An RNAi Screen Identifies New Genes Required for Normal Morphogenesis of Larval Chordotonal Organs

The proprioceptive chordotonal organs (ChO) of a fly larva respond to mechanical stimuli generated by muscle contractions and consequent deformations of the cuticle. The ability of the ChO to sense the relative displacement of its epidermal attachment sites likely depends on the correct mechanical properties of the accessory (cap and ligament) and attachment cells that connect the sensory unit (neuron and scolopale cell) to the cuticle. The genetic programs dictating the development of ChO cells with unique morphologies and mechanical properties are largely unknown. Here we describe an RNAi screen that focused on the ChO’s accessory and attachment cells and was performed in 2nd instar larvae to allow for phenotypic analysis of ChOs that had already experienced mechanical stresses during larval growth. Nearly one thousand strains carrying RNAi constructs targeting more than 500 candidate genes were screened for their effects on ChO morphogenesis. The screen identified 31 candidate genes whose knockdown within the ChO lineage disrupted various aspects of cell fate determination, cell differentiation, cellular morphogenesis and cell-cell attachment. Most interestingly, one phenotypic group consisted of genes that affected the response of specific ChO cell types to developmental organ stretching, leading to abnormal pattern of cell elongation. The ‘cell elongation’ group included the transcription factors Delilah and Stripe, implicating them for the first time in regulating the response of ChO cells to developmental stretching forces. Other genes found to affect the pattern of ChO cell elongation, such as αTub85E, β1Tub56D, Tbce, CCT8, mys, Rac1 and shot, represent putative effectors that link between cell-fate determinants and the realization of cell-specific mechanical properties.

several asymmetric cell divisions to generate the neuron, scolopale, cap, ligament and CA cells of a single organ (Brewster and Bodmer 1995). In parallel to the differentiation of the different cell types, which commences following the completion of cell divisions, patterning and localization of the organ as a whole take place. The LCh5 organ originates in the posterior dorsal region of each abdominal segment and it rotates and migrates ventrally to acquire its final position and orientation (Salzberg et al. 1994;Inbal et al., 2003;Kraut and Zinn 2004). The ligament cells lead the migration process and pull the organ ventrally (Klein et al. 2010). Upon reaching their final destination the ligament cells recruit a LA cell through an EGFR-dependent mechanism (Inbal et al. 2004). During larval stages, with the dramatic increase in body size, the LCh5 organ, which remains anchored to the cuticle on both of its sides, elongates dramatically and goes through major morphological changes (Halachmi et al. 2016).
Whereas early steps in ChO development, namely the recruitment and specification of ChO precursors and the pattern of cell divisions, have been studied extensively (e.g., (Jarman et al., 1993;Lage et al. 1997;Okabe and Okano 1997;Brewster and Bodmer 1995), our knowledge about the genetic basis of later aspects of cell-fate determination, differentiation, morphogenesis and attachment of these organs is very sparse. To start filling in the large gaps in our knowledge about ChO development we have conducted an RNAi-based screen for new determinants of larval ChO organogenesis. Previous genetic screens for genes required for normal patterning of the embryonic peripheral nervous system (PNS) in general, or the ChOs in particular, were based on phenotypic analyses of the sensory neurons only (Salzberg et al. 1994;Kania et al., 1995;Kolodziej et al., 1995;Salzberg et al., 1997). Thus, these screens could not identify genes that affect specifically the nonneuronal cell types (cap, ligament and attachment cells) or affect postembryonic aspects of ChO development. There are two reasons for which screening in larvae, rather than in embryos, is critical for the identification of genes required for ChO morphogenesis: first, it has been recently shown that ChO morphogenesis is not completed during embryogenesis and that terminal differentiation and patterning takes place during larval stages (Halachmi et al. 2016). Thus, developmental defects that only become evident in larval stages are expected to be identified. The second reason is that only after hatching the ChOs start to experience significant mechanical stresses caused by larval growth and locomotion. Thus, genes required for the ability of the ChO to resist mechanical stresses and maintain organ integrity would not be identified by screening in the embryo.
Here we describe for the first time a screen that was performed on second instar larvae and focused on the accessory and attachment cells of the ChO, rather than the sensory neurons. The screen included 918 RNAi strains directed against 547 candidate genes. The genes were selected based on their expression pattern (enriched in ChOs), or potential function in cellular processes that seem critical for normal morphogenesis of ChOs, namely, tubulin-related genes and genes involved in cell migration. The screen identified multiple candidate genes required for different aspects of ChO morphogenesis, including the correct differentiation of specific cell types within the organ, proper attachment between the cap and CA cells and the normal pattern of cell elongation. The latter aspect of ChO development is especially interesting, as cell elongation in response to stretching forces probably depends, among other things, on the mechanical properties of the cell. Thus, the genes identified to be required for the normal pattern of cell elongation may provide a first insight into the formation of ChO cells with unique mechanical properties.

Fly strains
Fly strains used in this study: dei ChO -GFP, dei attachment -RFP (Halachmi et al. 2016). The GFP-RFP marker chromosome was recombined to en-gal4 (A. Brand, personal communication to FlyBase; Gramates et al. 2017), ato-gal4 (Hassan et al. 2000) or P{GMR12D06-gal4} (Pfeiffer et al. 2008). For the analysis of aTub85E loss of function, we used the weak hypomorphic allele Mi{PT-GFSTF.0}aTub85EMI08426 (Bloomington #60267). RNAi strains from the GD and KK libraries were obtained from the Vienna Drosophila Resource Center (VDRC); RNAi strains from the TRiP collection were obtained from the Figure 1 The larval chordotonal organs. (A) Schematic illustration of a first instar larva showing the eight ChOs (black bars) that form a zigzag line of stretch receptors in each of the seven abdominal segments A1-A7. Five ChOs are clustered in the pentascolopidial organ (LCh5). LCh1 is a single lateral ChO. VChA and VChB are two ventrally located ChOs. (B) Schematic illustration of a larval LCh5 organ. The organ is stretched diagonally from a dorsal posterior to a lateral anterior position in each abdominal segment between the epidermis (shown in blue) and the body wall muscles (not shown). The cap cells of the LCh1 and VChB organs are also presented. (C) An LCh5 organ of a second instar larva from the en-gal4 UAS-GFP, dei ChO -GFP, dei attachment -RFP reporter/driver strain used for screening. The cap and ligament cells express GFP (green) and the cap-attachment and ligament attachment cells express RFP (red). GFP expression is also evident in the epidermal stripe of En-positive cells (double-headed arrow). The scale bar = 50 mm.
Bloomington Drosophila Stock Center, Indiana, USA. Two dei null alleles (dei KO-GFP and dei KO-mCherry ) were generated as part of this study (GenetiVision, Houston TX, USA). First, the dei KO-GFP allele was generated by replacing the dei coding sequence spanning amino acid 23-366 with a MiMIC-like cassette (Venken et al. 2011), by injecting two gRNAs (GGCCAGAGCGACGGACTCCAAGG and GAATGGA-TACCCATCCAGAGCGG) and a donor plasmid, containing 3XP3 GFP flanked by loxP sites and inverted attP sites, into nanos-Cas9 embryos (Bloomington #54591). The GFP-cassette was then replaced using Recombinase-Mediated Cassette Exchange (RMCE) with an mCherry cassette, using the plasmid pBS-KS-attB1-2-GT-SA-mCherry-SV40 (obtained from the Drosophila Genomics Resource Center, IN, USA) for generating the dei KO-mCherry allele. The two dei null strains are fully viable. The dei KO-GFP strain expresses GFP mainly in the cap and ligament cells. The dei KO-mCherry strain expresses mCherry in a dei-like pattern.
Collection and fixation of larvae for 2 nd instar larvae, virgin females of the dei ChO -GFP, dei attachment -RFP; ato-gal4, or the en-gal4, UAS-GFP, dei ChO -GFP, dei attachment -RFP strain were crossed to males of the desired RNAi strain ($30 females and 10 males). The flies were kept for 3-4 days at room temperature and then transferred to egg-laying chambers, put on grape juice plates with yeast paste and let to lay eggs for 24 hr at 29°. Adult flies were removed, and the progeny was left to mature at 29°for additional 20 hr. Larvae were washed once with phosphate buffered saline + 0.1% Tween-20 (PBT) and fixed overnight at 4°in 4% formaldehyde in PBT. Fixed larvae were washed twice with PBT (over 20 min) and twice with PBS (over 20 min) before mounting in Dako Fluorescent Mounting Medium (DakoCytomation, Glostrup, Denmark). Larvae were viewed using confocal microscopy (LSM 510, Zeiss) within a week from their fixation. Dissection and staining of 3 rd instar larvae were performed as previously described (Halachmi et al. 2012).

Data availability
The strains generated in this work are available upon request. The authors affirm that all data necessary for confirming the conclusions of the article are present in the article, figures, and tables. File S1 contains the list of all RNAi constructs tested in this study.

Genetic screen
Dissection and staining of large numbers of larvae is a slow and laborintensive process. To overcome this limitation, we took advantage of recently developed ChO-specific fluorescent reporters that allow rapid screening of whole-mount larvae without any need for dissection or immuno-staining (Halachmi et al. 2016). These reporter constructs are based on cis-regulatory modules from the dei locus (Nachman et al. 2015) that were used for driving cytoplasmic GFP expression in the cap and ligament cells of ChOs (dei ChO -GFP), and cytoplasmic RFP in the attachment cells of ChOs (dei attachment -RFP) ( Figure 1C). For the screening procedure the dei ChO -GFP, dei attachment -RFP chromosome was recombined to ato-Gal4, which drives expression specifically in the LCh5 lineage, and to en-gal4, which drives earlier and prolonged expression in the entire posterior compartment of the segment, including the LCh5 organs. Both of these drivers induce expression in all of the lineage-related cells of the ChO but do not induce expression in the LA cell, which is not derived from the lineage (shown schematically in Figure 5B). Flies from each of these strains were crossed to flies bearing UAS-RNAi transgenes from the VDRC collection and the ChO phenotype of the progeny was inspected in whole-mount 2 nd instar larvae. At least 10 larvae of each genotype were examined.
A collection of 918 RNAi strains directed against 547 candidate genes (Table S1) was selected and screened with both of the Gal4 drivers. The largest group of genes (240 genes, 379 RNAi lines) among this collection was selected based on gene expression pattern. It consisted of genes reported by (Cachero et al. 2011) or (Senthilan et al. 2012) to be enriched in ChOs during early stages of embryonic development or in antennal ChOs, respectively. The rest of the genes were selected based on potential functions rather than expression patterns. Since the accessory cells of ChOs are extremely microtubule-rich, we selected 112 genes (188 RNAi lines) identified in FlyMine (http://www.flymine. org) in a search for 'tubulin-related' genes. Since ChO morphogenesis in both the embryo and the larva requires extensive cell migration (Inbal et al. 2003;Halachmi et al. 2016), we selected additional 165 genes (280 RNAi lines) identified in FlyMine using the search term 'cell migration'. Additional 30 genes (71 RNAi lines) that were identified in previous screens for PNS development (Salzberg et al. 1994;Kania et al., 1995;Salzberg et al., 1997), or were identified as being expressed in ChOs in late developmental stages (A. Salzberg, unpublished observations), were also included. When possible, two independent RNAi strains from different libraries (GD and KK) were tested for each gene. RNAi strains identified in the primary large-scale screen were further analyzed using immunohistochemistry on dissected third instar larvae. Complementary RNAi strains from the TRiP collection (Perkins et al. 2015) were used for validating the specificity of the RNAi-induced phenotypes.
Phenotypic grouping the fluorescent markers used in the screen allowed us to identify phenotypes that could be grouped into three general and not mutually exclusive categories: 1. Loss or gain of GFP or RFP expression, often combined with abnormal morphology of cells. 2. Defective attachment or cell morphology without a major loss of marker expression. 3. Abnormal pattern of cell elongation. We assigned each of the identified genes into one of these three groups based on the most prominent phenotypic feature it presented (Tables 1, 2, and 3).
Loss or gain of GFP or RFP expression As outlined in Table 1 and Figure 2, seven genes were identified whose knockdown by RNAi led to a loss of GFP or RFP expression from specific ChO cells. The loss of marker expression could reflect a genuine loss of specific cell types, cell fate transformation, or specific loss of dei expression. Similarly, expansion of GFP/RFP expression could reflect gain of cells, cell fate transformation, or ectopic expression of the dei gene. Although the loss of marker expression does not necessarily reflect a true loss of specific cell types, we refer to the phenotypes as 'loss of cells' for the sake of simplicity, and group the phenotypes according to the type of the affected cell/s.

Loss of LA cells:
The phenotype caused by knocking down vein (vn) expression under the regulation of en-Gal4 was unique. vn is the only gene identified whose knockdown within the ChO lineage led to a nonautonomous loss of the LA cell ( Figure 2C-D). The LA cell is recruited from the epidermis via an EGFR-mediated pathway; the current observation validates the previously suggested notion that Vn is the ligand secreted by the ligament cells (Inbal et al. 2004). As a consequence of reducing Vn secretion from the ligament cells by means of vn RNAi expression, the EGFR is not activated in the target epidermal cell and therefore, LA cell differentiation does not occur. The observed vn RNAi phenotype also suggests that a crosstalk between the ligament cells and the LA cell is required for the convergence of the ligament cells' migrating tips onto a narrow attachment site. In the absence of a LA cell, the ligament cells' tips extend in different directions ( Figure 2C).

Loss of CA cells:
Expressing RNAi constructs directed against three genes, capricious (caps), Notch (N) and meru, led to a loss of at least one of the two CA cells ( Figure 2E-J), which was often accompanied with a collapse of the cap cells. In order to better characterize the phenotypes and distinguish between CA cell loss and cell fate transformation, we counted the number of cap and CA cells present in the affected LCh5 organs using anti-Blistered (Bs) immunostaining. This analysis demonstrated that in the N and caps knockdown larvae, the loss of CA cells was consistently accompanied by an increase in the number of cap cells. Whereas the LCh5 organs of control larvae contained five cap cells and two CA cells each, the LCh5 of N-or caps-RNAi larvae consisted of one CA and six (or, occasionally, seven) cap cells ( Figure 2E-H). These results suggest that the activity of both N and caps is required for the correct specification of CA vs. cap cell-fate by influencing the asymmetric division of the secondary ChO precursor that gives rise to the cap and CA cells. This finding corroborates findings of a previous RNAi screen that identified caps as a gene affecting asymmetric cell division in the external sensory lineage (Mummery-Widmer et al. 2009).
Unlike N and caps, the knockdown of meru led to the loss of one CA cell with no concomitant increase in the number of cap cells ( Figure 2I-J). The meru gene was identified by Reeves and Posakony (Reeves and Posakony 2005) as a direct target of the proneural genes and was implicated in the sensory perception of pain by (Neely et al. 2010). More recently, (Banerjee et al. 2017) have identified Meru as a modulator of cell polarity that connects planar cell polarity with apicalbasal polarity during asymmetric cell divisions within the external sensory organ lineage. The identification of meru in the current screen points to a possible role of meru in the ChO lineage as well. Whether its role in the internal sensory (ChO) lineage is similar to its role in the external sensory lineages remains to be elucidated. A more severe and variable phenotype was caused by downregulating the daughters against DPP (Dad) gene within the posterior compartment of the segment. Dad downregulation led to loss of the two CA cells, often collapse of the cap cells and expansion of the dei ChO -GFP expression into the region of the sensory unit ( Figure 2K-L).

Loss of cap and CA cells:
The expression of shaven (sv)-RNAi under the regulation of either ato-gal4 or en-gal4 caused a severe loss of cap and CA cells that was not accompanied by an obvious increase in the number of other types of cells ( Figure 2M-N). This observation suggests that Sv is required for the differentiation and/or survival of the cap and cap-attachment cells. Interestingly, the sv gene is required for the differentiation of shaft cells, which are equivalent to the cap cells in the adult external sensory (ES) lineages (Fu et al. 1998;Kavaler et al. 1999).
Expansion of GFP expression: The expression of prospero (pros)-RNAi under the regulation of either ato-gal4 or en-gal4 led to an n Loss of CA cells, abnormal organ shape, expansion of the dei ChO -GFP signal into the region of the sensory unit Loss of cap and CA cells, or loss of the dei ChO -GFP dei attachment -RFP signal Expansion of the dei ChO -GFP expression into the region of the sensory unit Table 1 lists the seven genes identified in the screen whose knockdown by RNAi led to loss or expansion of the dei ChO -GFP and/or dei attachment -RFP reporters. The RNAi strains directed against each of the genes, the phenotype they caused, and the ability of each RNAi strain to cause a phenotype when expressed under the regulation of ato-Gal4 and en-Gal4 are listed. The number of predicted off targets is indicated for RNAi strains whose phenotypes were not reproduced by additional RNAi strains directed against the same gene.
expansion of the dei ChO -GFP expression into the region of the sensory unit ( Figure 2O). This phenotype could indicate that loss of pros expression causes the scolopale cell, the only prosexpressing cells in the ChO lineage, to acquire an accessory (ligament or cap) cell identity. As mentioned above, the knockdown of Dad often led to a similar expansion of GFP expression into the sensory unit ( Figure 2K), and so did the knockdown of senseless (see below, 3N).
n NTnot tested. Table 2 lists the thirteen genes identified in the screen whose knockdown by RNAi led to defective pattern of attachment or cell morphology. The RNAi strains directed against each of the genes, the phenotype they caused, and the ability of each RNAi strain to cause a phenotype when expressed under the regulation of ato-Gal4 and en-Gal4 are listed. The number of predicted off targets is indicated for RNAi strains whose phenotypes were not reproduced by additional RNAi strains directed against the same gene.

Defective attachment or cell morphology
In wildtype larvae, the cap cells are stretched between the scolopale cells and the CA cells. During larval growth, the CA cells grow dramatically, extending numerous tubulin-rich extensions and forming a wide integrin-rich junction with the attached cap cells (Halachmi et al. 2016;Greenblatt Ben-El et al. 2017). These morphological changes are likely required for adjusting the ability of the CA cells to anchor the cap cells and remain attached to the cuticle under conditions of increasing mechanical stresses. Ten genes were identified in the screen whose knockdown caused an abnormal pattern of cap/CA cell attachment. Three additional genes affected the cap cells on their scolopale-facing side ( Table 2).
Defective attachment between the cap and CA cell: Down-regulation of the Drosophila EGF-receptor gene, Egfr, within the en domain resulted in the development of small, often slightly elongated, CA cells that expressed lower levels of the dei attachment -RFP marker as compared to control larvae. The contact area between the affected CA cells and the attached cap cells was greatly reduced and the bundle of five cap cells appeared abnormally thin near the attachment site ( Figure 3A). The LA cell, which depends on Egfr activity for its development, does not originate from the en domain and thus could develop properly in the en-Gal4/Egfr-IR larvae. The expression of RNAi constructs directed against six additional genes caused a Egfr-like phenotype: couch potato (cpo), which encodes for an RNA binding protein, CG13653, a gene with unknown function, furry (fry), which encodes for an actin cytoskeleton regulator, echinoid (ed), which encodes for a homophilic cell adhesion molecule, Eb1, which encodes for a microtubule-associated protein, and WASp (Wsp) the fly homolog of the Wiskott-Aldrich Syndrome family of actin nucleation factors ( Figure 3B-G). The expression of RNAi construct directed against pyramus (pyr), which encodes for one of the three known Drosophila Fibroblast Growth Factor (FGF) ligands led to the development of abnormally shaped CA cells and slightly elongated ligament cells ( Figure 3H).
Two other genes found to be important for proper attachment between the cap and CA cells were stripe (sr), which encodes for an early growth response-like transcription factor, and myospheroid (mys), which encodes for the prevalent variant of beta-integrin (bPS). Sr has been previously shown to be required for CA cell differentiation in the embryo (Inbal et al. 2004). Here we show that during larval stages the Sr-deficient CA cells fail to anchor the cap cells properly, leading to their detachment and collapse ( Figure 3I-J). The phenotype of the ato-Gal4/sr-IR larvae also validates the notion that the LA cell depends on the autonomous activity of Sr (Inbal et al. 2004) and could, therefore, develop properly in the ato-Gal4/sr-IR larvae. In addition to the defects in cap cell attachment, the sr knockdown larvae occasionally presented elongated CA or ligament cells ( Figure 3J).
Reducing the level of mys expression had no major effect on the differentiation of the CA cells, as suggested by their normal size and overall morphology as well as the normal level of dei attachment -RFP expression they presented. However, the loss of bPS integrin led to detachment and collapse of the cap cells. In segments in which the cap cells remained attached, the contact area between the cap and CA cell was greatly reduced and the cap cell appeared much thinner than normal close to the cap/CA contact point ( Figure 3K-L). Occasionally, the mys knockdown larvae presented elongated ligament cells in addition to the defects in cap cell attachment ( Figure 3L). This observation supports the idea that cap cell elongation depends on integrin-based interaction with the extracellular matrix (Greenblatt Ben-El et al. 2017).
n  Table 3 lists the eleven genes identified in the screen whose knockdown by RNAi led to abnormal pattern of ChO cell elongation. The RNAi strains directed against each of the genes, the phenotype they caused, and the ability of each RNAi strain to cause a phenotype when expressed under the regulation of ato-Gal4 and en-Gal4 are listed. The number of predicted off targets is indicated for RNAi strains whose phenotypes were not reproduced by additional RNAi strains directed against the same gene.
Abnormal alignment or attachment of the cap cells on their scolopale-facing side: The knockdown of three genes, senseless (sens), raw and Rac1 affected the cap cells on their ventral side where they normally attach to the scolopale cells. sens, which encodes for a zinc finger transcription factor, is an important regulator of neurogenesis in the embryonic PNS, where it is required for enhancement and maintenance of proneural gene expression in the sensory organ precursors (Salzberg et al. 1994;Nolo et al. 2000). Unlike normal larval ChOs, in which the ventral tips of all five cap cells are aligned, the cap cells of sens-depleted larvae vary in length and often appear shorter and detached on their ventral side ( Figure 3M-N). In other segments, the cap cell-specific GFP signal expanded into the scolopale cell ( Figure 3N). A closer examination of the affected organs in 3 rd instar larvae demonstrated that shorter cap cells remained attached to scolopale cells that were located in abnormal dorsal positions, possibly reflecting defects in ChO cell migration. The expansion of the GFP signal into the scolopale cell suggests a partial scolopale-to-cap cell fate transformation ( Figure  3O-P).
In raw deficient larvae, the cap cells are of varying lengths and some of them seem to extend into the region normally occupied by the scolopale cells ( Figure 3S). Raw, a membranous protein, was previously shown to be involved in cell movement, elongation and ensheathment e.g., (Jack and Myette 1997;Blake et al. 1998Blake et al. , 1999Byars et al. 1999;Bates et al. 2008;Jemc et al. 2012), thus the observed phenotype could reflect defects in the interactions between the cap and scolopale cells that lead to abnormal contact between the two cell types. In a previous PNS screen, insertional mutations in the raw/cyr gene caused a ChO phenotype of darkly stained (MAb22C10) elongated neuronal cell  bodies, thicker than normal axon bundles and mild pathfinding defects (Kania et al. 1995;Prokopenko et al. 2000). Irregularities in cap cell alignment and occasional expansion of the GFP signal into the scolopale cell was also evident in Rac1 knockdown larvae ( Figure 3Q-R). Rac1, a small GTPase is involved in regulating the dynamic rearrangements of the actin cytoskeleton and was shown to play a role in peripheral glia migration, nerve ensheathment and axon outgrowth (Luo et al. 1994;Sepp 2003). Additional phenotypes observed in the Rac1 knockdown larvae were longer than normal ligament cells and detachment of the cap from the CA cells ( Figure 3R). The identification of Rac1 as well as fry, WASp and shot (see below) point to the importance of the actin cytoskeleton in ChO morphogenesis.
Abnormal pattern of cell elongation During larval growth, the LCh5 organ, which is anchored on both its sides to the cuticle, stretches and elongates from approximately 70 microns at the end of embryogenesis to more than 300 microns at the 3 rd instar larva. Normally, most of this elongation is attributed to the cap cell, which increases its length nearly 13-fold and comprises 65-70% of the entire organ length. We have identified 11 genes whose knockdown led to an abnormal pattern of cell elongation within the ChO (Table 3). In larvae expressing RNAi constructs against any of the identified genes, the cap cells were shorter than normal, whereas the ligament cells and/or the CA cells, were longer than normal (Figure 4). The total length of the organ did not change. Ten of the genes included in this phenotypic category encode for different variants of a and b tubulin and for other types of microtubule-associated proteins: seven tubulin genes, two chaperones (tubulin-specific chaperone E (Tbce) and CCT8), and the spectraplakin-encoding gene shortstop (shot). The 11 th gene in this phenotypic group encodes for the basic helix-loop-helix transcription factor Taxi wings/Delilah (Dei). The knockdown of three additional genes included in other phenotypic categories, sr, mys and Rac1, lead to abnormal elongation of the ligament cells (see Figure 3J, L, Q-R, respectively).
a and b tubulin are encoded in the Drosophila genome by a small gene family comprised of five genes for a tubulins and five genes for b tubulins (Sánchez et al. 1980;Gramates et al. 2017). Although RNAi transgenes directed against seven of these genes (aTub85E, aTub67C, aTub84B, bTub60D, bTub56D, bTub97EF, bTub85D) caused defects in LCh5 cell elongation, we suspect that some of the phenotypes were caused by off-targeting effects and do not reflect a genuine requirement for that specific tubulin isoform. Based on information provided by the VDRC, off-targeting is common among RNAi transgenes directed against the various a tubulin isoforms and among RNAi transgenes directed against the various b tubulin isoforms (Table S2). We therefore refer here only to the two major tubulin isoforms that are expressed within the ChO, namely aTub85E and bTub56D (b1 tub).
Interestingly, the knockdown of a and b tubulin genes led to distinguishable phenotypes. Despite the fact that both aTub85E and b1Tub are expressed in all of the accessory and attachment cells, down-regulation of aTub85E led to shortening of the cap cells and concomitant elongation of, primarily, the ligament cells, whereas down-regulation of b1Tub led to shortening of the cap cells and elongation of the CA cells ( Figure 4B-C). By the time the affected larvae reached the 3 rd instar larval stage, the CA cells of the aTub85E knockdown larvae were often elongated as well (see Figure 5F), yet the phenotypic difference between the a and b gene was still evident. In contrast to the aTub85E knockdown larvae, in the b1-tub knockdown 3 rd instar larvae the ligament cells were not elongated ( Figure 4B'). Knocking down the expression of CCT8, Tbce, or shot led to a bTublike phenotype, whereas knocking down the expression of dei led to a pronounced aTub85E-like phenotype ( Figure 4D-G). The ligament cells of the shot knockdown larvae were occasionally elongated as well ( Figure 4G').

Keeping the ligament cells short
The pronounced cell-elongation phenotypes observed upon knocking down the expression of either dei or aTub85E was somewhat surprising since, previously, we have shown that a deletion of the aTub85E locus did not lead to any obvious defects in ChO's morphology in late embryos (Klein et al. 2010). Similarly, examination of the LCh5 organs of dei deficient embryos did not reveal any abnormal phenotypes, suggesting that Dei does not play a critical role in embryonic ChO development (A.Salzberg unpublished data). The current observations, however, implicate both aTub85E and dei in ChO morphogenesis and suggest for the first time that their loss affect the ability of the ChO cells to elongate properly in response to developmental organ stretching.
dei and aTub85E share the same expression pattern within the ChO, both being expressed in the cap, ligament, CA and LA cells. Thus, the excessive elongation of the ligament cells caused by their loss of function could stem from the inability of the cap cells to elongate properly in response to organ's stretching, or from the inability of the ligament cells to resist stretching and remain short. In order to test whether dei and aTub85E are required for preventing ligament cell elongation, we down-regulated their expression specifically in the ligament cells under the regulation of repo-gal4. The ligament-specific knockdown of either aTub85E or dei resulted in extremely elongated ligament cells, shorter than normal cap cells, and normally shaped CA and LA cells ( Figure 5D, G). These observations indicate that both aTub85E and dei are critical for the development of ligament cells that are able to remain short during organ elongation. A cap cell-specific Gal4 driver (currently not available) is needed for establishing whether these genes are additionally required for the inherent ability of the cap cells to elongate properly. To validate the RNAi-induced phenotypes of aTub85E and dei, and to examine the effects of eliminating or reducing their expression from the entire ChO, including the LA cell, we examined the phenotypes of larvae homozygous for the viable weak hypomorphic allele aTub85E MI08426-GFSTF.0 , or larvae homozygous for a dei null allele we have generated. Both of the mutants exhibited elongated ligament cells, validating the role of these genes in keeping the ligament cells short (Figure 5E, H). The aTub85E MI08426-GFSTF larvae displayed in addition slightly elongated LA cells ( Figure 5H).
Although dei affects ligament cell elongation similarly to aTub85E, its effect is probably not mediated through downregulation of aTub85E expression as suggested by the persistent aTub85E expression in dei knockdown or knockout larvae ( Figure 5C-E). Another transcription factor identified in the current screen, Sr, was previously found to be a positive regulator of aTub85E expression in the ligament cells during embryogenesis (Klein et al. 2010). The role of Sr in post-embryonic morphogenesis of the ChO could not be deduced from phenotypic analyses of sr mutants due to embryonic lethality, however, as a positive regulator of aTub85E, Sr is expected to affect ligament cell elongation. The occasional elongation of ligament cells seen in ato-gal4/sr-RNAi larvae ( Figure 3J) supports such a notion. To further test whether Sr is essential for the development of ligament cells that are resistant to stretching, we knocked down sr expression specifically in the ligament cells and examined the ChOs of 3 rd instar larvae. As shown in Figure 5I, the knockdown of sr within the ligament cells caused a loss of aTub85E expression and extensive elongation of these cells, comparable to the aTub85E knockdown phenotype. This observation indicates that Sr activity is critical for the development of ligament cells that avoid cell elongation, possibly through its positive effect on aTub85E expression.
Among the ChO cells, the ligament cells seem the most sensitive to the loss of aTub85E, as they abnormally elongate upon any reduction in its expression levels, while other cells maintain their normal length. In contrast to aTub85E, when b1Tub was knocked-down under the regulation of en-gal4 or ato-gal4, the CA cells, rather than the ligament cells, were abnormally long, suggesting the hypothesis that this tubulin may be required for maintaining rigid attachment cells. However, to the best of our knowledge, b1Tub is the only b-tubulin isotype expressed at high levels in the ligament cells. Thus, if this tubulin indeed affects attachment cell rigidity, it is expected to affect similarly the properties of the ligament cells. Indeed, when we knocked down the expression of b1Tub specifically in the ligament cells, it led to their extreme elongation ( Figure 5J). Similarly, when we knocked down aTub85E specifically in the CA and LA cells, under the regulation of the GMR12D06-GAL4 driver, it resulted in extreme elongation of these attachment cells ( Figure 5K). Altogether these observations suggest that both aTub85E and b1Tub are required for preventing cell elongation of both the ligament cells and the attachment cells, but the attachment cells are more sensitive to the loss of b1Tub whereas the ligament cells are more sensitive to the loss of aTub85E.

DISCUSSION
The proprioceptive larval ChOs respond to mechanical stimuli generated by muscle contractions and consequent deformations of the cuticle. Thus, their function likely depends on the correct mechanical properties of their accessory (cap and ligament) and attachment (CA and LA) cells that transform the deformation from the cuticle to the sensory neuron. Here we describe a genetic screen that focused, for the first time, on the development of the ChO accessory and attachment cells, rather than the sensory unit itself.
The screen identified 31candidate genes required for different aspects of cell fate determination, differentiation and morphogenesis of these cells ( Figure 6) and provided new entry points to the study of ChO cell mechanics. One important outcome of the cell-specific differentiation programs characterizing each of the ChO cell types is the differential response of the cells to forces imposed on them by larval growth and the consequent stretching of the organ. In this respect, perhaps the most interesting group of genes identified in the screen includes genes required for the differential elongation of ChO cells. Although the phenotypic analysis of the identified genes was restricted to morphological parameters, namely the extent of cell elongation, we assume that the observed morphological alterations reflect, at least in part, changes in cell mechanics and are thus expected to affect mechanosensing. Functional studies are required to test this assumption. Interestingly, two major aberrant cell-elongation phenotypes were observed in the screen: over-elongation of the ligament cells, or over-elongation of the attachment cells, both at the expense of cap cell elongation ( Figure 6). The loss of either dei or aTub85E led to extreme elongation of the ligament cells, and localized knockdown experiments pointed to the role of these genes in the development of ligament cells that do not elongate during organ stretching.
Another group of genes, consisting mainly of b-tubulins and microtubule-associated proteins (Tbce, CCT8, and shot) primarily affected the ability of the attachment cells to resist stretching and avoid cell elongation. Three other genes, sr, mys, and Rac1, affected ligament and/or CA cell elongation, in addition to affecting other aspects of ChO morphogenesis, such as the attachment between the cap and the CA cells.
The differential effect of knocking-down different tubulin genes is intriguing for two reasons. First, reducing the availability of either the a or the b tubulin monomers is expected to have a detrimental effect on the primary construction and maintenance of the microtubules. Second, the aTub85E and btub56D (b1Tub) genes are similarly expressed in both the ligament cell and the attachment cells and it is not clear why each of the cell types shows higher sensitivity to the loss of one of them. Several explanations, or a combination thereof, are possible. One obvious explanation could be the availability of additional tubulin isotypes expressed within the same cells that can compensate for the loss of the specific knocked-down isotype. Differences in the expression levels of various tubulin genes together with different efficacies of the RNAi transgenes could also affect the sensitivity of the different cells to knockdown of specific tubulin isotypes. Another point to be considered is that a and b tubulin molecules differ in the post-translational modifications they go through, such as the tyrosination/detyrosination and acetylation of a but not b tubulins, which are associated with stabilized, long-lived microtubules, or as was recently suggested, render microtubules mechanically resistant to compressive forces (Xu et al. 2017;Janke and Montagnac 2017). It is possible that the microtubule population and, moreover, their dynamics in the attachment cells differs from that of the ligament cells thus making these cells more or less sensitive to the loss of specific a or b isotypes and their unique modifications.
Even more puzzling is the very different behavior of the cap cell during the ChO's elongation phase. All four types of ChO accessory and attachment cells contain abundant microtubules and the cytoplasm of the cap cell, in particular, is densely packed with microtubules. Even though the cap cell expresses high levels of dei, aTub85E and b1Tub, which seems to protect the CA and ligament cells from stretching, this cell increases its length more than 10-folds during larval growth. Perhaps a key to the differential responses of the cap and ligament cells to stretching is the presence of the structurally divergent mesodermal variant, b3Tub, which is expressed in the ChO exclusively in the cap cells and only toward the end of embryogenesis shortly before larval hatching (Matthews et al. 1990;Kaltschmidt et al. 1991;Hinz et al. 1992;Buttgereit and Renkawitz-Pohl 1993;Dettman et al. 2001). It was previously suggested by (Dettman et al. 2001) that b3Tub reduces the level of cross-linking between microtubules, allowing for their sliding past each other and enabling cell elongation. Unfortunately, the one b3Tub-directed RNAi strain that caused a phenotype when expressed in the ChO lineage is cross-reactive with the b1Tub gene. Thus, it is impossible to conclude from the RNAi data about the unique role of b3Tub in cap cell morphogenesis. Given that the available loss-offunction alleles of b3Tub and b1Tub are lethal (Myachina et al. 2017), better genetic tools that allow cell-specific knockout of b3Tub or b1Tub within the ChO are required for distinguishing between the roles played by each of these tubulin isotypes. Additional tools are also required to allow for cap cell-specific knockdown of genes using RNAi transgenes. Such tools will enable us, for instance, to test whether dei is required in the cap cell for its ability to elongate, in addition to its role in keeping the ligament cells short, and will allow us to conduct an RNAi screen for genes that are required for cap cell elongation.