Gelsolin and dCryAB act downstream of muscle identity genes and contribute to preventing muscle splitting and branching in Drosophila

A combinatorial code of identity transcription factors (iTFs) specifies the diversity of muscle types in Drosophila. We previously showed that two iTFs, Lms and Ap, play critical role in the identity of a subset of larval body wall muscles, the lateral transverse (LT) muscles. Intriguingly, a small portion of ap and lms mutants displays an increased number of LT muscles, a phenotype that recalls pathological split muscle fibers in human. However, genes acting downstream of Ap and Lms to prevent these aberrant muscle feature are not known. Here, we applied a cell type specific translational profiling (TRAP) to identify gene expression signatures underlying identity of muscle subsets including the LT muscles. We found that Gelsolin (Gel) and dCryAB, both encoding actin-interacting proteins, displayed LT muscle prevailing expression positively regulated by, the LT iTFs. Loss of dCryAB function resulted in LTs with irregular shape and occasional branched ends also observed in ap and lms mutant contexts. In contrast, enlarged and then split LTs with a greater number of myonuclei formed in Gel mutants while Gel gain of function resulted in unfused myoblasts, collectively indicating that Gel regulates LTs size and prevents splitting by limiting myoblast fusion. Thus, dCryAB and Gel act downstream of Lms and Ap and contribute to preventing LT muscle branching and splitting. Our findings offer first clues to still unknown mechanisms of pathological muscle splitting commonly detected in human dystrophic muscles and causing muscle weakness.

www.nature.com/scientificreports/ demonstrated that Eve, Lb and Slou iTFs regulate the number of fusion events by setting expression levels of genes that act as identity realisators in a muscle-specific manner 13,15 . However, the identification of "realisator genes" has so far been limited to only a few examples, owing to the technical challenges of detecting gene expression in specific muscle populations during embryogenesis.
To further analyse diversification processes and identify genes acting downstream of iTFs, we optimised translating ribosome affinity purification (TRAP) 16 to small subsets of FCs and developing muscle precursors 17 . TRAP-purified mRNA profiling followed by bioinformatic analysis and generation of temporal transition profiles identified muscle subset-specific translatome signatures with Gelsolin (Gel) and dCryAB as new identity realisator genes controlling shape-and size-related properties of Lms-expressing muscles.
Intriguingly, dCryAB and Gel act downstream of Ap and Lms and their loss-of-function phenotypes recall dystrophic muscle branching/splitting in humans 18 .

Results
Translational profiling (TRAP) of muscle subsets identifies dCryAB and Gelsolin expressed predominantly in LT muscles. TRAP is based on the polysome capture of the GFP-tagged ribosomes with their associated mRNAs (Fig. 1A). Here we tagged polysomes with UAS-Rpl10A-GFP in two muscle subsets using Slou-GAL4 or Lms-GAL4 drivers (Fig. 1B) and in all embryonic muscles using Duf-GAL4. For each muscle subset, translational profiling was performed on embryos collected from three developmental time windows T1: 7-10 h AEL, T2: 10-13 h AEL and T3: 13-16 h AEL covering the main muscle development steps. To assess the specificity of TRAP-based muscle targeting, we analysed GO terms. We found that the up-regulated genes fitted muscle-related GOs while GO categories associated with the list of down-regulated genes were not related to muscle developmental processes (Fig. 1C). Gene expression profiling with total embryonic RNA (input fractions) as a reference was used to identify differential gene expression and perform spatial and temporal gene clustering. A significant portion of upregulated genes (FC > 2, p < 0.05) turned out to be common to the restricted muscle populations (Slou-and Lms-positive) and the overall (Duf-positive) population (dark colors in Fig. 1D). The percentage of specific transcripts (light colors) remained relatively constant until the latest time point where the proportion of Lms-specific transcripts increased slightly compared to Slou (29% versus 23% respectively , Fig. 1D). This difference may be due to Lms-expressing muscles being less heterogeneous than Slou -positive ones and expressing specific set of genes for their terminal differentiation. Finally, to test muscle type-specific gene expression, we generated volcano plots (Fig. S1A,B) and observed that genes with previously characterised expression patterns in muscle subsets displayed expected up-or down-regulation. For example, in Lms-positive muscles, ap and lms gene transcripts were enriched, whereas slou and org1 transcripts, specific for Slou-positive muscles, were depleted (Fig. S1B). We also tested whether TRAP would detect "low expression" genes. To do so, we crossed our lists of enriched and depleted transcripts with modEncode datasets 19 and found that more than 30% of up-regulated TRAP-ed genes entered the "low expression" modEncode category, whereas most depleted transcripts fitted the "high expression" modEncode category (Fig. S1C,D). TRAP-based translational profiling of muscle subsets was thus specific for the targeted muscle populations and sensitive enough to detect low transcript levels.
We then applied temporal transition profiling to identify clusters of genes showing similar dynamics of expression patterns, thus potentially under common upstream regulatory cues. We considered that this approach could be applied to TRAP datasets to identify novel muscle identity realisator genes acting downstream of iTFs 15 . The generated temporal transition heatmap for Slou-and Lms-positive muscles revealed clusters of genes with several expression behaviours (Fig. 1E).
Here we focused on gene clusters showing two contrasting transition behaviours, "down-down" (Cluster 1) and "up-up" (Cluster 2). Cluster 1 genes were characterised by enrichment of GOs associated with "chromatin binding" (Fig. 1F). Among Cluster 1 genes, we found twist and several Notch pathway-involved genes whose Heatmap was generated using R package Pheatmap version 1.0.12 ; URL: https:// cran.r-proje ct. org/ web/ packa ges/ pheat map/ index. html. Tr1 represents gene expression transition from the time window T1 to T2 and Tr2 the transition from T2 to T3. Cluster 1 (green) and cluster 2 (red) genes are indicated. (F) GO comparison of genes belonging to Cluster 1 and Cluster 2. Size of the circles represents the ratio of genes present in each GO category and color code the associated p-value representing the probability of seeing at least x number of genes out of the total n genes in the list annotated to that GO term. (G) A zoomed view of the transition profile heatmap restricted to genes from cluster 2 that belong to the "actin binding" GO category. Respective Tr1 and Tr2 fold changes are indicated on the heatmap. (H) In situ hybridisation showing that Gel and dCryAB that are part of the "actin binding" GO class are predominantly expressed in LT muscles. Lateral views of stage 15 embryos are shown. Anti-actin or anti-β3 tubulin antibodies are used to reveal muscle pattern. Schemes of muscles in an abdominal segment are shown with Gel and dCryAB expression indicated by a color code: red -high expression and light pink -low expression levels.   (Fig. 1G). We found this sub-cluster of particular interest as it contains genes with similar biological functions and differential Lms-versus Slou-transition profiles. Among them, Gel and dCryAB with "up-up" transition profile in the Lms subpopulation ( Fig. 1G) are both preferentially expressed in Lms-positive LT muscles (Fig. 1H). Drosophila Gel belongs to the conserved Gelsolin/Villin family of www.nature.com/scientificreports/ actin interactors 20 with actin depolymerisation activity 21,22 . dCryAB codes for a small heat shock protein (sHSP) carrying an actin-binding domain and known to interact with cheerio/filamin 23  dCryAB promotes regular shapes and prevents branching of growing LT muscles. To test whether dCryAB helps set LT muscle features, we generated null allele (dCryAB HR ) using CRISPR mutagenesis (Fig. 3A). dCryAB loss-of-function turned out to be homozygous lethal with mutants surviving until the late 3rd larval instar. Compared to wild-type, the late stage embryos devoid of dCryAB ( Fig. 3B-D) showed dissociations between LT1 and LT2 and/or LT2 and LT3 muscles (32% of segments) and irregular growth of LTs (28% of segments). The partial dissociation of LTs was in several instances associated with the irregular LT shapes (Fig. 3B, C) and branched LT muscle extremities (8% of segments) (Fig. 3C, right panel). We also noted a reduced number of LTs in 6% of segments (Fig. 3D). Branched LT muscles are also occasionally detected in ap and lms mutant contexts ( Fig. S4) indicating that dCryAB is involved in preventing LTs branching downstream of LT iTFs. The irregular LT growth in dCryAB mutant embryos raised the question of whether dCryAB could impact on LT interactions with tendon cells and with motor neurons. By the end of embryonic stage 15, ß-PS integrin accumulates at the extremities of muscles and at the surface of their cognate attachment sites, promoting the formation of myotendinous junctions (MTJs). Accordingly, in wild type stage 15 embryos, ß-PS integrin labels ventral and dorsal ends of LTs ( Fig. 3E-E"). However, in dCryAB mutants this ß-PS accumulation was hardly detected, suggesting impaired MTJs formation ( Fig. 3F-F"). Because in late dCryAB mutant embryos we do not detect an LT muscle detachment phenotype, observed in the ß-PS loss-of-function context, we conclude that dCryAB mutation impairs but does not prevent ß-PS accumulation at LT MTJs. Consistent with this, in 2nd instar dCryAB mutant larvae, LTs, including those with branched ends, appear attached (Fig. 3G,H).
We then tested whether the LTs in dCryAB mutants were properly innervated. In wild type embryos, the dorsal branch of the SNa nerve innervating LTs defasciculates at the level of LT2 and grows dorsally within a gap between the ventral extremities of LT2 and LT3 (Fig. 3I-I"). In dCryAB mutants, most segments have wild type LT innervation, but in those with forked LT ends (essentially seen at the ventral extremity of LT2), the gap between LT2 and LT3 is filled, preventing defasciculation of SNa and LT innervation ( Fig. 3J-J").
Thus, by coordinating LT muscle growth, dCryAB ensures timely accumulation of ß-PS integrin and optimal LTs attachment. The capacity of dCryAB to control LT shapes and prevent their branching facilitates proper LT innervation.
Gel mutant embryos bear an increased number of LTs through fibre splitting. The CRISPR mutagenesis in the 5' region of Gel resulted in two different null mutations, Gel9.3 and Gel9.8 (Fig. 4A), both leading to a premature stop codon. Because Gel9.3 and Gel9.8 exhibited similar molecular lesions, were homozygous viable, and showed equivalent LT muscle phenotypes (Fig. 4B), we chose one of them, Gel9.3, for further analyses.
Despite irregularity in LT growth (32% of segments) and some LT dissociation phenotypes (14% of segments), Gel mutants also showed an increased number of LT muscles (17% of segments) (Fig. 4B-D), a phenotype previously described (6) and detected in particular in ap mutants (Fig. S4B), however not observed in the dCryAB loss of function context (Fig. 3).
To follow formation of supplementary LTs in vivo, we recombined Gel9.3 with the Lms > LifeActinGFP (LAGFP) LT sensor line. We first confirmed that Gel9.3;Lms > LAGFP embryos form the supernumerary LTs, which become individualised with connectin-labeled cellular membranes (Fig. 4F,F'). We then performed time lapse experiments encompassing mid-embryogenesis on Lms > LAGFP (Fig. S5A, control) and on Gel9.3;Lms > LAGFP embryos in which splitting occurs (Fig. S5B). The example of splitting presented (Fig. 4E,  Fig. S5B) concerns LT3 which grows asynchronously, expands and subdivides progressively into two fibres with separate extremities. This aberrant morphogenesis appears to have a functional impact, as at the beginning of larval life, the striated sarcomeric pattern of split LT muscles is severely impaired (Fig. 4G,G'), indicating that their contractility is compromised. Thus, the supernumerary LTs in Gel mutants arise from enlarged fibres that eventually split. Another, interesting feature is that split LTs extremities accumulate ßPS-integrin (Fig. 5A), suggesting that LT identity information ensuring choice of attachment sites is transmitted during splitting.
Gel controls LT muscle size by preventing excessive myoblast fusion. On performing time lapse experiments, we observed that splitting occurred during the developmental period in which muscle fibres grow by fusing with surrounding myoblasts, and that split LTs were enlarged compared to non-split neighbours (Fig. 4E). Our previous findings showing that muscle size depends on the number of fusion events 15,24 thus raised the question of whether LT splitting could be associated with increased fusion. This appears to be the case since the LT-targeted increase in fusion by overexpressing Duf could lead to splitting (Fig. 5B,C). On the other hand we found that Gel is expressed in developing LT muscles but not in FCs (Fig. S2)  www.nature.com/scientificreports/ LT FCs in Gel mutants remains unchanged (Fig. 5D), suggesting that the supernumerary Gel-devoid LTs arise by an aberrant fusion-involving muscle morphogenesis. Indeed, the number of Mef2-positive nuclei in LT1-LT4 at embryonic stage 16 was significantly higher in Gel mutants including the transheterozygous Gel9.3/Gel9.8 context compared to controls (Fig. 6A-D). We then sought to determine whether the LT-specific increase in fusion events observed in Gel mutants could impact on the fusion programs of neighbouring muscles. The number of myonuclei in immediate LT neighbours in the SBM and LO1 muscles (but not in more ventrally located VT1) was reduced, indicating that local availability of FCMs could impact on fusion programs (Fig. 6E,F). Muscle splitting observed in Gel mutants thus results from excessive fusion, with late fusion events that could be detected associated with LTs showing a split phenotype (Fig. 6B, right panel). The capacity of Gel to negatively regulate fusion was confirmed by the reduced number of myonuclei in LTs in which Gel was prematurely activated (Fig. 6D) and by the large number of unfused myoblasts seen in embryos with ectopic Gel expression in all muscles (Fig. 6G). Thus, we propose that Gel triggers fusion arrest in LTs. In Gel loss of function context LTs continue to grow by fusion and eventually split (Fig. 6H).

Discussion
TRAP was first developed to isolate polysome-associated mRNA from a subset of neurons in mice 16 and was later adapted to other model organisms including Xenopus 25 , zebrafish 26 and Drosophila 17,27 . Here we applied TRAP to determine the first translatomic signatures underlying diversification of muscle types, and we identified Gel and dCryAB as new LT muscle identity realisator genes. dCryAB contribute to preventing LTs branching, and Gel plays a role in LT muscle size control by limiting the number of fusion events. Consequently, supernumerary myonuclei are present in Gel-devoid LTs, which eventually split. Both splitting and branching are low penetrance phenotypes (17% and 8% of segments, respectively), indicating that Gel and dCryAB are not the sole identity realisators that prevent these aberrant growth-related LTs features.
Formation of branched muscle fibres has recently been reported as a result of adversely affected muscle identity 28 , and supernumerary LTs were also detected in ap and lms mutant embryos 6 . Here we report that split and branched muscles are occasionally detected in both LT iTF and Gel or dCryAB mutant embryos indicating that a muscle identity-dependent shape and size control system operates in developing muscles.
In humans, CryAB mutations are associated with desminopathies in which aberrant muscle fibres with branched morphology are frequently detected 29 . Gelsolin mutations cause amyloidosis, characterised by the toxic accumulation of protein aggregates, which can lead to an inclusion body myositis (IBM)-like phenotype with necrotic and centronuclear split fibres 30,31 .
Functional analyses of dCryAB and Gel in Drosophila embryos indicate that muscle fibres branch or split when the identity realisators for these muscle are not properly activated.
Reduced levels of ß-PS integrin at the extremities of dCryAB-devoid LTs suggests weak interactions with attachment sites and could explain observed LTs overgrowth and dual attachment of branched fibers. These dCryAB loss of function phenotypes could result from inappropriate actin cytoskeleton dynamics in growing LTs and/or affected function of cheerio/filamin, a direct dCryAB interactor 23 . The observation that dCryAB protein accumulates at sub-membrane areas and at LT myotube ends suggests it could contribute to preventing non-polarized, branched LT growth.
In advanced stages of muscular dystrophies, a large subset of muscle fibers shows longitudinal splitting, described already more than forty years ago (34). After chemical or physical injury, split fibers form also in undergoing regeneration wild type muscles (35) suggesting a link between myoblast fusion-involving regeneration (chronic in dystrophic context) and muscle splitting. However, despite critical impact on dystrophic muscle cytoarchitecture, on vulnerability to contraction-induced damage and on muscle weakening, mechanisms of muscle fibers splitting remain unexplored (36,37).
Our finding in Drosophila that local increase in fusion occuring in Gel mutant embryos causes split fiber phenotype points to the excessive fusion as a prerequisit of splitting. Because Gel protein displays actin-severing properties and as we show here partially colocalises with actin in growing LTs, we speculate it could affect submembrane F-actin sheet and reduce the required for fusion myotube rigidity.  To establish KO lines, molecular characterisation of target loci was performed as described 33 . Briefly, genomic DNA was extracted from individual larvae by crushing them in 20 μL QuickExtract solution (Cambio) and releasing the DNA in a thermomixer according to the supplier's instructions. We used 1 μL (previously diluted five times) of the supernatant in 25 μL PCR reactions. PCR products were then sequenced by Sanger. Indels can be observed as regions with double peaks in heterozygous flies, corresponding to wild type and mutated allele respectively. In the case of dCryAB, homologous recombination events were recovered by selecting flies with red eye fluorescence.

RNA extraction and RT-qPCR. mRNA was extracted from Lms, Slou and Duf muscle populations using
TRIzol reagent (Invitrogen) following the manufacturer's instructions. RNA quality and quantity were then assessed using Agilent RNA 6000 Pico kit on Agilent 2100 Bioanalyzer (Agilent Technologies).
Microarray analysis. Agilent 8 × 60 K probe (60-mer) gene expression microarrays were used. We assessed the quality of triplicates using Pearson's correlation test. Correlations for all conditions were 85% or higher. The microarray data were quantile-quantile normalised. Gene expression data from Slou-, Lms-and Duf-positive cells were compared to the whole embryo datasets to generate lists of genes differentially expressed, fold change ≥ 2, p < 0.05. GO Princeton software was used to assign GO classification. We then compared Lms-and Slou-positive cells to Duf-expressing cells to make two lists of differentially regulated muscle-specific genes, fold change ≥ 2, p < 0.05. We computed and compared GO biological processes from these two lists using an R package cluster profiler.
Temporal transition heatmap. Translational temporal profiles from Slou and Lms at the three time points were converted to "transition values" defined as log ratios between T2 and T1, T3 and T2. Transition  www.nature.com/scientificreports/ values were considered as three discrete classes: upregulated (> 1), stable (between − 1 and 1), and downregulated (< 1). Thus expression profiles from Lms and Slou muscles, which contain three temporal windows, were converted into vectors of two transitions (Tr1, Tr2), which allow the determination of correct gene behaviour. For example, the profile "red-red" group genes whose RNA level increases between T1 and T2 (Tr1), and then continues to increase between T2 and T3 (Tr2).
In situ hybridisation and immunostaining. Embryos were dechorionated and fixed in 4% paraformaldehyde/heptane for all immunohistochemistry. Fluorescent in situ hybridisation with a TSA amplification system (Perkin-Elmer) and immunohistochemistry was as described previously 13 .To generate the RNA probe for Gel (primers used: 5'-5' AAT CGA CTC CGT GGT GAC TC-3' and 5'-GGG AGG CCA AAG ATG AGC TGTC-3') the corresponding DNA sequences were cloned by PCR in pCR II topo vector. The corresponding anti-sense RNAs were transcribed in vitro using T7 or SP6 RNA polymerase. For dCryAB, Gold collection clone GH01960 was used to generate RNA probes. For fluorescent staining, the following antibodies were used: rabbit anti-β3 tubulin TRAP experiment. RPL10aGFP-tagged embryos were collected, and messenger RNAs from the different muscle populations isolated as described 17 . Micro-array data generated in this study were deposited to GEO database : GSE137443: GSM4079386 to GSM4079439.
Live imaging. The Lms-Gal4; UAS-lifeActGFP (Lms>LAGFP) double transgenic line was generated and used for time lapse imaging of LT muscle formation in the gel mutant context. Image acquisition was performed on manually aligned living embryos at 21 °C using an inverted Leica SP8 confocal microscope. The time interval between acquisions was set to 3 min and the acquisition time was 3-4 h. Movies were generated and analysed using Imaris software (Bitplane).