Detecting New Allies: Modifier Screen Identifies a Genetic Interaction Between Imaginal disc growth factor 3 and combover, a Rho-kinase Substrate, During Dorsal Appendage Tube Formation in Drosophila

Biological tube formation underlies organ development and, when disrupted, can cause severe birth defects. To investigate the genetic basis of tubulogenesis, we study the formation of Drosophila melanogaster eggshell structures, called dorsal appendages, which are produced by epithelial tubes. Previously we found that precise levels of Drosophila Chitinase-Like Proteins (CLPs), encoded by the Imaginal disc growth factor (Idgf) gene family, are needed to regulate dorsal-appendage tube closure and tube migration. To identify factors that act in the Idgf pathway, we developed a genetic modifier screen based on the finding that overexpressing Idgf3 causes dorsal appendage defects with ∼50% frequency. Using a library of partially overlapping heterozygous deficiencies, we scanned chromosome 3L and found regions that enhanced or suppressed the Idgf3-overexpression phenotype. Using smaller deletions, RNAi, and mutant alleles, we further mapped five regions and refined the interactions to 58 candidate genes. Importantly, mutant alleles identified combover (cmb), a substrate of Rho-kinase (Rok) and a component of the Planar Cell Polarity (PCP) pathway, as an Idgf3-interacting gene: loss of function enhanced while gain of function suppressed the dorsal appendage defects. Since PCP drives cell intercalation in other systems, we asked if cmb/+ affected cell intercalation in our model, but we found no evidence of its involvement in this step. Instead, we found that loss of cmb dominantly enhanced tube defects associated with Idgf3 overexpression by expanding the apical area of dorsal appendage cells. Apical surface area determines tube volume and shape; in this way, Idgf3 and cmb regulate tube morphology.

identify genetic players that might be involved in tube formation during human development (Copp and Greene 2010;Wang et al. 2019), limited accessibility to the process impedes studies to understand how these genes influence morphogenesis. An important question that remains in the field is exactly how those, and other unknown genetic programs, drive tube formation at a cellular level. Studies in the Drosophila system could help us overcome these challenges.
The Drosophila laid egg and the egg chamber that produces it serve as great models to study tube development (Berg 2005;Osterfield et al. 2017). The laid egg has two eggshell structures, known as dorsal appendages (DAs), which facilitate gas exchange for the developing embryo by retaining pockets of air within the chorion (Hinton 1969). The dorsal appendages develop through the formation of two biological tubes that use the wrapping mechanism of tube formation (Dorman et al. 2004) (Figure 1). Although the dorsal appendages are not tubes themselves, cellular tubes mold their physical form (Dorman et al. 2004;Berg 2005). Therefore, by looking at the shape of the dorsal appendages on a laid egg, we can assess mistakes during the process of tube formation.
Dorsal appendage formation takes place before the egg is laid, while the egg chamber develops in the ovaries of the adult female ( Figure S1A). The egg chamber consists of 16 germ-line cells (a single oocyte connected to 15 sibling nurse cells) surrounded by a monolayered epithelium of somatic follicle cells ( Figure S1B). Within the ovary, the egg chambers are organized in assembly lines, ranging from the youngest, stage-1 (S1) egg chambers, at the anterior, to mature, stage-14 (S14) egg chambers, at the posterior. This organization and the presence of multiple egg chambers within a single female allow us to simultaneously observe tube formation at different stages of development (King 1970). By dissecting fly ovaries and fixing egg chambers, we can capture stationary processes of otherwise fastmoving events of morphogenesis (Hudson and Cooley 2014;Peters and Berg 2016A).
The morphological changes that form the dorsal-appendage tubes begin at S10B of egg chamber development ( Figure 1A and Figure  S1B). At this stage, the nurse cells occupy the anterior half of the egg chamber and the oocyte occupies the posterior half. The follicle cells that surround the nurse cells have become so thin and squamous they are called "stretch cells", while the follicle cells over the oocyte are columnar in shape. By S10B, among the columnar follicle cells, two patches of cells are differentially programmed to make the two tubes that mold the dorsal appendages ( Figure 1A). Each patch is made of 11 floor cells and 52 roof cells (Dorman et al. 2004), and these sub-populations will become the floor and the arched roof of the DA tube, respectively ( Figure 1A, arrows). From S10B to S14, these cells undergo cell-shape changes, cell intercalation, and cell migration in order to form two mature tubes (Dorman et al. 2004;Osterfield et al. 2013). While the tubes are forming, the follicle cells release chorion proteins into the tube lumen; these chorion proteins are later crosslinked to create an oar-shaped structure with a rounded stalk and a flat paddle ( Figure 1A and 1B). Once this morphogenetic process is finished, the follicle cells undergo apoptosis and detach from the egg (Nezis et al. 2002). Thus, the final shape of the dorsal appendages reflects the process of tube formation ( Figure 1B).
We have used the dorsal appendages model to identify and characterize genes involved in tube formation (Berg 2005). For example, we identified bullwinkle (bwk), which when mutated results in wide and short dorsal appendages resembling moose antlers [hence the name] ( Figure 1C; Rittenhouse and Berg 1995). In bwk egg chambers, the DA-forming tubes don't seal properly, and dorsal-appendage-making cells migrate more laterally than anteriorly (Dorman et al. 2004). Interestingly, bwk acts outside the dorsal-appendage-making cells, in the nurse cells, by regulating expression of genes that signal to the stretch cells (Rittenhouse and Berg 1995). The stretch cells act as mediators to then communicate with the dorsal-appendage-making cells, facilitating the proper formation and elongation of the DA tubes (Tran and Berg 2003). The signals involved in each of these processes are not completely understood.
To understand the molecular landscape of the stretch cells and how it drives dorsal appendage formation, we recently used proteomic analysis to evaluate stretch cells purified from wild-type and bwk egg chambers; we discovered that bwk mutants vastly over express a novel family of growth factors, the Imaginal disc growth factors (Idgfs). Lowering the expression of Idgf4, Idgf5, or Idgf6 ameliorates the bwk mutant phenotype, suggesting they are part of the bwk signaling pathway that regulates dorsal appendage formation. Up-regulating only a single member of this family, e.g., Idgf3, is sufficient to produce a phenotype similar to the bwk mutant ( Figure  1D) (Zimmerman et al. 2017).
There is limited knowledge about the Idgfs. The first Idgfs were identified from conditioned medium and shown to act as growth factors, playing roles in cell-shape changes, cell proliferation, and cell migration in cell lines cultured in vitro (Kirkpatrick et al. 1995;Kawamura et al. 1999). Supporting the hypothesis that Idgfs might influence cell behaviors, transcripts from Idgf1, Idgf2, Idgf3, Idgf4, and Idgf6 accumulate in sites of the embryo where major morphogenetic changes occur, such as the ventral furrow and midgut invaginations (Kawamura et al. 1999;Jambor et al. 2015). The Idgfs encode proteins containing a signal-peptide domain (Zhu et al. 2008) and a mutationbearing-chitinase catalytic domain (Varela et al. 2002), suggesting they are secreted molecules that evolved from chitinases but lack the ability to break down chitin. Their receptors and signaling pathway are not known.
In addition to helping us comprehend biological tube formation, elucidating the mechanism of action of the Idgfs could give insight into other human diseases since the human orthologs of the Idgfs, the Chitinase-Like Proteins (CLP's) (Zhu et al. 2008), are up-regulated in immune diseases (reviewed in Ober and Chupp 2009), in numerous cancers (reviewed in Libreros et al. 2013), and during infections (Erdman et al. 2014). In spite of their well-known association with these diseases, relatively little is known about their mechanism of action. Indeed, it is not clear whether the CLPs are pathogenic, protective, or both, depending on circumstances. Characterizing the cellular mechanisms of Idgfs in tube formation could facilitate studies of human CLPs by providing testable hypotheses on how chitinase-like proteins could be acting in these contexts.
The main goal of this study was to increase our understanding of the Idgfs and their role in tube formation by identifying a genetic pathway that interacts with the Idgfs during dorsal appendage formation. We designed an unbiased screen to uncover genes that suppress or enhance the DA defects produced by over-expressing Idgf3. We identified large regions of chromosome 3L that, when removed by half, showed a possible genetic interaction with Idgf3 for tube morphogenesis. Using the same approach with smaller, overlapping deletions, we narrowed down a subset of those possible interacting regions to a few candidate genes. Using RNAi lines and mutant alleles, we discovered a genetic interaction between Idgf3 and combover (cmb), a Rho-kinase substrate that physically interacts with a Planar Cell Polarity (PCP) pathway component (Fagan et al. 2014). Through immunostaining, we go on to show how this interaction influences tube formation at a cellular level. In this way, we report the identification and first cellular characterization of an Idgf-interacting gene. Additionally, we identify other potential Idgf3-interacting candidates, some of which might also interact in the cmb-Idgf3 pathway.

MATERIALS AND METHODS
Fly stocks used w 1118 and the other stocks used in this work are available upon request. w 1118 ; CY2-GAL4 (Queenan et al. 1997) was provided by Trudi Schüpbach and was used in lieu of a wild-type strain. w 1118 ; UAS-Idgf3/TM3,Sb was obtained from the Bloomington Stock Center (BL# 52658). Using these strains, we created the stock w 1118 ; CY2-GAL4; UAS-Idgf3/TM3, Sb. All deficiency lines were provided by the Bloomington Stock Center (Tables 1 and 2). UAS-RNAi lines were obtained from the Bloomington or Vienna Stock Centers (Table 2, Dietzl et al. 2007;Ni et al. 2011). The combover loss-of-function (LOF) strain, w 1118 ; +/+; cmb KO /TM6B, Hu, and an overexpression allele w 1118 P{UAS-cmb-RB} (Table 2) were generously donated by Andreas Jenny's laboratory (Fagan et al. 2014). The cmb KO LOF allele is null for both of the Cmb protein isoforms due to the deletion of an 1-kb fragment early in the coding region and its replacement with a white + marker. The overexpression allele used in this paper produces only the smaller of the two Cmb isoforms, Cmb-PB.

Modifier-screen crosses
Six-to-ten virgin females from the Idgf3-overexpression stock were crossed to four males of each stock from the deficiency kit, or to males from overlapping deletion lines, or to RNAi or mutant-allele strains. For the modifier screen, crosses were done at 25°, while for narrowing down regions, crosses were performed at 22°. From each cross, at least nine but usually 25 F1 females were used for the egg collection assay ( Figure S2).

Dorsal appendage analyses
One-day-old to four-day-old females of the desired genotypes were mated to males in nutrient rich vials at 25°and then transferred to 30°t o optimize the activity of our CY2-GAL4 driver. GAL4 encodes a yeast transcription factor, and past work in the lab showed that 30°p roduces strong GAL4 expression, as measured by a GFP reporter, without visibly affecting egg chamber development (Peters and Berg 2016B). Flies were transferred every day to fresh nutrient-rich vials for three days. On day four, flies were transferred to collection tubes that contained apple juice agar plates with fresh yeast where flies laid eggs. On day five, laid eggs were collected, alternately rinsed with water and embryo wash (0.7% NaCl, 0.05% Triton X-100), mounted on slides in 70 uL of Hoyer's medium (van der Meer 1977), and incubated overnight at 65°. Dorsal appendages were scored by using dark-field optics on a Nikon Labophot microscope at 10X magnification; n . 100 except in rare instances as specified in Table 1 and Tables S1 and S2. We grouped dorsal appendage phenotypes into three categories ( Figure 2A, Figure 2A' and Figure S3). Eggs with DAs closely resembling wild type were classified as Normal/Mild: the dorsal appendages were positioned just lateral to the dorsal midline, extended anteriorly 30% of egg length, and exhibited an oar-like shape with paddles that occupied about half the length of the entire dorsal appendage. These DAs had smooth edges. We classified eggs as Moderate when two of the wild-type features looked mildly defective, such as slightly shorter DAs with wavy paddles. Some eggs exhibited stronger defects, such as DAs that were triangular in shape. For others, the DAs were of normal size, but there was no clear separation of the paddles with the bases, as if the paddles were missing. In some instances, the edges of the dorsal appendages looked jagged or serrated. This category also contained proportionally normal-looking dorsal appendages but of increased size relative to the entire egg, which was of normal size. We scored eggs as Severe when the DAs were short (half the normal length) and wide, or when they exhibited defects in three or more of the normal features. In some eggs, the two dorsal appendages were linked by chorion protein in between them. In other instances, there was a small quantity of chorion protein extending out of the egg, but there was not a specific shape to the DA. In other eggs, the DAs were merged completely at their bases, or chorion protein accumulated on the dorsal side of the egg instead of forming DAs.

Chromosome annotation
The chromosome 3L map shown in Figure 2C was produced with the R package chromoMap (Anand 2019) by mapping the effect that each deficiency had on dorsal appendage phenotypes when Idgf3 is Figure 1 Dorsal appendage formation is a good model to study biological tube formation. (A) Schematic drawings of two later stages of oogenesis, S10B and S14, depicting the cells that make up the egg chamber. On the left, arrows indicate the cells programmed to build the dorsal appendages in red (floor cells) and blue (roof cells). On the right, an arrow points to the tube lumen, defined by a dotted line. Drawings obtained with permission from Dorman overexpressed, onto the breakpoint coordinates of the deficiencies. The gradation of color (red enhancer to blue suppressor) was calculated by subtracting the percentage of eggs with normal dorsal appendages for the appropriate control from the percentage of eggs with normal dorsal appendages of each specific deficiency.

Immunostaining
On the third day at 30°, F1 females from the desired genotypes were anesthetized on a CO 2 pad and their ovaries dissected. To limit variability between samples, dissections were performed simultaneously by three people in the lab and completed within fifteen minutes. Dissected ovaries were placed in phosphate-buffered saline (PBS) on ice and fixed in 4% EM-grade formaldehyde [Thermo Fisher Scientific,Catalog# 43368] in PBS with 0.1% Tween 20 for twenty minutes. Ovaries were then washed three times in PBS with 0.1% Tween 20. To ensure even staining, single egg chambers of stages S10B and S12 were then dissected out, permeabilized with 1% Triton X-100 in PBS followed by three washes in PBS with 0.1% Tween 20. Eggs were then blocked in 10% Western Blocking Reagent (WBR, Roche) in PBS with 0.1% Tween 20 and incubated with gentle shaking overnight at 4°with mouse anti-Broad-core (1:250 uL, 25E9.D7 concentrate, Developmental Studies Hybridoma Bank, DHSB; Oda et al. 1994) and rat anti-Ecadherin (1:50 uL, DCAD2-concentrate, DHSB; Dubreuil et al. 1987). Egg chambers were then washed four times in PBS with 0.1% Tween 20 and 10% WBR and incubated for three hours at room temperature with Alexafluor 488-conjugated goat anti-mouse (1:200), Alexafluor 568-conjugated goat anti-rat (1:200), 4', 6-Diamidino-2-phenylindole (DAPI) (1 ug/mL) in PBS with 0.1% Tween 20 and 10% WBR. The egg chambers were then washed three times in PBS with 0.1% Tween 20 and 10% WBR, once in PBS 0.1% Tween 20, and mounted in Aqua polymount [Polysciences, Catalog# 18606] for imaging.

Confocal image acquisition
Imaging and scoring of egg chambers were done blind by covering the genotype labels with tape and assigning letters A, B, C, and D. Their names were revealed after all data analyses. We used the Leica SP8X confocal microscope, with the 20X objective and then with a zoom of 2X focusing on the DA-forming patches. Wavelength emissions 488nm, 461nm, and 568nm were used with a PMT detector at 30%, 5% and 30% intensity power, respectively. The format of the acquired images was 1024 · 1024 at a speed of 600, with Z slices separated by 0.25 mm. We captured images moving basally (facing Figure 2 Modifier screen identifies regions on chromosome 3L that enhance or suppress the dorsal-appendage defects due to overexpressing Idgf3. (A) Representative images of dorsal appendage phenotypes observed in the modifier screen, grouped and color coded into normal/mild (gray), moderately defective (black), or severely defective (red). Images are oriented with anterior facing left and dorsal facing out of the page. Dotted line shows the midline of the egg chamber. Scale bar is 100 microns and applies to all the pictures. The DAs on the egg shown for the normal/mild category touch at their bases, i.e., the egg is slightly ventralized. We scored this phenotype as a "mild" defect. (A') Schematic diagrams outlining the DAs from a normal egg chamber (left) and the defective DAs shown in panel A (three schematics on the right). Orange lines on the normal DAs show the expected sizes of the paddle, base, and spacing between the two dorsal appendages. Orange arrows point to the defective features portrayed by each egg chamber. (B) Graph shows the proportions of eggs with normal/mild, moderately defective, and severely defective dorsal appendages produced by females of genotypes: w 1118 ; CY2-GAL4/+; +/+ (right) and w 1118 ; CY2-GAL4/+; UAS-Idgf3/+ (left). (C) Chromosome 3L, color coded to show regions that modified the DA defects resulting from Idgf3 overexpression. Key to the color code is on the right. Shading represents deletions that produced higher percentages of eggs that were defective (red) or normal (blue) compared to over-expressing Idgf3 on its own; the colors do not represent p-values but rather, they denote a calculated score (see methods). (D) log 10 of Chi-squared p values calculated from comparing the dorsal-appendage phenotypes between w 1118 ; CY2-GAL4/+;UAS-Idgf3/+ (control) and w 1118 CY2-GAL4/+;UAS-Idgf3/Df(3L) (tested). Significance threshold (y= -3.1) corresponds to the log 10 of 0.05, adjusted for multiple sampling using the Bonferroni correction. Filled circles indicate the deletion lines that we chose for further analysis, in magenta (enhancers) and in green (suppressors). Arrows indicate strong and weak interactions. outward in this tissue) to apically (facing the oocyte), starting from where the Br-positive nuclei were visible and ending below the apical E-cadherin staining, just after reaching the oocyte's cytoplasm.

Image analysis
Images were processed using ImageJ Version 2.0.0-rc-59/1.51n (FIJI) (Schindelin et al. 2012). To identify DA patches, we created a Z projection using all images captured in the Br-positive channel for each egg chamber. To use a standardized method to identify the patches and to uniformly distinguish high-Br staining (DA cells) from low-Br staining (posterior and lateral cells) among different S10B egg chambers, the images were smoothed and made binary using the method Max Entropy or Momentum. We only used egg chambers in which the entire DA patch was visible, that is, those egg chambers mounted with a dorsal or partially dorsal view of the patches. To measure aspect ratios, we traced the exterior boundary of the high-Br cells to create a shape that enclosed the entire patch. Measurements were set up to calculate shape descriptors: aspect ratio, circularity, and roundedness of the basal side of the cells. Although we did not quantify Br signal as a means of defining the patch and determining its aspect ratio, we did image all genotypes of each of the three separate experiments using the same confocal settings. For apical surface measurements, we used single slices in the E-cadherin channel at the apical-most region of the tube. We calculated length by tracing and recording a straight line from the base of the dorsal appendage tube to its tip.

Statistical analysis
For the modifier screen, we used a Chi-squared test for consistency (Kanji 2006) to compare control and test samples. Although we categorized DAs into three groups (normal/mild, moderate, or severe), some samples lacked sufficient numbers of severe eggs to conduct the test properly; we therefore combined the moderate and severe groups into a single "defective" class for all comparisons. We used the R statistical package (R Core Team 2017) to generate a list of p-values using the function chisq.test() with one degree of freedom. We calculated a threshold of significance using the Bonferroni correction test by dividing the 0.05 significance value by 72 (the number of samples we compared), resulting in a 6.9 X 10 24 cutoff. The code used to carry out these Chi-squared tests is available as supplementary material. To calculate p-values for image analyses in Figure 6, the R function for two-sided, unpaired, corrected t-tests, t.test(), was used with a confidence interval of 95% (R Core Team 2020; Welch 1947).

Data availability
Fly strains are available upon request. Supplemental materials include Figure S1, which provides a schematic overview of the fly ovary and the expression patterns of the GAL4 drivers used in this work. Figure  S2 shows the crosses used for the modifier screen. Figure S3 provides examples of the types of dorsal appendage defects observed in the screen. All the raw data necessary for confirming the conclusions of the modifier screen are available in Tables S1 and S2, and the R code used for Chi-squared tests is available in Supplemental file MSTableS1_Code. Confocal Z stacks used for image analysis of S10B and S12 egg chambers are available upon request. Supplemental material available at figshare: https://doi.org/10.25387/g3.12478892

Modifier-screen set up
Previous studies from our lab, using the GAL4-UAS system, demonstrated that overexpressing Idgf3 in the stretch cells of the egg chamber causes dorsal appendage defects about 50% of the time (Zimmerman et al. 2017). We used this knowledge to identify possible genetic interactions with Idgf3 by screening regions of the genome that, when reduced to a dosage half that of wild type, could suppress or enhance the frequency of dorsal appendage defects.
We chose chromosome 3L to scan for modifiers of the Idgf3overexpression phenotype because we have limited knowledge of 3L genes that might be involved in dorsal appendage formation (Tran and Berg 2003;Berg 2005;Boyle et al. 2010). We used the deficiency kit available at the Bloomington Stock Center (Cook et al. 2010) and identified 72 lines that were the minimum number of stocks that uncover virtually all of 3L, a chromosome arm that comprises one fifth of the genome (Table 1). All the lines have defined breakpoints and share the same genetic background (Parks et al. 2004;Ryder et al. 2007). Each deletion uncovers on average 58 genes, with a range of six to 159 genes (Table 1).
We planned to use UAS-RNAi constructs to test individual genes within possible interacting regions, but since we did not know which cells of the egg chamber respond to the Idgf3 signal, we needed a GAL4 driver that would express in all follicle cells. We created a stock that uses CY2-GAL4 (Queenan et al. 1997) to drive expression of UAS-Idgf3 in all follicle cells from stage 6 onward ( Figure S2B).
We examined the dorsal-appendage phenotype of this newly created stock. Flies that expressed GAL4 alone produced eggs with DA defects 4% of the time, while flies that overexpressed Idgf3 produced eggs with DA defects 29% of the time ( Figure 2B). The reduced frequency of DA defects observed with overexpression of Idgf3 using CY2-GAL4 compared with c415, a stretch-cell-specific GAL4 driver (50%, Zimmerman et al. 2017), could be due to spatial, temporal, or quantitative differences in GAL4 expression ( Figure  S1C). For example, overexpression of Idgf3 from the stretch cells could create a signal concentration gradient, while overexpressing Idgf3 from all the follicle cells could create a uniform signal cloud, exposing the DA-making cells to different amounts of Idgf3. Alternatively, early overexpression of Idgf3 using CY2-GAL4 might cause the activation of pathways that counteract the effects of Idgf3 overexpression, pathways that are ineffective if activated at S10.
We considered creating a strain that carried both the c415-GAL4 stretch-cell driver as well as the CY2-GAL4 all-follicle-cell driver to increase the percentage of Idgf3-overexpression defects. The CY2-GAL4 driver alone, however, created a sensitized background sufficient for our screening approach. Although more investigation is needed to discover the mechanism that produces different frequencies of DA defects, the 29% frequency produced by CY2-GAL4 allowed us to screen for modifiers of the Idgf3-overexpression phenotype.
Subsequently, we observed that maintaining the Idgf3-overexpression stock at 25°caused a gradual decline in the frequency of defective DAs over time. Since the CY2-GAL4 driver is moderately active at 25° (Queenan et al. 1997 ; Table S2), we hypothesized that this activity is enough to cause the accumulation of Idgf3-overexpressionsuppressing mutations. To remedy this problem, we rebuilt the CY2-GAL4 -. UAS-Idgf3 strain and maintained it at a lower temperature, 22°. This new strain (when shifted to 30°as described in the methods), continues to produce eggs with defective DA phenotypes at 29% frequency.

Screening for suppressors and enhancers
We crossed females from the Idgf3-overexpression stock with males of each of the 72 deficiency lines or with males of a control stock, w 1118 . By choosing non-balancer flies, we obtained females that n■ overexpressed Idgf3 while also being heterozygous for a deficiency uncovering one part of chromosome 3L ( Figure S2B); crosses to the control w 1118 males produced females that only overexpressed Idgf3 ( Figure S2A). We shifted the flies to 30°to optimize Gal4p activity, then collected eggs and scored DA defects from each cross.
To determine whether each 3L deletion removed genes that interacted with Idgf3, we calculated the frequency of dorsal appendage defects on eggs laid by the Idgf3-overexpressing-deletion-heterozygous females and compared that value with the frequency of DA defects on eggs from females overexpressing Idgf3 alone. We calculated p-values as an indicator of strength of interaction ( Figure 2C) and drew a threshold of significance at 6.9 · 10 24 after correcting for multiple testing (see methods). To our surprise, we found that 46% of the deficiencies significantly modified the Idgf3-overexpression phenotype: 38% of the deficiencies enhanced and 8% of the deficiencies suppressed (Table 1).
For this part of the screen, we did not test if each deficiency produced a dorsal-appendage phenotype on its own because we wanted to quickly identify possible interacting deficiencies. Therefore, some deletions might have had a dorsal-appendage phenotype independent of Idgf3 overexpression. If so, we over-estimated the number of enhancers in this first stage of analysis.
We considered other factors that could contribute to our data being skewed toward enhancers. One possibility is that the Idgfs are dosage sensitive for dorsal appendage formation: both down regulation and up regulation of each Idgf cause dorsal appendage defects (Zimmerman et al. 2017). Therefore, genes that play a role in either type of regulation could perturb this balance and enhance the frequency of dorsal appendage defects. Also, since each deficiency reduces the copy number of many genes at one time, these large deletions could cause a cumulative effect if more than one cellular pathway is affected. Finally, the dorsal appendages are not critical structures of the fly; we therefore do not expect to find internal mechanisms that ensure proper DA development under such genetic alterations. Nonetheless, in this large screen, we identified several potential sites on chromosome 3L that genetically interact with Idgf3.

Selecting regions to narrow down
We used three criteria to pick deficiencies for further analyses. First, we wanted to identify both suppressors and enhancers of the Idgf3overexpression phenotype since these opposite phenotypes could reveal opposing inputs into how Idgfs are regulated or received. Second, we wanted to identify candidate genes that through their interaction with Idgf3 would reveal critical processes for tube formation. In other words, removing only one copy of the gene while overexpressing Idgf3 would disrupt tube formation drastically. Those candidate genes would be uncovered by the deficiencies with small p-values ( Figure 2C, strong intensity). Alternatively, and our third criteria, we were interested in identifying important developmental pathways in which the Idgfs might play a role but that have back-up mechanisms to ensure robust function. In other words, removing one copy of a gene will produce only a mild effect on the Idgf3-overexpression phenotype because redundant genes or genes in parallel pathways could make up for the reduction of the gene product. Those genes would be uncovered by deficiencies with p-values near the threshold of significance ( Figure 2C, weak intensity).

Narrowing down the interacting regions
We narrowed down the regions of interaction by using overlapping deficiencies. First, we took advantage of a tiling effect obtained from the original screen. 72% of the deficiencies from the primary screen partially overlapped with other deletions adjacent to their ends (Table  1, overlapping coordinates). Additionally, we obtained smaller, partially overlapping deficiencies from the Bloomington Stock Center, deficiencies that were produced using the same technologies as those used in the original modifier screen (See examples in Figure 3A and 3B).
Our strongest enhancer, Df(3L)ED4674, overlapped with another deficiency from the original modifier screen, a deletion that suppressed the Idgf3-overexpression phenotype ( Figure 3A). This result eliminated those genes in the proximal region of Df(3L)ED4674. We tested six additional overlapping deficiencies in the region and found two that strongly enhanced the Idgf3-overexpression phenotype. Although the Df(3L)ED4674 and Df(3L)Exel6130 increased the frequency of DA defects in the Idgf3-overexpression background, they did not result in a phenotype when Idgf3 levels were normal (Table  S2), demonstrating that the defects depended on overexpressing Idgf3. By assessing the regions of overlap among these deletions, we ascertained the presence of an Idgf3-interacting gene among five candidate genes.
When investigating Df(3L)ED4502, we found one overlapping deficiency, Df(3L)BSC614, that enhanced the Idgf3-overexpression phenotype, confirming a genetic interaction with Idgf3 and narrowing down the region to 51 candidate genes ( Figure 3B). To refine the interaction further, we tested two additional deficiencies available in the region and found they did not enhance the Idgf3-overexpression phenotype. One of those deficiencies, Df(3L)Exel6119, resulted in a high number of dorsal appendage defects when Idgf3 levels were normal (Table S2), suggesting that a gene in that region plays a role in n■ dorsal appendage formation but does not interact with the Idgf3 pathway. Since this small deficiency is contained within the original deficiency Df(3L)ED4502, it explains why Df(3L)ED4502 also results in dorsal appendage defects when Idgf3 is at normal levels ( Figure 4, lane 3). In contrast to Df(3L)Exel6119, however, Df(3L)ED4502 interacted synergistically with Idgf3. We concluded that the genes uncovered by Df(3L)BSC614, but that fall outside Df(3L)Exel6119, were responsible for the enhancing effect seen with the large deficiency Df(3L)ED4502 in an Idgf3-overexpression background ( Figure 4, lane 4). In total, our mapping of this region suggested 37 candidate genes.

Identifying interacting genes
To identify the actual gene responsible for the interaction with Idgf3, we used UAS-RNAi constructs provided by the Bloomington or Vienna Stock Centers. Using the same DA assay, we tested those alleles for interaction with Idgf3. When evaluating the UAS-RNAi lines, we kept in mind that the amount of reduction of the gene product would likely differ from the original screen, in part because knockdown depends on the amount of Gal4p protein, which also drives expression of Idgf3. Since we were reducing transcripts only in the follicle cells, and the germ line produces a large amount of RNA to be loaded maternally into the embryo, we could not assess the level of transcript reduction easily. Furthermore, protein products could perdure even if transcripts were completely degraded. We therefore interpreted a lack of phenotype from RNAi knockdown with caution, and we used null alleles when available, as nulls more closely mimic the conditions of the original deletion screen.
We narrowed down the region uncovered by Df(3L)ED4674, our strongest enhancer, to five candidate genes ( Figure 3A, Table 2). RNAi alleles against four candidates, CG9705, CG9706, CG9674, and nudC, did not modify the Idgf3-overexpression phenotype. When we attempted to knock down expression of the fifth gene, eIF3e, by crossing a UAS-RNAi construct to either the Idgf3-overexpression stock or CY2-GAL4 alone, pupae did not eclose, and we were unable to obtain adult females to test their eggs for dorsal appendage defects. We also failed to obtain adults when using a temperature-sensitive gal80 with CY2-GAL4. Unfortunately, no mutant alleles exist. Our observations suggest that eIF3e, which encodes a translation initiation factor, is an important developmental gene in Drosophila. Consistent with this hypothesis, the mammalian eIF3e is essential for mouse embryonic development (Sadato et al. 2018). As more null alleles become available, we will be able to test eIF3e, and other individual genes, for interaction with Idgf3 with more confidence.

Idgf3 genetically interacts with combover
After refining the interacting region for Df(3L)ED4502, one of our modest enhancers, 37 genes remained to be tested ( Figure 3B, Table 2). We used a candidate gene approach and tested alleles of three genes whose function is relevant for development (Table 2). We identified a mild genetic interaction with a combover (cmb) RNAi line, which enhanced the Idgf3-overexpression phenotype. We tested a loss-of-function mutant and an overexpression line for this gene (Fagan et al. 2014) and found complementary effects: +/cmb KO enhanced the Idgf3-overexpression phenotype, and the overexpression of cmb-RB suppressed the Idgf3-overexpression phenotype in three separate replicas (Figure 4). Figure 3 Overlapping deficiencies across chromosome 3L were used to narrow down regions that interact with Idgf3. (A, B) Genome browser maps showing two examples of interacting regions and the deletions we used to refine the list of potential interacting genes. Each block indicates the coding and non-coding regions of a gene. Unfilled blocks are genes whose interaction was neither confirmed nor discarded; black blocks are genes that failed to interact with Idgf3; red or pink blocks indicate potential candidate genes. Magenta lines show the span of the deficiencies that enhanced the Idgf3-overexpression phenotype. A green line shows the span of a deficiency that suppressed the Idgf3-overexpression phenotype. Black lines show the span of the deficiencies that did not change the percentage of defects observed by overexpressing Idgf3. Dotted regions indicate deletions that extend distally or proximally from the region shown here. (A) Example of a simple narrowing-down process. Df(3L)ED4674 strongly enhanced Idgf3 overexpression. Df(3L)ED223 and Df(3L) Exel6130 confirmed and narrowed down the region of enhancement; these deletions did not produce a phenotype on their own. All the other deficiencies tested in the region did not have any effect on the Idgf3-overexpression phenotype (except for suppressor Df(3L)BSC414), narrowing down the region to five candidate genes, shown in red. (B) Example of a complex narrowing-down process. Df(3L)ED4502 identified an enhancer region, which was confirmed and narrowed down with Df(3L)BSC614. Two other deficiencies uncovered portions of the region; neither produced any effect on the Idgf3-overexpression phenotype. These results left 37 candidate genes remaining to be tested. The small arrow points to a gene that was discarded as a candidate gene because the coding regions for all its transcripts lay within the non-interacting deficiency. Ruler above each browser map shows the span of DNA in Megabases (Mb).
n■ Table 2 Overlapping deficiencies and RNAi/null alleles identify Idgf3-interacting

No effect
Tgi

UAS-cmbRB
Indicates deficiencies tested in the original screen and reported also in Table 1 We then asked: how are Idgf3 and cmb interacting at the cellular level to impact the shape of the DA tubes? We tested two hypotheses: 1) they modify the narrowing and lengthening of the tube by regulating cell intercalation; and 2) they control the surface area of the tube lumen.

Idgf3 and cmb do not affect cell intercalation during dorsal appendage formation
The gene cmb was first identified as a substrate of Rho kinase, in vitro (Fagan et al. 2014). This report suggested that cmb might play a role in the Planar Cell Polarity pathway (PCP) since the overexpression of cmb in the Drosophila wing causes the growth of multiple wing hairs, a phenotype that is characteristic of mutations in other PCP components. Moreover, Cmb physically interacts with one component of the PCP pathway (Multiple wing hairs) in both a yeast two-hybrid system and by immuno-precipitation (Fagan et al. 2014).
PCP genes set up planar directionality of the epithelium, defining the orientation of static tissues such as those that produce mammalian hairs, bird feathers, and fish scales (reviewed by Butler and Wallingford 2017). The PCP pathway also coordinates the behavior of cells in morphogenetic tissues, directing movements that drive cell intercalation during a variety of developmental processes (Keller et al. 2000;Wallingford and Harland 2001;Park and Moon 2001;Darken et al. 2002).
In our system, cell intercalation facilitates tube formation (wrapping) at stage 11, and it helps narrow and lengthen the tubes during S12 -S13 (Dorman et al. 2004;Osterfield et al. 2013;Ward and Berg 2005). When the DA patches are defined at S10B, they are longer along the Dorso-Ventral (DV) axis compared to the Anterior-Posterior (AP) axis ( Figure 5A and 5I). During S11, the floor cells zip up the tube underneath the roof cells, and the roof cells contract their apices. At the same time, more lateral cells move toward the dorsal midline, exchanging neighbors and altering the shape of the DA patch (Dorman et al. 2004;Osterfield et al. 2013;Ward and Berg 2005). During S12, as the roof cells release apical tension in a biased fashion (Peters and Berg 2016B), cell intercalation continues, producing a patch that is now longer along the AP axis than the DV axis ( Figure 5E and 5J; Dorman et al. 2004;Ward and Berg 2005).
To test if the Idgf3-cmb interaction affects cell intercalation during DA formation, we quantified this transition by measuring the aspect ratio of S10B and S12 DA patches. We compared egg chambers produced by CY2-GAL4 (control) females with those produced by +/cmb KO females, Idgf3-overexpressing females, and cmb KO /Idgf3overexpressing females. If Idgf3 and cmb affected cell intercalation, we expected to see similar aspect ratios among genotypes at S10B, but significant differences between control and experimental groups at S12. To mark the exact boundary of the dorsal-appendage patches, we used an antibody against Broad (Br), a transcription factor required to specify DA-forming cells (Tzolovsky et al. 1999): high levels of Broad ("High Br") define the DA patches, while moderate or low Broad ("low Br") marks lateral and posterior main-body follicle cells (Dorman et al. 2004). To avoid introducing any bias, we conducted the image acquisition and quantification blind to the genotypes we were analyzing (see methods).
High-Br staining revealed the basal location of the nuclei in the dorsal-appendage-making cells ( Figure 5A-H). As expected, at S10B, the shapes of the patches were similar for all genotypes ( Figure 5A-D), except for a small but significant difference between the Idgf3-overexpression group and the control. Since the means of the aspect ratios of all the genotypes were tightly clustered (0.66+/20.04; Figure 5I), however, this small difference (P = 0.03) might simply have resulted from slight timing differences in S10B stage. That is, some eggs from the Idgf3-overexpression genotype might have been at the end of S10B, transitioning to S11, when the patch nuclei condense slightly as roof cells begin to constrict their apices.
When looking at S12 egg chambers, we were surprised that the DA patches from both control and experimental egg chambers had elongated and narrowed to a comparable degree ( Figure 5E-H), with aspect ratios that again exhibited similar means (1.71+/20.06; Figure 5J). Moreover, the slight differences among genotypes at each stage were not significant (P , 0.05), and the small difference seen in the Idgf3-overexpression group at S10B was not present at S12. Based on these results, we concluded that the Idgf3-cmb interaction was not directly affecting cell intercalation in our system. This result was consistent with studies in the Drosophila testis demonstrating a non-PCP role for cmb in sperm individualization (Steinhauer et al. 2019).

Idgf3 and cmb affect apical area of dorsal appendage tubes
To examine the effect of the Idgf3-cmb interaction on tube lumen morphology, we stained egg chambers with an antibody against E-cadherin (E-cad) to reveal cell shapes ( Figure 6A-D). E-cad localizes on the apico-lateral sides of cells and is an important component of cell junctions, controlling cellular adhesion (Ratheesh et al. 2012).
We quantified the area of the apical side of the dorsal-appendage tubes at stage S12 ( Figure 6). Importantly, we had to account for the migration of the dorsal-appendage tubes because tube migration, with its accompanying cell rearrangements and apical surface expansion, is an ongoing process during stage S12 (Peters and Berg 2016B). Slight differences in the distance migrated during this stage of DA development markedly affects apical area. Therefore, we first used nurse-cell morphology, which was unaffected among genotypes, to identify egg chambers that were at mid-S12. We then quantified the area of the apical-most side of the dorsal-appendage tubes; in this tissue, the apical-most portion of the DA tube is at the lowest level of the roof cells, where roof and floor cells touch. Finally, we normalized this area by the length of the tubes ( Figure 6E). In this way, we determined, for each visible dorsal appendage patch of each egg chamber, what the area of the dorsal-appendage tube is at a specific length of migration and compared these data among genotypes.
While we found that the distances migrated by the dorsal-appendage tubes ( Figure 6E, left panel) were not significantly different among genotypes ( Figure 6F), we observed noticeable differences in tube area ( Figure 6E, right panel and Figure 6G). We found that the apical area of the dorsal-appendage tube, when controlled by the distance migrated at S12, averaged 16.50+/21.5 mm in eggs laid by control flies (Figure 6A and 6G). The normalized apical area was similar in eggs laid by +/cmb KO mutants (18.04+/21.2 mm; Figure 6B and 6G). In contrast, eggs laid by the Idgf3-overexpression females had significantly higher values (20.045 +/2 1.4 mm; Figure 6C and 6G) than the controls, and knocking out one copy of cmb in the Idgf3overexpression background significantly enhanced the Idgf3-overexpression effect (24.20 +/2 1.7 mm; Figure 6D and 6G). Since cmb KO Figure 5 The Idgf3-cmb genetic interaction does not affect cell intercalation during dorsal appendage tube formation. (A-H) Representative examples of dorsal or dorsolateral views of S10B egg chambers (A-D), and of S12 egg chambers (E-H) stained for Broad (Br) protein (white). The dorsal-appendage patches have high levels of Br protein from S10B on. The remaining mainbody follicle have low levels of Br protein. The intensity of all images was increased equally to delineate entire egg chambers. Images represent egg chambers from three different experimental replicas. Scale bar is 100 microns and applies to all pictures. (A and E) w 1118 ; +/CY2-GAL4. (B and F) w 1118 ; +/CY2-GAL4; +/cmb KO . (C and G) w 1118 ; +/CY2-GAL4; +/UAS-Idgf3. (D and H) w 1118 ; +/CY2-GAL4; UAS-Idgf3/cmb KO . Each image is a projection of 62 confocal slices showing the dorsal-appendage patches with high-Br protein levels (intense white nuclei). The number of egg-chambers-scored per genotype is indicated on each panel. Dotted lines show the midlines of each egg chamber. (I, J) Quantification of the aspect ratio of each dorsal appendage patch. Schematics on each panel show the outline area of the dorsal-appendage patch, enlarged to the right, that was considered for aspectratio calculation using FIJI (See methods). "a" indicates anterior, "p" indicates posterior. Box plots show the mean, quartile, and range of aspect ratios measured for each genotype. /+ did not produce a phenotype on its own, the significant change in apical area observed in the cmb KO /UAS-Idgf3 is dependent on the overexpression of Idgf3, similar to what we found in the genetic analysis of laid eggs. Interestingly, the lower half of the cmb KO /UAS-Idgf3 distribution lies within the range of the control data ( Figure 6G). These results are comparable to those for laid eggs, where 42% had normal dorsal appendages (Figure 4).

Role of Idgf3 and cmb during dorsal appendage formation
To explain how cmb and Idgf3 might interact together to regulate the apical area of cells during DA tube elongation, we propose two mechanisms. Both mechanisms rely on the fact that cmb is a substrate for Rho kinase (Rok) and that Rok directly affects actomyosin network tension (Munjal et al. 2015;Riento and Ridley 2003). Changes in actomyosin network tension likely control the behavior of DA-making cells during tubulogenesis.
During wrapping, in the first steps of tube formation, roof cells constrict their apices and floor cells elongate to seal the tubes (Dorman et al. 2004). These cell behaviors require high apical network tension (Osterfield et al. 2013). Following this step, tubes begin to elongate. Transcription factors that modulate expression of actin-regulatory genes control this transition, and changes in tension then allow biased apical expansion and anterior crawling (French et al. 2003;Boyle et al. 2010;Peters et al. 2013). Since cmb is a Rok substrate in other contexts, it is possible that cmb might be involved in regulating tension in dorsal-appendage-making cells during this transition. Experimental work measuring the actomyosin network tension in a cmb mutant could help test this hypothesis.
Under this hypothesis, how can Idgf3 interact with cmb to regulate dorsal-appendage-tube elongation? One idea is that Idgf3 acts upstream of the Rok-cmb pathway. That is, Idgf3 could interact with an unidentified receptor in the dorsal-appendage-making cells, activating Rok and thus inhibiting Cmb and reducing actin polymerization, as seen in wing hair formation (Fagan et al. 2014).
To explain our experimental observations, removing only one copy of cmb might not be enough to cause changes in actin polymerization on its own, but overexpressing Idgf3 might be enough to push the system over a threshold. In this way, the combination of Idgf3 overexpression and the removal of one copy of cmb might enhance the effect of Idgf3 overexpression alone. Analyzing the amounts and distribution of actin in the dorsalappendage-making cells of Idgf3 overexpression and cmb KO/+ /Idgf3-overexpression mutants could provide evidence to support or reject this hypothesis.
There is another mechanism that can explain our observations. While cmb regulates apical tension of the dorsal-appendage tube, Idgf3 could be acting in a parallel pathway to cmb. This alternative mechanism recognizes that apical expansion is normally coordinated with anterior crawling (Boyle and Berg 2009), which involves roof cells physically interacting with the extracellular matrix (ECM) (Dorman et al. 2004). We propose that Idgf3 influences this interaction by modulating the stiffness of the ECM (Zimmerman et al. 2017), possibly by activating enzymes that degrade the ECM around the elongating dorsal appendage tubes. Since the leading cells contact the ECM along their basal surfaces, lowering the stiffness of the ECM could lower the resistance the leading cells encounter and alter integrin-mediated intracellular actin dynamics. These changes would then trigger the Images are a single, 0.5-micrometer slice taken at the most apical region of the roof cells that make the dorsal-appendage tube. The number of egg chambers scored for each genotype is indicated on each panel. (E) Magnified views of the apical surface of a control dorsal-appendage tube at a plane where roof cells meet floor cells. Two factors affect apical tube measurements: 1) the temporal progression of dorsal appendage tube elongation (magenta line), since apical area increases as the tube cells move forward, and 2) the apical area of tube-making cells (filled magenta space). (F, G) Quantification of the distance migrated by each dorsal-appendage tube (F) and of the area of the tube normalized by tube elongation (G; See methods). Box plots show the mean, quartile, and range measured for each genotype.
coordinated expansion of apical surfaces regulated by cmb. A similar change in ECM stiffness is critical for the cell-shape changes that drive Drosophila wing elongation (Diaz-de-la-Loza et al. 2018).
In our experimental observations, removing one copy of cmb might not be sufficient to expand apical area because the ECM exerts a force containing the elongating tubes. Overexpressing Idgf3, which, in this second hypothesis, leads to an abnormal decrease of the ECM stiffness around the tubes, could lower the ECM force that counteracts the expansion of the tubes. Under these circumstances, removing one copy of cmb enhances cell expansion, resulting in the phenotype seen in the cmb KO/+ ; Idgf3 mutants. Experiments in which we artificially manipulate the stiffness of the ECM could help us understand if the ECM does play a role in dorsal appendage formation. In addition, quantifying the width of the ECM in Idgf3 and cmb mutants would provide evidence to support or reject our hypothesis.
Although more work is needed to fully understand the cmb-Idgf3 interaction, we have successfully identified one biological effect on tube shape when the expression of these genes is altered. It would be interesting to explore if human CLPs affect apical cell area of an epithelium or if the human ortholog of combover, PCM1, contributes to actin dynamics in metastatic cancer when the CLPs are up regulated (Libreros et al. 2013).

ACKNOWLEDGMENTS
Thanks to Andreas Jenny for providing us with the cmb alleles; to Lindsay Charley for identifying the method for patch recognition of the S10B egg chambers; to Ken Jean-Baptiste for statistics discussions; to the Berg lab for helpful discussions and input on experiments; to the Bloomington and Vienna Stock Centers for providing stocks; and to the Developmental Studies Hybridoma Bank for antibodies. This work was supported by NSF Predoctoral Fellowship # 14-590 to CYE and by NIH R01 GM079433 to CAB.