the second gene identified as affecting blue cone abundances in zebrafish (or any animal), after gdf6a.³⁵ Additionally, similar to morpholino knock down, the dnthrβ model has an excess UV cone phenotype; therefore, we used this line to test whether Gdf6a signaling and Thrβ are epistatic in their regulation of UV cone (and/or blue cone) development. With the conditional activation of dnthrβ expression, we bypassed the gdf6a/thrβ lamination phenotype by disrupting Thrβ activity after lamination. We heat shocked gdf6a^s327/s327; hsp70:dnthrβ embryos at 52 hpf, as done previously. But instead of increasing UV and blue cone abundances, expression of dnThrβ in gdf6a^s327/s327 mutants yielded slightly reduced UV and significantly reduced blue populations (by 62%, P = 1.2 × 10⁻⁴); thus, dnthrβ failed to rescue the gdf6a blue cone phenotype. Gdf6a^s327/s327 retinas also retained the phenotype of fewer red cones (up to 35% reduction compared to wildtype, P = 0.0001) regardless of dnthrβ expression (Fig. 4; average values of raw counts available in Supplementary Fig. S1). No opsin coexpression was detected in mutant retinas with dnthrβ expression. Thus, it appears that thrβ is not directly downstream of gdf6a in blue or red cone fate. 

Figure 4

 

Thrβ disruption with dnThrβ does not rescue the gdf6^s327/s327 blue cone phenotype. (A) Expression of dominant negative thrβ (dnthrβ) increases UV and blue cone abundances, but in gdf6a^s327s327 mutants (labeled gdf6a^−/− or “−/−”) the low blue cone abundance is not changed and UV cones are not increased with dnthrβ expression. Gdf6a^s327s327 mutants have reduced red cone abundance regardless of dnthrβ expression compared to nontransgenic wildtype siblings. Values with matching letters are not significantly different (Kruskal-Wallis test with Mann-Whitney pairwise comparisons, gray asterisks indicate *P < 0.05, **P < 0.01, ***P < 0.001 relative to WT sibling HS control). (B) Representative retinal images of wild-type sibling gdf6a^s327s327 and gdf6a^s327s327 with dnthrβ expression.

Discussion

In this work, we describe novel genetic roles among gdf6a, tbx2b, and thrβ that impact cone photoreceptor subtype determination, and which elaborate the cone determination regulatory network. We summarize the effects of gdf6a, thrβ, and tbx2b on cone development as found in our experiments here, and then explore the larger implications, including intersecting effects, of these factors on cone subtype development below. 

Gdf6a Signaling Positively Regulates tbx2b in UV Cone Development

Tbx2b is a downstream target of gdf6a in the early retina, contributing to dorsal identity and determination of dorsal retinal cells, and to UV cone development, where gdf6a also influences tbx2b.^{35,45,47,69,70} Tbx2b is required for UV cone identity during photoreceptor development; homozygous tbx2b mutant retinas have few or no UV cones (lor and fby alleles are thought to be hypomorphic and null alleles, respectively).³⁹ These mutants also present with an overabundance of rods (termed the lots-of-rods phenotype).³⁹ The lots-of-rods phenotype can be elicited in some tbx2b^+/lor heterozygous mutants if gdf6a is absent (i.e., gdf6a^s327/s327).³⁵ Here, we crossed tbx2b^fby mutants to gdf6a ^s327 mutants to quantify the effects of the combined mutations on cone subtypes more comprehensively. A 23% reduction in UV cone abundance was observed in the gdf6a^s327/s327; tbx2b^+/fby genotype compared to wild types, and this aligns with previous findings that these compound mutants are predisposed to excess rod generation.^35,39 Thus, gdf6a loss in tbx2b^+/− mutants likely enhances the sensitivity of progenitors to insufficient levels of tbx2b expression, which affects both UV cone and rod abundance (perhaps with some degree of stochasticity, as not all gdf6a^−/− ; tbx2b^+/− retinas show reduced UV cones and excess rods). The absence of a lots-of-rods phenotype in gdf6a mutants alone suggests that tbx2b expression and/or function can be sufficiently supported by factors independent from gdf6a, so long as two functional tbx2b alleles are present. Thus, the epistatic relationship in UV cone development wherein tbx2b would be downstream of gdf6a is not a closed system, and other factors may participate in regulating tbx2b in this pathway. Reduction in tbx2b wild-type gene expression dosage may also worsen gdf6a homozygous mutants' blue and red cone phenotypes, although the relevance of tbx2b to these cone types is less clear. In summary, gdf6a impacts upon both UV and blue cone development, plausibly via stimulating or supporting tbx2b expression for the generation of UV cones and possibly via a separate pathway for blue cones (Fig. 5). 

Figure 5

 

Regulatory actions of thrβ, gdf6a, and tbx2b in zebrafish cone photoreceptor determination. The proposed pathway shown is a summary of the effects of the factors studied on cone subtype development and does not represent order of expression, order of progenitor/precursor progression, or a chronologic sequence of events. Gdf6a promotes blue and red cones and indirectly influences UV cones (dashed arrow), and tbx2b promotes UV cones. Thrβ stimulates red cone identity and inhibits UV cones and blue cones (red dashed lines). Gdf6a's regulation of blue and red cone abundances supersedes thrβ; thus, the effects of gdf6a and thrβ should be considered independent in the pathway depicted. Work by other groups established that six7 is required for green cones, but it is not known whether the other factors, such as gdf6a, actively regulate green cones.

Thrβ Promotion of Red Cones and Suppression of UV and Blue Cones Depend on Photoreceptor Development Stage

Thrβ is known to be required for M-opsin and for long-wavelength-sensitive cone identity in mice and in zebrafish^18,19; this represents a rare shared node in photoreceptor development between these species, and few other such commonalities have been experimentally demonstrated to date (i.e., beyond commonalities in gene expression). Thus, thrβ is of deep interest for comparative retinal development, and we sought to extend the characterization of its role and position in the zebrafish photoreceptor regulatory network. With a dnthrβ conditional line, we were able to discern that thrβ influences cone progenitors differently depending on developmental age. UV and blue cone abundances were both enhanced, whereas red cones were unaffected when dnthrβ was expressed late in photoreceptor development. Thus, we discovered that thrβ can negatively regulate zebrafish blue cone fate (mice, like all eutherian mammals, lack this sws2 cone subtype), in addition to regulating red cones and UV cones. 

Reduced thrβ expression early in photoreceptor development (via MO knockdown) strongly affects red cone abundance (reduction) and UV cones (increase) but not blue cones. These differential and age-dependent responses to reduced Thrβ (in either expression or activity) suggest that the competency of cone progenitors regarding thrβ may change during retinal development. This interpretation of our data is consistent with observations of differential cone abundances when thrβ was ectopically expressed under various promoters with differential timing of expression.¹⁹ Red cones appear to have a critical window of induction by thrβ shortly after initiation of crx expression because in thrβ morphants the red cones can be rescued by thrβ expression driven by the crx promoter (i.e., as early as 19 hpf), as described by Suzuki et al.¹⁹ In the same article, induction of thrβ after the final cell division resulted in cones that coexpress red cone opsin with another cone opsin, an imperfect “rescue”.¹⁹ Thus, during the critical period for red cone determination, Thrβ induces red cone fate and suppresses UV cone fate, whereas after this period (in postmitotic photoreceptors), Thrβ seems to merely activate red cone opsin expression (and not change cone identity). Likewise, disrupting endogenous Thrβ activity with dnThrβ late in photoreceptor development, as done here, did not change the size of the red cone population. The dnthrβ expression line we created did not replicate the reduction of red cones observed in morphants, which may be due to one or more factors: expression levels of dnthrβ may not be sufficient to bind a large enough proportion of endogenous Thrβ proteins to effect measurable change in gene transcription during the critical window for red fate; the critical window for red cone fate was missed; or transcription feedback mechanisms in response to dnThrβ dampened its effects, possibly in part due to the ubiquitous expression of the dnthrβ transgene (e.g., enhanced expression of endogenous thrβ or use of alternative binding partners to facilitate transcription). 

Using our model of conditional disruption of the Thrβ protein, we determined that Thrβ's effect on blue cones is limited to late photoreceptor development. We speculate that thrβ's time-dependent effects on blue cone abundances may be a result of negative transcriptional actions because this phenotype is not observed when thrβ expression is merely reduced by MO (see Fig. 2). Thrβ proteins are known to bind and occupy regulatory elements without activating transcription, thereby preventing or reducing gene expression,^71–73 unless (or until) the requisite cofactors are present in adequate amounts, including active thyroid hormone T₃ and coactivators. Thus, Thrβ may limit the number of cone progenitors assuming certain fates (UV cones and blue cones later in development) and promote others (such as red cones) at the appropriate time (Fig. 5). Modification of Thrβ's actions may occur through differential expression of corepressors, such as silencing mediator for retinoid or thyroid hormone receptors (SMRT) and/or dimerization with other thyroid receptors versus retinoid X receptors. As Thrβ is known to interact with RXRs and RORs in mouse development, and zebrafish cone differentiation initiates in the ventral retina where retinoic acid signaling is active,^71,74–78 these interactions are plausible. 

Interestingly, inhibition of Thrβ in the adult mouse retina enhances cone photoreceptor survival,⁷⁹ which stands in contrast to its necessity for M-cone development and function in mice⁸⁰ and zebrafish¹⁹ and possibly similar roles in chicks^81,82 and Xenopus^83–85 (although more studies are needed). Thus, the role of thrβ is context dependent: thrβ promotes specifically red cone identity within a specific window of development; outside this window and outside of red cones, thrβ may have negative effects on cone development and survival. 

Actions of thrβ in Cone Development Are Superseded by gdf6a

We next examined gdf6a and thrβ, as both are implicated in UV cone development. By knocking down thrβ in gdf6a mutants, we encountered a surprising disruption to lamination across all three retinal layers, as well as near-total lack of blue or red cones. 

To resolve whether gdf6a and thrβ are linked in terms of cone development and to further understand the temporal nature of thrβ in cone subtype fate determination, we created a transgenic zebrafish with inducible dominant negative thrβ expression to disrupt endogenous Thrβ protein activity. 

We then returned to gdf6a and thrβ with the expanded question of whether they act in shared pathways toward UV or blue cone fate. Our new observation above that thrβ is required for suppressing blue cones is noteworthy in contrast to the requirement of gdf6a in promoting blue cone development. We, thus, expected to identify epistatic or other intersections between these factors by using combinatorial genetic disruption. Intriguingly, neither the thrβ MO nor our dnthrβ transgene changed the gdf6a phenotype of low blue and low red cone abundances. Although thrβ disruption in wild-type retinas increased UV cone abundance, UV cones were not increased in gdf6a mutants with thrβ disruption. Thus, disrupting thrβ failed to rescue blue or red cones or enhance UV cones, suggesting that the influence of gdf6a overrides that of thrβ in cone subtype determination. In other words, the data do not support that thrβ would be directly downstream of gdf6a in a shared genetic pathway. Alternative explanations are that thrβ and gdf6a act in parallel pathways, thrβ acts upstream of gdf6a, or the early effects of Gdf6a signaling (or lack thereof) on photoreceptor progenitors influences (or constrains) how gene expression may be affected by Thrβ or dnThrβ. For example, gdf6a mutant retinas exhibit decreased proliferation and increased apoptosis at 2 dpf.^69,86 The reduction or loss of progenitors, some of which are presumed photoreceptor progenitors, may limit the potential for dnthrβ to increase or rescue cone subtype populations. These speculations await further testing. For simplicity, the effects of gdf6a and thrβ on cone subtype fates are depicted as independent influences in Figure 5. 

Gdf6a^s327/s327 mutants exhibit numerous retinal phenotypes: increased apoptosis^44,69; a smaller progenitor pool,⁶⁹ reduced markers of proliferation, and reduced blue and red cone abundances yet normal UV cone abundances.³⁵ We speculate that these phenotypes may be interrelated and that the gdf6a cone phenotypes may reflect imbalances in progenitor proliferation or survival. Progenitor survival may be impacted further upon decreased thrβ expression, resulting in dramatic reductions in differentiated cells, including cones (yet only one cone type appears to persist—UV cones). We speculate the areas lacking photoreceptors in gdf6a^s327/s327; thrβ morphants may be artifacts of photoreceptor progenitors that received sufficient doses of thrβ MO to cease dividing or to die entirely. Comparing the blue and red cone phenotypes seen in gdf6a mutants with either thrβ knock down or dnthrβ expression, it appears that disruption to their gene functions negatively affects both types. Although Thrβ knockdown in gdf6a mutants caused photoreceptor loss in parts of the retina, the remaining intact areas contained UV cones but almost no red cones or blue cones (Fig. 2B). Thus, the blue and red cone populations were more severely affected than UV cones. In gdf6a mutants expressing dnthrβ, the blue and red cone abundances were reduced in an additive fashion as well, although not as dramatically (Fig. 4). These observations suggest that both gdf6a and thrβ are required to support photoreceptor progenitors (and perhaps progenitors of specific cone subtypes) and to promote red cones. For blue cones, the relationship is less clear: knock down of thrβ alone does not affect blue cones but it does in combination with gdf6a loss. 

Conclusions

The cone photoreceptor development relevant genes thrβ, gdf6a, and tbx2b interact in a fashion that is critical for the production of three cone types, and our work has found that their influences in cone subtype development intersect in intriguing ways, thereby adding new aspects to a model of vertebrate photoreceptor development. This and future work may provide new avenues to explore how the development of color vision may be modified to suit the multitude of visual environments, including diverse color environments, in which vertebrates live.⁸⁷ Furthermore, a detailed model of vertebrate photoreceptor development, especially one featuring multiple cone subtypes, is a valuable framework for regenerative medicine. This framework, once established, will guide the modification of existing stem cell protocols and other therapeutic strategies toward higher cone production for restoring daytime vision. 

Acknowledgments

The authors thank Andrew Waskiewicz and Josh Gamse for gifting the mutant fish lines; Andrew Waskiewicz and Valerie Sim for sharing imaging resources; Hao Wang for performing molecular work in generating the Tg(hsp70:dnthrb.V2A.eGFP)^ua3113 line; Elizabeth Hodges, Ramona Rosca, and Quinton Schmidt for their assistance in data collection; and Aleah McCorry and Xinyue (Amy) Zhang for animal care. 

Supported by MD/PhD studentships from Alberta Innovates Health Solutions and from Canadian Institutes of Health Research (MGD). Operating Funds were from the Natural Sciences and Engineering Research Council of Canada. 

Disclosure: M.G. DuVal, None; W.T. Allison, None