We had previously shown that precursors of the long tendons are clusters of epithelial cells selected among the cells of leg segmental joints. The Notch pathway initiates stripe expression, the key factor in tendon cell differentiation, in the leg tendon precursors (Soler et al., 2004; Laddada et al., 2019).

These Sr-positive cells invaginate and migrate to form the tube-shaped long tendon between the end of the larval stages and the first hours of pupal formation. To gain a better understanding of the molecular mechanisms underlying these dramatic cell morphological changes, we undertook a whole transcriptome RNA-seq analysis of these cells at the onset of pupation (0 h APF). Total RNAs were prepared from FACS-isolated sr-gal4>UAS-GFP cells from 0 h APF leg discs and processed for whole-transcriptome RNA-seq (GSE169313), followed by bioinformatic analysis (for cell-sorting and RNAseq detailed protocols and validations see Materials and Methods and Supplementary Material). With the FPKM cut-off at 10 for reliable detection of gene expression, we found that a wide array of genes 5,479) were significantly expressed in sr-gal4>UAS-GFP cells, probably reflecting the developmental plasticity exhibited by Drosophila imaginal disc cells (McClure and Schubiger, 2007). However, gene ontology (GO) analysis of these genes showed an over-representation of the GO terms related to “locomotion” and “muscle structure development” with numerous genes known to be involved in muscle attachment site development and/or the formation of myotendinous junctions (Table 1). Considering GOs related to pathways, we found an over-representation of Notch, Wnt and MAPK pathways, consistent with our previous results and those of others, showing that these pathways play a key role in tendon development in Drosophila (Yarnitzky et al., 1997; Soler et al., 2004; Lahaye et al., 2012; Laddada et al., 2019). Finally, this analysis also pointed to GO terms relative to tube morphogenesis such as “epithelial tube morphogenesis” or “regulation of tube size”, supporting our earlier assertion that long tendon development shared common features with salivary gland or tracheal tube morphogenesis (Girdler and Röper, 2014; Hayashi and Kondo, 2018; Laddada et al., 2019). In order to functionally validate this point, we crossed the sr-gal4 line with UAS-RNAi or UAS-Dominant Negative lines directed against genes known to regulate many aspects of cell adhesion, migration or maintenance of tissue integrity during morphogenetic processes such as tubulogenesis (Supplementary Table S1). Expression of most of these RNAi or dominant negative transgenes strongly impacted the normal development at larval or pupal stages and RNAi against genes such as the βPS integrin totally disrupted the long tendon formation (data not shown). Interestingly, knocking down the expression of genes controlling the length of developing tubes (Luschnig et al., 2006; Kerman et al., 2008) such as serpentine, vermiform or lolal, led to an excessive elongation of the long tendons in the mature leg (Supplementary Figure S1). Altogether, these results validate our approach to identifying new regulators of appendicular long tendons and support a pivotal role of tubulogenesis in their development.

Because transcription factors are fundamental in controlling the developmental program to build any structure during development, we carried out a lethality and climbing-based in vivo RNAi screen targeting the genes that encode for TFs that we found specifically enriched in our RNAseq data. Remarkably, of the 5,479 genes with FPKM>10, transcripts of 31 genes encoding for predicted transcription factors (Rhee et al., 2014) exhibited more than 1.5-fold higher levels of expression in the GFP + cells than in the whole leg disc cells (input), strongly suggesting that these TFs have a relevant specific role in the development of the leg long tendon (Supplementary Table S2). We performed this RNAi screen by crossing tendon-specific UAS-Dicer2; sr-gal4 line with UAS-RNAi lines targeting these 31 candidates from two independent libraries: TRiP (Bloomington Stock Center) and VDRC (Vienna Stock Center), when available. To limit the number of false positives due to off-target effects, we did not include UAS-line RNAi from the VDRC stock center, for which more than two off-targets were predicted (this was the case for three genes). In this way, we crossed 53 UAS-RNAi lines to target the 31 selected genes, and we could test two different lines for 71% of them (22/31) (Supplementary Table S2). F1 generation were screened for developmental lethality and/or climbing defect (for the detailed screen protocol, see Materials and Methods). Out of the 31 candidates analyzed, seven showed a significant embryonic (or early larval stage) lethality, with more than 20% of embryos unable to hatch or dying before their third larval stage, 12 displayed a pupal lethality score higher than 20%, and 14 showed a climbing defect. Thus, nearly 65% (20/31, including stripe itself) of the candidate genes were positive for at least one readout and for eight of them this result was obtained with two different RNAi lines (Table 2). Within these positive hits, we found odd and drm both members of the odd-skipped gene family that act downstream of the Notch pathway to promote leg tendon growth (Laddada et al., 2019). We also identified esg, which interferes with the Notch pathway during stem cell maintenance/differentiation (Korzelius et al., 2014), and which is required during tracheal morphogenesis (Miao and Hayashi, 2016). Moreover, it has been shown that esg misexpression in embryonic epidermal cells (including muscle attachment sites) can interfere with the development of muscle attachment sites (Staudt et al., 2005). These positive genes strengthen our strategy to identify new regulatory factors of leg long tendon development. Other candidates, including six uncharacterized predicted transcription factors (CG numbers), have never been reported or investigated in tendon development.

We then considered the vertebrate orthologs of these newly-highlighted TFs in order to identify functional evolutionary conservation. We note that of the 14 conserved genes, the orthologs of dar1 (KLF5 and KLF4) and CG9650 (BCL11 A) were shown to be differentially regulated in limb tendon cells during mouse development in two independent studies (Havis et al., 2014; Liu et al., 2015). Strikingly, for each of these genes, the expression of two different RNAi lines gave significant adult climbing defect and/or metamorphosis but no embryonic lethality (Table 2). These observations strongly suggest that these two genes are likely candidates for the specific development of appendicular long tendons. Below we analyze the expression and function of dar1 characterized by high transcript enrichment in our RNAseq datasets.

To confirm the expression of dar1 in the leg disc tendon, we performed immunostaining on sr-gal4>UASmcherryNLS leg discs from third larval instar to 5 h after pupal formation (APF) using Dar1 antibody (Ye et al., 2011). dar1 expression colocalizes with stripe expression domains, within the cells corresponding to the specified long tendons of the leg (Figure 2). dar1 was first found in the two earliest clusters of Sr-positive cells that would form the long tendon of the tarsi (lt) and in tibia levator tendon (tilt) in the dorsal femur (Figures 2A,B). Subsequently, during early pupation this expression was maintained in lt and tilt and new Sr-positive clusters corresponding to other long tendons in different leg segments started to be specified (Figures 2C–F). Importantly, dar1 expression was observed in all these clusters and correlated with our RNA-sequencing data generated from tendon precursors at early pupal stage (0,2 h APF), which showed a high specific enrichment of dar1 expression in sr-gal4>GFP cells compared to IP (FC = 4.7), strengthening the reliability of our data. To determine whether dar1 was a hallmark of all tendon cells in Drosophila, we also analyzed its expression in embryos and in wing discs (Supplementary Figure S2). In embryos, we could not detect Dar1 in sr-gal4>UASmcherryNLS positive tendon cells (Supplementary Figure S2A,B). Likewise, tendon precursors of flight muscles in the wing disc did not show any dar1 expression (Supplementary Figure. S2C,D).

All in all, these results strongly suggest that Dar1 plays a specific role in the establishment of these unique long tendons in Drosophila.

To determine whether the adult locomotion defects observed in tendon-specific dar1 KD could be explained by an abnormal development of the appendicular musculotendinous system, we analyzed adult leg myotendinous architecture in dar1 KD flies. In sr-gal4>UAS-Lifeact.GFP, UAS-dar1RNAi flies, we observed heterogenous results with only a few tendons partially affected per fly (data not shown). This observation is likely due to a residual expression of the targeted gene, which is expected when using RNA interference. We therefore decided to use the sr-gal4>UAS-Lifeact.GFP, UAS-dar1RNAi flies associated with a heterozygous null dar1³⁰¹⁰ allele to potentialize dar1RNAi efficiency and facilitate the subsequent analysis. As shown by longitudinal cryosections along the proximo-distal axis of the sr-gal4,dar1^3010/+>UAS-Lifeact.GFP, UAS dar1RNAi legs, long tendons in all segments were often shortened or even missing, and so muscle pattern was strongly affected (Figure 3). In control legs, major muscles labeled by phalloidin, displayed a feather-like pattern with each fiber attached on one side to a long tendon and on the other one to an individual cuticle muscle attachment site (cMAS). By contrast, when long internal tendons were affected, both ends of a muscle fiber were anchored to a cMAS, generating a transverse fiber. This is clearly visible in the ventral tibia where the long tendon appears much shorter in the visualized cryosections (Figure 3E). Thus, in the proximal region of this segment, both ends of the muscle fibers are attached to cMAS, whereas in the distal part, the fibers still connect to the shortened remaining ventral long tendon (Figure 3F). We could also observe a highly disorganized muscle pattern in the dorsal femur that coincided with no visible tilt in this cryosection. Our observations thus indicate that when dar1 expression is specifically lowered in leg tendon precursors, resulting long tendons are severely affected, leading to abnormal muscle patterning. Moreover, even in the absence of long tendons, the muscle fibers can still attach to cuticle via cMAS, suggesting that the latter, like those of flight and embryonic muscles, do not required Dar1 activity.

To better characterize the role of Dar1 in long tendon development, we focused our analysis on the time when tendon precursors started to express dar1, from the L3 larval stage to the early hours of pupation. We specifically analyzed the tarsal lt, extending from pre-tarsus to femur, and the tilt in the dorsal femur. These two tendons form earliest: they are specified at the early third larval instar and then invaginate and elongate until the first hours of metamorphosis (Laddada et al., 2019; Soler et al., 2004). Tendon specific depletion of dar1 led to apparent shortening of these elongated structures in early metamorphosis (Figures 4A–F). To quantify this phenotype, we measured the relative size of the lt and tilt (for measurement details see Materials and Methods) at three developmental time points (late L3, 2 and 4 h APF) in control (sr-gal4,dar1^3010/+>UASLifeact.GFP) and dar1 KD (sr-gal4,dar1³⁰¹⁰/+>UAS-Lifeact.GFP, UASdar1RNAi) leg discs. For each of these developing tendons, we observed a significant reduction of their length at the three different time points when dar1 expression was lowered (Figures 4G,H). The tarsal tendon was very significantly shortened as early as late L3, whereas the difference in size of the tilt in dorsal femur was more pronounced at later stages (4 h APF). This is probably because tarsal lt is specified and starts to elongate slightly earlier than the one in the dorsal femur (Soler et al., 2004) and is therefore longer at a given time point. Thus, the length difference compared with control was greater in earlier stages for the tarsal lt than for the tilt in dorsal femur. We infer that the apparent smaller size of the tendons after dar1 KD becomes more and more significant as the tendons elongate, supporting a potential function of Dar1 in this process. Likewise, 2 h APF leg discs from homozygote dar1 ³⁰¹⁰ rare escapers (less than 1% of the larva reach the pupal stage and none of them survive for more than a few hours APF), stained against discs large (dlg) septate junction marker, showed no visible elongating structure (Figures 5A,B). Remarkably, in the location of the most distal tarsal segment where lt normally starts to develop, we could still see an accumulation of dlg protein prefiguring the local folding of the epithelium and the lumen formation (Figures 5A,B). These observations indicate that Dar1 is not needed for initial steps of epithelium invagination but is then required for the tube-like tendon elongation.

Besides this elongation defect, dar1 RNAi expression in long tendon precursors results in a severe depletion of filopodia as revealed by UAS-Lifeact.GFP expression (Figures 5C–F). For instance, the total number of filopodia at the tip of the tarsal lt in control 2 h APF disc was systematically greater than 30 (n = 10), whereas in dar1 KD leg discs, tarsal lt systematically displayed fewer than five remaining protrusions (n = 17). This result indicates that the reduction of dar1 expression directly or indirectly affects the formation of actin-rich filopodia. Moreover, while the tendon lumen in control leg discs is underlined by apical Lifeact.GFP accumulation, in dar1 KD leg disc, we observed a fluorescent dotty pattern reflecting a general actin disorganization or mislocalization.

Taken together, our results show that Dar1 participates in the cellular events required for actin cytoskeleton organization and that dar1 KD reduces the number of cytoplasmic protrusions.

The fluorescence-activated cell sorting (FACS) protocol was adapted from (Harzer et al., 2013). Briefly, approximately 50 UAS-Lifeactin.GFP; sr-gal4/TM6b, tb white pupae (0 h APF) were dissected on ice to collect 250–300 leg imaginal discs in M3 (S3652 Sigma Aldrich) complemented medium. Leg disc cells were dissociated in collagenase (P4762-Sigma Aldrich) and papain (C267-Sigma Aldrich) solution for 1 h at 30°C, 300 rpm (Thermomixer Eppendorf), with additional mechanical stirring. After filtering, we used a FACSAriaTM device (4 °C, 20 psi, nozzle ∅ 100 μm) to first sort a batch of cells (roughly 50,000 cells per sample) independently of fluorescence, which we used as input (IP) and we then sorted the tendon cell (GFP+) population based on GFP fluorescence. Samples were directly collected in Trizol reagent (Invitrogen, 15596026). RNA extraction was performed with Zymo Quick-RNA microprep kit (R1051), sample RNA quality and quantity were estimated with QuBit (ThermoFisher, Qubit RNA HS Assay Kit, Q32852) and BioAnalyzer (Agilent, Agilent RNA 6000 Nano Kit 5067-1511), and sample specificity was analyzed by qPCR. Samples were stored at−80°C. For the detailed protocol, see Supplementary Material S1.

Extracted total RNA from GFP+ and IP cells from three different replicates were sent to Heidelberg genomic platform EMBL for high-throughput mRNA sequencing (NextSeq 500/Illumina). They generated mRNA libraries with NEB RNA Ultra kit (New England Biolabs E7770 L) and dToligos probe were used to target mRNA for cDNA synthesis. Single-end multiplex was performed. Bioinformatic analysis was performed by Dr Yoan Renaud using FastQC and Bowtie2 software (reference genome Dm6). FPKM and differential expression between IP and GFP + samples were determined using R script DEseq2 software. Sample correlations are given in Supplementary Material. Differential expression was performed using R-package DEseq2. Fold change (FC) between GFP+ and IP samples was computed using their normalized raw counts. GEO accession number: GSE169313.

53 UAS-RNAi lines (see Supplementary Table S2) from two separate stock centres, VDRC and BDSC (Dietzl et al., 2007; Ni et al., 2009) were crossed with UAS-dicer2, UAS-CAAXmcherry; sr-gal4/TM6b,tb, hu line and screened for lethality and/or climbing defect. 5–10 males carrying UAS-RNAi were mated with 15–20 females. Crosses and egg laying were performed at 25°C; 48 h after egg laying, larvae were transferred at 28°C.

Embryonic or early larval stage lethality rate was measured by counting the number of non-tubby over tubby third instar larvae. Percentage of lethality was calculated by applying formula (1 − (number of non-tubby larvae)/(number of tubby larvae)) × 100. Crosses were considered as positive for embryonic (early larval stage) lethality when lethality was higher than the 20% arbitrarily chosen threshold. To assess the percentage of lethality during metamorphosis, we applied formula (1 − (the number of non-tubby, non-humeral emerging adult flies)/(total number of initial non tubby pupae)) × 100. Crosses with a higher score than 20% were considered positive candidates for metamorphosis lethality.

When sr-gal4>UAS-RNAi adult flies survived, we performed a climbing test adapted from the RING (Rapid Iterative Negative Geotaxis) assay published by Gargano et al. (Gargano et al., 2005). Up to ten young non-balanced flies (48 h old) were transferred to an empty vial and allowed to recover at room temperature for at least 2 h. The vial was then rapped sharply on a table three times in rapid succession to initiate negative geotaxis responses. The flies’ positions in the vial were digitally captured after 5 s and we calculated the percentage of flies remaining in the bottom third of the vial. For each cross, two replicates (two vials of ten flies) were measured three times. We calculated the mean of these six trials. When this mean was equal to or greater than 30%, the corresponding cross was scored as “climbing defect”.

RNAis directed against stripe (100% embryonic lethality) and mcherry (0% embryonic lethality, 5% metamorphosis lethality, 15% climbing defect) were considered as positive and negative controls, respectively.

The following primary antibodies were used: guinea pig anti-Dar1 (1:250, courtesy of B. Ye), guinea pig anti-Stripe (1:500, from T. Volk), mouse anti-Fas III (1:500, Developmental Studies Hybridoma Bank [DSHB]), rabbit anti-Twist (1:500, our lab), mouse anti-Dlg (1:500, DSHB), chicken anti-GFP (1:500, Abcam), rat anti-DE-cadherin (1:500, DHSB), rabbit anti-Dcp1 (1:100, Cell Signaling), rabbit anti-pH3 (1:1000, Invitrogen), and anti-mcherry (rabbit, Abcam). Muscle fibers were visualized using cy3 or cy5-conjugated phalloidin (1:1000, Invitrogen) and nuclei using DAPI (ThermoFisher D1306). Secondary antibodies (Jackson) anti-rabbit, anti-guinea pig, anti-chicken and anti-mouse conjugated to Alexa488, cy3 or cy5 fluorochromes were used (1:500). Immunohistochemistry experiments were performed on samples fixed in 4% paraformaldehyde (PFA) for 20 min, rinsed in 0.1% PBS-Triton and blocked in 10% horse serum before immunostaining with primary antibodies (at 4°C, overnight). The samples were then rinsed and incubated with appropriate secondary antibody (at room temperature, 1 h). For cryosections, we dissected adult leg and fixed them from 40 min to 1 h in 4% PFA. Fixed samples were then incubated in sucrose solution (30%) at 4 °C overnight. Adult legs were then laid on the bottom of a plastic well and carefully covered with Neg-50TM gel (Richard-Allan Scientific). The preparation was frozen at −80°C to set the gel. Sections were cut with a cryostat at 4°C (thickness 18–20 um). The sliced samples were stained with phalloidin before imaging. Immunostaining was visualized on an inverted SP8 Leica confocal microscope, and images were analyzed with Imaris 7.6.5 software.

Tendon size was determined by normalizing the length of the tendon over the whole size of the imaginal disc. For this purpose, we used the bounding-box function of the Imaris software to semi-automatically generate a 3D mesh of the imaginal disc (using FasIII staining). The disc was viewed as an ellipse with major axis (length) X and minor axis (width) Y. Discs were considered as flat objects, since at the developmental stages of analysis the thickness (Z) dimension has no significant impact on the measurement. From the ellipse-disc we computed the radius R of each corresponding disc (R = √X × Y⁄2). The lengths of tendons were measured using the function MeasurementPoint by manually following their pathways. Finally, we normalized the tendon length by dividing the length by the disc radius R. Statistical analyses were carried out using Prism software with an analysis of variance (ANOVA) between the different conditions and the different stages. Statistical significance between control and knockdown conditions was estimated using the Bonferroni test.

0 and 5 h APF leg discs from sr-gal4,dar1^3010/+>UAS-mcherryNLS and sr-gal4,dar1^3010/+>UAS-mcherryNLS,UAS-dar1RNAi pupae were dissected and stained with DAPI for 1 h. 0.5 μm confocal sections of long tendon of tarsi and dorsal tendon in femur (tilt) were performed. Images were processed with Imaris software and numbers of mcherryNLS and DAPI-positive cells per tendon were determined automatically using the function Spots. The automatic cell counting was subject to manual adjustement.

Statistical significance was estimated using the Bonferroni or Welch test.