The Blimp-1 transcription factor acts in non-neuronal cells to regulate terminal differentiation of the Drosophila eye

In organogenesis, terminal differentiation is a vital step during which cell-cell interactions, cellular products and changes in cell morphology all contribute to producing a fully functional organ. Cell differentiation is regulated by transcription factors that are activated in a precise spatial and temporal sequence by extracellular signals, cross-regulatory interactions and epigenetic modifications (Arnes and Sussel, 2015; Gouti et al., 2015). The downstream targets of these transcription factors include proteins that autonomously execute the specialized function of each cell, as well as proteins that establish and maintain the cell-cell interactions that direct the formation of a functionally integrated organ. In general, the transcription factors that initially specify cell fates have been better characterized than those that regulate terminal differentiation.

The developing Drosophila eye is an excellent model system to study cell fate specification and differentiation, as it is a single epithelium in which cell signaling directs equivalent progenitors to produce a well-defined and highly differentiated set of cell types. The compound eye is made of about 800 repetitive units called ommatidia. Each ommatidium is composed of eight photoreceptors, four cone cells and two primary pigment cells, surrounded by a lattice of secondary and tertiary pigment cells and mechanosensory bristles (Fig. 1F). The photoreceptors detect light through their rhabdomeres, stacked membranes packed with rhodopsin that are functionally equivalent to the outer segments of vertebrate rod and cone photoreceptors (Zelhof et al., 2003). Non-neuronal cone cells and primary pigment cells secrete the major components of the corneal lens and the underlying pseudocone; in addition, cone cells are implicated in typical glial support functions such as the control of ion balance, energy resources and sustained neurotransmission (Cagan and Ready, 1989; Charlton-Perkins et al., 2017). Cone cells envelop photoreceptors as they extend their rhabdomeres and contribute to photoreceptor morphogenesis (Charlton-Perkins et al., 2017). As well as producing screening pigments, secondary and tertiary pigment cells are involved in the synthesis and recycling of the chromophore 11-cis-retinal and the neurotransmitter histamine, and secrete some components of the corneal lens (Cagan and Ready, 1989; Chaturvedi et al., 2014; Stahl et al., 2017; Wang et al., 2012).

Specification of the Drosophila eye is orchestrated by a set of retinal determination transcription factors, including two homologues of Pax6, which controls eye development throughout the animal kingdom (Glaser et al., 1994; Kumar, 2009; Wawersik and Maas, 2000). These factors activate glass (gl), which encodes a zinc finger transcription factor that induces photoreceptor differentiation (Bernardo-Garcia et al., 2017; Zelhof et al., 2003). In addition to its effect on photoreceptors, gl is cell-autonomously required for the differentiation of the non-neuronal cell types of the eye (Liang et al., 2016; Morrison et al., 2018); therefore, it is not known how each cell type proceeds along a distinct terminal differentiation pathway. Identifying other transcription factors that act combinatorially with Gl would shed light on the cell type-specific regulation of terminal differentiation genes. One candidate for such a factor is Blimp-1, which is predicted to bind to the regulatory regions of many Gl target genes (Morrison et al., 2018). We investigated the possibility that Blimp-1 might control the differentiation of specific retinal cell types.

Blimp-1 is a C2H2 zinc-finger transcription factor that has orthologs in many bilaterians (Bikoff et al., 2009; Nakamura and Extavour, 2016). Human BLIMP1 (also known as PRDM1) regulates the differentiation of various immune cell types, including cytotoxic T cells and antibody-secreting plasma cells (Sciammas and Davis, 2004; Sledzinska et al., 2020), and also specifies primordial germ cells (Fang et al., 2018; Sasaki et al., 2015). Drosophila Blimp-1 is known to act in the fat body to regulate the timing of pupation; it is induced by the hormone ecdysone and represses the transcription of ftz transcription factor 1 (ftz-f1), preventing premature pupation (Akagi et al., 2016; Akagi and Ueda, 2011). Blimp-1 is also important for cuticle deposition and tracheal tube maturation in the embryo (Ng et al., 2006; Ozturk-Colak et al., 2016, 2018). The function of Blimp-1 in retinal development has been examined in the mouse, where it is transiently expressed in rod photoreceptors and stabilizes the immature photoreceptor fate by preventing upregulation of bipolar cell-specific genes (Brzezinski et al., 2010, 2013; Wang et al., 2014). The only study of Drosophila Blimp-1 in the eye showed that RNAi knockdown of Blimp-1 disrupted corneal lens shape, suggesting that Blimp-1 may regulate corneal lens secretion or morphogenesis (Minami et al., 2016).

In order to investigate how Blimp-1 regulates cell specification and differentiation in the Drosophila eye, we characterized the site of Blimp-1 action by eliminating its function from specific cell types. We found that Blimp-1 is required in secondary and tertiary pigment cells for the normal bi-convex morphology of the corneal lens. We also discovered a separate function for Blimp-1 in cone cells, where it acts to promote normal photoreceptor morphogenesis, in part by increasing Perlecan levels. We identified potential Blimp-1 target genes by transcriptomic analysis of retinas with loss or gain of Blimp-1 function. In addition to genes that may directly contribute to corneal lens production and cone cell-photoreceptor interactions, we observed strong upregulation of the transcription factor Slow border cells (Slbo), a C/EBP homologue (Montell et al., 1992), when Blimp-1 was knocked down, and Slbo overexpression produced a similar phenotype to loss of Blimp-1. A subset of the functions of Blimp-1 may be mediated indirectly through repression of slbo. Our results show that Blimp-1, which we find to be transiently expressed under the control of the steroid hormone ecdysone, directs the terminal differentiation of the non-neuronal cells of the eye.

The transcriptional networks that control the specification and differentiation of different cell types in the eye are not fully understood. The Gl transcription factor was initially thought to specifically direct photoreceptor differentiation, but was recently shown to also have cell-autonomous functions in cone and pigment cells (Bernardo-Garcia et al., 2017; Liang et al., 2016; Morrison et al., 2018; Moses et al., 1989), implying that other transcription factors must contribute to cell type specificity. We used i-cisTarget (Herrmann et al., 2012; Imrichová et al., 2015) to search for sequence motifs in the regulatory regions of the set of genes that we found to be induced by ectopic expression of Gl (Morrison et al., 2018). One of the most highly enriched motifs matched the binding site for Blimp-1, a PR/SET domain transcription factor related to Gl that contains five zinc fingers (Agawa et al., 2007; Kuo and Calame, 2004; Ng et al., 2006). We therefore chose to investigate the role of Blimp-1 in cell type differentiation in the eye.

In clones of cells homozygous for Blimp-1^KG09531, a P-element insertion in the first intron, the external surface of the eye became smooth, without clearly demarcated corneal lenses (Fig. 1A-C). To confirm that this phenotype was due to loss of Blimp-1, we made Blimp-1 null mutants (Blimp-1¹² and Blimp-1¹⁷) by using the CRISPR/Cas9 system to delete the majority of the coding sequence including the zinc fingers (Fig. 1G). Clones homozygous for these deletion mutants also had a smooth external eye surface (Fig. 1D), and this phenotype was rescued by expression of UAS-Blimp-1 (Ozturk-Colak et al., 2016) within the clones (Fig. 1E). A similar phenotype has been reported for Blimp-1 RNAi expression in the eye (Minami et al., 2016). To determine how loss of Blimp-1 affects the structure of the eye, we first looked at its expression pattern during the pupal stages using an available antibody (Ng et al., 2006). We stained retinas that contained Blimp-1 null mutant clones, in order to distinguish true Blimp-1 staining from background cross-reactivity. Blimp-1 was present in all cell types in the retina (schematic in Fig. 1F) from 24 h after puparium formation (APF) to 30 h APF, and was highly expressed in cone and pigment cells at 46 h APF, but lost from photoreceptors at this stage (Fig. 1H-M; Fig. S1B). No expression of Blimp-1 was seen in third instar larval eye discs or 72 h APF pupal retinas (Fig. S1A,C).

In Blimp-1 mutant clones at mid-pupal stages, we found that photoreceptors labeled with anti-Elav and cone cells labeled with anti-Cut were specified normally (Fig. 1H,J,L). Even in large Blimp-1 mutant clones generated in a Minute (RpS17) background (Morata and Ripoll, 1975), patterning of the pigment cell lattice at 42 h APF appeared to be largely normal (Fig. S2B) despite the glossy surface of the adult eye (Fig. S2A). In contrast to its role in the mouse retina, Blimp-1 thus does not appear to be required for cell fate determination in the Drosophila eye.

We next wanted to determine whether Blimp-1 acts downstream of Gl. We observed an increased number of Blimp-1-expressing cells in gl mutant clones in the pupal retina (Fig. S3A), suggesting that Gl represses Blimp-1 expression. The extra Blimp-1-expressing cells did not express the neuronal marker Elav (Fig. S3A) and may be undifferentiated cells. To determine whether misexpression of Blimp-1 contributes to the gl mutant phenotype, we generated Blimp-1 mutant clones in a gl mutant background. However, loss of Blimp-1 did not improve the patterning defects seen in the gl mutant pupal retina (Fig. S3B). If Blimp-1 contributes to the regulation of Gl target genes, its effects may be insufficient to modify the overall gl phenotype.

In sections through the adult eye, we observed that regions mutant for Blimp-1 had an altered corneal lens shape (Fig. 2A). Instead of the symmetrical external and internal curvatures seen in wild-type regions (bi-convex lens, Fig. 2B), the external surface was flat (plano-convex lens, Fig. 2C). As Blimp-1 is expressed in multiple cell types, we determined the site at which Blimp-1 acts to control corneal lens shape by using cell type-specific drivers to express Blimp-1 RNAi. Because cone cells and primary pigment cells are centrally located (Fig. 2D) and secrete the most abundant corneal lens proteins (Stahl et al., 2017), we expected that loss of Blimp-1 function from these cells would explain the corneal lens defects. However, corneal lens shape was normal when Blimp-1 RNAi was expressed in cone and primary pigment cells with sparkling (spa; also known as shaven)-GAL4 (Jiao et al., 2001) (Fig. S4). In contrast, corneal lens shape was strongly affected when we knocked down Blimp-1 in secondary and tertiary pigment cells using 54-GAL4 (Lee and Luo, 1999) (Fig. 2E,F). These higher order pigment cells are named for their ability to produce screening pigments; however, they also secrete some components of the corneal lens (Cagan and Ready, 1989; Stahl et al., 2017). In horizontal sections, corneal lenses in whole eyes or in clonal regions that expressed Blimp-1 RNAi in pigment cells had a plano-convex shape (Fig. 2F,G). Overall, the flat appearance of each individual corneal lens gave the RNAi-expressing regions a smooth and shiny appearance (Fig. 2H). One possible explanation for the raised appearance of these facets could be deposition of excess chitin above the secondary and tertiary pigment cells. However, when we examined eyes that expressed Blimp-1 RNAi in pigment cells by electron microscopy, we did not observe extra corneal lens material in this position (Fig. 2I,J).

The outer curvature of the corneal lens (air-facing) accounts for most of its refractive power (Stavenga, 2003). To assess the curvature of the outer surface, we measured the outer angle between adjacent corneal lenses (Fig. 2K). Corneal lenses within Blimp-1¹⁷ mutant clones showed a significant increase in this angle to 182±10° (mean±s.d.), compared with the average angle for control clones of 134±6° (Fig. 2L). We also modeled control and Blimp-1 mutant corneal lenses as thick optic lenses and calculated their focal lengths based on previous studies (Stavenga, 2003) (Fig. 2M,N), with the assumption that the refractive indices of the corneal lens and pseudocone remain the same as measured for wild-type flies (Stavenga, 2003). The radius of the outer surface of the Blimp-1 mutant corneal lens was treated as infinite. We calculated an average focal length for control corneal lenses of 25.7±1.8 μm, consistent with the previously reported wild-type value of 26.8±2.3 μm (Stavenga, 2003) (Fig. 2O). The focal length for mutant lenses was increased almost fourfold to 99.4±8.6 μm (Fig. 2O), which is longer than the entire depth of the retina of the mosaic flies, measured at its longest point at the anterior of the eye (90±2.7 μm, n=10). In conclusion, the plano-convex corneal lenses formed in Blimp-1 mutant regions have a dramatically increased image focal length that we predict would make it impossible for them to form a clear image on the retina.

We next examined how loss of Blimp-1 affects photoreceptor development. In clones that were homozygous for Blimp-1 or in eyes in which Blimp-1 RNAi was expressed in all cell types with the lGMR-GAL4 driver, photoreceptor rhabdomeres visualized by staining for the leucine-rich repeat protein Chaoptin (Chp) (Krantz and Zipursky, 1990) or the channel Transient receptor potential-like (TrpL) (Niemeyer et al., 1996) were short and disorganized, and photoreceptor nuclei labeled with Elav (Koushika et al., 1996) were mislocalized (Fig. 3A; Fig. S5A). This does not appear to be due to an autonomous requirement for Blimp-1 in photoreceptors, as expressing Blimp-1 RNAi specifically in photoreceptors with elav-GAL4 produced no obvious phenotype (Fig. S6A,B). In contrast, driving Blimp-1 RNAi with prospero (pros^PSG)-GAL4 (Charlton-Perkins et al., 2017) or spa-GAL4, which overlap in cone cells, resulted in similar defects in rhabdomere extension and nuclear arrangement (Fig. 3B,C; Fig. S5B). These defects were autonomous to regions in which Blimp-1 was lost from cone cells (Fig. 3D). Photoreceptor nuclei, most likely the R8 nuclei that are normally located at the proximal side of the retina, appeared to be displaced to the ends of rhabdomeres, which did not extend as far basally as in wild-type ommatidia (Fig. 3A,C,D,F,G).

Cone cells envelop photoreceptors as they extend their rhabdomeres, and could contribute to photoreceptor morphogenesis (Charlton-Perkins et al., 2017). We therefore looked at whether autonomous loss of Blimp-1 from these cells alters their development. sallimus (sls) encodes a Titin-like protein that is enriched in cone cell feet (Morrison et al., 2018), which are located at the proximal side of the retina (Cagan and Ready, 1989). We confirmed that Sls colocalized with Fasciclin III in the apical regions of cone cells, and that the staining we observed in cone cell feet was abolished by expressing sls RNAi in the retina (Fig. S7). We found that Sls staining was lost from cone cell feet in Blimp-1 mutant regions, suggesting that cone cell differentiation is disrupted by loss of Blimp-1 function (Fig. 3E). However, removal of sls alone is not sufficient to disrupt photoreceptor rhabdomeres (Fig. S7B). terribly reduced optic lobes (trol), which encodes the extracellular matrix (ECM) protein Perlecan (Pcan) (Voigt et al., 2002), is another gene that we found to be regulated by Blimp-1 in cone cells (Fig. S8A). Pcan was also localized to cone cell feet (Fig. S8B), and knocking it down in the eye resulted in shortened rhabdomeres (Fig. S8C,D), suggesting that it may contribute to the defective photoreceptor morphogenesis that results from loss of Blimp-1 in cone cells.

Blimp-1 expression is known to be regulated by hormonal signaling in both Drosophila and mammalian development (Agawa et al., 2007; Ahmed et al., 2016; Akagi et al., 2016; Akagi and Ueda, 2011; Ozturk-Colak et al., 2018). To test whether the hormone ecdysone regulates Blimp-1 levels in the developing Drosophila eye, we reduced ecdysone signaling by knocking down the expression of the ecdysone receptor (EcR). Expressing EcR RNAi in all cells in the eye caused phenotypes resembling Blimp-1 loss-of-function, including a flat external corneal lens surface, shortened photoreceptor rhabdomeres and disordered photoreceptor nuclei (Fig. 4A,B). We next examined Blimp-1 expression in pupal retinas in which clones of cells expressed EcR RNAi. Although early expression of Blimp-1 at 30 h APF was only slightly affected in EcR RNAi clones (Fig. 4C-E), the stronger expression seen in cone and pigment cells at 41 h APF was dramatically reduced (Fig. 4F-H). These results suggest that upregulation of Blimp-1 expression in mid-pupal stages by ecdysone signaling is required for normal adult photoreceptor and corneal lens morphology.

Blimp-1 expression is transient in the eye and was lost from photoreceptors earlier than from the other cell types (Fig. 1J-M). Blimp-1 mRNA and protein are unstable, enabling it to control the timing of developmental processes (Agawa et al., 2007; Akagi and Ueda, 2011). We tested the effect of Blimp-1 overexpression in the eye by generating clones in which UAS-Blimp-1 was expressed in all cell types with lGMR-GAL4 (Fig. S9). In the adult eye, these regions lacked their normal structure and sometimes appeared to be black (Fig. S9B). Sectioning and immunostaining of these eyes revealed defects in both neuronal and non-neuronal cell types. Photoreceptor nuclei misexpressing Blimp-1 were displaced medially (Fig. S9A), the rhabdomeres in each ommatidium had increased levels of Chp and sometimes appeared to be fused together (Fig. S9E,G), and staining for the calcium channel TrpL (Niemeyer et al., 1996) was weaker in some Blimp-1-expressing ommatidia (Fig. S9D). Blimp-1 overexpression also affected pigment cell development; Photoreceptor dehydrogenase (Pdh), an enzyme involved in chromophore regeneration that is specific to pigment cells (Wang et al., 2010), showed reduced expression in regions of the adult eye that overexpressed Blimp-1 (Fig. S9C), and many extra pigment cells remained in the lattice at 42 h APF (Fig. S9F). The arrangement of photoreceptor nuclei was also disrupted when we overexpressed Blimp-1 specifically in photoreceptors, suggesting that extending the period of Blimp-1 expression in these cells has deleterious effects (Fig. S6C). These results show that timely removal of Blimp-1 is crucial for normal eye development.

The corneal lens and photoreceptor defects caused by loss and gain of Blimp-1 function suggested that terminal differentiation of multiple cell types in the eye was disrupted. To find the target genes of Blimp-1 that could mediate these phenotypes, we carried out a transcriptomic analysis of 48 h APF pupal retinas in which Blimp-1 was knocked down or overexpressed in all cell types with lGMR-GAL4. In retinas expressing Blimp-1 RNAi, we found 189 genes that were significantly downregulated and 184 genes that were upregulated (Fig. 5A). In retinas overexpressing Blimp-1, 514 genes were significantly upregulated and 196 genes downregulated (Fig. 5A). Although some genes were oppositely affected by gain or loss of Blimp-1, as expected for target genes, many of the genes that were induced by Blimp-1 overexpression were not reduced by Blimp-1 RNAi (Fig. 5B). Some of these genes may not be physiological targets of Blimp-1, and others may not normally be expressed at this developmental stage. The presence of predicted Blimp-1 binding motifs in the regulatory regions of genes that were upregulated as well as those that were downregulated by loss of Blimp-1 suggests that Blimp-1 may act as both a transcriptional repressor and an activator (Herrmann et al., 2012; Imrichová et al., 2015) (Table S1; Fig. S10A,B).

Among the genes with known or predicted functions that are regulated by Blimp-1, some of the larger classes encode enzymes, cell surface proteins, cuticle proteins and other secreted proteins (Fig. 5C,D; Table S1). Drosophila corneal lenses have a cuticular nature, and four major corneal lens proteins have been identified by mass spectrometry. Crystallin and Retinin are secreted from cone and primary pigment cells, whereas Cpr66D and Cpr72Ec are produced by higher order pigment cells (Stahl et al., 2017). The Blimp-1 corneal lens phenotype is likely to arise from misregulation of proteins secreted by higher order pigment cells, as loss of Blimp-1 function from cone and primary pigment cells did not affect the corneal lens (Fig. S4). However, Cpr66D and Cpr72Ec mRNA levels were not significantly altered by Blimp-1 knockdown. As 79 unique cuticle proteins were found in the adult Anopheles gambiae lens using mass spectrometry (Zhou et al., 2016), it is likely that the Drosophila corneal lens contains more protein components than have been identified to date. Blimp-1 loss-of-function may affect uncharacterized corneal lens components that are secreted by the secondary and tertiary pigment cells. The morphology of the corneal lens could also be affected by changes in the proteins that attach it to the pigment cells; for example, Blimp-1 affects the expression of a number of zona pellucida (ZP) domain proteins that have been shown to control the shape of cuticular structures in the late embryo (Fernandes et al., 2010). The photoreceptor phenotypes resulting from Blimp-1 loss-of-function in the cone cells could potentially reflect changes in the expression of secreted proteins such as Pcan, cytoskeletal proteins such as Sls and cell surface proteins that might promote interactions between photoreceptors and cone cells (Fig. 5C,D). Both sls and trol were significantly downregulated in retinas expressing Blimp-1 RNAi (Table S1).

Blimp-1 is thought to act downstream of ecdysone to implement temporal regulation of organogenesis as well as the development of the whole organism (Agawa et al., 2007; Akagi et al., 2016; Akagi and Ueda, 2011; Ozturk-Colak et al., 2018). We hypothesized that Blimp-1 could promote normal eye morphogenesis by ensuring a normal rate of development. We used high-throughput data from modENCODE (Graveley et al., 2011) to determine whether loss of Blimp-1 function systematically shifted gene expression earlier or later in development. Of the genes with reduced expression in 48 h APF Blimp-1 RNAi retinas, we found that 59 genes (31%) had their peak expression at 48 h APF, whereas smaller numbers were increasing or decreasing their expression, which is consistent with the notion that Blimp-1 regulates genes with precise developmental timing (Fig. 5E). Among the genes that show increased expression in Blimp-1 RNAi retinas, we found that almost three times as many were increasing their expression at 48 h APF (40 genes, 21.7%) as were decreasing it at this stage (14 genes, 7.6%) (Fig. 5E). This suggests that loss of Blimp-1 leads to the premature expression of genes that would normally reach their peak later in development, and that accelerated development may thus contribute to its loss-of-function phenotypes.

The gene that was most highly upregulated in retinas expressing Blimp-1 RNAi was slbo (Fig. 6A), which encodes a transcriptional activator homologous to human CEBPD that contributes to border cell migration in the Drosophila ovary (Borghese et al., 2006; Montell et al., 1992). Consistent with the RNA-seq data, we observed slbo-lacZ upregulation in Blimp-1¹² mutant clones, where it was confined to the non-neuronal cells of the retina (Fig. 6B). Overexpressing slbo in otherwise wild-type eyes with lGMR-GAL4 recapitulated many of the phenotypes of loss-of-Blimp-1. The eye assumed a flat and shiny appearance (Fig. 6C-E) with short rhabdomeres and abnormally arranged photoreceptor nuclei resembling those caused by Blimp-1 knockdown (Fig. 6H,I). Overexpression of slbo specifically in higher order pigment cells with 54-GAL4 was sufficient to make the corneal lenses become plano-convex (Fig. 6F,G). The similarity between the phenotypes caused by Slbo overexpression and Blimp-1 loss-of-function suggests that Blimp-1 could regulate the differentiation of cone and pigment cells in part by repressing slbo. Consistent with this model, i-cisTarget (Herrmann et al., 2012; Imrichová et al., 2015) predicted Slbo binding sites in a large subset of the genes that showed altered expression after Blimp-1 knockdown (Fig. S10A,B). However, expressing Blimp-1 RNAi in slbo mutant clones did not visibly rescue its corneal lens and photoreceptor phenotypes (Fig. S10C-F), indicating that loss of Blimp-1 has significant effects that are independent of slbo upregulation.

The Blimp-1 transcription factor has been shown to play important regulatory roles in the development of many cell types and organs in both flies and mammals (Agawa et al., 2007; Akagi et al., 2016; Brzezinski et al., 2010; Fu et al., 2017; Lin et al., 2014; Ozturk-Colak et al., 2016, 2018; Sciammas and Davis, 2004; Sledzinska et al., 2020). The only reported function for Blimp-1 in the mouse retina is to stabilize the photoreceptor cell fate (Brzezinski et al., 2010, 2013); however, we found that Drosophila Blimp-1 acts at a later stage of retinal development to promote the terminal differentiation of several cell types. It has no visible function in photoreceptors, but acts in the cone cells to enable normal photoreceptor rhabdomere extension and in the higher order pigment cells to promote the formation of a biconvex corneal lens (Fig. 6J). Blimp-1 controls the terminal differentiation of these cells by regulating a battery of genes that encode likely structural components of the eye as well as Slbo and other transcription factors. Upregulation of Blimp-1 expression in the retina requires ecdysone signaling, illustrating a role for steroid hormones in regulating sensory organ morphogenesis.

Although humans and Drosophila have shared common ancestors only in the very distant past and their lenses have distinct molecular compositions, both species evolved bi-convex lens shapes that enable them to focus light effectively onto the retina. The fly corneal lens cannot change its curvature through muscle contractions as the human lens does; to compensate for its fixed parameters, the fly retina is able to detect light throughout its depth (Zhu, 2013). However, the thickness of the retina is insufficient to compensate for the massive reduction in the refractive power of the corneal lens resulting from the plano-convex shape caused by Blimp-1 loss-of-function, especially as loss of Blimp-1 also reduces retinal thickness. Loss of curvature on the external surface of Blimp-1 mutant corneal lenses is more detrimental to their refractive power than flattening of the internal surface facing the pseudocone would be, as the refractive indices of the corneal lens and pseudocone are similar to each other, but quite different from that of air (Stavenga, 2003).

Although major corneal lens proteins such as Crystallin and Retinin are produced by the cone and primary pigment cells (Stahl et al., 2017), Blimp-1 acts in the more peripheral secondary and tertiary pigment cells to generate the external curvature of the corneal lens. Blimp-1 has been shown to regulate chitin deposition during embryonic cuticle formation and tracheal maturation (Ozturk-Colak et al., 2018), and chitin is a major component of the corneal lens that may determine its shape. Blimp-1 also controls the expression of numerous cuticle proteins that could be secreted from higher-order pigment cells to give the periphery of the corneal lens a composition distinct from the center. Besides regulating the expression of direct structural components of the corneal lens, Blimp-1 could contribute to the bi-convex shape by regulating the forces that pigment cells exert on the periphery of the corneal lens through its effects on the expression of cell surface and cytoskeletal proteins.

In addition to its effect on corneal lens shape, Blimp-1 loss-of-function causes a defect in photoreceptor morphology. Instead of extending throughout the depth of the retina, rhabdomeres are absent from the proximal retina, where the photoreceptors form stalk-like structures lacking rhabdomere markers. Photoreceptor nuclei are also displaced to the proximal ends of the rhabdomeres. During development, photoreceptors interact with cone cells as they undergo a 90° rotation of their apical membrane surface (Charlton-Perkins and Cook, 2010). Microvilli projecting from this apical membrane subsequently form rhabdomeres that extend proximally down the retina, growing to ∼90 μm in length (Charlton-Perkins and Cook, 2010; Longley and Ready, 1995). As cone cells are known to structurally and physiologically support photoreceptors during development (Charlton-Perkins et al., 2017), loss of Blimp-1 in these cells may affect their ability to contribute to photoreceptor development. Genes encoding cell surface, cytoskeletal and secreted proteins that we found to be regulated by Blimp-1 (Fig. 5C,D), including trol, could promote rhabdomere extension or attachment to an ECM produced by the cone cell feet.

We found that loss of Blimp-1 leads to a dramatic upregulation of the transcription factor Slbo in non-neuronal retinal cells. Overexpressing slbo in the eye results in a plano-convex lens and shortened rhabdomeres, similar to what is observed in Blimp-1 mutants. slbo overexpression could thus contribute to the Blimp-1 loss-of-function phenotypes. However, removing slbo is not sufficient to restore normal retinal development in the absence of Blimp-1, indicating that at least one crucial target gene requires direct input from Blimp-1 in addition to the absence of Slbo. Two distinct mechanisms, one regulated by Slbo and one by loss of Blimp-1 independently of Slbo, can thus lead to this constellation of phenotypes.

Most previous studies of slbo have focused on its requirement for normal border cell migration in the ovary (Montell et al., 1992) and have shown that its overexpression is detrimental to most other tissues (Rorth et al., 2000). As the eye is quite distinct from the ovary in its structure and morphogenesis, it is surprising that only a single transcription factor prevents slbo from being expressed there. It is more common to find a factor that is poised for expression when a single repressor is removed in tissues that share a common origin or developmental pathway. For example, knot is required for wing patterning but is repressed by Ultrabithorax in the haltere disc, which is a structure homologous to the wing disc (Hersh and Carroll, 2005). It is not known what activators drive slbo expression in the absence of repression by Blimp-1. In the ovary, slbo is activated by JAK/STAT signaling, which acts early in eye development to promote growth of the larval eye disc, but is not known to function in the pupal retina (Silver and Montell, 2001; Zeidler et al., 1999). The mouse slbo homologue Cebpa is activated by Blimp-1 in uterine tissues (Goolam et al., 2020), suggesting possible conservation of the regulation of C/EBP genes by Blimp-1, but not of the direction of this effect.

Given the presence of Blimp-1 at the periphery of the developing mouse cornea (Parfitt et al., 2015), our results raise the intriguing possibility that its role in regulating the curvature of refractive surfaces is conserved. Further investigation of this hypothesis may provide insight into the genesis of the refractive disorders that afflict a growing number of people.

The Blimp-1 sgRNA sequences GATCTTCGGCGCATGCAAGC (Blimp-1 sgRNA 1) and ATTGAGGCACCTCTCGAGTG (Blimp-1 sgRNA 2), identified on www.flyrnai.org/crispr2/, were cloned into pCFD5 (Port and Bullock, 2016) by PCR and Gibson assembly, and the construct was integrated into the attP40 site at 25C6. UAS-Blimp-1^FLAG-HA (Drosophila Genomics Resource Center clone UFO06327) was integrated into the VK22 site at 57F5. Injections and screening of transgenic flies were carried out by Genetivision. The sgRNA-expressing flies were crossed to nos-Cas9 [Bloomington Drosophila Stock Center (BDSC), #54591] and the progeny were screened for failure to complement the lethality of Blimp-1^KG09531 (BDSC, #15195). PCR analysis showed that two lines that failed to complement (Blimp-1¹² and Blimp-1¹⁷) carried a 2 kb deletion between the two sgRNA sites.

Drosophila melanogaster stocks used to generate Blimp-1 mutant clones were: (1) UAS-CD8-GFP, ey3.5-FLP; lGMR-GAL4; FRT80, tub-GAL80/TM6B; (2) ey-FLP; FRT80, Ubi-His2AvRFP /TM6B; (3) ey-FLP; FRT80, RpS17, Ubi-GFP; (4) FRT80, Blimp-1¹⁷/TM6B; (5) ey-FLP; FRT80, Blimp1^KG09531/TM6B; (6) slbo-lacZ; FRT80, Blimp-1¹²/SM6-TM6B. Stocks used to generate Blimp-1 knockdown clones were: (1) UAS-CD8-GFP, ey3.5-FLP; 54-GAL4, UAS-lacZ; FRT82, tub-GAL80/SM6-TM8B; (2) ey-FLP, spa-GAL4; UAS-myr-Tomato; FRT82, tub-GAL80; (3) UAS-CD8-GFP, ey3.5-FLP; lGMR-GAL4; FRT82, tub-GAL80/SM6-TM6B; (4) UAS-CD8-GFP, ey3.5-FLP; elav-GAL4; FRT82, tub-GAL80/SM6-TM6B; (5) UAS-Blimp-1 RNAi BL; FRT82. Stocks used to generate Blimp-1 overexpression clones were: (1) UAS-CD8-GFP, ey3.5-FLP; lGMR-GAL4; FRT82, tub-GAL80; (2) UAS-Blimp-1^72.1; FRT82/SM6-TM6B; (3) UAS-Blimp-1^FLAG-HA; FRT82/SM6-TM6B. Other stocks used were 54-GAL4, UAS-lacZ/SM6-TM6B (Lee and Luo, 1999); spa-GAL4 (Jiao et al., 2001); lGMR-GAL4 (Wernet et al., 2003); ey3.5-FLP, Act>CD2>GAL4 (BDSC, #35542 and #4780); pros^PCG-GAL4 (Charlton-Perkins et al., 2017); UAS-Blimp-1 RNAi BL (HMC04792, BDSC, #57479); UAS-Blimp-1 RNAi KK [P{KK107466}VIE-260B, Vienna Drosophila Resource Center (VDRC), #108374]; UAS-slbo (Rorth et al., 1998); UAS-slboHA (FlyORF F000038); slbo^8ex22 (Rorth and Montell, 1992); gl^60j (Moses et al., 1989); Blimp-1-GFP.FPTB (BDSC, #67656); UAS-Blimp-1^72.1 (Ozturk-Colak et al., 2016); UAS-sls RNAi (TRiP.JF01099); and UAS-trol RNAi (P{KK105502}VIE-260B, VDRC, #110494). Stocks used for rescue were: (1) UAS-CD8-GFP, ey3.5-FLP; lGMR-GAL4; FRT80, tub-GAL80/TM6B; (2) UAS-Blimp-1^72.1; Blimp-1¹²/SM6-TM6B. Genotypes used for RNA-seq were: (1) lGMR-GAL4/UAS-Blimp-1^72.1; (2) lGMR-GAL4/UAS-Blimp-1 RNAi BL; (3) lGMR-GAL4/+. Male and female flies were used interchangeably, as no sex-specific differences were observed.

Pupal retinas were fixed and stained as previously described (Tea et al., 2014). Retina-brain complexes were dissected out of pupae aged from the white prepupal stage and collected in ice-cold PBS in a glass plate. These were fixed for 30 min in 4% formaldehyde in PBS, washed in ice-cold PBS/0.2% Triton X-100 (PBT) for 15 min and incubated overnight at 4°C in primary antibodies in 10% donkey serum in PBT, followed by three 20 min washes in PBT, a 2 h incubation at 4°C in secondary antibodies in PBT and another three 20 min washes in PBT. Retinas were separated from the brain on glass slides and mounted in 80% glycerol in PBS. For cryosectioning, adult heads were fixed for 3-4 h at 4°C in 4% formaldehyde in 0.2 M sodium phosphate buffer (pH 7.2) (PB) after removing the proboscis. Fly heads were glued onto glass rods using nail polish and transferred through a sucrose gradient in PBS consisting of 20 min each in 5%, 10%, 25% and 30% sucrose. Heads were frozen in OCT on dry ice. Then, 12 µm cryosections were cut at −21°C and slides were fixed post-sectioning for 30 min in 0.5% formaldehyde in PB at room temperature. After three 10 min washes in PBT, slides were blocked for 1 h and incubated in primary antibodies overnight at 4°C in 1% bovine serum albumin (BSA) in PBT. After three 20 min washes in PBT, slides were incubated in secondary antibodies in 1% BSA in PBT for 2 h at 4°C and mounted in Fluoromount (Southern Biotech). A 1:5 dilution of Calcofluor White solution (25% in water; Sigma Aldrich, 910090) was included with the secondary antibodies where indicated. The primary antibodies used were: mouse anti-Chp [1:50; Developmental Studies Hybridoma Bank (DSHB), 24B10], chicken anti-GFP (1:400; Invitrogen, A10262), rat anti-Elav (1:100; DSHB, Rat-Elav-7E8A10 anti-elav), mouse anti-Cut (1:10; DSHB, 2B10), rat anti-Ecad (1:10, DSHB, DCAD2), mouse anti-FasIII (1:10, DSHB, 7G10 anti-Fasciclin III), mouse anti-Pros [1:10, DSHB, Prospero (MR1A)], rabbit anti-β-galactosidase (1:5000; MP Biomedicals, 55976), rat anti-Kettin/Sls (1:200; Abcam, ab50585), rabbit anti-DsRed (1:500, Takara Bio, 632496), rabbit anti-TrpL (1:200; Niemeyer et al., 1996), rabbit anti-Pdh (1:200; Wang et al., 2010) and guinea pig anti-Blimp-1 (1:400; Ng et al., 2006). All antibodies were validated either using mutant or knockdown conditions as shown, or by verifying that the staining pattern matched previously published descriptions. The secondary antibodies used were from either Jackson ImmunoResearch (Cy3 or Cy5 conjugates used at 1:200) or Invitrogen (Alexa488 conjugates used at 1:1000). Images were acquired on a Leica SP8 confocal microscope and processed using ImageJ and Adobe Photoshop. Outer lens angles were measured with ImageJ according to the diagram in Fig. 2K, and rhabdomere length and nuclear position were normalized to the depth of the retina measured using ImageJ. Significance was calculated using Welch's two-tailed t-tests. Sample sizes for quantifications were not predefined and no samples were excluded.

The plasmid pYZ205 (New England Biolabs), which encodes a 6XHis-SNAP-chitin binding domain (CBD) fusion construct, was transformed into Escherichia coli BL21DE3 and protein expression was induced with 1 mM IPTG at OD₆₀₀ of 0.6 at 37°C for 4 h in Luria Broth with 50 mg/ml ampicillin. The protein was isolated using nickel-NTA agarose resin and eluted in 1 ml fractions with elution buffer [50 mM NaH₂PO₄ (pH 7.2), 300 mM NaCl, 250 mM imidazole, 1 mM DTT]. The eluted proteins were quantified on a Nanodrop Spectrophotometer and their purity was checked by SDS-PAGE. The protein fractions with the highest purity and quantity were selected and dialyzed in phosphate buffer [50 mM NaH₂PO₄ (pH 7.2), 100 mM NaCl, 1 mM DTT] overnight at 4°C. Dialyzed SNAP-CBD protein was adjusted to a final concentration of 5 µM in phosphate buffer and incubated with 10 µM SNAP substrate fluorophore (SNAP-surface Alexa Fluor 488, New England Biolabs, #S9129 or SNAP-surface Alexa Fluor 546, New England Biolabs, #S9132) at 37°C for 1 h. The fluorescently labeled probes were dialyzed in phosphate buffer overnight at 4°C to remove unreacted substrates and stored in 50% glycerol at −20°C. SNAP-CBD-probes were used at a final dilution of 1:200 in immunohistochemistry experiments.

Adult heads were cut in half and fixed in freshly made fixative containing 2.5% glutaraldehyde, 2% paraformaldehyde and 0.05% Triton X-100 in 0.1 M sodium cacodylate buffer (pH 7.2) on a rotator for at least 4 h until all heads had sunk to the bottom of the tube, and then in the same fixative without Triton X-100 overnight at 4°C on a rotator. After washing in PBS, the heads were post-fixed in 1% OsO₄ in PBS for 1.5 h, dehydrated in an ethanol series (30%, 50%, 70%, 85%, 95%, 100%), rinsed twice with propylene oxide and embedded in EMbed812 epoxy resin (Electron Microscopy Sciences). Then 500 nm semi-thin horizontal sections were mounted on a glass slide and baked on a hot plate overnight at 60°C. The sections were stained with 0.1% Toluidine Blue, dried on a hot plate and cover-slipped with Permount^TM mounting medium (Electron Microscopy Sciences). Images were taken on a Zeiss Axioplan microscope and processed in Adobe Photoshop. For transmission electron microscopy, 70 nm ultra-thin sections were cut and mounted on formvar-coated slot grids and stained with uranyl acetate and lead citrate (Hess et al., 2006). Electron microscopy imaging was performed on a Talos120C transmission electron microscope (Thermo Fisher Scientific) and recorded using a Gatan (4k×4k) OneView Camera with Digital Micrograph software (Gatan).

For scanning electron microscopy, flies were adhered to tape on microscope slides with nail polish and the eye images were initially captured as light micrographs (Nikon E600 equipped with a Nikon Coolpix 4500 camera) and subsequently as natural scanning electron micrographs (Hitachi 2460N Scanning Electron Microscope, 180×).

Pupal retinas (with attached lamina and medulla) were dissected from triplicate samples of 30 animals of each genotype and RNA was extracted using Trizol (Invitrogen). RNA quality and quantity was assessed using the Bio-analyzer 2100 (Agilent). Library preparation and sequencing was carried out by the NYU Genome Technology Center. Libraries were constructed using Illumina TruSeq Stranded mRNA (#20020595), with 250 ng of total RNA as input, and 10 cycles of PCR amplification. Samples were sequenced using single lane Hiseq 4000, paired end read 50. RNA-seq data were analyzed using the sns/rna-star pipeline (https://igordot.github.io/sns/routes/rna-star.html) and the differential expression analyses were carried out using the sns/rna-star-groups-dge pipeline (https://igordot.github.io/sns/routes/rna-star-groups-dge.html). Sequencing reads were mapped to the reference genome (dm6) using the STAR aligner (v2.6.1d) (Dobin et al., 2013). Alignments were guided by a Gene Transfer Format (GTF) file. The mean read insert sizes and their standard deviations were calculated using Picard tools (v.2.18.20) (http://broadinstitute.github.io/picard). The read count tables were generated using subread (v1.6.3) (Liao et al., 2014), normalized based on their library size factors using DEseq2 (Love et al., 2014), and differential expression analysis was performed. To compare the level of similarity among the samples and their replicates, we used principal component analysis. Genes were considered significantly changed and were included in the heat map if the log₂ fold change for either overexpression or knockdown was >1, P<0.05, and standard deviation/mean of the three replicates was <0.5. The heat map was constructed in Matlab R2017a (https://www.mathworks.com) and the Euler diagram was made using Biovinci software (Bioturing).

We thank Sofia Araujo, Leonard Dobens, Craig Montell, the Bloomington Drosophila Stock Center, the Vienna Drosophila RNAi Center, the Drosophila Genomics Resource Center and the Developmental Studies Hybridoma Bank for fly stocks and reagents. Essential information for this study was provided by Flybase. We thank NYULH DART Microscopy Laboratory members Alice Liang, Chris Petzold, Kristen Dancel Manning and Joseph Sall for consultation and assistance with plastic sections and electron microscopy, and this core is partially funded by NYU Cancer Center Support Grant NIH/NCI P30CA016087. We are grateful to Yutong Zhang, Shaowen Jiang and Alireza Khodadadi-Jamayran for assistance with the RNA-seq experiment and analysis. We thank DanQing He and Ariel Hairston for technical assistance. The manuscript was improved by the critical comments of Jane Hubbard and Maria Bustillo. We thank Utpal Banerjee, UCLA (funded by a Howard Hughes Medical Institute Professors Grant) for support during the collection of some of the data.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200217.