Photoreceptors are highly specialized cell types in the retina that “see” light. Light photons captured by photoreceptors are converted to electrical signals that travel through the optic nerve to the brain and form vision. In vertebrates, photoreceptors come in two major classes−rods and cones. The genesis and development of rods and cones follow a stereotypical order programmed by a photoreceptor gene regulatory network. This regulatory network also operates in adult retinas to ensure robust photoreceptor functions and cellular integrity. Components of this network and early events that regulate photoreceptor cell fate determination have been reviewed extensively (Swaroop et al., 2010; Bassett and Wallace, 2012; Brzezinski and Reh, 2015; Cepko, 2015; Wang and Cepko, 2016) and are not covered here. Rather, this review summarizes the mechanisms that regulate rod and cone differentiation after their fate is acquired through the lens of CRX studies in development and diseases. Most findings are based on the mammalian model organism Mus musculus, which provides the most comprehensive evidence on CRX protein functions.

The Cone-rod homeobox (CRX, OMIM: 602225, UniProt: O43186) gene encodes a homeodomain transcription factor that regulates gene expression programs essential for photoreceptor development, function, and maintenance. Coding variants in CRX have been associated with at least three types of retinopathies that result in blindness, including Leber Congenital Amaurosis (LCA), Cone-rod Dystrophies (CoRD), and Retinitis Pigmentosa (RP). To date, CRX is the only gene known to be associated with all three conditions, underscoring its critical role in both cone and rod biology. It is, therefore, important to understand CRX’s mechanisms of action in photoreceptors. Here, we review recent progress in elucidating CRX molecular functions in photoreceptor development and diseases. We highlight an integrated approach that draws on quantitative in vitro biochemical models, functional genomics in variant knock-in mouse retinas, and high-throughput screens built on systems biology principles. This holistic approach uncovers complex and intricate CRX regulatory principles that are otherwise elusive using conventional methodologies. These newly identified CRX regulatory principles explain the distinct pathogenic mechanisms in animal models, facilitate the functional predictions of other CRX coding sequence variants identified in clinical studies of CRX-linked diseases, and inform gene therapy development and precision medicine. We envision such an integrated approach is readily transferable to the study of transcription factors in other retinal cell types and their associated diseases.

In 1997, three laboratories independently reported the cloning of the CRX/mCrx gene using complementary methods, including yeast one-hybrid system, cDNA hybridization, and degenerative RT-PCR (Chen et al., 1997; Freund et al., 1997; Furukawa et al., 1997). These studies demonstrated that CRX/Crx encodes a 299 amino acid sequence-specific DNA-binding protein, and it recognizes regulatory elements in the promoter of rhodopsin, a gene that encodes the rod-specific photopigment. The predicted human CRX and mouse CRX protein sequences only differ by 10 amino acids with 100% identity in the DNA binding domain (Figures 1, 2A). In addition, CRX protein shares sequence similarity in the DNA binding domain with many other homeobox family members implicated in early brain and eye development, including CHX10 (VSX2), OTX1, and OTX2 (Chen et al., 1997; Furukawa et al., 1997; Zheng et al., 2023). DNase I footprinting and transcription reporter assays identified CRX binding sites at photoreceptor gene regulatory sequences and demonstrated CRX’s primary function as a transcription activator. Multiple sequence alignment of the CRX bound and activated promoter sequences revealed an enriched DNA motif - CTAATC[C/T] – similar to that of the well-characterized D. melanogaster Bicoid homeodomain protein (Hanes and Brent, 1989, 1991). Protein truncation studies identified a C-terminus transcription effector domain (Figure 2A) responsible for CRX-mediated gene activation (Chau et al., 2000; Chen et al., 2002). Collectively, these early studies demonstrated that CRX is a homeodomain transcription factor that regulates photoreceptor gene expression and laid the foundation for CRX studies in animal models.

In vivo, Crx’s spatial and temporal expression patterns correlate with its roles in photoreceptor development and maintenance. In the mouse, cone genesis starts on embryonic day 10 (E10), and the last cones are born in the periphery on E18 (Carter-Dawson and Lavail, 1979; Young, 1985). Rod genesis partially overlaps that of cones, spanning from E13 to post-natal day 7 (P7). The peak of rod genesis is around the time of birth of the animal (P0). The expression of Crx transcripts is first detected at E12.5, localized to the outer aspect of the neural retina, corresponding to developing cones (Chen et al., 1997; Furukawa et al., 1997; Aavani et al., 2017). As cone genesis continues to increase after E12.5 and the initiation of rod genesis, the expression of Crx becomes stronger and remains restricted to the prospective photoreceptor layer. In the post-natal retina, Crx expression is observed throughout the prospective photoreceptor layer and reaches a peak at early post-natal ages (P6-7 in the CD1 strain and P3-5 in the C57BL/6J strain). Crx expression then slightly decreases before settling at a high level maintained in mature rods and cones throughout adult life. It is important to note that these measurements are taken at the tissue level, and the slight expression drop after the peak might be a consequence of programmed cell death during normal development (Young, 1984; Braunger et al., 2014). Crx expression dynamics at a single cell level remains an important question to be addressed.

The disruption of CRX expression or function profoundly impacts photoreceptor development and survival. In the developing mouse retinas, retrovirus-mediated ectopic expression of Crx in P0 progenitor cells increases the number of rod photoreceptor-only clones, suggesting an instructive role of Crx in rod photoreceptor fate during retinal development (Furukawa et al., 1997). Genetic ablation of Crx in mice (Crx-/-) does not affect the genesis of photoreceptors but prevents their terminal differentiation and leads to rapid degeneration of the immature photoreceptor cells on or before P21 (Furukawa et al., 1999). CRX likely functions in cells that already adopt photoreceptor fate, where it promotes and maintains photoreceptor-specific gene expression programs. Indeed, lineage studies confirm that Crx expression is activated in post-mitotic photoreceptor precursors (Muranishi et al., 2011). Interestingly, ectopic expression of a dominant-negative form of CRX in P0 progenitor cells, with its homeodomain fused to the repressor domain of the D. melanogaster Engrailed protein, completely blocked rod terminal differentiation (Furukawa et al., 1997). Since the dominant negative CRX was ectopically expressed at a cell state when endogenous CRX is not activated, it might have perturbed additional programs not normally regulated by CRX and changed the intrinsic potentials of these cells to fully develop. Nevertheless, it emphasizes that photoreceptor development – and the underlying photoreceptor gene expression – is very sensitive to small quantitative differences in CRX regulatory activity. Mutations that either increase or decrease CRX activity can lead to diseases.

To understand CRX’s mechanisms of action in vivo, its DNA binding sites genome-wide were identified by chromatin immunoprecipitation followed by sequencing (ChIP-seq) in the developing mouse retinas (Corbo et al., 2010; Zheng et al., 2023). Recently, by comparing the high-resolution CRX ChIP-seq data from WT and Crx mutant retinas, Zheng et al. identified a set of “CRX-dependent genes” that rely on CRX binding at their regulatory elements for expression (Zheng et al., 2023). Many of the CRX-dependent genes are essential for photoreceptor structures and functions. The capacity of CRX to bind these regulatory elements also depends on the interactions with nucleosomes, which are structural units of the chromatin (Luger et al., 2012; Ahmad et al., 2022). Nucleosomes can inhibit the binding of many transcription factors, including CRX, by occluding their binding sites. Chromatin remodeling is a critical step to ensure transcription factors and transcriptional machinery have physical access to DNA. A comparison of the chromatin landscape in WT and Crx KO mouse retinas revealed that CRX is required for chromatin remodeling at a subset of its binding sites (Figure 3A; Ruzycki et al., 2018). During photoreceptor development, this subset of CRX binding sites increases accessibility and undergoes retinal-specific acquisition of epigenetic modifications associated with active promoters and enhancers, likely through CRX-dependent recruitment of chromatin remodeling complexes. DNA motif discovery analysis revealed that the CRX binding-dependent accessible sites tend to have a single enrichment of CRX consensus motifs while CRX-binding independent accessible sites have enrichment of additional neuronal transcriptional factor motifs. CRX likely adopts both independent and collaborative modes of action in different genomic contexts to regulate photoreceptor gene expression.

Although no chromatin remodeling defects were observed at CRX binding-independent accessible sites, the possibility of genetic compensation cannot be ruled out. The absence of Crx transcripts and/or proteins may activate the transcription of related genes, partially compensating for the loss of CRX. Accumulated evidence demonstrates that genetic compensation is a highly regulated process such that it is only triggered by certain types of genetic lesions (El-Brolosy and Stainier, 2017; El-Brolosy et al., 2019; Ma et al., 2019). A mutant Crx encoding a defective CRX protein that does not trigger such a compensation mechanism may affect the CRX-independent accessible sites and lead to more severe perturbations in photoreceptor differentiation than Crx KO animals. This model may explain why the loss of one CRX allele seems tolerated in heterozygous carriers while some CRX coding variants are associated with severe dominant phenotypes in humans.

Crx is one of the earliest expressed photoreceptor-specific transcription factors, and its expression is essential for maintaining the expression of many downstream transcription factors and co-factors. It is important to note that in the Crx KO mouse retina, many of these downstream factors are expressed at early stages of photoreceptor development but diminish later, suggesting that initiation of their expression is independent of CRX, but their maintenance requires CRX. In addition to being a transcription activator, CRX also interacts – both directly and indirectly – with an array of transcription (co-)factors in stimulating photoreceptor gene expression (La Spada et al., 2001; Chen et al., 2004; Peng and Chen, 2005, 2007; Peng et al., 2005; Hennig et al., 2008; Onishi et al., 2009; Tomohiro et al., 2014; Andzelm et al., 2015). Extensive discussions on CRX and transcription (co-)factors in photoreceptor-specific gene regulation can be found in (Hennig et al., 2008; Swaroop et al., 2010). Here, we highlight two pairs of interactions – CRX-OTX2 and CRX-NRL – that are at the core of photoreceptor development.

In the most recent ClinVar release, 338 CRX coding variants have been documented – 80 annotated as pathogenic/likely pathogenic, 192 as uncertain significance/conflicting, 77 as benign/likely benign, and 11 as other. CRX coding variants are associated with at least three forms of inherited retinal disorders (IRDs) that cause blindness, including Leber congenital amaurosis 7 (LCA7, OMIM: 613829), Cone-rod dystrophy 2 (CoRD2, OMIM: 120970), retinitis pigmentosa (RP, OMIM: 268000). Figures 2C, D summarize our curated CRX coding variants identified in individuals with vision problems. Figure 2C presents the number of unique CRX variants at each amino acid position, and Figure 2D shows the type of protein sequence change at each position. A complete list of our curated set of CRX coding variants and accompanying references can be found in Supplementary Table 1.

CRX-linked retinopathies vary greatly in the age of onset, rate of progression, and severity, reflecting the complexity of CRX’s mechanisms of action and highlighting challenges in evaluating sequence variants in different genetic backgrounds. Despite heterogeneity in clinical phenotypes, most CRX coding variants arise de novo, appear completely penetrant, and cause diseases in heterozygotes (autosomal dominant). Multiple reports have described putative CRX null variants to be tolerated in heterozygous carriers or are associated with variable phenotypes in the family, preventing a conclusive genotype-phenotype correlation from being drawn (Silva et al., 2000; Jones et al., 2017; Ibrahim et al., 2018; Yahya et al., 2023). These patterns suggest that haploinsufficiency is not the key mechanism of pathogenesis for dominant CRX variants. Consistently, heterozygous deletion of Crx only produces very mild phenotypes in the Crx^+/– mouse retinas (Furukawa et al., 1999). Therefore, the Crx KO mouse does not provide an appropriate model for severe dominant CRX diseases.

To study CRX disease variants effectively, a categorical approach has been employed (Tran and Chen, 2014). In this paradigm, disease variants are categorized into four major classes based on their locations in CRX functional domains and the impacts on CRX biochemical properties (Figure 2E). For each class, one or more representative human variant knock-in mouse models (Figure 2F) were created and subjected to in-depth molecular and cellular characterizations (Supplementary Table 2). This approach has yielded invaluable insights into different pathogenic mechanisms, both developmental and degenerative, revealed the multifaceted roles CRX plays in regulating photoreceptor biology, and laid the foundation for developing targeted gene therapies against different disease mechanisms. In the following four sections, we briefly describe the major findings from mouse models of different pathogenic classes and discuss the lessons learned on CRX functions during normal development.

Once we understand how CRX regulates gene expression in normal conditions and representative variant knock-in animal models, we can make functional predictions of newly identified coding variants. Instead of testing one variant at a time, deep mutational scanning (DMS) is an emerging strategy to efficiently assay the functional consequences of hundreds to thousands of different protein variants in parallel (Fowler and Fields, 2014). The development of machine learning methods in protein structure prediction and pattern discovery has also significantly reduced the barriers to comprehending the enormous amount of data generated by typical DMS experiments. With deep mutational scanning, we can generate a lookup table of all possible single amino acid substitutions in CRX and even combinatory variants. We can then build on the categorical approach and associate every uncharacterized variant to one of the characterized variants by their similarity in affecting CRX’s intrinsic properties. Insights gained from Crx animal models will guide the design of DMS experiments, for example, by selecting the most biologically meaningful readouts for variants in different pathogenic classes.

Coding variants in CRX homeodomain impact CRX DNA-binding interactions and perturb target gene expression. Similarly, non-coding variants that change CRX DNA binding site sequences could perturb CRX target gene expression. Interactions between CRX coding variants and CRX binding site non-coding variants constitute another layer of complexity and may account for the missing heritability and phenotypic heterogeneity of a subset of disease-linked CRX coding variants (Zhang and Lupski, 2015). Since perturbation of many CRX target genes alone can lead to severe vision problems, understanding CRX regulatory functions globally and at the gene-specific level are both necessary. A comprehensive and quantitative CRX model built on in-depth animal studies and systems biology principles should fulfill such a need, considering both CRX intrinsic activities and additional factors, such as fluctuations of CRX protein levels, interactions with other factors, and the chromatin environment. One example is the application of Massively Parallel Reporter Assays (MPRAs) to assess the regulatory activities of CRX-bound genomic sequences in ex plant WT and disease variant knock-in mouse retinas (White et al., 2016; Hughes et al., 2018; Friedman et al., 2021; James et al., 2023). When combined with rationally designed mutagenesis libraries, one can start to understand the importance of CRX in different genomic contexts and predict the degree of impact on different genes in response to mutant CRX activity. This information could help prioritize non-coding variants−whether it protects or sensitizes a gene to CRX coding variants. Besides MPRAs, many assays have been developed to interrogate other aspects of transcription factor-gene expression relationships and can be readily adapted to the retinal system (Wang et al., 2012; Arnold et al., 2013; Dixit et al., 2016; Muraro et al., 2016; Maricque et al., 2019).

As a master transcriptional regulator, CRX controls many aspects of photoreceptor biology. It can be more challenging to treat diseases caused by variants in CRX than in a gene with a discrete function. With many years of studies on CRX in vitro and in animal models, we are now starting to explore different therapeutic strategies to target different pathogenic mechanisms. For example, supplementing the normal gene – known as gene-augmentation therapy – may be sufficient to treat loss-of-function variants, while simultaneous silencing or removal of the mutant allele will be critical for treating antimorphic variants. To the extent of data published, at least four groups have attempted proof-of-concept strategies both in animal models (Roger et al., 2022; Sun and Chen, 2023) and in organoid models (Chirco et al., 2021; Kruczek et al., 2021) targeting mutations in all four pathogenic classes. To advance gene therapy to treat CRX-associated disorders, there are many important questions await careful evaluation, such as developmental vs. degenerative pathogenic mechanism, effective treatment window, toxicity of CRX overexpression, and neuroplasticity of diseases photoreceptor cells (Sun and Chen, 2023). Nevertheless, these early studies demonstrate that, with careful design, gene therapy could be a viable strategy for CRX-associated diseases.

From 25 years of CRX research, we now know that photoreceptor gene regulation is a highly coordinated process that requires fine-tuned CRX transcription factor activities. Expanded from the simplest model – CRX binds its consensus motif 5′-TAATCC-3′ and activates gene expression, it is now clear that CRX DNA binding affinity and specificity, interaction with collaborating transcription (co-)factors, and interplay with local and broader chromatin environment collectively contribute to the precise gene expression programs that constitute the molecular foundations of photoreceptor development, functions, and long-term survival. Variants that differentially disrupt CRX activities lead to different disease manifestations, underscoring the intricate connections of CRX functions in various aspects of photoreceptor biology.

These observations also open new research avenues of CRX functions and photoreceptor biology. For example, a large subset of CRX disease mutations is associated with severe early-onset dominant CoRD (e.g., E80A), suggesting cone and rod photoreceptors may rely on different CRX regulatory principles or have different degrees of dependency on CRX activity. For example, cone genes and rod genes may rely on different types of homeodomain motifs that are differentially bound by CRX WT and CRX E80A. The limited cone population in the mouse retina has been prohibitive to high-throughput quantitative studies (Jeon et al., 1998). A cone-dominated retina (chicken and ground squirrel) will be better suited to tackle this question. Relatedly, mutations in the same CRX variant class can be associated with progressively worsened phenotypes in knock-in mouse models, pointing to a quantitative connection between CRX transcription factor functions and the sensitivity/resilience of different CRX-regulated genes. For instance, a gene controlled by multiple copies of consensus CRX motifs is more likely to buffer against fluctuations in CRX activity than a gene regulated by sub-optimal CRX motifs. A systematic comparison between different Crx animal models may offer important insights into this model. Lastly, proteins involved in DNA sequence-independent interactions with CRX are a less explored area. As discussed above, the heterogeneity of CRX transcription effector domain variants may be attributed to differential impacts on CRX interacting factors. High-throughput, quantitative systems that combine proximity labeling and mass spectrometry are now available to answer these questions (Roux et al., 2012; Rhee et al., 2013; Lam et al., 2015; Kim et al., 2016; Schopp et al., 2017; Branon et al., 2018; Ramanathan et al., 2018).

In the past two decades, the dramatic increase in genetic testing has led to the identification of many coding variations in transcription factors important for retinal development and homeostasis. To this date, it remains a significant challenge to identify specific disease-causing variants and to make an accurate prognosis. We believe the integrated approach of CRX research provides one solution to these challenges and will accelerate the development of personalized medicine for rare genetic diseases affecting the retina and other tissues.

YZ: Conceptualization, Data curation, Visualization, Writing–original draft, Writing–review and editing; SC: Conceptualization, Funding acquisition, Writing–review and editing.