Abstract
During development, retinal progenitors navigate a complex landscape of fate decisions that culminates with an array of unique cell types that are required for proper vision. Here, we aim to identify factors that are required for fate decisions in photoreceptors. These factors help create a diversity of photoreceptor subtypes that sustain vision in day and night, enable the detection of colors, of prey and predators, and other aspects of vision. To identify these factors, we generate a high-quality and deep transcriptomic profile of each photoreceptor subtype in zebrafish. From these profiles, we focus on transcription factors—key players in cell-fate decisions. We apply CRISPR-FØ screening as a versatile platform to explore the involvement of transcription factors in photoreceptor subtype-specification. We find that three differentially-expressed transcription factors (Foxq2, Tbx2a and Tbx2b) play unique roles in controling the identity of photoreceptor subtypes within the retina. Our results provide novel insights into the function of these factors and how photoreceptors acquire their final identities. Furthermore, we have made our transcriptomic dataset openly available and easy to explore. This dataset and the screening method will be valuable to the scientific community and will enable the exploration of genes involved in many essential aspects of photoreceptor biology.
Introduction
The specification of cell types is a fundamental and complex process that underlies the formation of tissues and organs. Cell specification is especially critical for the primary sensors in sensory systems. Primary sensory cells not only have to acquire distinct and elaborate specializations, to be able to detect and transduce particular physical stimuli, but also have to wire with specific circuits to faithfully relay sensory information. In the retina, for instance, photoreceptor specification must coordinate with the distinct developmental timelines of other retinal cell types to form functional visual circuits (1). In addition, photoreceptors have clearly defined photoreceptor subtypes which differ in spectral sensitivity, morphology, density across the retina, wiring, and gene expression. Comprehensively, the factors required to specify photoreceptor subtypes remains unclear. Our aim is to identify these factors, exploiting technical advantages of the zebrafish model.
Evolutionary, fish and birds have retained an ancestral and diverse set of photoreceptor subtypes. In zebrafish, there are five photoreceptor subtypes that can be readily distinguished by morphology and opsin expression: rods expressing rhodopsin (rho), UV cones (or short single cones) expressing a UV-sensitive opsin (opn1sw1), S cones (or long single cones) expressing a short-wavelength-sensitive opsin (opn1sw2), M cones (or accessory members of double cones) expressing mid-wavelength-sensitive opsins encoded by genes that have been duplicated twice (opn1mw1 - opn1mw4), and L cones (or principal members of double cones) expressing long-wavelength-sensitive opsins encoded by duplicated genes (opn1lw1 - opn1lw2) (2–5). In comparison, most mammals possess only three photoreceptor sub-types: rods, a short-wavelength sensitive cone (S cone), and a mid/long-wavelength sensitive cone (M/L-cone). The mammalian S and M/L cones are evolutionary related to the ancestral UV and L cones, respectively, while the ancestral S and M cones were lost. Even more recently in evolution, duplication and subsequent mutations of the L-opsin gene led to the emergence of an additional functional subtype in Old-World Primates, which supports human trichromatic vision (6).
The correct specification of photoreceptor subtypes is critical for proper vision. Foundational work across vertebrates has identified transcription factors that are required for the generation of photoreceptors progenitors during development, before subtypes are clearly defined(7–12). Numerous studies have investigated how photoreceptor subtypes are specified from these progenitors. Studies in mice have shown that the transcription factor NRL and its downstream effector NR2E3 are required for rod specification (13–15). Interestingly, in humans, mutations in NR2E3 cause enhanced S-cone syndrome, where failures in rod specification ultimately lead to impaired visual acuity, abnormal color vision, night blindness and retinal degeneration (16). In addition to conserved factors involved in rod specification, THRb is required for L-cone specification in mice (17), zebrafish (18) and, most likely, birds (19). In the absence of these “key gatekeepers”— NRL, NR2E3 or THRb—mouse photoreceptor progenitors acquire an S-cone fate, and S-cone specification is frequently assumed to be a passive process (12). Yet, mounting evidence derived mainly from work in nonmouse species challenges this simplistic model. For example, while Nr2e3 is also required to generate rods in other vertebrates (20), Nrl is dispensable (21, 22). In addition, UV- and S-cone specification in zebrafish is far from passive and requires the action of key transcription factors—Tbx2b and Foxq2 (23, 24). Moreover, factors that regulate opsin expression—e.g. Rora, Rorb or Rxrg in mouse (25–27)— or survival of retinal progenitors—six6, six7 and gdf6a in zebrafish (28–33) —can also cause subtype-specific alterations in photoreceptor development. Together these studies highlight that our understanding of how photoreceptor subtypes are specified remains incomplete.
Our study seeks to identify additional genes involved in this complex process and is divided into three sections. First, we obtain a deep and high-quality transcriptomic profile (RNA-seq) of the five zebrafish photoreceptor subtypes. Second, we explore this RNA-seq dataset and identify multiple transcription factors that are potentially involved in the specification of photoreceptor subtypes. Third, we use a CRISPR-based FØ-screening approach (34, 35) to directly explore the role of three transcription factors—Foxq2, Tbx2a and Tbx2b. We demonstrate that Foxq2 is required for S-cone specification exclusively. UV-cone specification requires both Tbx2a and Tbx2b. Additionally, Tbx2a and Tbx2b respectively maintain the identity of L cones and S cones by repressing M-cone cell fate.
In the future, our dataset can be applied to explore other important aspects of photoreceptor biology that also differ between subtypes (phototransduction, metabolism, synaptic wiring, etc.). To facilitate future studies, we provide open and easy access to our transcriptomic dataset and analysis (https://github.com/angueyraNIH/drRNAseq/). This dataset can be used to further our understanding of how photoreceptors acquire their final identities. This knowledge can be used to inform strategies to control the photoreceptor differentiation in organoids—a potential gateway for cell-replacement therapies in retinal degenerations.
Results
Transcriptomic analysis of adult zebrafish photoreceptors
Identifying the transcription factors required to specify photoreceptor subtypes is critical to understand the normal de-velopment of the retina and to inform cell-replacement therapies to restore vision. RNA-seq is a powerful way to identify novel genes expressed in particular cell subtypes. Although RNA-seq approaches have been used to identify genes differentially expressed between photoreceptor subtypes in many species, the limited transcriptome depth derived from single-cell techniques (36) constitutes a barrier in the reliable detection of transcription factors, which are frequently expressed at low levels (37). To obtain a deep, high-quality RNA-seq dataset from zebrafish photoreceptors, we manually collected pools of photoreceptors of a single subtype (38). We identified photoreceptors using well-characterized transgenic lines that express fluorescent proteins in each subtype with high specificity, including rods—Tg(xOPS:GFP), UV cones—Tg(opn1sw1:GFP), S cones—Tg(opn1sw2:GFP), M cones—Tg(opn1mw2:GFP) and L cones—Tg(thrb:tdTomato) (18, 39–42) (Figure 1A). We collected dissociated photoreceptors from these transgenic lines under epifluorescence. Manual collection allowed us to focus on fluorescent and healthy photoreceptors, with intact outer segments, cell bodies and mitochondrial bundles and to avoid cellular debris and other contaminants (Figure 1B).
For each sample, we collected pools of 20 photoreceptors of a single subtype derived from a single adult retina. After collection, we isolated mRNA and generated cDNA libraries for sequencing using SMART-seq2 technology (Figure 1C). In total, we acquired 6 rod samples and 5 UV-, 6 S-, 7 M- and 6 L-cone samples. On average, we were able to map 86.4% of reads to the zebrafish genome (GRCz11; range: 76.3% - 90.4%), corresponding to 10.19 million ± 1.77 million mapped reads per sample (mean ± s.d.) (Figure 1D). This quantity of reads demonstrates that our technique provides substantially deep transcriptomes—especially when compared to single-cell droplet-based techniques where the number of reads per cell is largely 20000 or less (36, 43, 44). Using unsupervised clustering (t-distributed Stochastic Neighbor Embedding or tSNE), we found that samples correctly clustered by the subtype they were derived from. Proper clustering provides evidence that differences in gene expression captured in our RNA-seq data stem mainly from distinctions between photoreceptor subtypes (Figure 1E).
The expression of opsin genes is unique between photoreceptor subtypes and under normal conditions it is a reliable marker of subtype identity. Consistent with this idea, each sample had a high number of reads for the appropriate opsin. Furthermore, there were very low reads for other opsins, corroborating the purity of our samples (Figure 1F). We also found that reads for phototransduction genes were high and consistent with the known differences in gene expression between rods and cones (e.g., gnat1, rod transducin, had high reads only in rod samples while gnat2, cone transducin, had high reads in all cone samples) and between cone subtypes (e.g., expression of arr3a in M and L cones and arr3b in UV and S cones) (Figure 1 – Figure supplement 1) (44, 45). To expand our analysis to other genes, we first used principal component analysis (PCA) as an unbiased approach to determine how variability in gene expression defines photoreceptor subtypes. PCA revealed that most of the differences in gene expression were between rods and cones. When cones were considered separately, the biggest differences in gene expression arose from two groupings: UV and S cones vs. M and L cones. Subsequent analysis revealed a clear separation of M and L cones, with UV and S cones showing the least differences (Figure 1G). Guided by this analysis, we performed differential gene-expression analysis by making pairwise comparisons following the directions of the principal components, revealing a diverse set of differentially expressed genes (Figure 1 – Figure Supplement 2 and Supplementary Data 1).
In summary, our manual, cell-type specific, SMART-seq2-based approach yielded high-quality zebrafish photoreceptor transcriptomes, with ~2000-fold more depth than published single-cell RNA-seq studies in the retina (36, 43, 44, 46). Our RNA-seq data is in good agreement with current knowledge of photoreceptor-expressed genes, but importantly also uncovered novel differences in gene expression between photoreceptor subtypes. This RNA-seq dataset constitutes a useful resource to explore genes that are generally or differentially expressed by photoreceptor subtypes which could be involved in multiple aspects of photoreceptor biology. Our subsequent analyses center on transcription factors.
Expression of transcription factors in zebrafish photoreceptors
Transcription factors play a critical role in photoreceptor specification. Therefore, we isolated all RNA-seq reads that could be mapped to transcription factors. When ranked by average expression levels across all samples (in Fragments Per Kilobase of transcript per Million mapped reads or FPKM), neurod1 was revealed as the transcription factor with the highest expression (~5-fold). High expression in adult photoreceptors suggests that neurod1 plays a role not only during development or regeneration but also in the mature retina (10, 47, 48). Among the 100 most highly expressed transcription factors, we identified many genes known to known to be important during photoreceptor development or specification including crx, otx5, rx1, rx2, nr2e3, six6b, six7, meis1b, egr1 and thrb (Figure 2A) (7–9, 28, 32, 49–52).
Next, we used PCA to understand how transcription-factor expression differs between photoreceptor subtypes. Like our whole-transcriptome analysis, we found that most of the differences in transcription-factor expression can be attributed to differences between rods and cones (Figure 2B). By performing pairwise comparisons of transcription factors based on rod vs. cone expression, we identified three relevant groups: (1) consistently expressed across all subtypes, (2) rod-enriched and (3) cone-enriched (Figure 2C). Consistent with previous studies, expression of crx and otx5 was similar across subtypes (7, 8), nr2e3, samd7 and samd11 showed clear rod-enrichment (15, 53, 54), while expression of nrl, which is only transiently required to specify rods in zebrafish (22), was low and only detected in a third of the rod samples. Lastly, our results verified that six6a, six6b, six7, and neurod1 were cone enriched transcription factors (10, 28, 32). By expanding our analysis beyond previously characterized genes, our dataset revealed a total of 75 transcription factors with significant differential expression between rods and cones (Figure 2C and Supplementary Data 2).
We next examined the variance in transcription-factor expression between cone subtypes. PCA revealed that both L and M cones could be distinguished by differences in transcription-factor expression alone, while UV and S cones again showed the fewest differences in transcription factor expression (Figure 2D). By analyzing the differentially expressed transcription factors between cones subtypes, we found a group of seven transcription factors significantly enriched in both L and M cones compared to UV and S cones (Figure 2E). These included ahrlb, a gene associated with Retinitis Pigmentosa in humans (55) and six7, known to be involved in cone progenitor development and survival (28). We found twelve L-cone enriched transcription factors (Figure 2E); these included thrb, known to be critical for L-cone identity across vertebrates (17, 18) and rxrga, a regulator of L-opsin expression in mouse (25). Amongst the ten M-cone enriched transcription factors, we identified thraa, another thyroid hormone receptor, known to be expressed by photoreceptors (56). A small group of just five genes was enriched in both UV and S cones compared to L and M cones and included tbx2b. Seven transcription factor were enriched in S cones—including foxq2—and two were enriched in UV cones—including tbx2a (Figure 2E and Supplementary Data 3).
In summary, our analysis has revealed novel patterns of expression of transcription factors between photoreceptor subtypes. Notably, a considerable fraction of these transcription factors has no known function in photoreceptors in zebrafish or in other species, making them clear targets for follow-up studies aimed at understanding photoreceptor specification and other subtype-specific functions.
FØ screening as a platform to explore the involvement of transcription factors in photoreceptor identity
Our dataset revealed an extensive collection of transcription factors potentially involved in photoreceptor differentiation. Therefore, we next established methods to identify which transcription factors are involved in subtype specification. We focused on a small subset of these transcription factors, and evaluated their function using a highly efficient CRISPR-Cas9-guided mutagenesis approach that enables FØ screening (34, 35). In these FØ screens, single-cell embryos are injected with Cas9 and multiple guide RNAs (gRNAs) targeting a gene of interest. Just days after injection, in the FØ generation, injected larvae can be assessed phenotypically. Although these FØ larvae are genetic mosaics (some cells may not carry mutations and mutations between cells are not identical), each larva can be genotyped to establish a robust link between gene function and phenotype. In addition, FØ larvae can be created in the context of any combination of existing transgenic or mutant lines. We used this FØ-screening approach to study the role of three transcription factors in the determination of photoreceptor identity: Foxq2, Tbx2a and Tbx2b. The results presented below provide novel insights into the function of these genes and help establish this approach as a versatile method for future screens.
Foxq2 is required for S-cone specification
First, we explored the function of Foxq2. Our RNA-seq data revealed that foxq2 is expressed at relatively high levels, ranking 33rd among transcription factors (Figure 2A), and specifically enriched in S-cones, with negligible expression in other photoreceptor subtypes (Figure 3A). Based on this Scone specific expression, we used the FØ-screening platform to test whether foxq2 was involved in S-cone specification.
For our FØ analyis, we designed two gRNAs targeted against the DNA-binding forkhead domain of foxq2 (57). Preliminary experiments established that we were able to produce mutations in foxq2 in ~95% (21/22) of injected FØ larvae. All analyses presented here correspond to FØ larvae that have been genotyped to confirm mutations in foxq2 (FØ[foxq2]). To evaluate changes in photoreceptors, we used subtype-specific photoreceptor reporter lines and evaluated changes in photoreceptor densities in the central retina of larvae at 5 days post-fertilization (5 dpf). Compared to wildtype controls, FØ[foxq2] larvae displayed a dramatic decrease in the density of S cones. We were able to clearly identify a decrease in S cones through their mCherry expression in Tg(opn1sw2:mCherry) (Figure 3B), or by the lack of S-opsin antibody labeling (Figure 3 – figure supplement 1A). The density of other photoreceptor subtypes remained unchanged, with a decrease in S cones being the only significant change (Figure 3C). This demonstrates that the loss of S cones in FØ[foxq2] larvae is selective and not compensated by an increase in another photoreceptor subtype. Our results are close in agreement with a recent study that characterized a complete foxq2 loss-of-function (see discussion) (24), demonstrating the power of the FØ-screening approach. In conclusion, foxq2 is a S-cone specific gene that is required for the early specification of S cones but not of other photoreceptor subtypes.
Tbx2a and Tbx2b are independently required for UV-cone specification
To further expand our FØ analysis, we explored the role of Tbx2 in photoreceptor differentiation. As a teleost duplicated gene, there are two versions of tbx2 in the zebrafish genome: tbx2a and tbx2b. We chose to evaluate the function of Tbx2 for three main reasons. First, our RNA-seq data revealed that both tbx2a and tbx2b show high expression in UV cones (Figure 4A). Second, Tbx2 is known to be differentially expressed in cones of many species, including cichlids (58), chickens (59) and primates (46). Third, Tbx2b in particular, has been shown to be involved in the determination of UV cones in zebrafish (23).
For our FØ analysis, we injected embryos with Cas9 protein and 3 gRNAs targeting exon 3 of tbx2b or 3 gRNAs targeting exon 3 of tbx2a. In both genes, exon 3 contains critical DNA-binding residues that are completely conserved across verte-brates (60). Initial experiments showed that, using this approach, we were able to produce mutations in tbx2b in ~80% (19/23) of injected larvae and in tbx2a in ~90% (30/33) of injected larvae. Our gRNAs were selective for their respective gene, and we did not observe any mutations in the non-targeted paralogue (0/48 larvae tested). Moving forward, we restricted all subsequent analysis to larvae with confirmed mutations in tbx2b (FØ[tbx2b]) or in tbx2a (FØ[tbx2a]). As tbx2a and tbx2b are enriched in UV-cones, we first assessed changes in UV-cone fate, using a UV-cone reporter line, Tg(opn1sw1:nfsB-mCherry), (61). We used this UV-cone reporter line along with a rod reporter line, Tg(xOPS:GFP) (39) as loss of Tbx2b function increases rod density (23). In wild-type control larvae at 5 dpf, UV cones are numerous and densely distributed across the retina, while overall rod density is low, with most rods concentrated in the ventral retina and with lowest density in the central retina (23, 62, 63). Using these transgenic lines, we first quantified UV cone and rod density in the central retina of control and (FØ[tbx2b])larvae. In agreement with previous studies, loss of Tbx2b function produced a marked decrease in UV cones and an increase in rod density (Figure 4B and D) (23). The robust phenotype in FØ[tbx2b] larvae further demonstrates the efficiency and flexibility of our approach. After replicating the described loss-of-function phenotypes of tbx2b mutants in FØ[tbx2b] larvae, we examined FØ[tbx2a] larvae. Surprisingly, we found that FØ[tbx2a] also displayed the same phenotype as FØ[tbx2b]—a marked loss of UV cones and an increase in rods, albeit this increase in rods was lower in FØ[tbx2a] than in FØ[tbx2b] (Figure 4C and D).
To confirm the phenotypes of tbx2 mutants revealed through imaging of reporter lines, we quantified opsin expression using real-time quantitative PCR (qPCR). We found that, in comparison to controls, FØ[tbx2b] showed a clear decrease in UV-opsin expression and a significant increase in rhodopsin expression. FØ[tbx2a] also showed a clear decrease in UV-opsin expression, but without significant changes in rhodopsin expression (Figure 4 - figure supplement 1). Together our reporter lines and qPCR analyses suggest that, despite 87% protein-sequence similarity and expression of the two genes in the same cell, both Tbx2a and Tbx2b are independently required for the specification of zebrafish UV cones. Loss-of-function of either gene leads to a decrease in UV cones and a concomitant routing of photoreceptor progenitors towards a rod fate. In FØ[tbx2b] routing towards a rod fate appears to be stronger than in FØ[tbx2a].
Tbx2a inhibits M-opsin expression in L cones
After ascertaining the requirement of Tbx2a and Tbx2b in UV-cone specification, we examined whether either of these transcription factors impacted the specification of other photoreceptor subtypes. In addition to expression in UV cones, we detected significant enrichment of tbx2a in L cones, albeit with expression levels lower than in UV cones (Figure 4B). Furthermore, our qPCR quantification of opsins revealed a significant increase in M-opsin expression in tbx2a mutants—specifically of opn1mws2—along with a small decrease in L-opsin expression—specifically of opn1lw2 (Figure 3 – Figure Supplement 1). Based on these results, we tested whether Tbx2a is involved in M-cone or L-cone specification.
To examine M and L cones, we assessed FØ[tbx2a] larvae using an M-cone reporter line, where GFP expression is under direct control of the M-opsin promoter, Tg(opn1mws2:GFP), in combination with an L-cone reporter line, Tg(thrb:tdTomato). In control larvae, the expression of GFP and tdTomato is non-overlapping, reflecting the distinct fate of M cones and L cones (Figure 5A, top). In FØ[tbx2a], we found no significant changes in the number of L cones (identified by their tdTomato expression) (Figure 5B), and a dramatic increase in the number of GFP-positive cells (pre-sumptive M cones). Interestingly, in FØ[tbx2a], many GFP-positive cells co-express tdTomato (L-cone marker)—a phenotype which is not present in control larvae (Figure 5A, middle). We quantified the fraction of tdTomato-positive L cones with significant GFP expression in control larvae (see methods) and found that, only a small fraction of L cones are classified as double positive (mean ± s.d.: 5.2% ± 6.0). In comparison, in FØ[tbx2a], this fraction is significantly higher (mean ± s.d.: 37.5% ± 18.9%, p < 0.01) (Figure 5C). This abnormal expression of GFP in L cones in FØ[tbx2a] larvae suggests a loss of inhibitory control over the M-opsin promoter, and is supported by the increase in M-opsin expression identified through qPCR. As a control, we repeated this M-cone and L-cone assessment in FØ[tbx2b] larvae—despite no detectable expression of tbx2b in M or L cones (Figure 3B). While FØ[tbx2b] also had a significant increase in the number of GFP-positive cells (see next section) (Figure 5A, bottom), there were no changes in the number of tdTomato-positive L cones (Figure 5B) or in the fraction of L cones with significant GFP expression (mean ± s.d.: 2.6% ± 1.8%, p = 0.27) (Figure 5C).
These results suggest that Tbx2a, but not Tbx2b, is important for L-cone identity. While not required for their initial specification, upon maturation, Tbx2a helps preserve L cone identity. Without Tbx2a, L cones are unable to suppress M-opsin expression (58). Overall, analysis of FØ[tbx2a] larvae revealed that Tbx2a is important for both UV-cone specification and maintaining L-cone identity.
Tbx2b inhibits M-opsin expression in S cones
After identifying an additional role for Tbx2a, we turned our analysis to Tbx2b. Our RNA-seq revealed that in addition to UV cones, tbx2b is expressed in S cones (Figure 4A). Furthermore, our qPCR quantification also showed an increase in M-opsin expression in FØ[tbx2b]—specifically of opn1mw1 and opn1mw2 (Figure 4 - figure supplement 1)—and in S-opsin expression. Based on these results, we tested whether Tbx2b is involved in S- or M-cone specification.
For our analysis of S and M cones in FØ[tbx2b] larvae, we used the M-opsin reporter line, Tg(opn1mws2:GFP), in combination with an S-cone reporter line, Tg(opn1sw2:nfsB-mCherry). In control larvae, expression of the reporter proteins is largely non-overlapping, except for a small fraction of S cones that consistently expresses GFP (Figure 6A, top) (42). In FØ[tbx2b], we did not find significant changes in the number of S cones (identified by mCherry expression) (Figure 6B). As described above, in FØ[tbx2b], we observed a clear increase in the number of GFP-positive cells (presumptive M cones). Furthermore, in FØ[tbx2b] larvae, we found that this increase in GFP expression was restricted to S cones, which are double-positive for GFP and mCherry expression (Figure 6A, bottom). We quantified the fraction of mCherry-positive S cones with significant GFP expression, and found that, in control larvae, this fraction is low (mean ± s.d.: 9.2% ± 10.2%). In comparison, in FØ[tbx2b] this fraction is significantly higher (mean ± s.d.: 57.6% ± 32.2%, p < 0.01) (Figure 6C). This abnormal increase in GFP expression in S cones in FØ[tbx2b], combined with the increase in M-opsin expression found in our qPCR analysis, indicates a loss of inhibitory control over the M-opsin promoter. As a control, we repeated this S- and M-cone assessment in FØ[tbx2a]. We again observed an increase in GFP-positive cells in FØ[tbx2a] but without any significant changes in the number of S cones (Figure 6B) or in the fraction of mCherry-positive S cones with significant GFP expression (mean ± s.d.: 8.3% ± 4.9, p = 0.81) (Figure 6C). These results again corroborate our RNA-seq data showing that Tbx2a is not expressed by S cones and therefore is not involved in S-cone determination.
In summary, further analysis of FØ[tbx2b] revealed that, while not involved in the initial specification of S cones, upon maturation, Tbx2b helps to maintain S-cone identity. Without Tbx2b, S cones are unable to suppress the expression of M-opsin. Overall, analysis of FØ[tbx2b] larvae revealed that Tbx2b is important for both UV-cone specification and maintaining S-cone identity.
Discussion
Our work provides a valuable resource to accelerate discovery in photoreceptor and retinal biology. We have generated transcriptomic profiles from photoreceptors with unmatched depth and purity. These transcriptomes can be used to explore gene expression differences across photoreceptor subtypes. Importantly, we are able to reliably identify transcription factors that play a central role in controlling fate decisions during photoreceptor development. We also demonstrate how FØ screening can be applied as a rapid, efficient, and flexible platform to create and study loss-of-function phenotypes relevant to photoreceptor specification. In our study, we apply FØ-screening to investigate loss-of-function phenotypes, associated with the transcription factors Foxq2, Tbx2a and Tbx2b, and find a strong correlation between phenotype and expression pattern (Figure 7). Together these methods provide an excellent in vivo setting to discover the function of other novel genes identified in our RNA-seq dataset.
Relation to other transcriptomic datasets
Three recent studies have derived transcriptomes from zebrafish retinal cells and contain information from adult photoreceptors that provide an excellent resource to benchmark the quality of transcriptomes presented here. In our study, we derived samples using manual collection for a cell-type specific, SMART-seq2-based approach (38). The other recent studies used a variety of methods to segregate cell types in the retina. Rod transcriptomes were obtained by fluorescent-activated cell sorting (FACS) (Sun et al., 2018) (64). Retinalcell transcriptomes were obtained using a single-cell dropletbased (dropSeq) approach in adults and at several time points during development (Hoang et al., 2020) (43). Finally, transcriptomes from adult zebrafish photoreceptors were obtained by enrichment through FACS followed by dropSeq (Ogawa and Corbo, 2021) (44).
We find that there are general consistencies across these datasets, which can be exemplified by focusing on photo-transduction genes: we identify rod-enrichement in 26 out of the 27 phototransduction genes that are known to be rodspecific (the exception, cngb1b, previously named si:dkey-44k1.5, was found as expressed in all photoreceptors), and 22 are also identified as rod-enriched in Sun et al., 2018. Our datset also identifies cone-enrichement in 31 out of the 35 phototransduction genes known to be cone specific, with sparse or null counts for rcvrnb, gnb3a, pde6hb and opn1mw4. We did not detect opn1mw4 because we used Tg(opn1mw2:GFP) to collect M cones. Out of the total 62 known phototransduction genes, 58 display very similar subtype-specific expression patterns in Ogawa and Corbo, 2021. We found that these subtype-specific expression patterns are obscured in Hoang et al., 2020 due to contamination with rod transcripts in all the retinal cells derived from adults. In this study many known rod-specific genes are present in all photoreceptor subtypes (Figure 1 - figure supplement 3A). Rods are the predominant cell type in the zebrafish adult retina—constituting ~40% of all photoreceptors (39). In our experience, rods are fragile during dissociation and rod contamination presents a challenge to obtaining pure, subtype-specific datasets. Rhodopsin (rho) detection in nonrod samples is a simple way to assess contamination. We find that samples in Sun et al., 2018 and in Hoang et al., 2019 have significant rod contamination (> 15%), while in Ogawa and Corbo, 2021 and in the data presented here, the rod contamination is low (< 5%) (Figure 1 – figure supplement 3B). Transcriptome depth, quantified as the number of reads per sample for our study and for Sun et al., 2018 or reads per cell for the dropSeq studies, was considerably higher in our study (Figure 1 – figure supplement 3C).
We then focused on expression of the targets of our FØ-screen—foxq2, tbx2a, and tbx2b. The phenotypes of our targets relate directly to the expression patterns derived from our transcriptomes. We find that similar to our rod samples, transcriptomes from FACS-rods have low counts for these three transcription factors. The transcriptomes obtained from dropSeq show enrichment of: foxq2 in S cones (but also significant expression in UV cones), tbx2a in UV cones (but not clearly in L cones), tbx2b in S cones (but not clearly in UV cones). Finally, transcriptomes from FACS-dropSeq photoreceptors matched our data for foxq2 and tbx2b, but the expression of tbx2a in L cones was missed (Figure 1 – figure supplement 3D).
In summary, we find that the methods presented in this study are especially useful to generate high-quality transcriptomes of targeted cells. High depth and low contamination increase the statistical confidence and allow the detection of genes expressed at relatively low levels (e.g., tbx2a expression in L cones). Our method nicely complements dropSeq approaches which sample many more cells, which is especially advantageous for discovering new cell types or tracking developmental trajectories. In our view, these techniques are complementary and integration across datasets is important. To facilitate such comparisons, we have created an interactive plotter that integrates the reanalysis across the datasets outlined here. Our interactive plotter is openly available for the community and allows easy exploration and direct comparisons of all datasets (https://github.com/angueyraNIH/drRNAseq/), including code and data needed to replicate our analyses.
Reliability and efficiency of FØ screening
The RNAseq data and analyses presented here provide a substantial number of candidate genes potentially involved in the specification of photoreceptor subtypes. The generation of loss-of-function mutants remains a cornerstone in the study of gene function. Nevertheless, creating true mutants for all candidate genes would require excessive effort and resources, especially given the need for crosses with the relevant reporter lines. We propose that using an FØ screen in this context is an advantageous method to accelerate discovery and will enable future resources to be focused adequately. To validate this method and to quantitatively benchmark its reliability and efficiency, we first focused on foxq2, an Scone specific transcription factor with high expression and reliable detection in S cones across datasets (Figure 1 - figure suppelment 3C). A recent report, published as we completed this study, independently identified foxq2 as a cone-enriched transcription factor and characterized a foxq2 full knockout in zebrafish (24), allowing us to make direct comparisons between true germline mutants and genetic mosaic FØ[foxq2] larvae. In germline foxq2 mutants, there is a negligible expression of S opsin and a complete loss of S cones, without any apparent change in the number of any other photoreceptor subtype (Figure 7). In our FØ[foxq2] larvae, we found an average loss of S cones of 80%, without changes in the densities of the other photoreceptor subtypes. This demonstrates that the methods presented here are able to create mutations in a targeted gene with high efficiency and, to reproduce the phenotype of a full knockout without noticeably affecting related cell types. Moreover, the ability to evaluate the effect of gene mutations in less than a week, in any genetic background—transgenic lines or other mutant lines to investigate genetic pathways and networks—makes FØ screening a versatile technique to identify other transcription factors that may be involved in photoreceptor development and subtype specification.
Tbx2 is a master regulator of photoreceptor fate
After the success of uncovering phenotypes in FØ[foxq2], we explored the effects of tbx2 mutations in photoreceptor identity. Our analyses revealed that Tbx2 is connected to the fate of all photoreceptor subtypes in zebrafish (Figure 7).
First, we showed that tbx2a and tbx2b are both expressed in UV cones, and the loss of either gene impairs UV-cone specification. The high conservation in the amino acid sequence of TBX2 across vertebrates and the specific expression in evolutionarily related cone subtypes (opn1sw1-expressing photoreceptors in zebrafish, chicken and primates) (46, 59) suggests that TBX2 may play a similar role across vertebrate species. We find that loss of UV cones in either tbx2a or tbx2b mutants is associated with an increase in the number of rods during development. The switch in fate from UV cone to rod suggests that Tbx2a and Tbx2b play a role in an early fate decision in photoreceptor progenitors, allowing the acquisition of UV-cone identity by actively repressing the expression of rod genes. Interestingly the increase in rods (or rhodopsin expression) was not equal between FØ[tbx2a] and FØ[tbx2b], suggesting that the two transcription factors regulate downstream targets differently. In addition, in vitro experiments have shown that Tbx2 binds to DNA as a monomer (60), which makes the possibility of Tbx2a/Tbx2b dimers unlikely. It still remains a mystery why the specification of UV cones in zebrafish would require a “two-factor authentication” system that relies on two highly-homologous but independent transcription factors.
Second, we show that tbx2a and tbx2b are expressed in L cones and S cones, respectively. Further, Tbx2a and Tbx2b help maintain L-cone and S-cone identity by repressing the expression of M opsin in vivo. A recent study in cichlids demonstrated that Tbx2a can bind and directly regulate the M-opsin promoter in vitro (58). This work also found that expression of tbx2a correlated strongly with the relative expression of M and L opsins, which cichlid species use to adjust their overall spectral sensitivity and match the requirements imposed by their habitats. These findings highlight that Tbx2 plays a late role in L-cone and S-cone fate, helping to maintain their identity after specification.
Outlook
While conducting the experiments described in this paper, we learned a few lessons worth highlighting. First, we find that manual picking targeted cell types allowed us to focus on collecting healthy cells and generate transcriptomes of high depth and quality. Another important advantage of this method is that barriers imposed by a cell type with a low density can be largely ignored, as long as the targeted cell types can be recognized. For this reason we think it would be interesting to apply this technique to fully understand further subdivisions of each photoreceptor subtype including the differences between opn1lw1- and opn1lw2-expressing L cones (65) or between opn1mw4-expressing M cones and other M cones in zebrafish (44). Furthermore it would be interesting to explore regional specializations across the retina like the one proposed for UV cones in the acute zone (66), and for fovea vs. periphery differences in primates (46). This technique is likely to also be useful beyond photoreceptors to dissect differences between subtypes of other retinal cells.
Second, we find it is critical to create fast and easy access to multiple transcriptomic datasets. Eliminating technological barriers is important to ensure data can be accessed by all users. By ensuring proper access, new hypotheses pertaining to factors involved in photoreceptor development and other aspects of photoreceptor biology can be more readily explored. For example, many orthologs of human genes associated with retinal degenerations show high expression in zebrafish photoreceptors. For these reasons, we have taken a special effort to provide an interactive plotter that allows open exploration of all datasets in a single place. We hope that this tool is valuable to the scientific community.
Third, the results of our FØ-screen highlight some important features of how photoreceptors acquire their final identity. The process of specification seems to require several stages: defects in early stages can lead to a loss of subtypes (e.g. S cones in foxq2 mutants) or to a change in identity (e.g. rods and UV cones in tbx2 mutants), while defects in later stages can lead to specification deficits without loss of cells (e.g. misexpression of M opsin in tbx2 mutants). In addition, while some transcription factors (like Foxq2 but also Thrb for example) mainly play a role in activating a particular fate, it is clear that others (like Tbx2, but also shown for Nr2e3 or Prdm1) play mainly a role in inhibiting the fate of other cell types (49, 67). Because of its conserved sequence and expression, TBX2 may play a similar role in mammalian S cones—actively repressing the fate of rods and M/L cones. Such active repression is most likely a fundamental mechanism to maintain subtype identity throughout the life span of an organism. These mechanisms of cell identity echo beyond photoreceptors into the context of specification of any cell type.
Finally, although we focused on the differential expression of transcription factors because of their central role in controlling fate decisions, similar approaches can be used to identify and study the function of genes involved in phototransduction, metabolism, ciliary transport, synaptic machinery, or any other aspect of photoreceptor biology. The dataset and methods described here are an excellent resource to propose hypotheses, to generate an initial list of candidate genes and to perform efficient screening.
Materials and Methods
Animals
We grew zebrafish larvae at 28°C in E3 embryo media (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl2, and 0.33 mM MgSO4) and kept them under a 12 h:12 h light-dark cycle. At 1 dpf, we added 0.003% 1-phenyl-2-thiourea (PTU) to the embryo media to block melanogenesis. All work performed at the National Institute of Health was approved by the NIH Animal Use Committee under animal study protocol #1362-13. For RNAseq samples with adult zebrafish, animals of both sexes were used. For FØ, larvae were examined at 5 dpf. At these ages, sex cannot be predicted or determined, and therefore sex of the animals was not considered. Transgenic lines used in this study are listed in Table 1.
RNA-seq sample collection
We euthanized adult zebrafish by immersion in ice-cold water (below 4°C) followed by decapitation. We pierced the cornea with a 30-gauge needle and removed the cornea and lens before performing enucleation. Once the eye was isolated, we gently separated the retina from sclera and RPE using fine forceps or electrically-sharpened tungsten electrodes (69) and immediately started incubation in papain solution (5 U/mL papain Calbiochem#5125, 5.5 mM L-Cysteine, 1 mM EDTA in divalent-free Hank’s balanced salt solution) for 35 minutes at 28°C. After a brief wash in DMEM supplemented with 5% bovine serum albumin, we performed mechanical trituration of the retina with the tip of a 1 mL pipette and used a cell-strainer polystyrene tube to obtain a single-cell suspension. After spin-down (2000x G for 2 min), we resuspended cells in 500 μL of enzyme-free fresh DMEM and diluted the cell suspension into three serial 10-fold dilutions before plating in glass-bottom petri dishes. The dilutions ensured that we could find a preparation where the density of cells and debris was low and most photoreceptors were truly isolated. We inspected the cell suspension using an epifluorescence microscope (Invitrogen EVOS cell-imaging system) and collected 20 photoreceptors per retina based on their fluorescence and morphology (prioritizing cells that looked healthy, had intact outer segments, visible mitochondrial bundles and undamaged cell membranes) using an oil-based microinjector system (Eppendorf CellTram 4R) and glass pipettes with a 15 μm opening (Eppendorf TransferTip-ES). After collection, we resuspended photoreceptors in 1 μL of fresh PBS, rein-spected cells for fluorescence, collected them in a PCR tube containing 8 μL of lysis buffer of the RNA kit and kept the tube on ice until cDNA libraries were prepared. We used the SMART-seq v4 ultra-low input RNA kit for sequencing (Takara #634897) using the manufacturer’s instructions for single-cell samples, followed by the Low Input Library Prep Kit v2 (Takara #634899). We pooled up to 12 samples (with different barcodes) in one lane of a flow cell for sequencing (Illumina HiSeq 2500) and used a 150 bp paired-end read configuration. The first sequencing batch contained 4 UV-cone and 4 S-cone samples in a single flow cell, and the second sequencing batch contained the rest of the samples divided across 2 flow cells (6 rod, 1 UV-cone, 2 S-cones, 7 M-cones and 6 L-cone samples).
RNA-seq data analysis
After an initial quality control and trimming of primer and adapters sequences using Trimmomatic (70), we aligned reads to the zebrafish genome (Danio rerio.GRCz11) using HiSat2 (71) and we assembled and quantified transcripts using Stringtie (72). We performed differential expression analysis using DEseq2 and pcaExplorer for initial visualizations (73, 74). Genes were considered as differentially expressed if |fold-enrichment|> 1.5, p-value < 0.01 and the estimated false positive rate or p-adjusted <0.1. In addition, genes were required to have positive reads in > 50% of the enriched samples. To be able to detect differences that relied on expression on just 1 or 2 cone subtypes, we removed the requirement on fold-enrichment in rod vs. cone comparisons. To further explore the data, we transformed read numbers into fragments per kilobase per million reads (FPKM) (Supplementary Data 01) and developed custom routines in Python for plotting. We subselected transcription factors by selecting genes identified with “DNA-binding transcription factor activity” in ZFIN (75) and repeated principal component and differential expression analyses (Supplementary Data 02 and 03). To ensure broad access to our transcriptomic data, we provide access to the raw data (GEO accession number GSE188560), and after analysis in several formats including as a plain csv file, as a Seurat object for easy integration with dropSeq datasets (76), and finally as an interactive database for easy browsing and visualization (https://angueyraNIH.github.io/drRNAseq/). In order to make direct comparisons between our data and other RNAseq studies, we have integrated visualizations that use their publicly available data. For rod transcriptomes obtained using FACS we used the provided analyzed data, which includes gene log counts per million (cpm) for 4 rod samples (GFP-positive cells) and 4 non-rod samples (GFP-negative retinal cells) (GSE100062) (64). For transcriptomes from adult photoreceptors obtained using FACS followed by dropSeq (44), we used the Seurat object provided by the authors (GSE175929) and we used custom scripts in R using RStudio and Seurat (76–78) to export average expression values and percent of cells with positive counts of each gene for each cluster. For trancriptomes of retinal cells obtained using dropSeq (43), we used the Seurat object for zebrafish development provided by the authors (http://bioinfo.wilmer.jhu.edu/jiewang/scRNAseq/), and we updated the object to Seurat v03 (79), extracted cells that corresponded to adult rods and cones, performed clustering and used the expression of opsins and other markers to identify cone subtypes (including arr3a for L and M cones, arr3b and tbx2b for UV and S cones, thrb and si:busm1-57f23.1 for L cones and foxq2 for S cones). All results and scripts necessary to recreate this analyses are also provided openly (https://github.com/angueyraNIH/drRNAseq).
Quantitative PCR (qPCR)
We euthanized groups of 20 to 30 zebrafish larvae by immersion in ice-cold water (below 4°C) and immediately performed RNA extraction using the RNeasy Mini Kit (Qiagen) and reverse transcription using the High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher), which relies on random primers. Samples were kept frozen (−20°C) until use. For qPCR assays we used the PowerUp SYBR Green Master Mix (Thermo Fisher) and a 96-well system (CF96, Biorad) following manufacturer’s protocols. We estimated expression levels using the relative standard curve method, using 5 serial standard dilutions of cDNA obtained from wild-type larvae. To calculate fold differences in gene expression, we normalized transcripts levels to the levels of β-actin 2 (actb2), and all measurements were repeated in triplicates. We performed statistical testing using Mann-Whitney rank sum tests, with a p-value < 0.05 required for significance. All primers used for qPCR are provided in Table 2.
FØ-CRISPR screening
We designed guide RNAs (gRNAs) using the online resource CHOPCHOP (80). We picked guides that targeted exons that encode the DNA-binding domains of transcription factors, had no self-complementarity, and that had 3 or more mismatches with other regions of the zebrafish genome. We used purified Cas9 protein (Alt-R® S.p. Cas9 nuclease, v.3) and chemically synthesized AltR-modified crRNA and tracr-RNA (Integrated DNA technologies) for injections (35) (Table 3). We prepared 1 μL aliquots of a 25 μM stock solution of Cas9 protein diluted in 20 mM HEPES-NaOH (pH 7.5), 350 mM KCl and 20% glycerol, and stored them at −80°C until use. We diluted each target-specific crRNA and the common tracrRNA using the provided duplex buffer as a 100 μM stock solution and stored them at −20°C. We prepared a 50 μM crRNA:tracrRNA duplex solution by mixing equal volumes of the stock solutions followed by annealing in a PCR machine (95°C, 5 min; cooling 0.1°C /s to 25°C; 25°C for 5 min; rapid cooling to 4°C), then we used the duplex buffer to obtain a 25 μM stock solution, before mixing equal volumes of the guides targeted to a single gene (3 guides for tbx2a and tbx2b, 2 guides for foxq2), making aliquots (3 μL for tbx2a and tbx2b, 2 μL for foxq2) and storing at −20°C until use. Prior to microinjection, we prepared 5 μM RNP complex solutions by mixing 1 μL of 25 μM Cas9, 1 μL of 0.25% phenol red and 3 μL of the tbx2a or tbx2b duplex solution, or 1 μL of pure water and 2 μL of the foxq2 duplex solution. We incubated the RNP solution at 37°C for 5 minutes and kept at room temperature for use in the following 2 – 3 hours. We injected ~1 nL of the 5 μM RNP complex solution into the cytoplasm of one-cell stage zebrafish embryos.
Genotyping
We extracted DNA from the bodies of larvae (5 dpf) after enucleation by placing them in 25 μL of 25 mM NaOH with 0.2 mM EDTA, heating to 95°C for 30 minutes, and cooling to 4°C. Then we neutralized the solution by adding 25 μL of 40 mM Tris-HCl and vortexed the samples. For genotyping, we used a fluorescent PCR method (81), and all primers and expected sizes are provided in Table 4. We added the M13F adapter sequence (5’-TGTAAAACGACGGCCAGT-3’) to forward primers and the PIGtail adapter sequence (5’-GTGTCTT-3’) to reverse primers and used incorporation of fluorescent M13F-6FAM for detection. Our PCR mixture (1x), for a 20 μL reaction, contained forward primer (0.158 μM), reverse primer (0.316 μM), M13-FAM (0.316 μM, IDT), Phusion HF PCR Master Mix (1x, BioLabs), water (6.42 μL), and 2 μL of DNA. We used the following PCR protocol: 1) 98°C denaturation for 30 seconds, 2) 34 cycles of 98°C for 10 seconds, 64-67°C for 20 seconds, 72°C for 20 seconds 3) final extension at 72°C for 10 minutes, 4) hold at 4°C.
Immunohistochemistry
We fixed zebrafish larvae at 5 dpf in 4% paraformaldehyde in phosphate buffered saline (PBS) for 1 hour at room temperature, followed by washes with 1% Triton X-100 PBS (3 x 10 min). We incubated larvae in primary antibodies diluted in 2% normal donkey serum (Jackson ImmunoResearch) and 1% Triton X-100 PBS for five days at 4°C with continuous and gentle shaking. To label S cones, we used a rabbit polyclonal anti-blue opsin (Kerafast EJH012) in a 1:200 dilution. After incubation with primary antibodies, we performed washes with 1% Triton X-100 PBS (3 x 15 min). We incubated larvae in donkey polyclonal secondary antibodies labeled with Cy5 (Jackson ImmunoResearch) in 1% Triton X-100 PBS overnight at 4°C with continuous and gentle shaking and performed washes in 1% Triton X-100 PBS (3 x 15 min) before mounting.
Imaging
Sample Preparation and Image Acquisition
For imaging, we enucleated eyes from fixed larvae using electrically sharpened tungsten wires (69). We placed isolated eyes on a coverslip and oriented eyes to place photoreceptors closest to the coverslip before using a small drop of 1.5% low-melting point agarose to fix them in place. Upon solidification, we added a polyvinyl-based mounting medium (10% polyvinyl alcohol type II, 5% glycerol and 25 mM Tris buffer, pH 8.7) and placed the coverslip on a glass slide, separated by a spacer (Grace Biolabs and/or duct tape) to avoid compression. We used the bodies of the larvae for genotyping and imaged the corresponding larval retinas using a Nikon A1R resonant-confocal microscope with a 25x, 1.10 NA waterimmersion objective. We acquired z-stack images from a 64 μm x 64 μm square area of the central retina (dorsal to optic nerve) for photoreceptor quantification every 0.4 – 0.5 μm at a 1024 x 1024 pixel resolution.
Image Analysis
Photoreceptor quantification: We imported confocal z-stacks of the central region of the retina (64 μm x 64 μm) into Napari (82). We created maximum intensity projections (MIPs) using a small subset of the z-stack (2 – 10 planes) that ensured that we captured all photoreceptor cells in the region into a single image. We then used the Napari plugin of Cellpose, a machine-learning based segmentation algorithm, to segment photoreceptors in each image, using the cyto2 model (83). Finally, we manually corrected the segmentation to ensure all photoreceptors were properly counted. We performed statistical comparisons for counts of each photoreceptor subtype between wt and FØ larvae using Mann-Whitney rank sum tests, with a p-value < 0.01 required for statistical significance.
Identification of double-positive cells in tbx2 mutants: The increase in the Tg(opn1mws2:GFP)+ cells in FØ[tbx2a] and FØ[tbx2b] larvae made segmentation of the green channel difficult and unreliable, as these additional cells did not conform to the normal spatial separation between M cones. For this reason, we exploited the more accurate segmentation of L cones and S cones using the red channel, when imaging Tg(thrb:tdTomato) or Tg(opn1sw2:nfsB-mCherry), respectively, and used it to create masks for the gren channel. We normalized the GFP signal across the whole image to span a 0 – 1 range (to be able to make comparison between images) and used a 10-pixel erosion (to avoid effects due to optical blurring during imaging of the GFP signal) before calculating the average normalized GFP signal contained within each S-cone or L-cone. By plotting the distribution of GFP signal in L cones, we were able to establish a threshold of 0.195 that was exceeded by only 5.2% of L cones in wt larvae and used it to classify L cones as GFP+ in both wt and FØ larvae. In the original work that established the Tg(opn1mws2:GFP) line, it was noted that a subset of S cones in wt larvae are GFP+ (42). We were able to identify these cells using a GFP signal threshold of 0.255 (9.2% of wt S cones), and again we used this same threshold to quantify the fraction of GFP+ S cones in both wt and FØ larvae.
Statistical analyses
We performed statistical analyses and data plots using Python in Jupyter notebooks (84). Values of data and error bars in figures correspond to averages and standard deviations, and for statistical comparisons we used Mann-Whitney U tests. Samples sizes and significant levels are stated in the figure captions. No randomization, blinding, or masking was used for our animal studies, and all replicates are biological. For RNAseq, we performed an initial sequencing run after collecting dissociated photoreceptors in squirrel (38) and zebrafish and established that a minimum of four samples per subtype were required to establish reliable statistical significance in differential gene-expression analysis. For FØ screening, our initial experiments were aimed at replicating the loss of UV cones and the increase in rods reported for tbx2b mutants (23), and we established that a minimum of 7 injected larvae per group were needed to provide enough statistical power in photoreceptor quantifications in FØ larvae. Injected larvae that had normal (wild-type) genotypes— a sign that CRISPR mutagenesis was not successful—were excluded from analysis.
Supplemental Materials
Supplementary data 01: Differential gene expression in zebrafish photoreceptors. Collection of CSV files containing output of differential gene expression analysis using De-SEQ2, along relevant directions (rods vs. cones; (UV+S) vs. (M+L); M vs. L; UV vs. S) and including counts (in FPKM) for all detected genes.
Supplementary data 02: Differential transcription factor expression in zebrafish photoreceptors: rods vs. cones. CSV file containing transcription factors with significant differential expression between rod and cone samples.
Supplementary data 03: Differential transcription factor expression in zebrafish photoreceptors: cone subtypes. Collection of CSV files containing transcription factors with significant differential expression between cone subtypes.
Data Availability
Sequencing data have been deposited in GEO under accession GSE188560. Data can be visualized at https://github.com/angueyraNIH/drRNAseq.
Declaration of Interests
The authors declare no competing or financial interests
Acknowledgements
This work was supported by the National Eye Institute Intramural Research Program (W.L.), the National Institute on Deafness and Other Communication Disorders Intramural Research Program (1ZIADC000085-01, K.S.K), and National Eye Institute Pathway to Independence Award (K99EY030144-01, J.A.). We would like to thank Jamie Sexton, Alisha Beirl and Katherine Pinter for all the animal care and technical support, members of the Kindt Lab and the Li Lab for useful discussions, and Matthew Brooks, Linn Gieser and Anand Swaroop for sequencing services. We are very grateful to Rachel Wong, Takeshi Yoshimatsu, Ralph Nelson, James Fadool, Steven Leach, Brian Perkins and Xiangyun Wei for providing the transgenic zebrafish lines used in this study.
Bibliography
- 1.↵
- 2.↵
- 3.↵
- 4.
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.
- 12.↵
- 13.↵
- 14.
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.
- 27.↵
- 28.↵
- 29.
- 30.
- 31.
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.
- 41.
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.
- 51.
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵