Buffered EGFR signaling regulated by spitz-to-argos expression ratio is a critical factor for patterning the Drosophila eye

The Epidermal Growth Factor Receptor (EGFR) signaling pathway plays a critical role in regulating tissue patterning. Drosophila EGFR signaling achieves specificity through multiple ligands and feedback loops to finetune signaling outcomes spatiotemporally. The principal Drosophila EGF ligand, cleaved Spitz, and the negative feedback regulator, Argos are diffusible and can act both in a cell autonomous and non-autonomous manner. The expression dose of Spitz and Argos early in photoreceptor cell fate determination has been shown to be critical in patterning the Drosophila eye, but the exact identity of the cells expressing these genes in the larval eye disc has been elusive. Using single molecule RNA Fluorescence in situ Hybridization (smFISH), we reveal an intriguing differential expression of spitz and argos mRNA in the Drosophila third instar eye imaginal disc indicative of directional non-autonomous EGFR signaling. By genetically tuning EGFR signaling, we show that rather than absolute levels of expression, the ratio of expression of spitz-to-argos to be a critical determinant of the final adult eye phenotype. Proximate effects on EGFR signaling in terms of cell cycle and differentiation markers are affected differently in the different perturbations. Proper ommatidial patterning is robust to thresholds around a tightly maintained wildtype spitz-to-argos ratio, and breaks down beyond. This provides a powerful instance of developmental buffering against gene expression fluctuations.


Introduction
Receptor tyrosine kinases (RTKs) are key regulators of diverse cellular processes and development. Mutations or aberrant activation/inactivation of RTKs lead to different anomalies including cancers [1]. RTKs trigger a cytoplasmic signaling cascade involving Ras-MAPK pathway [2]. EGF receptor signaling is extensively studied for its role in cellular homeostasis and in relation to human diseases [2]. Drosophila melanogaster has one EGF receptor (DER) and four activating ligands [3]. The difference in expression of the ligands in a tight spatiotemporal pattern was suggested to bring about difference in EGF responses. Among Drosophila EGFR ligands, Gurken [4][5][6], Spitz [7][8][9] and Keren [10] are homologous to TGFα and Vein [11][12][13] is homologous to neuregulin. Spitz is the canonical EGF in the fly. Beyond different ligands, positive and negative feedback loops are known to tightly regulate EGFR signaling [14,15]. The downstream targets of EGFR include activators like Vein and Rhomboid [16,17] and inhibitors like Argos [18-20], Kekkon-1 [21,22] and Sprouty [23,24]. Of the feedback molecules, Sprouty, Rhomboid and Kekkon-1 act in a cell autonomous manner, while Argos is a diffusible factor like cleaved Spitz, and can potentially act both in a cell autonomous and nonautonomous manner. The competitive and stoichiometric sequestration of Spitz by Argos has been shown to be critical for patterning different tissues [18,25]. Spitz and Argos are thought to be short-range activator and long-range inhibitor respectively [26].
A strong instance of EGFR-mediated patterning occurs in the Drosophila compound eye with its periodic units called ommatidia. A single ommatidium comprises of eight neuronal photoreceptors (PR) accompanied by twelve non-neuronal cells [27][28][29]. The photoreceptor cell fate specification starts in the early 3 rd instar larvae along an anteriorly progressing wave of Hedgehog in the eye disc leaving a morphogenetic furrow behind [30]. Posterior PR cells are older in developmental time than clusters just behind the morphogenetic furrow. EGFR signaling is a prerequisite for photoreceptor specification (except for founder photoreceptor, R8) and also cone and pigment cells until pupal stage [9,31].
Spitz binds to EGFR and activates the cytoplasmic Ras-MAPK pathway [32,33]. This cascade activates the transcriptional activator, PntP1 and degrades transcriptional repressor, Yan [34][35][36]. Downstream of the signaling pathway, argos is expressed and the protein product is secreted out of the cell. Argos binds Spitz in a 1:1 ratio by clamping to the EGF domain and restricts the amount of free ligand available for EGFR activation [37,38]. Dosage of EGFR components is known to maintain a biochemical balance, which dictates the final strength of the signaling pathway. Reducing Spitz levels in the background of Argos hypomorphic mutant reverted the rough eye phenotype to near-wildtype whereas in the background of Argos overexpression, Spitz dosage reduction enhanced the rough eye phenotype [18,25]. The relative strength of signaling decides the cellular choice, which in turn contributes towards proper pattern formation. But while relative expression dose of Spitz and Argos early in photoreceptor cell fate determination has been shown to be critical in patterning the Drosophila eye, the exact identity of the cells expressing these genes or the degree of differential expression of these genes in the larval eye disc has been elusive.
Argos is both a negative regulator and a target of EGFR signaling. argos expression is a specific proxy for strong EGFR signaling [20,39]. To quantitatively understand the level of expression of diffusible EGFR components (Spitz and Argos), following their protein products can be problematic. Immunofluorescence staining or enhancer trap lines to track diffusible proteins is challenging as quantitative inference on the type of cells which have secreted them is not possible [39]. The dual phosphorylated ERK (dpERK) staining is classically used to read the strength of the EGFR pathway. Since most of the RTK pathways converge on the MAPK pathway, dpERK staining may not specifically and quantitatively report on EGFR strength. These problems can be circumvented by detecting endogenous mRNA in situ, which additionally could also indicate the identity of cells that express these genes. In recent years Single-molecule Fluorescence In Situ Hybridization (smFISH) has emerged as a powerful method to sensitively and quantitatively report the expression of more than one gene simultaneously in wholemount tissues, along with cell-to-cell variability [40][41][42]. We have previously shown that such methods can be used to detect even low levels of gene expression in the wholemount Drosophila tissues [40].
In this paper, using smFISH for EGFR pathway genes, we reveal an intriguing differential expression of spitz and argos in photoreceptor and non-photoreceptor cells of the larval eye disc. We show that relative expression levels of spitz and argos is important for the generation of proper ommatidial pattern rather than their absolute expression level. By systematically tuning the expression of EGFR pathway genes, we analyze the biochemical buffer range where the relative gene expression levels of the ligand spitz and the negative feedback regulator argos contribute towards pattern formation during the Drosophila eye development.

EGFR signaling is directional during morphogenetic furrow progression
EGFR and the principal EGF, Spitz in Drosophila are expressed uniformly throughout multiple tissues during the development [43]. In the 3 rd instar eye disc, Spitz is responsible for activation of EGFR signaling posterior to the morphogenetic furrow. Downstream targets of EGFR signaling like argos are expressed as a downstream response to PntP1 activation [44]. As Spitz and Argos are diffusible factors that can act in both a cell autonomous and non-autonomous manner and act in a 1:1 stoichiometry to modulate EGFR signaling [37], we reasoned that their relative expression may be key for tissue patterning. We investigated their transcription status quantitatively in cells behind the morphogenetic furrow in larval eye discs. In-situ hybridization (ISH) for spitz has been performed in eye discs at this stage [9], and patterns of dpERK signal also described by immunofluorescence [39]. Despite this, clear cell type-specific differences have not been highlighted most likely due to lower sensitivity of the methods used. We therefore performed smFISH (S1 Fig) to quantify spitz and argos mRNA numbers combined with immunofluorescence for Elav, a pan-neuronal marker [45] to distinguish photoreceptor (PR) cells from other non-photoreceptor (non-PR) and undifferentiated pools of cells. Elav staining revealed a rosette-like pattern of DAPI stained nuclei in PR cells that were used for RNA counts (S2A Fig). Since the nuclei are densely packed and individual cells are hard to segment in the eye imaginal disc, we quantified the mRNA expression on 3D stacks and normalized to a 1000μm 3 tissue volume after delineating PR cells from non-PR cells. The membranes between PR and non-PR nuclei could be separated using Dlg, a membrane marker (S2B Fig). To our surprise we found that spitz is clearly expressed in higher levels in the photoreceptor cells (Elavpositive cells) compared to neighboring non-photoreceptor cells (Figs 1A, 1B, 1C and S1B). The negative feedback regulator, argos, on the other hand, is expressed exclusively in the non-photoreceptor cells that have not yet made a cell fate choice. Because argos expression is direct target of EGFR signaling, argos expression in neighboring cells indicates an exclusively cell non-autonomous effect of Spitz secreted by photoreceptor cells on their neighboring cells. The directionality in signaling can also be visualized by dpERK staining, a classical marker for high EGFR signaling, in the non-PR cells, and like argos mRNA higher dpERK levels are seen in non-PR . spitz mRNA is highly expressed only in the Elav-positive photoreceptor cells, whereas argos mRNA is highly expressed in the neighbouring non-neuronal cells. To clearly see the exclusive expression of spitz and argos, the image is zoomed into the yellow box and presented in (B). Scale bar is 10μm in (A) and 5μm in (B). These are z-projected images and hence individual transcripts get merged. A single z-slice is shown in S1B  (Figs 1D and S2D) indicating that the expression patterns are stable in time, because morphogenetic furrow progression is a proxy for time in development. We use eye discs from larvae of similar age (i.e. similar number of columns posterior to the furrow), but this indicates that otherwise too, the spitz and argos numbers are stable posterior to the furrow.

Modulating absolute transcript numbers of spitz and argos may not translate to an effect on eye phenotype
Knocking down components of the DER pathway during development in the eye imaginal disc is known to affect the adult eye phenotype. Hypomorphic allele of argos and argos overexpression both give rise to rough eye phenotypes [18]. Halving Spitz in the background of hypomorphic Argos reverted the rough eye phenotype to normal whereas in the background of Argos overexpression, it further enhanced the rough phenotype [18]. We used the UAS-Gal4 system [46,47] to knockdown argos and spitz. We used GMR-Gal4 [48] to drive UAS-dsRNA in all cells posterior to the morphogenetic furrow and Elav-Gal4 [49] to drive UAS-dsRNA in photoreceptor (neuronal) cells respectively. smFISH showed the decrease in respective mRNA transcripts compared to wildtype CantonS eye discs. As expected, we observed decreased argos and spitz expression when GMR-Gal4 was used to drive argos-dsRNA and spitz-dsRNA respectively and the adult flies showed a fully penetrant rough eye phenotype (Fig 2A). To our surprise, while the Elav-Gal4 driver did show knockdown of the spitz expression when driving the spitz-dsRNA, there was absolutely no discernible defect in the phenotype of the adult eye (Fig 2A). The defect in the adult eye phenotype is not due to the GMR-Gal4 as the absolute spitz and argos transcript counts in the GMR/+ control (with and without heatshock at 29˚C) are similar to CantonS (S3A Fig), and the adult eyes show regularly patterned ommatidia (S3B Fig). argos and spitz mRNA was quantified in all the crosses along with CantonS and the numbers reflected the respective gene knockdowns (Fig 2B and 2C). The quantified gene expression could not explain the absence of effect on adult eye phenotype in the spitz knockdown driven by Elav-Gal4. As expected, eye discs did not show any change in transcript numbers when argos dsRNA was driven by Elav-Gal4, which is expressed exclusively in PR cells As relative dosage of EGFR pathway is known to be important for photoreceptor specification [18,25], we wondered if the ratio of expression of spitz to argos may be the critical determinant of the final eye phenotype instead of absolute levels of expression. This does assume that mRNA level expression translates directly to levels of final activated protein products, but given that Spitz and Argos act in a 1:1 stoichiometry, we thought this hypothesis is well worth testing. Thus, we analyzed the spitz-to-argos expression ratio through smFISH in the photoreceptor cells ( Fig 2D) and eye fields with no distinction of PR and non-PR cells ( Fig 2E). Remarkably, the spitz-to-argos ratio in the progeny of Elav-Gal4 with spitz dsRNA was similar to wildtype CantonS flies, while it was either significantly higher or lower in other genotypes (GMR-Gal4 driving argos-dsRNA and spitz-dsRNA). Thus, the gene expression ratio of the activating ligand, Spitz, and the negative feedback regulator, Argos, might dictate the final availability of free active ligand for pathway activation rather than their absolute expression levels. Cell-type specific transcript counting (PR cells) and full-field counting (no distinction of PR or non-PR cells) showed similar trends of spitz-to-argos ratios and we will describe fullfield counts in all the further experiments.

Fig 2. spitz-to-argos ratio is important for proper ommatidial patterning. (A) smFISH was performed in eye
imaginal discs from represented crosses for spitz and argos mRNA in the two rows. In the third row, adult eye phenotypes are shown. The Elav > UAS spitz dsRNA progeny shows perfectly arranged ommatidia like the wildtype CantonS, while the other two crosses show rough eye phenotypes. Scale bar for widefield fluorescence images is 10μm and SEM images is 100μm. (B) and (C) mRNA numbers for spitz and argos were counted in photoreceptors and nonphotoreceptors in respective crosses. spitz in PR cells and argos mRNA in non-PR cells show variation in number as expected from the respective crosses (spitz expression has p-value > 0.05 ( � ) in a Student's t-test when compared between CantonS and GMR-Gal4 driving spitz dsRNA. Other comparisons of spitz expression in PR cells and argos expression in non-PR cells are highly significant with p-values < 0.001 ( ��� ) in a Student's t-test). spitz-to-argos mRNA ratio in the photoreceptor cells (D) was calculated as dosage was known to be important for ommatidial pattern formation (N = 9 tissues in all the crosses). Most remarkably, knocking down spitz in PR cells with a Elav driver, knocks down argos numbers too, and overall the ratio is unchanged. When a GMR driver is used to knockdown either argos or spitz in the full field, the ratio is higher or lower than wildtype. spitz-to-argos ratio in the eye field irrespective of cell type (PR or non-PR) did not show any difference and is used henceforth in the paper (E). spitz-to-argos ratios

Buffered spitz-to-argos ratio is important for proper ommatidial patterning
To test the hypothesis that the ratio of expression of spitz-to-argos is critical for determining the final eye phenotype we attempted to use other cassettes to modulate their levels and also to tune their expression in a systematic manner. EGFR CA is a constitutive active form of EGFR, and hence the downstream targets will be expressed throughout the development. We used a GMR-Gal4 driver to express this constitutive EGFR. As GMR expresses in all the cells posterior to the morphogenetic furrow [31], the patterned expression of argos is lost ( Fig 3A). spitzto-argos expression ratio was quantified for the EGFR CA progeny, the balancer control (non-EGFR CA ) and CantonS larvae. The balancer control had spitz-to-argos ratio levels similar to the CantonS around 2-2.2, whereas the EGFR CA expressing larvae had decreased ratio around 0.5 given high expression of argos and dramatically smaller and rough eye in adults ( Fig 3B  and 3C). To investigate if patterning defects in the adult eye vary continuously with the spitzto-argos ratio or show up beyond a specific threshold indicative of developmental buffering, we tuned the expression of spitz and argos. We generated a GMR-Gal4 line with a temperature sensitive Gal80 (Gal80 ts ) [50,51] to tune the expression of spitz and argos. Gal80 ts inhibits Gal4 activity at 18˚C but turns inactive at higher temperatures allowing the control on UAS-Gal4 mediated gene expression. EGFR CA was crossed with Gal80 ts ; GMR-Gal4 and shifted to 29˚C for different time points. The larvae were processed for smFISH and the absolute gene expression for spitz and argos ( Fig 3D) and the spitz-to-argos ratios ( Fig 3E) were quantified. The corresponding SEM images of the adult eyes were compared. The adult eye phenotype started to show patterning defects when the spitz-to-argos gene expression ratios neared 1 upon 40 minutes of heat shock at 29˚C (Fig 3F). The intermediate spitz-to-argos expression ratio did not lead to roughening of the adult eye, indicating the phenotypic effects start to appear beyond a critical biological range.

Threshold switch of spitz-to-argos ratio regulates phenotype
To recapitulate the buffer range with another gene, we crossed UAS-argos to the Gal80 ts ; GMR-Gal4 driver line to overexpress argos. Absolute gene expression ( Fig 4A) and ratio ( Fig 4B) were quantified. The corresponding SEM images ( Fig 4C) were observed for patterning defects. The adult eyes showed patterning defects when the spitz-to-argos ratio neared 1. This confirmed that the adult eye phenotype is indeed contingent on the spitz-to-argos ratio. The phenotype did not show any defects in the adult eye even though the spitz-to-argos ratio dropped from 2.2 in the no-heat shock control till it reaches 1. We did not capture any defects in the ommatidial patterning in the flies with intermediate ratio. This suggests that patterning defects show up only when the spitz-to-argos ratio crosses a certain threshold and the intermediate ratios are buffered in the development. Having investigated the effects of decreasing the spitz-to-argos ratio, we wondered how far the buffering will persist when we increased it from wildtype levels. For this, we expressed an EGFR DN (dominant negative allele) with the Gal80 ts ; GMR-Gal4 driver, where ratios would be driven to values greater than wildtype. spitz and (D and E) in the GMR-Gal4 driving argos dsRNA and spitz dsRNA are significantly different (p-values < 0.001 ( ��� ) in a Student's t-test) when compared to wildtype CantonS while there was no significant difference in progeny of Elav-Gal4 driving spitz dsRNA (n.s. indicates p-values > 0.05 in a Student's t-test). Error bars in (B), (C), (D) and (E) are standard errors of mean. The images within the figure panels are created by the authors. Scale bars in Scanning Electron Microscopy images have been provided from knowledge of pixel size and converting appropriately. Asterisks denoting significance of observed changes have been added to relevant graphs.
https://doi.org/10.1371/journal.pgen.1010622.g002 The ommatidial pattern is lost when EGFR is constitutively active and the size of the eye is also reduced. Heatshock was administered for 6 hrs to EGFR CA and non-EGFR CA larvae for maximal activation of the cassette. Scale bar is 10μm. (D) Absolute spitz and argos mRNA numbers from Gal80 ts ; GMR-Gal4 > UAS EGFR CA progeny after different timepoints of heatshock at 29˚C. N = 8 for all time points. (E) spitz-to-argos ratios in eye discs corresponding to different time points of heat shock at 29˚C. Quantification of absolute mRNA numbers and ratios represented in all the above plots were calculated in the eye field irrespective of the cell type. Error bars in all the plots are standard errors of mean. The change in spitz-to-argos ratio with heatshock is significantly different irrespective of timepoints when compared with no-heatshock control (p-values < 0.001 ( ��� ) in a Student's t-test). (F) Adult eyes corresponding to different time points of heat shock at 29˚C. Scale bar is 100μm. No rough eye phenotype is seen for 20 min of heatshock, though a difference in the spitz-to-argos ratio is already seen at this point. Beyond this the eyes are fully rough, and the phenotype is fully penetrant. This is indicative of developmental buffering. The images within the figure panels are created by the authors. Scale bars in Scanning Electron Microscopy images have been provided from knowledge of pixel size and converting appropriately. Asterisks denoting significance of observed changes have been added to relevant graphs.
https://doi.org/10.1371/journal.pgen.1010622.g003 argos mRNA transcripts were quantified ( Fig 4D) along with their ratio (Fig 4E) at different time points of heat shock at 29˚C. The respective SEM images (Fig 4F) showed ommatidial defects when the ratio approached 3. The buffer range of spitz-to-argos for contributing towards proper ommatidial patterning might lie between 1-3 while the wildtype ratio lies between 2-2.2. While the term 'rough eye' can allude to a broad range of phenotypes, the patterning defects in the adult eye were quite different at both ends of the spitz-to-argos buffer range. Low ratios broadly showed irregularly spaced ommatidia with decreased number of bristle cells and high ratios showed bristle cells with complete loss of ommatidial clusters (Figs 3F, 4C and 4F). The differences could arise due to the effect of receptor signal strength on the photoreceptor cell fate specification or cellular division in the disc. We also used Elav-Gal4 to drive UAS Argos to ectopically overexpress argos in the photoreceptor cells ( S5A Fig).

Different genetic perturbations affect proximate EGFR signaling outcomes differently
While EGFR signaling is important for photoreceptor specification in the larval stage, further EGFR-dependent identity determination also happens in the pupae; this finally results in a patterned eye phenotype in the adult, which is a distal outcome. To test the proximate signaling mechanisms that give rise to the final 'rough eye' phenotypes in the adults with different perturbations of EGFR signaling, we compared wildtype CantonS eye discs with discs from the progeny of Elav-Gal4 driving UAS-spitz-dsRNA and GMR-Gal4 driving UAS-spitz-dsRNA, UAS-EGFR CA and UAS-EGFR DN (Fig 5A). Yan, a negative transcription factor downstream to the EGFR pathway is known to be present in higher levels in the eye disc when EGFR is active [36]. High Yan in our experiments may be indicative of more sustained EGFR activity and hence was quantified. Yan intensity in discs of CantonS and progeny of Elav-Gal4 driving spitz dsRNA were similar compared to decreased intensity in discs from GMR-Gal4 driving spitz dsRNA and EGFR DN . Eye discs from GMR-Gal4 driving EGFR CA have significantly higher Yan compared to wildtype (Fig 5B). Previous studies have indicated a stochastic pulsatile behavior in EGFR signaling [52] and a higher Yan staining in a wider number of cells may indicate a longer or more frequent pulse duration for EGFR signaling, and clearly EGFR CA and EGFR DN affect this in opposite directions compared to wildtype. Further, we stained for neuronal markers as a proxy for proper differentiation of the photoreceptors. In wildtype eye discs, Elav is expressed in all PR cells and Prospero, under the control of EGFR, is expressed in R7 cell only [53,54]. This differentiation pattern is captured in CantonS and Elav-Gal4 driving spitz dsRNA discs. UAS-spitz-dsRNA driven by GMR-Gal4 showed a loss of neuronal cells, with very low signal from the Elav antibody. The residual Elav signal could be from the R8 founder photoreceptor neuron whose specification is not under the control of EGFR pathway. GMR-Gal4 driving EGFR CA shows more than one Prospero-positive cell per ommatidial cluster, whereas GMR-Gal4 driving EGFR DN shows complete loss of Prospero-positive R7 cells (Fig 5A). The number of Prospero-positive cells per ommatidial cluster in the above crosses is quantified (Fig 5C). We also stained the discs with EdU to capture the re-entry of G1 arrested cells in the morphogenetic furrow into the second mitotic wave. A band of S-phase cells just beyond the morphogenetic furrow is captured in CantonS discs along with discs from Elav-Gal4 driving spitz dsRNA. All crosses characterized by skewed ratios, showed complete loss of

Fig 5. Abnormal EGF receptor activity disrupts cell differentiation and cell division in the developing eye discs.
Markers for proximate EGFR signaling (Yan), neuronal differentiation (Elav and Prospero) and cell division (EdU) are stained in eye discs from CantonS and discs with altered EGFR activity. (A) Yan intensity (first row) is similar in CantonS and Elav-Gal4 driving spitz dsRNA discs, whereas Yan intensity decreases when GMR-Gal4 drives spitz dsRNA or EGFR DN and increases with EGFR CA . In CantonS eye discs, Elav stains all eight neuronal cells whereas Prospero stains only R7. The differentiation pattern is maintained in discs from Elav-Gal4 driving spitz dsRNA. Prospero-positive cells per cluster increased in discs when EGFR CA is driven by GMR-Gal4 whereas the signal was completely abrogated in discs when spitz dsRNA or EGFR DN was expressed. Eye discs were stained with EdU to visualize the band of S-phase cells posterior to the MF. Discs from CantonS and Elav-Gal4 driving spitz dsRNA showed the S-phase band which was absent when GMR-Gal4 was driving spitz dsRNA, EGFR CA and EGFR DN  S-phase band, implying abnormal EGF receptor activity completely disrupts the second mitotic wave. The number of EdU-positive cells per 100 nuclei after morphogenetic furrow were scored (Fig 5D). Together these markers for cell proliferation, differentiation and EGFR activity could explain the variety of the rough eye phenotype in the above crosses (Fig 5E). The adult eye of wildtype CantonS fly have perfectly patterned ommatidial clusters and bristles cells. Elav-Gal4 driving spitz dsRNA do not have any effect on the adult eye phenotype. Adult eye from GMR-Gal4 driving spitz dsRNA has reduced ommatidial clusters with randomly distributed bristle cells. GMR-Gal4 driving EGFR CA has no ommatidial clusters and drastically reduced number of bristle cells, whereas EGFR DN has smaller eyes with an increased number of bristle cells and complete loss of ommatidial clusters. GMR-Gal4 driving UAS-Argos have non-patterned ommatidial clusters and bristle cells. High EGFR activity contributes to higher photoreceptor cell specification with loss in the expansion of uncommitted cell pool resulting in loss of non-neuronal cell types in the ommatidia like the interommatidial bristle cells (eye phenotype of EGFR CA flies). On the other hand, downregulation of EGFR activity inhibited photoreceptor cell specification leading to predominant bristle differentiation (eye phenotype of EGFR DN flies). This also indicates that beyond the spitz-to-argos ratio, other signaling factors regulate the final eye phenotype. In fact, GMR-Gal4 driving spitz dsRNA and EGFR DN are more similar in terms of Yan intensity and number of prospero-positive cells, while they drive the spitz-to-argos ratio in different directions. And while GMR-Gal4 driving spitz dsRNA and EGFR CA drive the ratio in the same direction, at the molecular level how these affect EGFR signaling is different between these two manipulations. The final adult phenotype are also somewhat different between these two (Fig 5E), though they are more similar than the bristle-rich EGFR DN phenotype. The tight range for wildtype and balancer controls is perhaps indicative of the importance of relative Spitz and Argos doses, but the system is developmentally buffered to thresholds beyond this wildtype range. We never saw mixed phenotypes, and the flies always showed either largely proper ommatidial arrangements or fully penetrant rough eye phenotypes.

Discussion
Developmental pathways have evolved mechanisms to monitor positional information in order to generate reproducible organismal patterns. These pathways are robust and insensitive to small changes in individual processes involved. Spatial differentiation, where a population of cells undergo deterministic molecular differentiation, brings about spatial patterns [55]. Redundancy of mechanism and negative feedback are two ways in which reliability in pattern formation is brought about [56,57]. Lateral inhibition by diffusible molecules is another mechanism that can be used to generate patterns [58]. For systems which do not depend on developmental history, environmental makeup determines their molecular differentiation contributing towards generating a pattern [59].
In this paper, we propose a mechanism where relative expression levels of principal EGFR ligand, Spitz and negative feedback regulator Argos determines the extent of EGFR activation which is crucial for the periodic ommatidial pattern. Our data suggests that it is not the absolute gene expression but the balance between gene networks on the whole which may contribute towards pattern formation. GMR-Gal4 is expressed in all cells posterior to the morphogenetic furrow [48] and any expression cassette under the UAS is expressed strongly. bristles cells in different crosses. Scale bar in all the images is 10μm. The images within the figure panels are created by the authors. Scale bars in Scanning Electron Microscopy images have been provided from knowledge of pixel size and converting appropriately. Asterisks denoting significance of observed changes have been added to relevant graphs. https://doi.org/10.1371/journal.pgen.1010622.g005 Whereas, Elav-Gal4 is expressed only in neuronal cells and the expression is strongest towards the posterior end of the eye disc [49]. While Elav-Gal4 expression occurs in differentiated neurons starting with R8 in the eye disc (for which EGFR signaling is not needed), but beyond R8 specification, Spitz and Argos levels are important for subsequent PR cell differentiation and also in the pupal stages [60]. Although we observe an equally drastic reduction in argos expression when UAS-spitz dsRNA driven by GMR-Gal4 and Elav-Gal4 (Fig 2C), the eye discs show reduced dpERK staining in both cases (S6A Fig). The reduction was lower when Elav-Gal4 driver was used, corresponding to the absence of phenotype in the adult eye (S6B Fig). The eye discs expressing EGFR CA construct under GMR-Gal4 Gal80 ts with spitz-to-argos ratio near 1, showed a discontinuous S-phase band after the morphogenetic furrow indicating a lower population of cells entering the second mitotic wave (Fig 5). Fewer cells entering the second mitotic wave leaves the tissue field with fewer uncommitted cells to make cell fate decisions. This can affect pattern formation to a great extent. This could also explain fewer number of bristle cells in the rough adult eyes (Fig 3) as bristles cell fate is assigned from cells arising from the second mitotic wave [61]. It should also be noted that the rough eye phenotype for EGFR DN is rather different from EGFR CA and shows a profusion of bristles (compare 60 min time-points of Figs 4F and 3F). This mechanism of relative expression determining phenotype supports older work on the importance of spitz-to-argos dose as a critical determinant of eye patterning [18,25]. Indeed mRNA levels may not be always predictive of protein levels, and both translational regulation and post-translational modifications may well affect biological function. However, under conditions of stress [62] or where specific transcriptional programmes bring about developmental outcomes, transcript levels may be thought to be wellcorrelated to protein levels. In addition, experiments using smFISH lets us clearly identify the cells that are expressing specific genes, where the diffusible protein end-products may not provide as conclusive answers. We did attempt performing antibody staining for Spitz and Argos but such relative measures cannot be used to comment on expression stoichiometry (S7 Fig). Our study shows a clear differential expression of spitz and argos mRNA in the early eye field contributing to photoreceptor fate determination and also addresses the sensitivity of the system to the heterogeneity in the expression levels of gene networks and makes developmental programs robust. It has to be noted of course that signaling via EGFR is not the only pathway determining the ommatidial pattern in the eye. For example, Notch is known to play an important role in the initiation of neural development and also in ommatidial rotation [63,64]. Buffered regulation of genes in different developmental pathways that crosstalk can decrease sensitivity to variations in a gene network and can help explain other reproducible and stereotypical patterns generated throughout the development.

Drosophila stocks and crosses
All fly strains were grown on standard cornmeal agar at 25˚C. CantonS line was used as the wildtype fly. The cassettes used for modulating the EGFR pathway were UAS-argos dsRNA [65], UAS-spitz dsRNA [66], UAS-EGFR CA (FBst0305944), UAS-Argos (FBst0005363) and UAS-EGFR DN (FBst0005364). Tissue specific GMR (FBst0300831) and Elav (FBst0008765) drivers were used to express the above constructs. Heatshock to activate GMR-Gal4 and Elav-Gal4 was given to late 2 nd instar larvae overnight in 29˚C incubator. Gal80 ts /FM7 (FBst0007016) was used to generate Gal80 ts ; GMR-Gal4 with a temperature sensitive Gal80. The crosses with Gal80 ts line were set at 18˚C and shifted to new vials every alternate day. Towards the end of the 2 nd instar, the larvae were shifted to 29˚C for particular time points and immediately shifted back to 18˚C and continued to grow. Larvae were collected for dissection around mid-to-end 3 rd instar and processed for smFISH. Adult progeny from these crosses were allowed to emerge at 18˚C to assess the effect on ommatidial patterning through SEM.

Tissue preparation and smFISH
3 rd instar larvae were collected in nuclease-free 1X PBS (Ambion, AM9624) and washed once with the same. The larvae were flipped and transferred to a microcentrifuge tube containing 4% Paraformaldehyde (PFA, Sigma, P6148) in 1X PBS for 25 minutes at room temperature. The fixative was aspirated and washed twice with 1X PBS. The tissue was permeabilized using 0.3% Triton-X 100 (Sigma, T8787) in 1X PBS for 45 minutes at room temperature. The permeabilizing agent was aspirated and the tissues were washed twice with 1X PBS. The tissues were kept in 70% ethanol at 4˚C overnight. The tissues were washed twice with a wash buffer (20% Formamide (Ambion, 9342) and 2X SSC (Ambion, AM9763) in nuclease-free water) for 30 minutes each at 37˚C. The wash buffer was aspirated and the tissues were incubated with a hybridization mix overnight at 37˚C. The probe sequences for spitz and argos is given in the previous paper [40]. The mix was removed and tissues were washed twice with wash buffer at 37˚C. DAPI (Invitrogen, D1306; 2μg/ml) in wash buffer was added to the tissues and incubated for 30 minutes at 37˚C. The tissues were then washed with 2X SSC twice for 5 minutes. The eye imaginal discs were dissected in 2X SSC and mounted on a clean slide with a drop of Vectashield (Vector Labs). A cartoon depicting the brief protocol and singly labeled 20-mer oligonucleotide probes binding to mRNA target is shown in S1A Fig. StarSearch software (https://www.seas.upenn.edu/~rajlab/StarSearch/launch.html) was used to count transcripts in 3D as done before [40]. The absolute counts and ratios were calculated in regions of interest on the eye field and then averaged.

Antibody staining
The tissues were fixed and permeabilized as described earlier. The tissues were then washed twice with 1X PBS. Blocking solution (5% BSA in 1X PBS) was added to the tissues and incubated for 1 h at room temperature on a rotating mixer. For simultaneous smFISH-IF, the primary antibody is added into the hybridization mix and incubated overnight at 37˚C. The secondary antibody was diluted to 1:1000 in nucleasefree blocking solution and incubated with tissues for 3 hours at room temperature on a rotating mixer.

Image acquisition
The tissues were imaged on an Olympus BX63 upright widefield fluorescence microscope with a Retiga 6000 (Qimaging) CCD monochrome camera. The slides were kept at 4˚C for 1 hour at least before imaging. The images were acquired using a 60X, 1.42 N.A. oil immersion objective with a z-step size of 0.3μm. Narrow band-pass filters (ChromaTechnology-49309 ET-Orange#2 FISH for Quasar 570 labelled probes; 49310 ET-Red#2 FISH for CAL fluor 610 labelled probes) were used to spectrally separate single transcripts imaged in two colors.

Analysis of mRNA counts in the 3 rd instar eye disc
The smFISH protocol works perfectly in the wholemount tissues (~40μm). 30-40 z-slices corresponding to the photoreceptor focal planes (based on nuclear arrangement of DAPI-stained PR cells) are separated from the raw image stack and used for representation and analysis unless otherwise mentioned. For representation, the stacks were z-projected, merged wherever necessary. For mRNA counts, StarSearch software was used as mentioned earlier. The ROIs were drawn free-hand, area roughly in the order of 30000-35000 pixels. The counts are analyzed in 3D stack and converted to volume as described in our previous paper [40]. For counts along the A-P axis of the tissue, ROIs were drawn from the morphogenetic furrow to the posterior end. Equal number of columns were analyzed and represented in the all the analysis and henceforth narrow-window staging of larvae was not necessary. All numerical data underlying the graphs are provided in the S1 Data.xlsx file. GraphPad (https://www.graphpad.com/quickcalcs/ttest1/) was used to calculate p-values in a Student's t-test for statistical significance.

SEM imaging of adult eyes
Adult fly heads were dissected in 1X PBS. PBS was aspirated and 4%PFA was added for 25 minutes at room temperature. Fly heads were then washed twice with 1X PBS for 5 minutes each. Conducting carbon adhesive tape was spread on the SEM specimen stub. The fly heads were aligned on the carbon tape and were air-dried. The images were captured using a lower electron detector in a JEOL JSM7200F microscope at 170X magnification. dpERK along with Elav is stained in eye discs from CantonS, Elav-Gal4 driving spitz dsRNA and GMR-Gal4 driving spitz dsRNA. Unlike smFISH which affords absolute mRNA counts, immunofluorescence is a relative measure and it is more difficult to compare across experiments in different strains. But under identical staining and imaging conditions, both the crosses seemed to show significantly reduced dpERK staining compared to CantonS. The reduction was lower for the Elav driver, corresponding to the absence of a phenotype in the adult eye. Residual dpERK signal is observed around the Elav-positive neuronal cells with Elav-Gal4 driving spitz dsRNA. In GMR-Gal4 driving spitz dsRNA, patterned Elav and dpERK staining is lost. (B) Mean Intensities of dpERK in a fixed ROI is plotted here clearly showing the decreased dpERK staining when Elav or GMR-Gal4 drives spitz dsRNA. ( ��� indicates pvalues < 0.001 in a Student's t-test). N = 6 in all genotypes. Error bars are standard errors of mean. Scale bar is 10μm. The images within the figure panels are created by the authors. Asterisks denoting significance of observed changes have been added to relevant graphs. Scale bar is 10μm. Since cleaved Spitz and Argos are diffusible molecules, the staining does not recapitulate the spitz and argos mRNA expression patterns exactly. Moreover, unlike single molecule FISH (smFISH) for RNA that yields absolute mRNA counts, immunofluorescence experiments show only relative changes. The signal intensities captured are dependent on the affinity of the antibodies, quantum yield of the fluorophores used in secondary detection, image acquisition parameters etc. and are different for the two antibodies used. Thus while for a given antibody relative changes can be followed, there is no easy way of comparing staining intensities for antibodies directed against different antigens. Therefore such staining cannot be used for determining expression ratios of Spitz and Argos, unlike with smFISH, which both clearly marks the source cells and also yields absolute transcript counts. To our mind this provides two clear examples of the power of smFISH. The images within the figure panels are created by the authors. (TIFF) S1 Data. All numerical data underlying the graphs are provided in the S1 Data.xlsx file (XLSX)