Glass promotes the differentiation of neuronal and non-neuronal cell types in the Drosophila eye

Transcriptional regulators can specify different cell types from a pool of equivalent progenitors by activating distinct developmental programs. The Glass transcription factor is expressed in all progenitors in the developing Drosophila eye, and is maintained in both neuronal and non-neuronal cell types. Glass is required for neuronal progenitors to differentiate as photoreceptors, but its role in non-neuronal cone and pigment cells is unknown. To determine whether Glass activity is limited to neuronal lineages, we compared the effects of misexpressing it in neuroblasts of the larval brain and in epithelial cells of the wing disc. Glass activated overlapping but distinct sets of genes in these neuronal and non-neuronal contexts, including markers of photoreceptors, cone cells and pigment cells. Coexpression of other transcription factors such as Pax2, Eyes absent, Lozenge and Escargot enabled Glass to induce additional genes characteristic of the non-neuronal cell types. Cell type-specific glass mutations generated in cone or pigment cells using somatic CRISPR revealed autonomous developmental defects, and expressing Glass specifically in these cells partially rescued glass mutant phenotypes. These results indicate that Glass is a determinant of organ identity that acts in both neuronal and non-neuronal cells to promote their differentiation into functional components of the eye.


Introduction
Cell fate specification is achieved by integrating extrinsic signals with intrinsic transcriptional and epigenetic networks [1][2][3]. Certain transcription factors have been described as "master regulators" for their ability to specify entire organs [4][5][6], while downstream "terminal selectors" activate genes necessary to confer a specific cellular identity [7][8][9]. At intermediate levels of the hierarchy, other transcription factors endow progenitor cells with the ability to give rise to broad categories of cell types distinguished by their function or position [4,10,11]. In general, these mechanisms are thought to gradually narrow the fate choices available to each progenitor cell. It is not known whether distinct cell types within the same organ may share a common transcriptional signature.
The Drosophila eye consists of photoreceptor neurons and non-neuronal support cells that develop from a single field of uncommitted progenitor cells in the larval eye imaginal disc [12,13], which is specified by a network of transcription factors encoded by the retinal determination genes [14]. Differentiation of the eye disc proceeds in a posterior to anterior wave driven by Hedgehog (Hh) signaling and led by an indentation known as the morphogenetic furrow [12,15,16]. This process gives rise to regularly spaced clusters of cells in which progenitor cells are sequentially specified as photoreceptors R1-R8 and the four lens-secreting cone cells [17,18]. The remaining cells rearrange their positions and reduce their numbers during pupal development to produce a lattice of optically insulating pigment cells and mechanosensory bristles [19,20]. Signaling between these differentiating cells, most notably through the Epidermal Growth Factor Receptor (EGFR) and Notch pathways, plays an important role in assigning uncommitted cells to photoreceptor, cone or pigment cell fates [21]. Each cell's transcription factor content is also critical in determining how it responds to these signals [22][23][24].
The zinc finger transcription factor Glass (Gl) has been assigned a central role in specifying photoreceptor identity downstream of the retinal determination gene network [25][26][27]. In gl mutants, presumptive photoreceptors acquire neuronal characteristics, but fail to differentiate and do not express Rhodopsins or the phototransduction machinery [25,27,28]. These studies support the model that Gl drives the transition from a neuronal to a photoreceptor cell fate. Nevertheless, Gl expression is not confined to photoreceptors; it is present in all cells in the eye disc posterior to the morphogenetic furrow, and is dynamically expressed in all cell types during pupal development [26,29]. While cone cells and pigment cells are still present in gl mutants, their numbers and morphology are abnormal, giving rise to the "glassy" phenotype that was the basis for the original identification of gl, and pigmentation is reduced [25,30]. These defects have been considered to be a secondary consequence of abnormal photoreceptor development, as early studies suggested that Gl function was negatively regulated in non-neuronal cells [29]; however, its possible autonomous role in the development of these cell types has not been addressed.
Misexpression of Gl in the larval central nervous system has been shown to induce the expression of a subset of photoreceptor-specific genes [27]. As a first step towards investigating whether Gl function is restricted to neuronal lineages, we compared the effects of misexpressing it in different developmental contexts. We found that Gl was able to induce expression of the photoreceptor-specific gene chaoptin (chp) both in neuroblasts in the brain and in nonneuronal epithelial cells of other imaginal discs, independently of the retinal determination gene network. Transcriptomic analysis revealed distinct but overlapping sets of genes induced in these two cellular contexts, including genes characteristic of non-neuronal cone and pigment cells. Gain-of-function, rescue and loss-of-function experiments all supported autonomous roles for Gl in the normal differentiation of these non-neuronal cells of the eye as well as of photoreceptors. Furthermore, we identified several elements of a transcription factor network that cooperates with Gl to induce genes characteristic of cone and pigment cells. These results indicate that Gl is not a photoreceptor-specific factor, but a determinant of organ identity that is reiteratively used to promote the terminal differentiation of multiple cell types in the eye.

Gl is sufficient to induce ectopic Chp expression in both neuronal and nonneuronal cells
Gl is expressed in differentiating cells in the eye imaginal disc ( Fig 1A) and is required for the expression of photoreceptor-specific genes such as chaoptin (chp) (Fig 1B and 1D) in cells that have been specified to become neurons [25]. However, Gl expression has also been noted in the non-neuronal cone cells and pigment cells of the eye [26,29]. To test whether Gl activity is limited to neuronal cells, we generated two transgenic lines, UAS-gl-RA and UAS-gl-RB, in which the UAS promoter drives each of the gl isoforms annotated in Flybase (Fig 1C). gl-RA encodes a protein of 604 amino acids, containing five zinc fingers of which only the final three are essential for DNA binding [26,31], and could also produce a 679 amino acid protein, Gl-RC, by readthrough of the stop codon. gl-RB is predicted to retain an intron that would introduce a frameshift leading to protein termination in the middle of the last zinc finger. However, our RNA-seq data indicated that the intron in gl-RB was spliced out in vivo (Fig 1C), suggesting that its apparent retention was an artifact of cDNA preparation. Expressing either Gl isoform in gl mutant clones restored Chp expression, confirming that both transgenes are functional (Fig 1E and 1F). We used the two transgenes interchangeably in subsequent experiments.
As Gl was thought to drive the transition from a neuronal to a photoreceptor cell fate, we first tested the effects of misexpressing it in neural progenitors in the developing embryo and the larval brain. Endogenous Chp expression in the embryo is confined to larval photoreceptors in the Bolwig organ (S1A Fig), and the protein is present in photoreceptor axons and their terminals in the optic lobes of the third instar larval brain (Fig 2A). We found that misexpressing Gl in neuroblasts using asense (ase)-GAL4, deadpan (dpn)-GAL4 or inscuteable (insc)-GAL4 drivers, together with a temperature-sensitive form of GAL80 to bypass early lethality, led to induction of Chp in some cells in the central nervous system (CNS) of the developing embryo (S1B Fig) and in the central region of the larval brain and ventral nerve cord (Fig 2B,  2C and 2E).
The expression of Chp and Gl was not confined to cells that expressed UAS-CD8-GFP from the neuroblast driver (S2B- S2D Fig). As Gl can autoregulate its expression [26,32], this observation suggested that Gl could be maintained in differentiating cells derived from the neuroblast lineage, and continue to induce ectopic Chp expression in these cells. To determine wild type eye discs. Gl is expressed earlier than Chp. Arrowheads mark the morphogenetic furrow. (C) shows a diagram of the transcripts annotated on Flybase. gl-RB is predicted to retain an intron. Reads from RNA-Seq analysis of wild type eye discs and wing discs expressing UAS-gl-RB in clones of cells are aligned to the diagram and show no reads corresponding to this intron, indicating that it is spliced out in vivo (D-F) gl 60j homozygous mutant clones are marked by expression of GFP (green) and also express UAS-gl-RA (E) or UAS-gl-RB (F). Chp expression is absent from whether target gene induction in differentiating neuroblast progeny required sustained expression of Gl, we expressed Gl in neuroblasts but blocked its expression in differentiating neurons with GAL80 driven by the pan-neuronal elav promoter. Under these conditions, we no longer observed ectopic Chp expression in the ventral nerve cord (Fig 2F), indicating that Gl expression must be maintained to induce Chp in this region. Driving Gl expression specifically in differentiated neurons with elav-GAL4 also led to ectopic expression of Chp in the embryonic and larval CNS (Fig 2G, S1C Fig). In contrast, misexpressing Gl in glial cells with reversed gl mutant clones (D) but restored by either transgene (E, F). Arrows indicate representative clones. Scale bars: 30μm in D-F; 50μm in A,B.
(D) shows the condition used for RNA-seq analysis; endogenous Chp is absent from the photoreceptor axons. Scale bars: 100μm.
https://doi.org/10.1371/journal.pgen.1007173.g002 polarity (repo)-GAL4 or in sensory organ precursors with atonal (ato)-GAL4 did not result in ectopic Chp expression (Fig 2H and 2I, S1D and S1E Fig). These results indicate that Gl is not sufficient for chp transcription, but acts in combination with other sequence-specific transcription factors or chromatin regulators that are present in a subset of neuronal cell types.
A previous study suggested that Gl function was negatively regulated in non-neuronal cells, as ubiquitous Gl expression activated a reporter driven by Gl binding sites derived from the proximal enhancer of Rhodopsin 1 (Rh1) specifically in neurons [29]. However, during normal eye development Gl is expressed earlier than neuronal markers such as Elav [26] (S4B Fig), and its expression is maintained in non-neuronal cone and pigment cells through early pupal stages [26,29]. We therefore tested the effect of misexpressing Gl in developing leg and wing imaginal disc cells, which are epithelial progenitors similar to the first cells that express Gl in the eye disc. When we used Ultrabithorax (Ubx)-FLP and the MARCM system to express Gl in clones of epithelial cells in the wing imaginal disc (Fig 3A), we detected Chp expression within the clones (Fig 3B). Although not all of the clones misexpressing Gl activated Chp, we did not observe any consistent positional bias; the variability may have been related to the timing of Gl induction, as Ubx-FLP is active throughout wing disc development [33]. The cells that expressed Gl and Chp did not misexpress Elav (Fig 3A and 3B), indicating that they were neither originally fated to become neurons, nor converted to a neuronal fate by Gl. Similarly, when we misexpressed Gl in the leg disc using the Distal-less (Dll) driver, we observed ectopic Chp expression in Elav-negative cells ( Fig 3E). Gl is thus capable of activating the transcription of photoreceptor-specific genes in non-neuronal epithelial cells.
During normal eye development, gl expression is under the direct transcriptional control of Sine oculis (So), a retinal determination factor that forms a compound transcription factor with Eyes absent (Eya) [27,34]; gl can also be induced by the combination of Eya and Dachshund [35]. Consistent with the absence of Eya expression in the wild type wing disc epithelium [36], we found that Gl was still able to induce Chp when expressed in eya mutant clones ( Fig  3C). This result is consistent with Gl acting as a downstream effector of the retinal determination gene network.

Gl induces common and distinct sets of genes in neuronal and nonneuronal cells
To determine how a neuronal or non-neuronal cell context might influence Gl activity, we used RNA-Seq to examine changes in the transcriptome induced by Gl misexpression. For the neuronal misexpression condition, we used gl mutant third instar larval brains in which Gl expression was driven in neuroblasts by insc-GAL4 in combination with tubulin (tub)-GAL80 ts (Fig 2D). Using a gl mutant removed the contribution of clock cells in the brain that express Gl [37] and of mRNAs present in the photoreceptor axons that innervate the optic lobes. For epithelial misexpression, we used larval wing discs in which clones of cells generated with Ubx-FLP expressed Gl under the control of tubulin-GAL4 (Fig 3A and 3B). For comparison, we also sequenced the transcriptomes of wild-type larval eye discs, wing discs, wild-type and gl mutant brains. Fig 4A shows a heat map of 188 genes that were induced at least two-fold (with p<0.01) in Gl-misexpressing wing discs and/or brains, with additional cutoffs to eliminate genes that were expressed at very low levels or had a large variance between the triplicate samples (S1 Table). A core set of Gl target genes were induced in both tissues, enriched in wild-type eye discs compared to wing discs or brains, and reduced in gl mutant eye discs [38]. These include genes involved in photoreceptor differentiation, phototransduction, neurotransmitter reception, and pigment synthesis. In addition, we detected induction of genes with no previously described function in eye development (Table 1, S1 Table), but that are highly enriched in photoreceptors or cone cells in the developing or adult Drosophila eye [39].
Surprisingly, Gl induced many more genes in the wing disc than in the brain (Fig 4A and  4B). Some of the genes induced only in the wing disc are already as highly expressed in wildtype brains as in eye discs ( Fig 4C). However, other genes were induced in only one of the two tissues even though endogenous expression was low in both ( Fig 4C; Table 1), presumably reflecting regulation by tissue-specific activators or repressors. Two previous studies predicted a set of genes likely to be direct targets of Gl activation based on the presence of clustered Gl  Gl induces more genes in the wing disc than the brain. (A) is a heat map of 188 significantly induced genes, indicating fold changes (log 2 ) in Ubx-FLP>gl wing disc/wild type wing disc (induced in wing), insc>gl in gl brain/gl brain (induced in brain), wild type eye disc/wild type wing disc (eye/wing), wild type eye disc/wild type brain (eye/brain) and gl mutant eye disc/wild type eye disc (data from [38]). (B) Euler diagram drawn with APE showing the overlap of the 158 genes induced by Gl in the wing disc, the 44 genes induced by Gl in the brain, and the 89 Gl target genes predicted by [40], not including gl itself. The stringent cutoffs we used to classify genes as induced probably underestimate the actual overlap. (C) Heat map showing examples of different patterns of gene induction. Pph13 is induced in both tissues, enriched in the eye disc compared to the wing disc and brain, and reduced in gl mutants, making it a core Gl target gene. dyschronic (dysc) is induced by Gl in the wing disc but not the brain, and its levels are high in the wild binding sites in their regulatory regions and either reduced expression in gl mutant eye discs, or covariation with Gl across a variety of cell types and genetic conditions [38,40]. Although some of these predicted target genes were induced by Gl misexpression in one or both tissues, many were not. We found only 13 of the 89 Gl targets predicted by [40] were significantly induced in either condition ( Fig 4B, S3 Fig, S1 Table). One reason for this could be that the region searched for Gl binding sites included the first intron [38], which varies dramatically in length across different genes. The presence of potential Gl binding motifs is more likely to be significant in short than long sequences, and among the predicted target genes the length of the first intron was inversely correlated with the probability that the gene was induced in our type brain compared to the eye disc. CG18765 and CG9150 are induced specifically in the wing and the brain, respectively, despite low basal levels in the other tissue. Cyp4p2 is induced by ectopic Gl but is not enriched in the eye disc or dependent on endogenous gl. (D) A motif that was enriched in the regulatory regions of genes induced by Gl with p<10 −11 (top) is a close match to the binding site for Grh taken from Fly Factor Survey (bottom). Numbers are log 2 fold changes in conditions labeled as in Fig 4. Core target genes are induced in both tissues and enriched in eye discs compared to wing discs or brain with p<0.02, and reduced in gl mutant eye discs. Wing-specific genes are induced in the wing disc and unchanged or reduced in the brain, enriched in eye discs compared to wing discs, and reduced in gl mutant eye discs. Brain-specific genes are induced in the brain and unchanged or reduced in the wing disc, enriched in eye discs compared to brains, and reduced in gl mutant eye discs. We also note that this prediction method missed the validated direct target genes PvuII-PstI homology 13 (Pph13) and prospero (pros) [27,28,30]. The ability of Gl to induce genes that contain Gl binding motifs is also likely to depend on their responsiveness to other transcriptional activators and repressors present in the cell. A search for motifs that were enriched in the regulatory regions of genes induced by Gl, relative to predicted targets that were not induced, identified a motif matching the binding site for Grainyhead (Grh) (Fig 4D), a transcription factor that correlates with sites of open chromatin in the eye disc [40]. This suggests that chromatin structure controls the access of Gl to its binding sites.

Gl acts autonomously in non-neuronal cells in the eye disc
Our observation that Gl could activate gene expression in non-neuronal cells raised the possibility that its expression in the non-neuronal cell types in the eye [26,29] might correlate with a functional role there. Further evidence suggesting a function for Gl in cone and pigment cells came from the nature of the genes that were induced by Gl misexpression. While many of these genes are known to act in photoreceptors, including Pph13, Eye-enriched kainate receptor (Ekar) and the phototransduction component inactivation no afterpotential C (inaC) [41-43], Gl also induced genes that are highly enriched in cone cells, such as sallimus (sls), obstructor-B (obst-B), peste (pes) and the glial marker wrapper [39,44]. We confirmed that Gl-expressing clones in the wing disc autonomously induced Cut, a transcription factor specific to cone cells and bristle cells in the retina [45] (Fig 5A), and Sls (Fig 5C), which is enriched in cone cell feet in the pupal retina and lost from them in gl mutant clones (Fig 6G). Expression of these markers was not due to secondary induction of cone cells by Gl-expressing photoreceptors through Notch signaling [46], as it did not require the function of the Notch ligand Delta (Dl) in the Gl-expressing cells (Fig 5B and 5D). In addition to inducing cone cell markers, Gl misexpression led to the induction of pigment synthesis genes such as cardinal (cd) and rosy (ry) [47,48]. Consistent with induction of these genes, misexpression of Gl in clones of cells with eyeless (ey)-FLP or Ubx-FLP led to the appearance of ectopic red pteridine pigment, normally only produced by pigment cells in the eye [19], on the adult legs and abdomen (Fig 5E-5G).
gl mutants exhibit numerous defects in cone and pigment cell differentiation. The number of cone cells is reduced (Fig 6C and 6D), there is significantly less eye pigmentation (Fig 7D  and 7F) and the morphology of cone and pigment cells is abnormal [25,28] (Fig 6C), giving the eye a glassy appearance. However, these defects have been thought to arise as a secondary consequence of abnormal photoreceptor differentiation, and a possible autonomous requirement for Gl in these cells has not been examined. To test whether endogenous Gl has an autonomous function in cone and pigment cells, we used our UAS-gl transgene to restore Gl to specific cell types in a gl mutant background. Expressing gl in cone cells and primary pigment cells with sparkling (spa)-GAL4 [49] or in interommatidial secondary and tertiary pigment cells with 54-GAL4 [50] each produced a partial rescue of the mutant phenotype. Restoring Gl only in cone and primary pigment cells increased the number of cone cells per ommatidium (Fig 6B and 6D), consistent with an autonomous function of Gl within the cone cells that is independent of its role in photoreceptors. Restoring Gl only in the interommatidial pigment cells increased the level of pteridine eye pigments compared to gl mutant eyes with no rescue construct (Fig 7C, 7D and 7F). These results support an autonomous effect of Gl on the differentiation of the non-neuronal cell types of the eye. Restoring Gl specifically to photoreceptors with elav-GAL4 produced a similar partial rescue of cone cell number and pigmentation ( Fig  6D, Fig 7B and 7F), indicating that Gl also acts upstream of the signals produced by photoreceptors that induce the differentiation of other cell types.
We next wished to remove gl function from specific cell types. As the gl transgenic RNAi lines available did not fully reproduce the gl phenotype even when expressed ubiquitously throughout eye development, we used the CRISPR-Cas9 method to generate somatic gl mutations in a cell-type specific manner [51]. We generated a transgenic line that expressed two synthetic guide RNAs (sgRNAs) targeting gl (S4A Fig), and crossed it to UAS-Cas9 in a heterozygous gl mutant background. Expression of Cas9 throughout the eye disc beginning early in development with ey3.5-FLP and Act>CD2>GAL4 [52] resulted in mosaic eyes with patches resembling the gl mutant phenotype (S4C and S4D Fig). A weaker phenotype was obtained when we removed Gl specifically from photoreceptors using elav-GAL4 (S4E-S4G Fig). A majority of Elav + photoreceptors lost Gl expression, but Gl was still detected in the interommatidial cells, confirming that our CRISPR approach was cell type-specific (S4E and S4F Fig). Defects in the development of the non-neuronal cells were nevertheless observed (S2 Table), again consistent with reduced non-autonomous signaling by gl mutant photoreceptors.
To test the role of Gl in cone cells and primary pigment cells we used spa-GAL4 to express Cas9. Although Gl is not detectable in cone cells by 42h after puparium formation (APF) in the wild-type pupal retina, it is present during cone cell specification and in pupal primary pigment cells [29]. Gl expression was lost from a subset of presumptive cone cells marked by Pax2  in the third instar larval eye disc in spa-driven CRISPR mutagenesis (S4I Fig). At 42h APF we observed defects in the number and arrangement of cone cells and primary pigment cells, and perhaps as a secondary consequence, defects in the interommatidial cells (Fig 6E and 6H,  S5A-S5E Fig, S2 Table). Although the expression of cone cell markers such as N-cadherin (Ncad) [53] and Fasciclin III (FasIII) [39] was unaffected (Fig 6E), Sls was depleted from cone outlines (green in F) and anti-Sls (F', red in F). Mutating gl in cone cells can cause defects in their patterning (E) and loss of Sls in cone cell feet (F, arrows). Some staining is still visible in the photoreceptor axons that surround the cone cell feet. (G) 42 h APF retinas with GFP-labeled gl 60j homozygous mutant clones. Retinas are stained with anti-GFP (green), anti-Arm (red) and anti-Sls (G', blue in G). Sls is strongly expressed in cone cell feet and is lost in gl 60j clones. (H) quantification of patterning defects due to loss of Gl in cone cells and primary pigment cells (n = 143), compared to controls in which UAS-Cas9 and gl sgRNAs are present without a GAL4 driver (n = 295) or Cas9 is expressed in the absence of gl sgRNAs (n = 198). ÃÃÃÃ p<0.0001, Ã p<0.05, Fisher's exact test. Scale bars: 10μm.
https://doi.org/10.1371/journal.pgen.1007173.g006 cell feet when gl was mutated in these cells by driving Cas9 with spa-GAL4 (Fig 6F). Expression of Cas9 with 54-GAL4 resulted in loss of Gl from the interommatidial cells (S4K and S4L Fig), defects in the pigment cell lattice (Fig 7G and 7H, S5F-S5J Fig, S2 Table), and loss of pigment in patches in the adult eye (Fig 7E). These GAL4 lines were specific to the non-neuronal cells, as very few photoreceptor defects were observed (S5K Fig). These results demonstrate that Gl autonomously promotes the differentiation of all the cell types of the eye.

Gl cooperates with other transcription factors to induce cell type-specific genes
Our finding that Gl acts autonomously in photoreceptors, cone cells and pigment cells raises the question of how it activates a different set of genes in each cell type. Gl has been shown to cooperate with the downstream transcription factor Pph13 to induce and maintain photoreceptor-specific gene expression [27,32], and Orthodenticle also plays a role in regulating terminal differentiation of photoreceptors [41]. We therefore focused on identifying transcription factors that might help Gl to activate cone or pigment cell genes. Pax2 is a transcription factor that is specifically expressed in cone and primary pigment cells [54], independently of Gl (S4J Fig), and is required for their development [24,54]. It also shares many predicted target genes with Gl [40]. We found that coexpressing Pax2 with Gl in clones in the wing disc prevented induction of the photoreceptor-specific gene chp (Fig 3D), consistent with the previously reported anti-neuronal function of Pax2 [24], and enabled the induction of eya, which was induced in very few cells by either factor alone (Fig 8A-8C). Eya is specifically expressed in non-neuronal cells at pupal stages [55] and has a late function in their terminal differentiation, independent of its early role in retinal determination [56,57]. Consistent with this function, coexpressing Eya with Gl enabled it to induce lozenge (lz) (Fig 8D-8F), which encodes a transcription factor that is expressed in late-differentiating cells including the cone and pigment cells and is necessary for their differentiation [23,58,59]. Gl is necessary for lz expression [60] but not alone sufficient to induce it in the wing disc (Fig 8D).
Gl and Lz are known to cooperatively activate the expression of Pros, which cooperates with Pax2 to promote cone cell differentiation [24,30], and are predicted to co-regulate many additional targets [40]. We found that in combination but not individually, Gl and Lz were able to induce ectopic pros and homothorax (hth) expression in the wing disc (Fig 8G-8I). Interestingly, Pros staining appeared largely cytoplasmic, a localization seen in mature cone cells but not in R7 photoreceptors, where it is nuclear [39,61]. Hth is specific to pigment cells at pupal stages [62]. In clones misexpressing both Gl and Lz, Pros-expressing cells were often surrounded by Hthexpressing cells (Fig 8I), reflecting the relative positions of cone and pigment cells in normal development. Gl was still able to induce ectopic pigment when expressed in hth mutant clones ( Fig 8L); however, Escargot (Esg), another transcription factor specific to pupal pigment cells [63] that is also predicted to coregulate Gl target genes [40], showed a clear cooperative interaction with Gl. Coexpressing Esg with Gl in clones generated with Ubx-FLP strongly enhanced ectopic pigment formation (Fig 8M-8O), while Gl failed to induce pigment when expressed in esg mutant clones (Fig 8K). Esg is generally thought to be a repressor [64], so its effect on the induction of pigment synthesis genes is likely to be indirect. Together, these results show that other transcription factors influence the set of target genes that Gl is able to activate, and suggest a preliminary model for the transcriptional control of cell type differentiation in the eye (Fig 8P).

Discussion
Previous studies of Gl have described it as a determinant of photoreceptor cell fate [25,27,28]. Our data suggest that this characterization does not fully capture its function. We examined whether Gl was sufficient to drive uncommitted cells toward a differentiated photoreceptor cell identity. Although Gl could induce a subset of photoreceptor-specific genes, it did not initiate the entire photoreceptor differentiation program. Gl activated overlapping but distinct sets of genes in neuronal and non-neuronal cells, indicating that its function is not limited to neurons, but its effects are dependent on the cellular context. The results of loss-of-function, gain-of-function and rescue experiments show that in addition to its role in photoreceptors, Gl acts cell-autonomously to promote the normal differentiation of non-neuronal cone and pigment cells in the eye. Gl thus appears to have a parallel function in all the cell types of the eye, in combination with distinct transcription factors in each cell type, to give them their identities as components of an integrated organ (Fig 8P).

Cell context influences the transcriptional activity of Gl
Context-dependent effects on transcription factor activity have been well documented [65][66][67][68][69][70][71]. Our transcriptomic analysis shows that Gl can activate some of its target genes in cells that have either a neuronal or an epithelial identity, while others are activated in only one of the two contexts. As Gl is expressed earlier than neuronal markers [26], it may act in epithelial progenitors in the eye disc prior to their neuronal differentiation. The developing eye and wing discs are distinguished by the expression of the retinal determination genes in the eye disc primordium [14]. One important function of this gene network is to induce Gl, which then directs eye disc cells to differentiate as components of the retina [27,52]. The wing disc shares common signaling pathways with the eye disc, such as Hh, EGFR and Notch, which may enable Gl to activate some of its target genes in this context. Similarly, Decapentaplegic (Dpp) pathway activity appears to aid the ability of the "master transcriptional regulator" Eyeless/Pax6 to induce ectopic eye formation [35,72].
The effect of Gl also depends on other cell type-specific transcription factors. A Gl binding site from the Rh1 proximal enhancer drives more restricted reporter expression when adjacent sequences are included, supporting the existence of a repressor that can counteract activation by Gl [29]. The presence of such repressors or the absence of coactivators probably explains why only a few of the previously predicted Gl target genes [40] were induced by Gl misexpression in the wing disc or brain, and tissue-specific cofactors are likely to contribute to differential induction in the two contexts. We identified Pax2, Eya, Lz and Esg as transcription factors that can alter the spectrum of genes induced by Gl misexpression in the wing disc. In addition, the chromatin landscape may affect the availability of Gl binding sites. The enrichment of a Grh binding motif in genes that were induced by Gl suggests that Gl requires an open chromatin state to activate transcription [40], and is thus not a pioneer transcription factor [73]. Similarly, the C. elegans Gl homologue CHE-1, which is required for the expression of genes specific to the ASE gustatory neurons [74], can reprogram other cells into ASE neurons only when factors that promote chromatin-mediated repression are removed, indicating that its ability to activate target genes is influenced by their chromatin state [71,[75][76][77].
The duration of Gl expression may also influence its ability to activate target genes. Previous studies which concluded that Gl was not sufficient to activate photoreceptor-specific genes in non-neuronal cells used transient misexpression from a heat shock promoter [29]. However, we found that maintenance of Gl expression in the differentiated progeny of neuroblasts was necessary to induce ectopic chp expression. Extended expression may allow Gl to induce its target gene Pph13, which is necessary for the ectopic induction of genes such as Rh1 and the phototransduction components inactivation no afterpotential D and Arrestin 1 [27]. This feedforward mechanism could contribute to the specificity of photoreceptor determination.

Gl promotes the differentiation of multiple eye cell types
In general, cell fate specification is viewed as a series of decision points at which the expression of different transcription factors directs cells towards progressively restricted fates [10].
Downstream of factors that define broad spatial or temporal identities [78], "master regulators" such as Pax6 specify a field of multipotent progenitors within which patterned differentiation of all the cell types of the organ can occur [6]. Terminal selector genes are thought to be confined to specific cell types and directly induce their differentiated properties [8]. Gl was previously viewed as a terminal selector gene for photoreceptor identity [25,27,28]. However, Gl is present in cone and pigment cell precursors in addition to photoreceptors [26,29] and regulates the expression of genes such as lz and pros that are expressed in both photoreceptors and cone cells [30,60]. Our cell type-specific rescue and loss-of-function experiments show that the abnormal arrangement and differentiation of non-photoreceptor cells in gl mutant eyes is not simply a secondary consequence of the effect of gl on photoreceptors; instead, gl also acts autonomously in cone and pigment cells to regulate their number, arrangement and gene expression. In addition, ectopic Gl is capable of inducing genes specific to cone cells and the synthesis of red pteridine pigments that are normally made by secondary and tertiary pigment cells [19]. Gl thus appears to be a terminal differentiation factor for multiple cell types of the eye rather than a photoreceptor determinant. Few factors that confer organ identity on multiple distinct cell types have been described; for instance, differentiation of endocrine and exocrine cells in the pancreas appears to involve entirely distinct transcriptional regulatory networks [79]. In plants, however, different combinations of transcription factors produce different floral components, such that each transcription factor contributes to specifying multiple cell types of the flower [80].
As Gl itself does not distinguish photoreceptors from non-neuronal eye cells, its transcriptional targets must depend on other factors that control the identity of each cell type. The EGFR and Notch signaling pathways are important for recruiting both photoreceptor and non-photoreceptor cell types, and the transcription factors downstream of these pathways directly regulate cell type-specific genes [24,81]. Although EGFR signaling has a direct or indirect effect on the differentiation of all cell types except R8 [17,82,83], the level of signaling can influence cell fate specification. For example, activation of the Sevenless receptor increases Ras-MAPK signaling in R7 relative to cone cells; this allows the repressor Tramtrack to be degraded, leading to high level expression of genes that promote neuronal identity [23,84,85]. Combinatorial effects of the two pathways contribute to cell type-specific expression of genes such as Pax2 [59]. Our experiments suggest that Pax2 and Esg can modify the effects of Gl to promote the induction of cone and pigment cell genes respectively through a network of other downstream transcription factors.
The level of Gl could also affect which target genes it activates; Gl expression is reduced in cone cells during pupal development [29], and different levels of a transcription factor can specify different cell fates in other systems [86,87]. In addition, the time of cell differentiation could play a role; for instance, neuroblasts express a temporal series of transcription factors that specify different identities in their progeny [88,89]. Lz, a transcription factor that contributes to specifying R1, R6, R7 and the cone and pigment cells [22,23,58,59], is a target of Gl regulation and is also predicted to act in combination with Gl to control many common target genes, suggesting that it functions as a feed-forward temporal determinant [40, 60,90]. We found that Lz can indeed bias transcriptional activation by Gl towards genes expressed in later-differentiating cell types. The distinct effects of different combinations and levels of transcription factors and signaling pathways demonstrate how unique cell fates can be specified from a common pool of progenitors using few factors. Nevertheless, our understanding of cell type specification in the eye is probably still incomplete; for instance, analysis of the spa enhancer that drives Pax2 expression in cone cells revealed many essential inputs in addition to the previously established regulation by EGFR signaling, Notch signaling and Lz [59,91]. A fuller understanding of the network that modifies the transcriptional activity of Gl may hold the key to the problem of cell fate specification.

Drosophila genetics
UAS-gl-RB was made by cloning the gl cDNA from clone GH20219 (Drosophila Genomics Resource Center) into pUASTattB using EcoRI and XhoI. A 1.5 kb NdeI/XhoI fragment lacking the intron was generated by overlap PCR and used to replace the corresponding region of UASgl-RB to generate UAS-gl-RA. Both constructs were integrated into the VK37 attP site at position 22A3. The gl sgRNA sequences identified on www.flyrnai.org/crispr2/ [92], GCAGGATAGG-CAGCCGACGC (gl gRNA 1) and TACCCACCGCTGCTGAGTCC (gl gRNA 2) were cloned into pCFD4 [51] by PCR and Gibson assembly, and the construct was integrated into the attP40 site at 25C6. Injections and screening of transgenic flies were carried out by Genetivision.
Stocks used to generate clones were (1)

Pigment quantification
Pigment quantification was performed as described [97] with the following modification: for each sample, 30 heads from 3-5 day old adult females were homogenized in 0.5mL of 0.1M HCl in ethanol. OD 480 values were obtained using the homogenizing solution as a blank.

RNA-Seq
Larval tissue was isolated from 30 animals of each genotype in triplicate and RNA was extracted in Trizol (Invitrogen). Library preparation and sequencing was carried out by the NYU Genome Technology Center. RNA-Seq library preps were made using the Illumina TruSeq RNA sample Prep Kit v2 (Cat #RS-122-2002), using 500 ng of total RNA as input, amplified by 12 cycles of PCR, and run on an Illumina 2500 (v4 chemistry), as single read 50. For each RNA-seq sample, sequence quality was assessed with FastQC (http://www.bioinformatics.babraham.ac. uk/projects/fastqc) and sequencing adapters were removed with Trimmomatic [98]. Cleaned reads were aligned to the Drosophila reference genome (dm3) with Tophat2 v2.1.1 1. The Picard CollectRnaSeqMetrics program (https://broadinstitute.github.io/picard/picard-metricdefinitions.html#RnaSeqMetrics) was used to generate QC metrics including ribosomal RNA content, median per-gene coverage, bases aligned to intergenic regions, 5'/3' biases, and the distribution of the bases within exons, UTRs, and introns. Per sample gene expression profiles were computed using Cufflinks v2.2.1 1 and the RefSeq genome annotation for the Drosophila reference genome dm3 [99]. For multi-sample comparison, Principal Component Analysis and hierarchical clustering were used to verify that the expression profiles of the sequenced samples clustered as expected by sample tissue and genotype. Differential gene expression was computed for various contrasts between genotypes with the Cufflinks protocol [100] with default thresholds. Reads on the gl gene were visualized with Integrative Genomics Viewer (IGV, Broad Institute). Genes were considered significantly changed and included in the heat map if the log 2 fold change for either wing or brain misexpression was >1, p<0.01, average cpm for that misexpression condition was >1, and standard deviation/mean of the three replicates was <0.5. Intron sizes were obtained from Flybase and compared for targets predicted with high confidence by [40] that were induced (log fold change >1 in either or both tissues) or not induced (log fold change <0.1 in both tissues). Homer was used to identify motifs enriched in the genes included in the heat map compared to predicted targets that were not induced in either tissue, using a region that extended 2 kb upstream and downstream of each gene. RNA-Seq data have been submitted to NCBI GEO (reference number GSE99303). presence of Gl binding motifs in a region consisting of 5 kb upstream and the first intron are binned according to the size of their first intron and plotted as induced by Gl in one or more tissues (black) or induced in neither (grey). (B) A plot of the log fold change in Gl-expressing wing discs (non-patterned bars) or brains (patterned bars) for the 13 predicted targets that were induced by Gl, divided according to the size of their first introns (intron size 0-500 (red), 501-1000 (green), 1001-5000 (yellow), >5000 (blue)). Interestingly, 5 of these genes are highly enriched in mature photoreceptors (HisCl1, Ekar, inaC, chp, nrm) and 6 in cone cells (CG9150, cd, CG5653, sls, CG11498, Cad88C).  Table. Lists of the genes shown in the heat map in Fig 4A and the Euler diagram in Fig  4B. Separate sheets show the genes induced when gl was misexpressed with insc-GAL4 in gl mutant larval brains compared to gl mutant brains, genes induced when gl was misexpressed in clones made with Ubx-FLP in wing discs compared to wild type wing discs, genes more highly expressed in wild type third instar eye discs than wing discs, and genes more highly expressed in wild type third instar eye discs than brains, all using the same cutoffs (fold change>2, p<0.01, average counts in eye discs >1 and standard deviation/mean of eye disc samples<0.5). The final sheet shows predicted direct targets of Gl according to [40].