The core-promoter factor TRF2 mediates a Fruitless action to masculinize neurobehavioral traits in Drosophila

In fruit flies, the male-specific fruitless (fru) gene product FruBM plays a central role in establishing the neural circuitry for male courtship behavior by orchestrating the transcription of genes required for the male-type specification of individual neurons. We herein identify the core promoter recognition factor gene Trf2 as a dominant modifier of fru actions. Trf2 knockdown in the sexually dimorphic mAL neurons leads to the loss of a male-specific neurite and a reduction in male courtship vigor. TRF2 forms a repressor complex with FruBM, strongly enhancing the repressor activity of FruBM at the promoter region of the robo1 gene, whose function is required for inhibiting the male-specific neurite formation. In females that lack FruBM, TRF2 stimulates robo1 transcription. Our results suggest that TRF2 switches its own role from an activator to a repressor of transcription upon binding to FruBM, thereby enabling the ipsilateral neurite formation only in males.

T he female and male of a sexually reproducing animal are, in principle, different from each other in structure and function at the molecular, cellular, and organismal levels. Sexual traits often represent the most striking variations within a species, as a result of strong pressure from sexual selection 1 . The fruit fly Drosophila melanogaster, an excellent model for studying the genetic organization of complex traits, shows remarkable sexual dimorphisms in many aspects of its structure and function, including its behavior 2 . To court females, males of D. melanogaster engage in a series of sophisticated behavioral actions, which include orientation toward the target, chasing, tapping, wing extension/vibration for love song generation, licking and attempted copulation, culminating in copulation [3][4][5][6] . Normal females do not show any of these male-typical behaviors and, instead, signify rejection of male courtship by decamping, fending, flickering the wings, kicking, curling the abdomen and extruding the ovipositor, or signify their acceptance by slowing down locomotion, raising their wings, and opening the vaginal , and the fruB + -induced eye phenotype was partially rescued by GS9128 (c) or Trf2-S + overexpression (d). Enlarged views are shown in a′-d′. e Schematic drawings of the genomic organization of the Trf2 locus with the GS9128 P-insertion (upper panel) and TRF2-L and TRF2-S isoforms (lower panel). f-i Trf2-S knockdown in fru GAL4 -expressing cells reduced the courtship index (f), wing extension index (g), and copulation success (h), but had no effect on locomotor activity (i) in male flies, and the effects of Trf2 knockdown were compensated by Trf2-S + overexpression (f-h). Statistical differences were evaluated by one-way ANOVA followed by Tukey's multiple comparison test (f, i), by the Kruskal-Wallis test followed by the Dunn's multiple comparison test (g), or by the χ 2 -test (h). ***P < 0.001, *P < 0.05; ns not significant. The number of flies examined is indicated near the base of each bar.
Data shown in f-h were obtained from the same fly group, whereas data shown in i were derived from a fly group different from that used in f-h. Error bars indicate SEM. Scale bars: 200 µm (a), 30 µm (a′) plate [7][8][9][10][11] . These sexual differences in mating behavior reflect sexual differences in the nervous system, which activates a sexspecific motor program under the circumstances where the repertoire for mating behavior rather than that for other behaviors (e.g., aggression, feeding, and sleeping) is appropriate 12 . A significant portion of the neural circuitry for mating behavior is composed of sexually dimorphic or sex-specific neurons in this species 13 . The neural sex differences in Drosophila are generated under the hierarchical control of two sex-determinant genes, fruitless (fru) and doublesex (dsx) 14,15 . dsx plays a key role for sex determination in both neural and non-neural cells, whereas fru exerts its sex-determinant function only in neural cells 14,15 . This neuron-specific sex-determinant role of fru provides us with an outstanding opportunity to unravel the causal relationship between the single neuron sex difference and the behavioral sex difference with minimal possible disturbances in the non-neural sex determination process. The sexual function of fru is attributable exclusively to the P1 promoter-derived transcripts [16][17][18] , which produce five isoforms of Fru in males but no protein in females 19,20 . Thus the P1 promoter-derived transcripts encode male-specific Fru isoforms, FruAM, FruBM, FruCM, FruDM, and FruEM, where M stands for male-specific and A-E indicates the C-terminal variant type (isoforms A, B, and E by our nomenclature correspond to isoforms A, C, and B by the nomenclature adopted by von Philipsborn et al. 21 , respectively), among which FruBM (FruCM in von Philipsborn et al. 21 ) is the isoform with the strongest impact on neurobehavioral masculinization [20][21][22][23] . FruBM has an Nterminal BTB domain and two zinc-finger motifs at the C-terminus, suggestive of its role as a transcriptional factor 17,24 . Indeed, FruBM forms a complex with proteins known to function as chromatin factors, i.e., heterochromatin protein 1a (HP1a), histone deacetylase 1 (HDAC1), and the TIF1 homolog Bonus (Bon) 25 . A large number of potential transcriptional targets of FruBM have been proposed based on DamID, ChIP, and transcriptome analyses 23,26 . However, there exists only one established target gene of FruBM, in the sense that it has a defined cis element for FruBM binding, and the in vivo outcome of FruBMbinding to the cis element has been firmly demonstrated 27 . This target gene, robo1, encodes a key effector that specifies the neurite sex-type of a group of fru-expressing interneurons called the mAL cluster 27 . The mAL cluster is sexually dimorphic in three respects 28 : the number of cells composing the cluster is 5 in females vs. 30 in males; the contralateral neurite bifurcates in females but not in males; and the ipsilateral neurite exists only in males. Loss of fru in males transforms the mAL cluster from the male-type into the female-type in all three respects 28 . robo1 has been shown to inhibit the formation of the ipsilateral neurite in females, whereas, in males, male-specific FruBM represses robo1 transcription, thereby allowing the ipsilateral neurite to form 27 . In contrast, Hunchback (Hb) is required for specification of the branching pattern of the contralateral neurite with no role in the ipsilateral neurite formation 29 . The sex difference in the mAL cell number is a result of female-specific cell death in which neither robo1 nor Hb plays a role 28 . Thus, FruBM appears to regulate a distinct set of genes for establishing the sex-type of each of three sexually dimorphic structures of the mAL cluster. However, the mechanism whereby FruBM regulates a unique set of genes for a particular sexual trait while controlling another set of genes for the other sexual trait remains an enigma. In the present study, by searching for genetic modifiers of a fru-induced phenotype, we recover Trf2, a gene encoding a core promoter factor, and find that it suppresses the male-specific mAL ipsilateral neurite formation in females. We further demonstrate that TRF2 contributes to a FruBM-containing complex and regulates robo1 transcription. Surprisingly, we find that TRF2 strongly enhances the transcription repressor activity of FruBM when it is recruited to the FruBM-containing protein complex, whereas TRF2 on its own activates robo1 transcription in the absence of FruBM. We propose that TRF2 governs sex-specific function of a FruBM target gene, robo1, from that of an activator to a repressor upon detection of the presence of FruBM in males.
Isolation of Trf2 as a phenotypic modifier of fru. In screening a large collection of GS P-element insertion lines 30 for possible phenotypic interactions with fru, we took advantage of the visible phenotype induced upon overexpression of fru + in the eyeantennal disc, which distorted the adult compound eye structure (Fig. 1a, b). In this screen, we chose to express FruB, one of the non-sex-specific Fru proteins. Although FruBM is the most prevalent isoform in neural masculinization 31 , we found that FruBM overexpression via GMR-GAL4 led to lethality. To circumvent this problem, we used FruB. Non-sex-specific Fru proteins that are not endogenously expressed in postembryonic neurons have been shown to masculinize neurons when they are artificially expressed 20,32 . The GS system allows one to overexpress a gene flanking the GS P-element insertion in the presence of a GAL4 driver 30 . In our screen, GMR-GAL4 acted to drive both UAS-fruB and the second target in the genome, which flanked a GS Pelement. After examining 1364 GS lines, we recovered 40 dominant suppressors of the eye distortion phenotype induced by fruB overexpression, including GS9128 (Fig. 1c). The GS P-element in GS9128 had impinged immediately 5′ to the Trf2 transcription unit ( Fig. 1e) 33 , suggesting that it was Trf2 that was involved in the observed suppression of the fru-induced eye distortion. The Trf2 gene produces the TRF2-long (TRF2-L) and TRF2-short (TRF2-S) protein isoforms 34 . The effect of GS9128 in suppressing the fruB-induced distorted eye phenotype was recapitulated by overexpressing TRF2-S in the absence of GS9128 (Fig. 1d), indicating that Trf2 was indeed responsible for suppression of the fruinduced eye phenotype.
Trf2 supports sex-specific neurobehavioral traits. To determine whether Trf2 plays a role in generating sex-specific traits as fru does, we measured male courtship behavior in flies in which the Trf2 activity was genetically reduced by the action of Trf2 RNAi. We ascertained that TRF2 immunoreactivity vanishes upon the expression of Trf2 RNAi in cells that otherwise accumulate a large amount of TRF2 ( Supplementary Fig. 1). Trf2 knockdown in fru GAL4 -expressing neurons resulted in a marked reduction in the courtship index (CI) (Fig. 1f), wing extension index (WEI) (Fig. 1g), and copulation success rate (Fig. 1h), without affecting locomotion (Fig. 1i). All of the courtship defects induced by Trf2 knockdown were completely rescued by the overexpression of a transgene encoding TRF2-S (Trf2-S + ; Fig. 1f-h). Female flies with Trf2 knockdown in fru GAL4 -expressing neurons exhibited reduced mating success and deposited fewer eggs ( Supplementary  Fig. 2), indicating that Trf2 plays a role in these female-specific functions.
Sexually dimorphic circuitries underlie gendered behavior 13 . We have therefore examined the possible effects of Trf2 knockdown on sexually dimorphic neural structures. Here we focus on a particular neural cluster, mAL, which exhibits conspicuous sex differences in three respects, all of which depend on the presence (for masculinization) or absence (for feminization) of FruM: the number of neurons composing the cluster is 5 in females compared with 30 in males; none of the mAL neurons in females has the ipsilateral neurite, whereas some of the mAL neurons in males have it; and the contralateral neurite in the suboesophageal ganglion bifurcates near the tip in females but not in males 28 . We confirmed that all mAL neurons express TRF2 as well as FruM ( Supplementary Fig. 3). To visualize the entire structure of mAL neurons, we employed the mosaic analysis with a repressible cell marker (MARCM) technique 35 , which allows one to label the mAL cluster only in the left or right hemisphere without any interference from the mAL counterpart on the other side. When the mAL cluster was visualized as a neuroblast clone, the male-specific ipsilateral neurite appeared shorter in males with Trf2 knockdown than in control males (Fig. 2a, b). mAL neuroblast clones in fru hypomorphic mutant males exhibited a similar shortening of the ipsilateral neurite, which was ascribed to a reduction in the number of mAL neurons with a long ipsilateral neurite as resulting from sexual transformation of single cells 25 .
To determine whether Trf2 knockdown similarly increases the mAL neurons without the male-specific neurite at the expense of those with the male-specific neurite, we generated single-cell clones of mAL neurons. Notably, Trf2 knockdown in mAL neurons of male flies resulted in a marked increase in the proportion of mAL neurons without the male-specific neurite, compared with control flies: 66.67% in Trf2 knockdown vs. 33.33% in control flies ( Fig. 2e-g). This effect of Trf2 knockdown is reminiscent of the effect of a reduction in functional of FruM doses by the fru hypomorphic mutation (e.g., the proportion of single clones without the male-specific neurite in fru NP21 /fru 2 was 50% in ref. 25 ). We conclude that Trf2 knockdown in mAL neurons phenocopies the fru mutant effect in that the mAL neurons without the male-specific neurite are produced at the expense of those with the male-specific neurite.
In contrast to the robust effect of Trf2 knockdown on malespecific neurite formation, the shape of the contralateral neurite was marginally affected by Trf2 knockdown. In females, Trf2 knockdown resulted in formation of the male-specific ipsilateral neurites in three out of seven brains (Fig. 2c, d), implying that Trf2 exerts a feminizing effect on mAL neurons in wild-type females that lack FruBM.
To determine whether Trf2 is involved in the development of fru-dependent sex-specific characteristics of other neurons, we labeled foreleg sensory afferents, the central projections of which cross the midline in males but not females 36,37 . The extent to which the sensory fibers cross the midline was quantified by the midline crossing score (MCS) 36 , which was based on the fluorescent intensity measured at the midline relative to that at "blank" regions where no sensory fibers run (Fig. 3f). This analysis revealed that the MCS was significantly decreased in males by Trf2 knockdown as directed by poxn-GAL4, a driver for gustatory receptor neurons (Fig. 3a-c, g). Conversely, Trf2 knockdown in females increased the midline crossing ( Fig. 3g), suggesting the feminizing ability of Trf2 in the absence of FruBM. In hypomorphic fru (fru 2 /fru sat ) mutant males where the midline crossing was reduced (Fig. 3d, e, g), Trf2-S + overexpression partially restored the midline crossing (Fig. 3h). We conclude that Trf2 participates in establishing neural sex-specific characteristics in two opposing ways, i.e., masculinizing neurons in the presence of FruBM and feminizing neurons in the absence of FruBM.
TRF2-S enhances Fru-induced repression of robo1. TRF2-S activates transcription from a large number of promoters that lack a TATA-box 38-42 , whereas FruBM was recently shown to repress transcription from the promoter of robo1, a gene encoding a transmembrane receptor that governs the presence or absence of the ipsilateral neurite in mAL neurons 27 . We therefore examined the effect of TRF2-S on robo1 transcription in the absence or presence of FruBM with reporter assays in S2 cells. We used a reporter expressing luciferase under the control of a 1.7 kb robo1 promoter segment that carries a FruBM-binding site and thus is repressible by applying FruBM (Fig. 4a). We found that Trf2-S transfection significantly elevated the reporter activity (Fig. 4b). fruBM transfection (at 3 ng per 2 × 10 6 cells) markedly repressed the reporter activity as expected (Fig. 4b). The FruBM effect to repress the reporter activity was dose-dependent and no discernible effect was obtained when one-third of the amount (i.e., 1 ng) was used for transfection (Fig. 4c). Remarkably, TRF2-S, which otherwise activates transcription, repressed transcription from the robo1 promoter in the presence of a small amount of FruBM, even though this amount of FruBM alone was unable to repress transcription of the reporter (Fig. 4c). We conclude that TRF2-S exerts two contrasting effects on robo1 transcription, i.e., activation and repression, dependent on the absence or presence of FruBM.
TRF2-S binds to the robo1 promoter. The question arises as to whether a TRF2-S-containing complex binds to the robo1 promoter to activate and repress its expression. To clarify this point, Midline crossing score a-(c1+c2)/2 (b1+b2)/2 = Midline crossing score Midline crossing score we conducted electrophoretic mobility shift assays with robo1 promoter fragments in the presence and absence of TRF2-S, which was tagged with V5. A 101 bp fragment (the A fragment, Fig. 5a) derived from the region further upstream of FruBMbinding motif (FROS) revealed a retarded band, which disappeared in the presence of unlabeled fragment-A or exhibited a supershift upon the addition of an anti-V5 antibody that recognizes TRF2 (Fig. 5b). The mobility shift was not detected in the presence of FruBM, provided that TRF2-S was not co-transfected. Importantly, the A fragment contains a palindrome stretch, 5′-TATCGATA TATCGATT-3′, each moiety of which coincides with the binding consensus sequence DNA-replication element (DRE)-association motif (DREAM), to which DREF binds 43 . Furthermore, the polypyrimidine initiator motif TCT exists immediately 5′ to DREAM (Fig. 5a). Notably, TRF2 has been shown to associate with DREF 44 and requires TCT for activating transcription 42 . This raises an intriguing possibility-namely, this palindrome may serve as the site for TRF2-robo1 promoter interactions via DREF and TCT.
To evaluate the possibility that TCT and/or DREAM mediate the action of TRF2 to regulate robo1 transcription, the effects of deleting either or both of these motifs from the 1.7 kb promoter fragment on the reporter activity were examined. When TCT was deleted (ΔTCT), TRF2 was unable to increase the reporter activity, while it retained the ability to enhance the repressor action of FruBM (Fig. 5d). In contrast, loss of DREAM (ΔDREAM) impaired the ability of TRF2 not only to increase the reporter activity but also to enhance the repressor activity of FruBM (Fig. 5d). As expected, deleting TCT and DREAM altogether (ΔTCT and ΔDREAM) made the reporter unresponsive to TRF2 (Fig. 5d), as was ΔDREAM. We conclude that both TCT and DREAM are required in the robo1 promoter for increasing the transcriptional activity in response to TRF2, whereas DREAM alone mediates the action of TRF2 to enhance the repressor activity of FruBM.
TRF2-S associates with FruBM in vivo. Importantly, the A fragment has been shown not to contain a FruBM-binding site.
An intriguing possibility is that the TRF2-containing complex interacts with the FruBM-containing complex upon binding to the respective target sequences, which are positioned close to each other on the genome (Fig. 5a). In an attempt to elucidate the molecular basis for this phenotypic similarity between Trf2 and fru, we conducted immunoprecipitation assays to test the possibility that TRF2-S and FruBM interact with each other as components of a transcriptional machinery. To improve the power of the immunoprecipitation assays to detect protein-protein interactions, we conducted immunoprecipitation assays with lysates of S2 culture cells that had been transfected with tagged versions of Trf2-S and fruBM, confirming the association of TRF2-S and FruBM. We found that precipitates with an anti-TRF2 antibody contained not only TRF2 but also FruBM, as detected in western blotting by the antibodies that recognize the respective proteins (Fig. 5c). We suggest that the TRF2-S-containing complex interacts with the FruBM-containing complex in vivo, while each complex binds to a distinct site on the robo1 promoter.
Phenotypic outcome of Trf2, fru, and robo1 interrelationships. To validate the importance of TRF2 actions on robo1 transcription in establishing the FruBM-dependent sex-specific traits, we examined the effect of robo1 knockdown on the Trf2-dependent induction of mAL neurons with the male-specific neurite. The increase in the proportion of single-cell mAL clones that lack the male-specific neurite by Trf2 knockdown was, though statistically non-significant, ameliorated by the additional knockdown of robo1 ( Supplementary Fig. 4). We also examined whether Trf2 knockdown results in the precocious wing switching, a change in the male courtship posture uniquely induced in male mutant flies (e.g., robo1 Δ1 /robo Δ3 ) whose robo1 gene lacks a few nucleotides that compose the core binding motif for FruBM binding 27 . Strikingly, Trf2 knockdown in fru GAL4 -expressing neurons phenocopied the robo Δ1 /robo Δ3 mutants by exhibiting the precocious wing switching (Supplementary Fig. 5a, c). Furthermore, the effect of Trf2 knockdown to induce the precocious wing switching was completely abrogated by simultaneous robo1 knockdown ( Supplementary Fig. 5b, c). These results are consistent with the All experiments were conducted as triplicate and the mean ± SEM of relative luciferase activities are shown. Statistical differences were evaluated by one-way ANOVA followed by the Bonferroni's multiple comparison test. **P < 0.01, *P < 0.05; ns not significant NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01623-z ARTICLE idea that the disturbance of sex-specific traits by Trf2 knockdown is, at least in part, expressed through the misexpression of robo1 in fru-positive neurons including mAL neurons. It remains to be determined whether TRF2 regulates transcription of the FruMtarget genes other than robo1 and, if so, what aspects of sexspecific neurobehavioral traits are controlled by such mechanisms.

Discussion
Although it is well recognized that FruBM plays a key role in establishing neural sexual dimorphisms, it remains largely unknown how the presence or absence of FruBM affects the transcription of target genes that induce neuronal sex differences 13 . By focusing on the fru-expressing neuronal cluster mAL, which displays sexual dimorphisms in the cell number, the contralateral neurite branching pattern, and the presence or absence of the ipsilateral neurite 28 , we successfully demonstrated that the core promoter regulator TRF2-S preferentially supports FruBM in the male-specific ipsilateral neurite formation, with only minor effects on two other aspects of mAL sexual dimorphisms. TRF2-S was found to assist FruBM by enhancing the repressor effect of FruBM on the promoter of robo1, the gene encoding a receptor for the axon guidance factor that inhibits the male-specific neurite formation in mAL neurons 27 . This finding was unexpected because TRF2 typically functions as a transcriptional activator rather than a repressor, being associated with the TATA-less promoter present in over 1000 genes in the genome [39][40][41] . Our results further demonstrated that TRF2-S is immunoprecipitated with the FruBM-protein complex, which was previously shown to contain chromatin regulators such as Bonus, HP1a, and HDAC1 25 . It is therefore plausible that TRF2-S is involved in chromatin organization in addition to its role as a core promoter recognition factor. In keeping with this idea, mouse mutants null for Trf2 have been shown to exhibit a defect in chromatin condensation in early spermatids, leading to impaired spermiogenesis and male infertility 45 . Transcriptional repression by overexpressed TRF2 has been reported in TATA-dependent transcription, where TRF2 is postulated to absorb TFIIA 46 , a general transcription factor required for the initiation of transcription from the TATA-dependent promoter. Here we demonstrated that TRF2 acts on the TATA-box harboring the robo1 promoter as a transcriptional activator in the absence of the masculinizer protein FruBM, or as a transcriptional repressor in its presence. Our analysis demonstrated that TRF2 may interact with the robo1 promoter as mediated by the DREF-binding motif DREAM to enhance the repressor action of FruBM. Although DREF is generally considered to be a transcriptional activator, the DREFcontaining complex turns into a transcriptional repressor when DREF recruits α-thalassemia and mental retardation X-linked syndrome (ATRX) to DREAM 47 .
Another important finding in this study was that, despite the striking effect of Trf2 manipulation on the ipsilateral neurite formation, two other aspects of mAL sexual dimorphism, i.e., the different cell number and the branching pattern of the contralateral neurites, were only moderately affected by the manipulation of Trf2. Notably, all three aspects of the sexual dimorphisms of mAL are known to be FruBM-dependent 28 , implying that FruBM assembles different sets of transcriptional regulators to masculinize each of the three sex-specific traits. Therefore, the sex-type specification of a neuron likely proceeds under a layered system of transcriptional control that is spatially and temporally dynamic, and FruBM probably orchestrates this sophisticated network of transcription to render the brain circuitry distinct between the two sexes.
Behavioral assays. Flies were collected upon eclosion and reared individually under a 12 h:12 h light-dark cycle at 25°C. About 5-7-day-old test males and 3-5day-old Canton-S virgin females were used for the behavioral assays. To measure the CI, a male of each genotype and a Canton-S virgin female were paired in a small round chamber of 8 mm in diameter and 3 mm in height. The behaviors of paired flies were video recorded for 1 h. The video segment for the first 5 min after introducing the virgin female was used to calculate the CI, which was defined by the time spent by a test male for displaying courtship elements, including orientation, tapping, following, wing extension/vibration, and attempted copulation. For the WEI, only the time spent for vibrating/extending the wing was measured during the first 5 min after introducing the virgin female. The copulation success was measured based on whether the test male copulated within the 1-h observation period. Locomotion activity was estimated by quantifying the circular walk along the wall of a round chamber: the number of complete circles in both the clockwise and anti-clockwise directions made by a test male within 3 min after introduction to the chamber was counted and the values were transformed into the locomotion rate in mms −1 . The wing switching index was estimated by the method as described in ref. 27 . For every experiment, sample size was chosen based on preceding publications ensuring that it has adequate power to detect a meaningful differences across data sets.
Dissection, immunohistochemistry, and imaging of tissues. For immunostaining, the central nervous system (CNS) of 3-5-day-old files was dissected in cold phosphate-buffered saline (PBS) with sharp forceps (Dumont #5). After dissection, the CNS was fixed in 4% paraformaldehyde for 1 h, followed by two 30 min washings in PBS with 0.2% Tween 20 (0.2% PBT). Then the CNS was kept in blocking buffer containing normal goat serum and 0.2% PBT for 1 h at room temperature. Immunostaining was performed using a rabbit anti-TRF2 antibody (a generous gift from James Kadonaga; at a dilution of 1:100), a guinea pig anti-FruMale antibody (a product of our laboratory; 1:500 dilution), a rabbit anti-GFP antibody (Invitrogen; 1:500 dilution), and a mouse anti-nc82 antibody (Developmental Studies Hybridoma Bank (DSHB), University of Iowa, Iowa City, IA; 1:50 dilution). CNS tissues were incubated with the primary antibody for 3 days, then subjected to six 10-min washings in 0.2% PBT. Secondary antibodies used were Alexa Fluor 488 anti-rabbit IgG and anti-guinea pig IgG, Alexa Fluor 546 antimouse IgG and anti-rat IgG, and Alexa Flour 647 anti-mouse IgG (Invitrogen, Carlsbad, CA; 1:200). The CNS was stained for 2 days with the secondary antibody and then washed for 30 min in 0.2% PBT twice. Finally, the CNS was mounted on a slide glass with 80% (v/v) glycerol in PBS. For the observation of germ cells ( Supplementary Fig. 1), ovaries were dissected in PBS and blocked for 30 min in PBS containing 4% paraformaldehyde. The ovaries were washed three times in PBT, blocked for 1.5 h in a blocking buffer, and then incubated with a primary antibody at 4°C overnight. The primary antibodies used were the rat anti-TRF2 (current study; 1:100 dilution) and rabbit anti-Vasa (d-260, Santa Cruz Biotechnology; 1:50 dilution) antibodies. Images were acquired with a Zeiss LSM 510 META confocal microscope using ZEISS LSM Image Browser software. All images were acquired with either 20× Plan-Apo/0.8 or 40× Plan-Apo/0.95 lenses. Images were acquired at a resolution of 512 μm × 512 μm with 1 μm intervals.
Clonal analysis of mAL neurons. We used a fru NP21 -GAL4 line to label mAL neurons. The somatic clones were produced using the MARCM method 35 . The flies with the genotype y hs-flp/Y or w; FRTG13 UAS-mCD8::GFP/FRTG13 tub-Gal80; fru NP21 ,UAS-Dcr2/+ were used as the control males or females. The genotype of flies used in clonal Trf2 knockdown experiments was y hs-flp/Y (for males) or w (females); FRTG13 UAS-mCD8::GFP/FRTG13 tub-Gal80; fru NP21 /UAS-Trf2 RNAi (VDRC# 10443). For the production of neuroblast clones in mAL neurons, embryos at 0-24 h after egg-laying (AEL) were heat shocked at 37°C in a water bath for 1 h. For the production of single-cell clones of mAL neurons, larvae at 4-6 days AEL were heat shocked at 37°C for 1 h. Flies to be tested were reared at 29°C after the heat shock in order to enhance the expression of transgenes.
TRF2-S expression vectors. TRF2-S was cloned from Drosophila embryo complimentary DNAs and expressed with an N-terminal HA tag and C-terminal V5 tag in S2 cells using either the pact or pMT/V5-His C expression vector (Invitrogen). The sequence encoding HA-Trf2-S-V5-His was cloned independently in the pUAST and pJFRC81 vectors. These vectors were used to generate flies expressing UAS-HA-Trf2-S-V5-His by standard injection protocols.
Reporter assays. Reporter assays were carried out with the robo1 promoter luciferase reporter as described previously 27 . The pGL3-promoter vector carrying a 1.7 kb robo1 promoter fragment was used as a reporter construct. The pRL-TK Renilla luciferase vector (Promega) served as an internal control. About 100 ng of a reporter construct and 10 ng of an internal control were co-transfected into S2 cells (2 × 10 6 cells) with either pact-FLAG-fruBM and/or pact-HA-Trf2-S-V5-His and pact-MCS 25 using FugeneHD (Roche Diagnostics, Indianapolis, IN). Cells were lysed after 36-48 h of transfection with a passive lysis buffer (Promega), and luciferase activity was measured using a Dual-Luciferase Assay System (Promega). To standardize the transfection efficiency, the reporter luciferase activity of each sample was normalized to the corresponding control Renilla luciferase activity: the relative luciferase activity of a reporter construct was calculated relative to that carrying only an empty pact-MCS plasmid. All experiments were carried out in triplicate; the relative luciferase activities are shown as the mean ± SEM.
Electrophoretic mobility shift assays. pMT-FLAG-fruBM and pMT-HA-Trf2-S-V5-His were transfected together or individually into S2 cells, and the expression of proteins was induced by the addition of copper sulfate. Electrophoretic mobility shift assays experiments were carried out using a DIG Gel Shift Kit (Roche Diagnostics). Binding reactions were established as follows: 1× binding buffer (20 mM Hepes, pH 7.5, 10 mM (NH 4 ) 2 SO 4 , 0.2% Tween 20, 30 mM KCl, and 50 μM ZnSO 4 ), 0.1 μg µl −1 poly [d(I-C)], 5 ng µl −1 poly L-lysine, 5 µl of nuclear extract, and 1 ng of a digoxigenin (DIG)-labeled probe in a final volume of 20 µl. A 25-to 200-fold excess of unlabeled double-stranded oligonucleotide was included in an experiment to detect competing reactions. Reactants were incubated at room temperature for 20 min. For antibody-binding assays, the mouse anti-V5 antibody was added to the sample solution and incubated for 15 min prior to binding assays. Binding reactions were resolved on 5% polyacrylamide gels and transferred to a nylon membrane (Life Technologies, IB801001). The membrane was then UV cross-linked, blocked and probed with anti-DIG antibodies. DIG-labeled DNA-protein complexes were detected by chemiluminescence using an Image-Quant LAS 4000 system (Fujifilm).
Midline crossing score analysis. The MCS was calculated as described previously 25 . Briefly, stacked images of each sample were summed up using ImageJ, and a circle of 8-10 μm (marked as "a" in panel f of Fig. 3) was drawn on the resultant image so that the circle was centered at the prothoracic midline, where trans-midline axons are expected to run in males. The fluorescent intensity within the circle marked "a" was measured to quantify the level of midline crossing by fibers. Similarly, the fiber tracts locating lateral to the midline were quantified within circles "b1" and "b2" for the fluorescent intensity. To normalize for the background fluorescent level, the areas with no fibers delineated by circles "c1" and "c2" were also sampled. The MCS was calculated as: MCS = [a − (c1 + c2)/2]/[(b1 + b2)/2].
Antibody production. To generate the anti-TRF2 antibody, which recognizes a stretch of 14 amino acids (KLQNKRPRYNDPGT) in the C-terminus of TRF2, a 14-amino-acid peptide was synthesized, and rat antiserum against this peptide was raised and affinity-purified.
Statistical analysis. Statistical analyses were done by GraphPad Prism 7.0b software.
Data availability. The data sets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.