The Drosophila RNA Helicase Belle (DDX3) Non-Autonomously Suppresses Germline Tumorigenesis Via Regulation of a Specific mRNA Set

DDX3 subfamily DEAD-box RNA helicases are essential developmental regulators of RNA metabolism in eukaryotes. belle, the single DDX3 ortholog in Drosophila, is required for fly viability, fertility, and germline stem cell maintenance. Belle is involved both in translational activation and repression of target mRNAs in different tissues; however, direct targets of Belle in the testes are essentially unknown. Here we showed that belle RNAi knockdown in testis cyst cells caused a disruption of adhesion between germ and cyst cells and generation of tumor-like clusters of stem-like germ cells. Ectopic expression of β-integrin in cyst cells rescued early stages of spermatogenesis in belle knockdown testes, indicating that integrin adhesion complexes are required for the interaction between somatic and germ cells in a cyst. To address Belle functions in spermatogenesis in detail we performed cross-linking immunoprecipitation and sequencing (CLIP-seq) analysis and identified multiple mRNAs that interacted with Belle in the testes. The set of Belle targets includes transcripts of proteins that are essential for preventing the tumor-like clusters of germ cells and for sustaining spermatogenesis. By our hypothesis, failures in the translation of a number of mRNA targets additively contribute to developmental defects observed in the testes with belle knockdowns both in cyst cells and in the germline.


Introduction
DEAD-box RNA helicases are important developmental regulators of gene expression. They operate by remodeling the local secondary structure of RNAs and RNA-protein complexes in an ATP-dependent manner to provide subsequent association of RNA-binding proteins to their RNA targets. This implicates a crucial role of DEAD-box RNA proteins in regulating ribonucleoprotein (RNP) complexes. In line with this, DEAD-box RNA helicases participate in practically every stage of RNA metabolism, including transcription, processing, translation, ribosome biogenesis, splicing, RNA localization, RNA decay, and turnover [1][2][3]. DEAD-box proteins of the large DDX3 subfamily are found to be conserved from yeast to mammals. The human genome contains two paralogous DDX3 genes, one located in the X chromosome (DDX3X) and another in the non-recombining region of the Y-chromosome (DDX3Y or DBY) [4]. Expression of DBY is restricted to male germ cells [5]. DBY function is required for early stages of testis development in human fetal germ cells, including the

Immunofluorescence Staining and Confocal Microscopy
Testes of adult males were dissected in phosphate-buffered saline (PBS) at 4 • C, washed with PBT (1×PBS, 0.1% Tween 20), and fixed in 3.7% formaldehyde and PBT for 30 min at room temperature. All the following procedures were carried out as described previously [36]. Staining was detected by laser scanning confocal microscopy using a Carl Zeiss LSM 510 META machine (Carl Zeiss). All images were taken with a z-resolution of 1 µm. The obtained pictures were imported into Imaris ® 5.0.1 (Bitplane AG) for subsequent processing. Counting of spectrosomes and the number of specific testis cells on the confocal images was carried out using Imaris software. Data were analyzed using Statistica version 10 (StatSoft) and RStudio.

RNA Extraction, Reverse Transcription, and Real-Time Quantitative PCR
For identification of the transcriptional level of basic adhesion complex components RT-PCR analysis was performed with primers shown below. Briefly, total RNA was isolated from sets of 50-100 pairs of dissected testes, using TRIzol Reagent (Invitrogen) according to the manufacturer's recommendations. cDNA was synthesized using random hexamers and SuperScript II reverse transcriptase (Invitrogen). cDNA samples were analyzed by real-time quantitative PCR using the incorporation of SYTO-13 (Invitrogen). Thermal cycling consisted of 5 min at 95 • C, followed by 45 cycles of denaturation (94 • C, 20 s), annealing (64 • C, 20 s), extension (72 • C, 20 s), and a final extension of 5 min at 72 • C. All experiments were performed with at least three independent RNA samples; each sample was analyzed in duplicate. Statistical analysis for the group comparison was performed by a paired Student's t-test.

Electron Microscopy
Testes of 0-to 3-day-old males were dissected and fixed in sodium cacodylate buffer with 2.5% glutaraldehyde at room temperature for 2 h. Samples were washed in sodium cacodylate buffer, fixed in 2% OsO 4 for 2 h at 4 • C, washed, and, dehydrated by transferring through solutions of increasing alcohol concentration (30%, 50%, 70%, 90%, and 100%) for 15 min at room temperature followed by acetone for 15 min. Resin (Epon 812 with DDSA and MNA) infiltration was carried out according to the following protocol: resin-acetone (1:3), resin-acetone (1:1), resin-acetone (3:1) at room temperature for 1 h, resin alone for 12 h, and resin with catalysator DMP-30 at 37 • C for 24 h and at 60 • C for 48 h.

CLIP-seq Analysis
Freshly dissected testes of 0-3 day-old yw males (500 pairs per one library) were cross-linked on ice with UV radiation (254 nm) three times at 600 mJ/cm 2 in the presence of 200 mkl cold PBS (in a 1 cm 2 glass dish). The testes were transferred to 1.7-mL polypropylene microcentrifuge tubes and pelleted at 300 g at 4 • C. The supernatant solution over the organs was carefully removed by a tip and cross-linked material was stored at -70 • C until further use. Cells lysates were obtained using pre-chilled Dounce homogenizer on ice in a cold lysis buffer (containing 50 mM Tris-HCl, pH 8.0, 200 mM NaCl, 2 mM MgCl 2 , 1 mM DTT, 0.5% Nonidet P-40 (NP-40), in the presence of protease and phosphatase inhibitors (Sigma) in diethyl pyrocarbonate (DEPC)-treated water with the addition of Ribolock (Thermo Fisher Scientific)) by two series of 15 strokes of B pestle with a 1-2 min incubation between the series. The lysates were transferred to 1.7-mL polypropylene microcentrifuge tubes and were treated by RQ1 RNase-Free DNase (Promega) 40 U/mL for 10 min at 37 • C with vortexing. To generate RNA fragments of appropriate length we treated our lysates by RNAse A for 7 min at 37 • C. Immediately after that the lysates were supplemented by Ribolock. The lysates were subjected to centrifugation at 13,200 rpm at 4 • C for 20 min. The supernatants were transferred to new tubes and diluted to protein concentration of 8-10 mg/mL with lysis buffer (containing 50 mM Tris-HCl, pH 8.0, 200 mM NaCl, 2 mM MgCl 2 , 1 mM DTT, 0.5% Nonidet P-40, protease and phosphatase inhibitors (Sigma), Ribolock). We used 150 µL of mixed 50%-slurry of anti-mouse IgG beads (Invitrogen) for each tube of the cross-linked lysate. The beads were previously washed two times with 500 µL cold phosphate-buffered saline (PBS) supplied by Tween 20 (1 × PBS, 0.1% Tween 20 (PBST)) and incubated with 8 µL anti-Belle antibodies (described in [25]) for CLIP library preparations or 8 µL normal mouse serum for negative control in 200 mkl PBST during 20 min at room temperature on a rotator. The antibody-coated beads were washed and pre-equilibrated with lysis buffer. Then the clarified lysates were added to the experimental and control tubes. After incubation for 40 min at RT on a rotator, washing was performed four times with 1 mL washing buffer NT2 (50 mM Tris-HCl, pH 8.0; 300 mM NaCl; 1 mM MgCl 2 ; 1 mM DTT; 0.05% NP-40) and once with 200 µL (without Ribolock). The beads were pre-equilibrated with buffer for T4 RNA ligase 2 truncated (New England Biolabs (NEB)). Dephosphorylation of RNA 3 -ends was performed on-bead by Calf Intestinal Alkaline Phosphatase (Roche) 1U/µL for 10 min at 37 • C with frequent agitation. After that the beads were washed by 1 mL NT2 buffer with Ribolock two times. A previously pre-adenylated 3 -adapter (5 DNA Adenylation Kit (NEB) was used for preparation) was ligated with RNAs on-bead by T4 RNA ligase 2 (NEB) 400 U in 30 µL of ligation reaction 2 h at 25 • C on a thermal mixer. The beads were washed three times by 1 mL NT2 buffer with Ribolock. 5 -end RNA labeling was done by treating the samples with 4 µL PNK enzyme (10 U/mkl; NEB), 1× PNK B buffer (NEB) and 2 µL γ32P-ATP (10 mCi/mL) for 10 min at 37 • C with mixing at 1000 rpm for 15 s after each 3 min period of incubation. After that 10 µL of cold 1 mM ATP was added to the beads and incubation was continued for 5 min. The beads were washed three times with NT2 buffer. Belle-RNA complexes were eluted with 30 µL 2x sample buffer supplemented by 200 mM DTT 5 min at 70 • C, and the complexes were resolved in 8% Bis-Tris gels using 3-(N-morpholino)propanesulfonic acid (MOPS)-based running buffer (50 mM MOPS, 50 mM Tris base, 1 mM EDTA, 0.1% SDS, DEPC-treated water). The complexes were transferred from the gel into a nitrocellulose membrane (BA-85 S&S). The membrane was exposed to a storage phosphor screen for 1-2 h. Storage phosphor screen was scanned using Typhoon FLA700 Phosphoimager (GE Healthcare), and an image was obtained. Fragments of the membrane containing the main radioactive signals over 85 kDa were cut, fragmented with a clean scalpel blade, transferred to tubes, and incubated with 100 µL proteinase K solution (4 mg/mL proteinase K dissolved in buffer PK (100 mM Tris-HCl, pH 7.5, 10 mM EDTA and 50 mM NaCl) for 20 min at 37 • C with agitation at 1000 rpm. After that 100 µL PK/7M-urea buffer were added to each tube and incubated for additional 20 min at 37 • C at 1000 rpm. Then 200 µLTrisol LS (Invitrogene) for RNA isolation and 200 µL chloroform were added to each tube and mixed vigorously for 30 s. After incubation 5 min at 4 • C the tubes were centrifuged at 13,000 g and 4 • C for 10 min. The upper phases were transferred to new microcentrifuge tubes and precipitated in the presence of 300 mM sodium acetate, 1 µl Glycoblue (from stock 15 mg/mL, Thermo Fisher Scientific), and 2.5× volume of EtOH overnight at −20 • C. After that ligation of 5 -adapter RA5 was performed using 1 µL T4 RNA ligase (NEB) in manufacturer's buffer supplemented by 1mM ATP solution for 2 h at 37 • C with subsequent precipitation. cDNA libraries were prepared after reverse transcription (SuperScript III Reverse Transcriptase kit (Thermo Scientific Fisher)) and subsequent PCR tests for optimal amplification with forward RP1 primer and reverse Illumina index primer and HS-Taq DNA pol (Thermo Fisher Scientific). The CLIP libraries were sequenced on the Illumina HiSeq2500 platform. CLIP sequencing data has been deposited to the NCBI GEO under accession number GSE137956.

Computational Analysis of CLIP Libraries
Preprocessing of the CLIP libraries involving adapter removal, filtering of reads according to their quality (MAPQ score > 15) and collapsing of reads with exact matching sequences was performed using bioinformatics tools that based on the Galaxy framework [41]. Reads that were duplicated due to PCR amplification step were collapsed in silico to one read by removing excess of identical sequences associated with the same barcode. Eventually each read represents an individual RNA molecule cross-linked with Belle. Then the reads having survived the preprocessing stage were mapped on the last assembly of Drosophila melanogaster genome dm6 (UCSC Genome Browser, https://genome.ucsc.edu/) using Bowtie 2 [42] or TopHat [43] algorithms. The reads were mapped allowing unique hits and up to two mismatches. The reads were annotated as falling in the following categories: 5 -UTR; 3 -UTR; CDS; introns; ncRNAs including snoRNAs, snRNAs, small RNAs and transposable elements; unannotated regions of genome. Peak calling to determine specific signals over background reads were performed using Piranha software [44] by the zero-truncated negative binomial likelihoods method. The whole Drosophila genome was split into 30-nt intervals (bins) and coverage in these windows was counted for maximum coverage greater than or equal to the 0.87 cut-off (selection conditions p > 0.87, bin is equal 30 nt). Only peaks with the coinciding or overlapping genome coordinates determined by Piranha in both CLIP libraries were selected as high confidence ones (Table S1). The peaks that were determined by Piranha in the above-mentioned conditions for only one CLIP library and present in the second CLIP library as at least two overlapping reads were considered as medium confidence ones (Table S1;  Table S3). Pearson's correlation analysis for comparison of CLIP1 and CLIP2 libraries was performed using Libre Office software. To analyze the distribution of the reproduced peaks in protein-coding mRNAs, four hierarchical categories were created, as follows: 5'UTRs > 3'UTRs > CDS > introns. Peaks were manually reviewed using dm6 genome assembly in a local mirror of the UCSC genome browser. To predict and expose the functional networks of proteins encoded by mRNAs found in the CLIP-seq libraries as candidate Belle targets we used the DAVID v6.8 [45] and STRING [46] databases.

Functioning of Belle in Cyst Cells is Required for Differentiation of Early Germ Cells
Recently we have shown that RNA helicase Belle (DDX3) is a critical cell-autonomous factor for survival and divisions of GSCs and spermatogonial cells in the Drosophila testes [25]. Complete loss of germ cells occurs in the testes of newly eclosed bel 6/neo30 mutants, as well as in the germline-specific RNAi-mediated knockdown of bel (RNAi belKD). In contrast, in RNAi belKD in somatic testis cells (the cyst cells) using the tj-GAL4 driver leads to the maintenance of the germline content. However, we observe that about 53% analyzed tj-GAL4>UAS-bel RNAi testes exhibit the accumulation of multiple small germ cells arranged in tight clusters [25].
Here we found that the usage of another soma-specific driver, c587-GAL4, for RNAi belKD also had a profound effect on testis morphology and gametogenesis. This driver is mostly active in CySCs and early cyst cells of the testes [47]. Consistent with our previous observations 55.9% (57 of 102) of the analyzed testes of newly eclosed c587-GAL4>UAS-bel RNAi males were significantly reduced in size and filled with tightly packed clusters of small Vasa-positive germ cells ( Figure 1B,D). These testes practically did not contain spermatocytes and germ cells of subsequent spermatogenesis stages. Another fraction of the RNAi belKD testes (38.2%, 39 of 102 cases) exhibited the so-called mosaic phenotype with several small germ cell clusters in the neighborhood with spermatogonial cells and spermatocytes ( Figure 1B,E). We observed only rare cases of a wild-type-like phenotype for the RNAi belKD testes (5.9%, 6 of 102 cases), whereas all analyzed control testes displayed a wild-type phenotype (100%, n = 80) ( Figure 1B,C).
To examine the relationship between germline and somatic cyst cells in RNAi belKD testes, we used antibodies against Traffic jam (Tj), the soma-specific transcription factor that marks nuclei of CySCs and early cyst cells [38]. In the control c587-GAL4 testes we observed a wild-type distribution of Tj-stained CySCs (in the next row from the hub behind GSCs) and cyst cells which enveloped mitotically active spermatogonia in the testis anterior ( Figure 1C). However, RNAi belKD testes contained a certain excess of Tj-stained early cyst cells which were not associated with germ cells ( Figure 1D,E). The number of Tj-positive cyst cells in the c587-GAL4>UAS-bel RNAi testes was found to be insignificantly greater in comparison with the ones in the control testes (average number is 76.5 ± 48.3 cells, n = 68 versus 56.4 ± 13.4 cells, n = 51 in control; p = 0.073 (Wilcoxon/Mann-Whitney test)) ( Figure 1F). We also analyzed the testes of c587-GAL4>UAS-white RNAi male as independent control ( Figure S1); however, the number of Tj-positive cyst cells also did not differ significantly.
Using flies with the c587-GAL4-induced transgenic UAS-gfp construct along with UAS-bel RNAi we clearly established that GFP-marked somatic cyst cells mainly did not encapsulate early germ cells in the testes, but rather located separately from germ cell clusters being pushed to the testis shell ( Figure S2A). These somatic cells had a round shape in contrast to their flat elongated shape in the control testes c587-GAL4>UAS-gfp ( Figure S2B). The nuclei of Tj-positive cyst cells in the belKD testes also conform to a spherical shape; these nuclei have an increased volume and Tj protein exclusion from a certain internal space, presumably from the nucleolus ( Figure 1D,E). The significance of changes in nuclear morphology is not clear to date and requires further researches. We proposed that RNAi belKD in cyst cells led to a cessation of their internal differentiation program.
These data indicate that Belle deficiency in somatic cyst cells does not lead to severe disturbances in the maintenance and proliferation of CySCs. Nevertheless, belKD in early cyst cells often leads to a cell-autonomous effect of their accumulation. Most of Tj-stained cyst cells failed to envelope germ cells and instead are located separately. They do not undergo differentiation to mature cyst cells. In summary, our data indicate that RNAi belKD in cyst cells generally lead to the segregation of early germ cells and cyst cells from each other, the accumulation of tightly packed clusters of small germ cells near the testis apical tip, and certain enrichment in early cyst cells for subset of the RNAi belKD testes.

Proliferation Activity of Germ Cells in the Clusters
To study the peculiarities of proliferation of germ cells in the clusters of the RNAi belKD testes, we analyzed the distribution of spectrosomes and fusomes in the testes using antibodies against α-spectrin. Spectrosomes are spherical organelles that are inherent for GSCs and goniablasts, whereas fusomes are specific branched structures penetrating all spermatogonia or spermatocytes within an individual cyst through syncytial ring canals [33] ( Figure 1A). Control testes of c587-GAL4 males displayed a wild-type pattern of spectrosomes and fusomes (Figure 2A). In contrast, in the RNAi belKD testes, germ cells in the clusters contained mainly dot-like spectrosomes which are characteristic structures of GSCs ( Figure 2B). We also found that germ cells in the clusters maintained proliferation activity. Immunostaining of testis preparations with mitotic marker phospho-Ser10H3 (PH3) antibodies revealed that germ cells in the clusters divided independently of each other and asynchronously ( Figure 2D) in contrast to the synchronous divisions of spermatogonial cells in the cysts of the control testes ( Figure 2C). The number of individual asynchronous divisions of germ cells in the RNAi belKD testes was significantly higher, and reached 1.65 divisions per testis on the average compared with the control ones, where their number was only 0.27 per testis (p = 0.0093, Wilcoxon/Mann-Whitney test) ( Figure 2E). At the same time, the frequency of synchronous signals in germ cells did not significantly differ in the experimental (0.45 signals per testis, n = 20) and control testes (0.44 signals, n = 18) (p = 0.8886) ( Figure 2E). Because the meiotic cell division in the wild-type testes appears to be a rapid process, at any given moment we can see only a small number of dividing cells. Note that divisions of individual germ cells in the clusters occurred independently of their distance from the hub ( Figure 2D).
The founder protein of the PIWI subfamily of the ARGONAUTE family Piwi is known to participate in co-transcriptional repression of transposons and contribute to the control of GSC proliferation in the Drosophila testes [48,49]. Analyzing expression of Piwi we observed that in the wild-type testes nuclear protein Piwi brightly stains hub cells at the topmost part of the testes. Strong Piwi staining is also detected in GSCs that tightly adhere to the hub, as well as in their immediate daughters, goniablasts. Expression of Piwi sharply drops in germ cells at later stages of differentiation. Piwi is also detected in CySCs and cyst cells throughout all premeiotic stages ( Figure S3B) according to previous papers [36,48,49]. However, we found a significant overexpression of Piwi in the RNAi belKD testes in comparison with the control testes by immunostaining and Western blot analysis ( Figure  S3A-C). In the RNAi belKD testes, Piwi strongly marked nuclei of all germ cells of the clusters ( Figure  S3A) and was also expressed in somatic cyst cells ( Figure S3A, yellow arrowheads). The pattern of Piwi expression additionally confirms that germ cells in the clusters are of GSC-like nature.
Taken together, our results indicate that germ cells in the clusters retained stem-like properties and supported asynchronous mitotic divisions with complete cytokinesis, independently of proximity to niche structure, the hub. Germ cells in the clusters expressed GSC markers and did not undergo subsequent differentiation.

Ectopic Overexpression of A Transgenic Arm s10 COPY in Cyst Cells Did Not Restore the Belle Knockdown Phenotype in the Testes
We speculated that the observed phenotypic abnormalities in the cyst cell RNAi belKD testes, such as the generation of tumor-like clusters of non-differentiated germ cells, could be caused by a disruption of intercellular adhesion between early germ and somatic cyst cells. Cell adhesion is mediated by specialized adhesion molecular complexes that are exposed on the surface of the interacting cells. However, our RT-qPCR analysis did not detect significant differences in the expression of transcripts encoding proteins of basic testis adhesion complexes, E-cadherin (E-cad), arm (armadillo, β-catenin in Drosophila), and β-integrin, in the testes with RNAi belKD in cyst cells compared to the control ( Figure S4A). Western blot analysis also showed that total levels of E-cad, Arm, and β-integrin proteins in the RNAi belKD testes did not change significantly from those in the control testis (p > 0.05, Student's test) ( Figure S4B). We assumed that failures in the cyst formation observed in the RNAi belKD testes may be associated with improper traffic or exposure of the adhesion complexes on the surface of early cyst cells.
To examine whether the deficiency of Belle function in somatic cyst cells actually leads to the disruption of the formation of cadherin-catenin adhesion complexes we used the fly line carrying a transgenic construct UAS-arm S10 [50] encoding a non-degradable form of Arm. We generated flies with the c587-GAL4-induced UAS-arm S10 transgenic construct along with UAS-bel RNAi and confirmed an increased level of Arm protein in their testes ( Figure S4C). However, the testes of experimental males did not differ morphologically from the testes with RNAi belKD alone and contained the clusters of small germ cells having escaped their enveloping by somatic cyst cells ( Figure S4D-G). Thus, ectopic overexpression of a transgenic arm S10 copy in cyst cells was not able to restore wild-type testis phenotype.

Ectopic Expression of β-Integrin Rescued Early Stages of Spermatogenesis in RNAi belKD Testes
Integrin adhesion complexes have been shown to play a pivotal role in the balance and competition between CySCs and GSCs for niche occupancy [51]. We examined the influence of integrin-based adhesion complexes on the repression of the tumor-like germline clusters in the background of RNAi belKD in cyst cells ( Figure 3A-D, Figure S5A). We found that ectopic overexpression of β-integrin (encoded by myo (βPS) gene) led to a significant rescue of the germline tumor phenotype. In total, 61.2% (n = 30) experimental testes exhibited wide-type size and morphology including the proper formation and maintenance of the cysts and the differentiation of premeiotic germ cells ( Figure 3D). Another fraction of the testes, 38.8% (n = 19), exhibited a mosaic phenotype characterized by the presence of small germ cell clusters along with the cysts of spermatogonial cells and spermatocytes (not shown). We found that the level of Piwi, marker of GSCs, returned to wild-type and the marker of spermatocytes βNACtes [52] was restored to the wild-type level indicating the maintenance of the spermatocyte population in the "rescued" testes ( Figure 3E). In the wild-type testes the expected number of spectrosomes does not exceed 30. We found that in most of the experimental testes (belKD + β-int) the number of spectrosomes did not exceed 30 per testis ( Figure 3F). In sum, these data confirm the restoration of premeiotic stages of spermatogenesis. Thus, expression of integrin adhesion complexes is important for the maintenance of interaction between cyst cells and germ cells in the testes and for the prevention of germ cell tumor formation. However, the rescue of early stages of spermatogenesis owing to the expression of an additional copy of β-integrin did not lead to the restoration of male fertility ( Figure S5B). Using electron microscopy, we observed developmental defects in the experimental testes (belKD + β-int) at the stage of spermatid elongation ( Figure S5C). Apparently, the ectopic expression of β-integrin in early cyst cells does not allow rescue of the post-meiotic stage of spermatogenesis in the testes with belKD in cyst cells.
of β-integrin did not lead to the restoration of male fertility ( Figure S5B). Using electron microscopy, we observed developmental defects in the experimental testes (belKD + β-int) at the stage of spermatid elongation ( Figure S5C). Apparently, the ectopic expression of β-integrin in early cyst cells does not allow rescue of the post-meiotic stage of spermatogenesis in the testes with belKD in cyst cells.

CLIP-seq Analysis for Identification of Belle mRNA Targets
To identify direct mRNA targets of Belle in the testes, we performed cross-linking immunoprecipitation assay with subsequent deep sequencing (CLIP-seq) as described in Materials and Methods. We used UV-cross-linked testes of young yw males and anti-Belle antibodies raised previously [25] for the preparation of two biological replicates of CLIP-libraries. The recovery of cross-linked target RNAs in RNA-Belle complexes was monitored by autoradiography ( Figure 4A). The recovery of endogenous Belle protein in the immunoprecipitation procedure was monitored by Western blot analysis ( Figure S6A). Protein-RNA complexes were practically absent when anti-Belle antibodies were replaced with normal mouse IgG, indicating that the Belle-RNA complexes immunoprecipitated in these experiments were specific complexes ( Figure 4A; Figure S6A). The resulting cDNA libraries were barcoded and subjected to high-throughput sequencing. For the two CLIP-seq experiments we obtained 8.1 × 10 6 and 4.8 × 10 6 raw reads. However, we failed to amplify background material in the control experiment with normal mouse IgG instead of anti-Belle antibodies (not shown). This indicates a high signal-to-noise ratio for the Belle CLIP-libraries. Bioinformatics data analysis was performed as described in Materials and Methods. The number of reads for the Belle CLIP-libraries before and after processing and genome mapping are presented in ( Figure 4C). Then, peak calling was performed using Piranha software [44] to identify a list of Belle target RNAs (Table S1, Table S2). Two biological replicates of the CLIP-libraries showed high reproducibility in the numbers of unique-mapped reads per gene in the CLIP experiments (Pearson's correlation coefficient 0.88, Figure 4B). Only the peaks with coinciding or overlapping genome coordinates found in both of the CLIP libraries were selected for subsequent analysis. The distribution of unique peaks revealed that 60% of them maps to annotated regions of the genome ( Figure 4D). Across transcripts of protein-coding genes, Belle was found to be enriched in general in coding sequences and both in 5 -and 3 -UTRs (see Figure S6B-E for several examples of peaks mapped in 5 -and 3 -UTRs of protein-coding genes), and exhibited comparatively low binding to introns, 28.6% ( Figure 4E). This circumstance points out that Belle interacts preferably with mature mRNAs in the testes. However, according to published data, DDX3 helicases can be involved in a wide range of intracellular processes associated with RNA metabolism, including transcription and splicing [3]. Thus, we cannot exclude the participation of Belle in mRNA processing in the nucleus. Overall, we mapped Belle binding sites in mature transcripts of more than 300 protein-coding genes (Table S1).
Gene ontology (GO) analysis using DAVID software to search for overrepresented GO categories in comparison to the entire fly protein-coding transcriptome revealed that Belle targets were significantly enriched in genes encoding proteins which were involved in cell cytoskeleton organization, ubiquitin-dependent protein catabolic processes, sperm individualization, mitochondrial functions, intracellular signal transduction, transmembrane transport, and nucleic acid binding ( Figure S7A,B; Table S2). GO functional annotation clustering revealed at least seven functional clusters among candidate Belle mRNA targets ( Figure S7B; Table S2). Gene ontology (GO) analysis using DAVID software to search for overrepresented GO categories in comparison to the entire fly protein-coding transcriptome revealed that Belle targets were significantly enriched in genes encoding proteins which were involved in cell cytoskeleton organization, ubiquitin-dependent protein catabolic processes, sperm individualization, Transcriptome-wide identification of Belle mRNA targets by CLIP-seq analysis. (A) Autoradiogram of 32 P-labeled cross-linked RNA-Belle complexes. Two independent immunoprecipitations using anti-Belle antibodies and UV cross-linked testes and a control experiment with non-immune mouse IgG were performed. The recovery of cross-linked target RNAs in RNA-Belle complexes was monitored by autoradiography after RNAse A treatment and 5 -end labeling with γ 32 P fragments of the membrane containing the main radioactive signals above 85 kDa (arrow indicates position non-crosslinked Belle) that are outlined by red squares were cut for further library processing.

Belle is Involved in Translational Regulation of Multiple mRNA Targets
We speculated that Belle binding within 5 -UTRs may facilitate translation initiation by unwinding secondary structures of mRNAs owing to its helicase activity, whereas Belle interaction with 3 -UTR may contribute to either activation or repression of translation, according to previously published data [28]. The peaks determined in the coding regions suggest that Belle may participate in the elongation steps of translation as well [26,28].
Recently, several groups reported RNAi-mediated knockdown screens to identify candidate genes that are required in somatic cyst cells or in the germline of D. melanogaster testes for specific stages of spermatogenesis [53][54][55]. Having conducted comparative analysis of these datasets, we recovered a subset of 22 genes that coincide with the ones obtained in our CLIP-seq analysis as Belle targets (Table  S3). It has been shown that the RNAi knockdowns of 17 genes from this subset determined in our CLIP-seq analysis as potential targets of Belle regulation lead to similar phenotypes as belKD in cyst cells or in the germline, according to the results of at least one screen ( Figure 5A, Table S3). Among them we selected four genes, not1, rpn7, rho1, and caf1-55, knockdowns of which result in the same phenotypic defects in the testes based on data from all three independent screens ( Figure 5A).
We performed Western blot analysis of testis lysates for examination of expression levels of three proteins encoded by Belle-targeted transcripts in the testes of c587-GAL4>UAS-bel RNAi males, nos-GAL4>UAS-bel RNAi males, and the corresponding driver lines as controls. For Not1, the key component CCR4-Not deadenylation complex, we repeatedly observed a decrease in the intensity of bands corresponding to the major long (280-kDa) isoforms and the minor short (260-kDa) isoform of Not1 in the backgrounds of RNAi belKDs both in the soma and germline of testes ( Figure 5B). We also found a reproducible reduction of nucleosome remodeling factor Caf1-55. However, we did not detect significant changes in the level of small GTPase Rho1.
To confirm these results and validate the relationships between not1 transcripts and Belle in spermatogenesis, we carried out RNAi knockdown of not1 using soma-specific driver tj-GAL4. Immunostaining of fixed testis preparations revealed that the control testes (n = 13) contained a wild-type array of cells at all stages of spermatogenesis, whereas the not1KD testes were reduced in size and exhibited obvious phenotypic disorders ( Figure 5C-E). In most of the cases (n = 9), the germline content in the experimental testes was represented by only small undifferentiated cells assembled in tumor-like clusters with somatic Tj-positive cells located separately ( Figure 5D). In the remaining testes (n = 5), clusters of small germ cells were observed along with differentiating germ cells, including spermatocytes ( Figure 5E). Thus, not1KD in cyst cells reproduced the GSC tumor-like phenotype observed in the testes of males with belKD in cyst cells (see Figure 1D-E, Figure 2B and D). Taken together with the CLIP-seq data and Western blot analysis, these results indicate that Belle positively regulates translation of not1 transcripts. We also confirmed that RNAi knockdowns of caf1-55KD, rpn7 and Fs(2)Ket in cyst cells led to tumor-like germ cell cluster formation ( Figure S8), but with relatively low penetrance compared to not1KD. Note that we performed pilot experiments using the c587-Gal4 driver to reveal spermatogenesis defects in case of knockdowns of the selected genes. However, we did not observe any differences from the wild-type testis phenotype in these experiments (not shown), possibly due to low knockdown efficiency. For these reasons we used here the tj-GAL4 driver line supplemented by an additional dicer copy.
Using the nos-GAL4 driver we also performed RNAi knockdowns of not1, caf1-55, rpn7, Fs(2)Ket and rhoI in the germline ( Figure S9A-G). With aid of immunostaining we found a total loss of testis germ cells with the maintenance of somatic cells in cases of RNAi knockdowns of caf1-55, rpn7, and Fs(2)Ket, whereas the RNAi knockdown of not1 led to a rapid decrease in both somatic and germ cells ( Figure S9).
Using the STRING database [46], we clustered the genes from the full list of 22 candidate Belle CLIP-targets shared with the results of the screen studies [53][54][55] (Table S3) for potential interactions and found a significant enrichment (p-value = 0.00449) of clusters of interacting genes with high confidence settings ( Figure 5F; Table S4). Among them the clusters of genes involved in protein ubiquitination and proteolysis, Golgi-Endoplasmic Reticulum vesicle transport of macromolecules, small GTPase Rho1 activity, and RNA processing factors were identified (Table S4).   [53][54][55], and which were determined in our CLIP-seq analysis as putative mRNA targets of Belle. Knockdowns of four genes, not1, rpn7, rho1, and caf1-55, provide the same morphological defects in the testes, according to data from all three independent screens. (B) Western blot analysis of testis lysates of c587-GAL4>UAS-bel RNAi males, nos-GAL4>UAS-bel RNAi males, and the corresponding driver lines as controls. Anti-actin antibodies were used as a loading control. The values (obtained by using ImageJ software) were normalized to the loading control level (actin) and then to the values obtained for the control lysates. Mean values and standard errors are presented for at least three independent experiments. We revealed a decrease in the level of Not1 and Caf1-55 proteins in the testes with belKDs; however, we did not detect significant changes in the level of Rho1. (C-E) Immunostaining of fixed testis preparations of males with RNAi knockdown of not1 using soma-specific driver tj-GAL4. Testes of newly eclosed males were stained with anti-Vasa (red) and anti-Tj (violet) antibodies; chromatin was stained by DAPI (4 ,6-diamidino-2-phenylindole, blue). Scale bars are 30 µm.  (Table S3) was performed by STRING database tools. Only 17 of 22 genes form five interacting clusters with high confidence settings (interaction score 0.7). The thickness of the line between genes (color balls) reflects the relative probability of interaction. See Table S4 for a detailed description.
To analyze whether any signaling pathways were modulated by RNAi belKD in somatic cyst cells of testes we performed a mini-screen using antibodies to key components of different pathways ( Figure S10). We found that belKD in somatic cyst cells led to a significant decrease in the testes of transcription factor STAT92E, one of the main participants of the Janus kinase-signal transducer and activator of transcription (JAK-STAT) pathway, which is responsible for the attachment of GSCs and CySCs to the hub [31,[56][57][58]. We often observed that germ cells in the clusters did not maintain tight contacts with the hub. Thus, the survival of germ cells in the clusters does not depend on the level of activation of the JAK-STAT pathway in them. We found that the level of the key effector of the Hedgehog pathway in cyst cells, the 155-kDa form of Ci [59], significantly decreased in the belKD testes ( Figure S10). However, epidermal growth factor receptor (EGFR) pathway signaling activities in CySCs of the belKD testes were indistinguishable from those in the control testes, as revealed by the intensity of signals of the phosphorylated form of MAP kinase ERK, pERK, and of small GTPase Rac1, despite the fact that disruption of the EGFR pathway has been shown to exhibit a testis phenotype [60][61][62] similar to belKD in cyst cells. We determined that the level of the Glass bottom boat (Gbb) 55-kDa precursor, the signaling molecule of the main GSC self-renewal regulator bone morphogenetic protein (BMP) (transforming growth factor β (TGF-β)) pathway [63,64], was significantly increased in the testes of males with belKD in cyst cells ( Figure S10). The level of the second BMP signaling molecule, Decapentaplegic (Dpp), was found to be unchanged. Note that only transcripts of gbb, but not dpp, were identified as direct targets of Belle in our CLIP-seq analysis (Table S1). However, we did not detect significant changes in the level of phosphorylated Smad molecule (pSmad), a key component of the BMP signaling cascade in early germ cells in the belKD testes ( Figure S10).

Discussion
The RNA helicase Belle is known to be essential for fly viability, fertility, and germline stem cell maintenance [17,25,26]. However, the regulatory and mechanistic roles of Belle in spermatogenesis are still largely obscure. Here we revealed function of RNA helicase Belle in the testes as a germ cell tumor suppressor acting in-trans in early cyst cells. We found that RNAi-induced belKD in cyst cells resulted in the generation of large tumor-like clusters of GSC-like germ cells. Germ cells in the clusters were maintained over time; they spontaneously and asynchronously divided, but did not undergo subsequent differentiation ( Figure 1B-E, Figure 2B and D, Figures S2 and S3). Cyst cells failed to envelope these germ cells and instead were accumulated separately ( Figure 2D, Figure S2). They were also arrested in their differentiation, since the transition of stage-specific differentiation markers did not occur ( Figure 1F-G, Figure S2C). Previously it has been shown that mutations in EGFR pathway components resulted in a similar tumor-like germ cell phenotype [60][61][62]. However, the molecular mechanisms underlying these spermatogenesis disorders are not clearly understood; and we found no evidence of a direct contribution of Belle to the regulation of the EGFR pathway in the testes ( Figure  S10).
During wild-type spermatogenesis, coordinated morphogenetic processes occur both in the germline and somatic cyst cells and their physical interactions are essential for maintenance and self-renewal of GSCs in the niche and also for differentiation of their progeny [30][31][32]34]. We determined that belKD in somatic cyst cells led to the disruption of contacts between early germ cells and cyst cells. According to our results, premeiotic stages of spermatogenesis in the belKD testes were significantly rescued by the ectopic expression of β-integrin (βPS), but not non-degradable arm S10 (β-catenin) in early cyst cells (Figure 3 and Figure S4). It has been shown previously that integrin adhesion complexes in the testes are important for anchoring the hub on the surface of extracellular matrix [65,66] and also for the attachment of cytoplasmic protrusions of CySCs to the hub to obtain instructive signals that promote their maintenance and self-renewal divisions [51,67,68]. It is reported that SOCS36E protein represses the JAK-STAT signaling in CySCs cells. SOCS36E prevents these cells from displacing GSCs from their position near the hub by up-regulating integrin-based adhesion complexes [67,68]. The EGFR signaling pathway also contributes to the control of integrin expression in CySCs cells; and integrins play a crucial role in the regulation of competition between GSCs and CySCs for the niche occupancy [51]. Integrin heterodimeric complexes are known to mediate two-sided signaling. Intracellular signals cause conformational changes of integrin complexes, which contribute to their activation and increased affinity for extracellular ligands [69,70]. In turn, binding of integrin complexes with their extracellular ligands or clustering of integrin complexes on the cellular membrane contributes to conformational changes in their cytoplasmic domains, modulating their interactions with cytoskeletal proteins and activation of some signaling pathways [69][70][71]. We revealed here that transgenic overexpression of β-integrin provided significant restoration of early stages of spermatogenesis in the background of RNAi belKD in cyst cells (Figure 3). Our results confirm a more universal role of integrin complexes in CySCs in the maintenance of early stages of spermatogenesis, including supporting proper contacts between cyst cells and germ cells during cyst formation and preventing GSC tumorigenesis ( Figure 6A). Despite the fact that we did not observe integrin transcripts among Belle mRNA targets in our CLIP-seq analysis (Table S1), we speculate that Belle indirectly regulates unknown steps of the accumulation of integrin heterodimers on the surface of CySCs and early cyst cells, ensuring integrin complex localization or stability. There are at least three adaptor proteins that mediate the attachment of integrin complexes to the cellular actin cytoskeleton, including Talin (Rhea in Drosophila), spectraplakin Shot stop (Shot), and integrin-linked kinase (ILK) [72]. According to the GO analysis of the CLIP-seq data, factors responsible for actin filament organization and adherens junctions including Shot, are enriched among Belle mRNA targets (Table S1; Figure S7A; Table S2). This observation elucidates a possible mode of regulation of integrin-based adhesion complexes in cyst cells; however, this requires further investigation.
Despite the restoration of early stages of spermatogenesis by the expression of a transgene copy of β-integrin in somatic cyst cells, these males were sterile owing to developmental disorders at the spermatid stage ( Figure S5B,C). It has been found in the screen study [54] that RNAi-induced knockdowns of ter94 and arf102f in somatic cyst cells lead to spermatid defects. ter94 and arf102f mRNAs were determined in our CLIP-seq assay as potential Belle targets with peaks in the 5 -UTR of ter94 transcript and the coding region of arf102f ( Figure S6B; Table S1, and Table S3). However, the involvement of these factors in the development of spermatids requires further study.
Our data indicate different functional consequences of belKDs in the germline and in somatic cyst cells of the testes ( Figure 6B). Whereas belKD in the germline leads to a total germ cell loss with preservation of somatic testis cells as described previously [25], belKD in cyst cells induces a generation of tumor-like clusters of early germ cells which actively proliferate, but are not capable to further differentiation ( Figure 1D-E). Thus, we found that Belle functions in the testes were required cell-autonomously for the maintenance of early germ cells including GSCs and non-autonomously in cyst cells to ensure germ cell differentiation. This dual role of Belle appears to reflect the diverse functions of human orthologs DDX3X and DBY in the soma and germline of the human testes.
Being involved in a wide range of cellular processes, DEAD-box RNA helicases have proved to be challenging to study. The interaction of DEAD-box RNA helicases with their mRNA targets is considered non-processive and transient [1,3]. Here we used the powerful CLIP-seq approach based on in vivo RNA-protein crosslinking to firstly identify a range of potential targets of Belle regulation in the testes. CLIP-seq analysis revealed that Belle interacts with more than 300 transcripts of protein-coding genes in the testes (Figure 4, , Table S1; Table S2). According to the CLIP-seq data, we revealed the binding of Belle with transcripts of protein-coding genes both in 5 -and 3 -UTRs and the coding regions ( Figure 4E, Figure S6B-E). This indicates that Belle interacts predominantly with mature mRNAs and preferentially acts as a translational regulator. In accordance with previously published data, Belle/DDX3 can be involved both in translational repression and translational activation of target transcripts [18,[26][27][28][73][74][75][76]. Representation of the different defects of early stages of spermatogenesis caused by RNAi belKDs in the germline and somatic cyst cells. Whereas belKD in the germline leads to total germ cell loss with the maintenance of somatic cyst cells [25], belKD in cyst cells non-autonomously induces the generation of tumor-like clusters of early germ cells. Below: knockdowns of selected Belle targets lead Our CLIP-seq analysis provides entry points in attempting to understand the molecular mechanisms underlying the spermatogenesis defects observed in both in somatic cyst cells and in the germline with failures of Belle expression. Using comparative analysis of the recently published screens, we selected a subset of genes that overlap with the ones detected in our CLIP-seq analysis (Table S3). These screens focused on the identification of genes that are essential in cyst cells or in the germline of Drosophila testis during specific stages of spermatogenesis [53][54][55]. Comparative analysis revealed that RNAi knockdowns of 17 genes, which were determined in the CLIP-seq analysis as potential Belle targets, led to shared testis abnormalities with belKDs in cyst cells (tumor-like clusters of early germ cells) or in the germline (germ cell loss) ( Figure 5A; Table S3). Among them we selected four genes, not1, rpn7, rho1, and caf1-55, knockdowns of which provide the same morphological defects in the testes, according to data of all the three independent screens ( Figure 5A). We revealed a decreasing level of Not1 and Caf1-55 in the testes with RNAi belKD in cyst cells ( Figure 5B). We also experimentally confirmed that RNAi not1KD ( Figure 5C-E) and caf1-55KD ( Figure S8) in cyst cells can lead to the generation of the tumor-like germ cell clusters. Not1 is a key component of the CCR4-Not deadenylation complex involved in translational regulation of multiple mRNAs in development [28,77]. Caf1-55 is a component of the chromatin assembly factor 1 (CAF-1) complex, and also the nucleosome remodeling and deacetylase (NuRD) and polycomb repressive complex 2 (PRC2) chromatin complexes; it is essential for control of GSC self-renewal and proliferation both in the ovaries and testes [78,79]. We suggest that Belle positively regulates translation of not1 and caf1-55 mRNAs both in somatic cyst cells and germ cells. We verified the RNAi knockdown phenotypes for two other genes of the set, rpn7 and Fs(2)Ket. Rpn7 is involved in the proteasome-mediated ubiquitin-dependent protein catabolic process [55], whereas β-importin Fs(2)Ket is responsible for the nuclear import and export [80]. We found that both rpn7KD and Fs(2)KetKD in cyst cells caused germline cluster formation with reproducible but relatively low frequency ( Figure S8). We revealed that germline-specific RNAi knockdowns of at least four Belle target candidates, not1, caf1-55, rpn7, and Fs(2)Ket, led to a total loss of testis germ cells. Interestingly, in the cases of knockdowns of caf1-55, rpn7, and Fs(2)Ket, we observed the maintenance of somatic cells in the testes, whereas RNAi knockdown of not1 caused a premature decrease both in somatic and germ cells (Figure S9A-D,G). Note that failures of expression of Caf1-55, Rpn7, and Fs(2)Ket proteins in germ cells accurately reproduced the corresponding cell-autonomous maintenance defects observed in the most testes of males with the germline-specific belKD according the our previous study [25].
According to the screen data and our results, knockdowns of several selected Belle targets lead to similar developmental disorders in the testes ( Figure 5A,D,E; Table S3; Figure S8, Figure S9, Figure 6B). Assuming that Belle is an upstream translational regulator of these transcripts, we propose that failures of translation of a number of mRNA targets could cooperatively contribute to the morphological and developmental testis defects observed in belKDs both in somatic cyst cells and the germline ( Figure 6B). Comparison of this subset with the data of CLIP-seq analysis of human DDX3 mRNA targets in HEK293T cells [75] revealed transcripts of orthologous human genes, including the not1, caf1-55, hyd, and beta'COP orthologs Cnot1, RBBP4, UBR5, and COPB1 (Table S3).
Although we found that Belle facilitated translation of not1 and caf1-55 mRNAs ( Figure 5B), it appears to be a negative regulator of translation of gbb transcript in cyst cells ( Figure S10, Table S1). The hub and CySCs are known to secrete two BMP ligands, Gbb and Dpp, in the wild-type testes. Gbb appears to be more important for GSC maintenance in the testes, while Dpp is more significant in the ovarian niche [63,64]. This short-range signaling accomplished by these ligands is critical for the maintenance and self-renewal of GSCs. Diffusion of BMP ligands in wild-type testes is tightly restricted by a one-cell distance from the niche. However, it is shown that ectopic overexpression of BMP ligands itself does not cause abnormal expansion of GSCs in the testes [63,64]. Accordingly, we found no significant changes in the level of the key germline component of BMP cascade pSmad in the belKD testes ( Figure S10). So, the increased concentration of Gbb secreted by cyst cells into the internal testis space is unlikely to be the main causative factor of the formation of GSC-like clusters of germ cells. We speculate that Belle can normally negatively regulate gbb translation in differentiating cyst cells, preventing inappropriate activation of BMP signaling outside the niche.
In our previous paper we demonstrated that Belle is required cell-autonomously for mitotic progression and survival of GSCs and spermatogonia as the upstream regulator of mitotic cyclin expression [25]. A disruption of Belle expression leads to a strong germ cell loss in the testes; the phenotype mimics that of SCOS in humans with DDX3Y deletion [25,29]. Here we dissected a non-autonomous role of Belle (DDX3) in the prevention of tumorigenesis of early germ cells. To date, DDX3 helicase has attracted increasing attention owing to its essential but contradictory functions in the progression of human cancers in different tissues. The role of DDX3 as an oncogene or tumor suppressor is associated with activity in the tissue of predominant signaling pathways in which DDX3 is involved [11][12][13]. Our data augment and enrich our current understanding of the testis-specific functions of Belle/DDX3 as an essential translational regulator. Considering that somatic cyst cells of Drosophila testes appear to be the functional analogs of Sertoli cells of mammalian testes [34], the employment of the Drosophila model may contribute to the elucidation of DDX3 functions in the maintenance of spermatogenesis and shed light on tumorigenesis and cancer treatment in the human testes.
Supplementary Materials: The following are available online at http://www.mdpi.com/2073-4409/9/3/550/s1. Figure S1. Comparative immunostaining analysis of the testes of males c597-GAL4>UAS-bel RNAi and c587-GAL4>UAS-white RNAi. Figure S2. Peculiarities of cyst cell morphology of the testes with belKD in cyst cells. Figure S3. Piwi protein is overexpressed in the clusters of germ cells. Figure S4. Analysis of expression of components of cell adhesion complexes. Figure S5. Analysis of testes with ectopic expression of βPS (β-integrin) copy in cyst cells in the background of RNAi belKD. Figure S6. Preparation and analysis of Belle CLIP-seq libraries. Figure S7. Gene ontology analysis of Belle mRNA targets in the testes using DAVID Bioinformatics Resource v6.8. Figure S8. Immunostaining analysis of the testes of flies with RNAi knockdowns of Belle target gene expression in somatic cyst cells using the tj-GAL4 driver. Figure S9. Immunostaining analysis of the testes of flies with RNAi knockdowns of Belle target gene expression in the germline using the nos-GAL4 driver. Figure S10. Western blot analysis of testis lysates with antibodies to key components of different signaling pathways. Table S1. List of Belle RNA targets determined by Piranha software with selection conditions p > 0.87 for both CLIP1 and CLIP2. Table S2. GO analysis of Belle RNA targets using DAVID Bioinformatics Resource v6.8. Table S3. List of genes, knockdowns of which provide the morphological defects in the testes according to data of three independent screens and which determined in our CLIP analysis as putative mRNA targets of Belle. Table S4. STRING-based interaction clusters.