The Drosophila hnRNP M homolog Rumpelstiltskin regulates nanos mRNA localization

The anterior-posterior body axis of the Drosophila embryo is established through the asymmetric distribution of Bicoid (Bcd) and Nanos(Nos) proteins, which form gradients emanating from the anterior and posterior poles of the embryo, respectively (Driever and Nüsslein-Volhard, 1988; Gavis and Lehmann, 1992). Bcd functions as a transcriptional activator and a translational repressor to regulate expression of genes required for head and thorax development(Driever and Nüsslein-Volhard,1989; Dubnau and Struhl,1996; Rivera-Pomar et al.,1996; Struhl et al.,1989). As a translational regulator, Nos represses synthesis of the transcriptional repressor Hunchback (Hb) in the posterior of the embryo,permitting expression of genes required for abdominal development(Hülskamp and Tautz,1991; Tautz and Pfeifle,1989). More recently, additional roles for Nos have been identified in germline stem cell divisions, germ cell development, and dendrite morphogenesis (Forbes and Lehmann, 1998; Kobayashi et al., 1996; Wang and Lin,2004; Ye et al.,2004).

The Bcd and Nos protein asymmetries are generated through the asymmetric localization and consequent translation of their respective mRNAs(Berleth et al., 1988; Driever and Nüsslein-Volhard,1988; Gavis and Lehmann,1992; Wang and Lehmann,1991). Both bcd and nos mRNAs are synthesized maternally and become localized to opposite poles of the egg during oogenesis. Whereas anterior localization of bcd depends on active transport along microtubules (Cha et al.,2001; Weil et al.,2006), posterior localization of nos occurs by passive diffusion and entrapment at the posterior by the specialized posterior cytoplasm or germ plasm (Forrest and Gavis, 2003; Weil et al.,2006). Localization of nos to the germ plasm is essential to activate nos translation and produce the critical concentration of Nos required to repress translation of hb mRNA(Gavis and Lehmann, 1992; Gavis and Lehmann, 1994). Thus, when nos localization is abolished by mutations in germ plasm components such as oskar (osk) or vasa(vas), Nos is not produced and the resulting embryos lack abdominal segments. Although nos is highly concentrated at the posterior of the embryo, its localization is inefficient, such that the majority of nos mRNA remains unlocalized(Bergsten and Gavis, 1999). The obligate linkage of nos mRNA localization and translation prevents synthesis of Nos in the anterior of the embryo, where it is deleterious to head and thorax development (Gavis and Lehmann, 1992; Gavis and Lehmann, 1994; Wharton and Struhl, 1989). Recently, posterior localization of noshas also been shown to be essential for nos function in germ cell development (Gavis et al.,2008).

bcd and nos are two of a growing number of mRNAs known to be localized within Drosophila oocytes and embryos. Since different mRNAs are targeted to different destinations, specificity for localization pathways must be encoded in the transcripts and conferred on the cellular localization machinery by factors that recognize these signals. Cis-acting sequences that direct intracellular localization have been identified in a number of localized mRNAs and are usually located in 3′ untranslated regions (3′UTRs) (Gavis et al.,2007; St Johnston,2005). Posterior localization of nos mRNA is directed by a large localization signal within the nos 3′UTR that can be subdivided into four partially redundant localization elements(Gavis et al., 1996a). Individual localization elements exhibit varying degrees of competence to mediate posterior localization, but at least three contiguous elements are required to produce wild-type levels of Nos and proper abdominal development(Bergsten and Gavis, 1999). Such functional redundancy is characteristic of a variety of mRNA localization signals including the bcd, Xenopus Vg1 and fatvg, and yeast ASH1 signals (Chan et al.,1999; Gautreau et al.,1997; Gonzalez et al.,1999; Macdonald and Kerr,1998).

Previous genetic and biochemical studies suggest that different nos localization elements are recognized by distinct cytoplasmic factors that package nos into a ribonucleoprotein (RNP) complex competent to interact with the germ plasm components(Bergsten and Gavis, 1999; Bergsten et al., 2001). However, the specific localization factors that recognize the noslocalization elements and target nos to the germ plasm have remained elusive. Although mutations that cause defects in nos localization have been identified in numerous genetic screens, the effects of these mutations on nos are most likely to be indirect, as they disrupt localization or assembly of germ plasm. A recent genetic screen for nos localization factors identified the chaperone Hsp90 (also known as Hsp83 - FlyBase) as being important for nos localization, though its role also appears to be indirect (Song et al., 2007). Given the redundancy within the noslocalization signal and, consequently, the potential for functional redundancy among proteins that interact with different nos localization elements, genetic approaches may fail to identify direct-acting localization factors.

We have therefore taken a biochemical approach to identify noslocalization factors, starting with an individual nos localization element. Using a new RNA-affinity purification strategy, we have isolated Rumpelstiltskin (Rump), a member of the heterogeneous nuclear ribonucleoprotein (hnRNP) family. Rump appears to be the Drosophilahomolog of hnRNP M (also known as HNRPM in mouse), which has been previously implicated as a splicing factor in both invertebrates and vertebrates(Gattoni et al., 1996; Hase et al., 2006). A combination of biochemical and genetic data presented here show that Rump is a direct-acting nos mRNA localization factor. Rump associates with nos mRNA in vivo and binding of Rump to nos mRNA in vitro requires a repeated motif within the nos localization signal. Moreover, binding of Rump in vitro correlates with the ability of this motif to confer localization function in vivo. Through the analysis of a rump null mutant, we demonstrate a specific role for rump in posterior localization of nos. To our knowledge, this is the first example of an hnRNP M homolog regulating the localization of an mRNA target.

Three tandem copies of the nos +2′ localization element(Bergsten and Gavis, 1999) were cloned into the BglII site of pTRAP6 (gift of H. Krause, University of Toronto, Toronto, Canada) to generate pTRAP6-nos+2′-3X. pTRAP6 and pTRAP6-nos+2′-3X were linearized with ScaI and HindIII, respectively, for in vitro transcription by the mMessage mMachine Kit (Ambion) supplemented with 625 nM [α-³²P]UTP. Radiolabeled capped TRAP+2′-3X (779 nucleotide) and TRAP-ctrl (1261 nucleotide) RNAs were purified by G-50 Sephadex Quick Spin Column (Roche). MCP was cloned from pCaS-P_hsp83-MCP-GFP(Forrest and Gavis, 2003) into pGEX-4T-1, fusing MCP to the C-terminus of GST. GST-MCP was purified after expression in E. coli using glutathione-agarose (Sigma) according to the manufacturer's instructions.

Extract was prepared from 0- to 2-hour Oregon R embryos in TPB [60 mM HEPES pH 7.5, 10 mM MgCl₂, 150 mM NaCl, 0.1% Triton X-100, 10% glycerol,1× EDTA-free Complete Protease Inhibitors (Roche)] as described(Bergsten et al., 2001). All subsequent purification steps were performed at 4°C. For each sample, 1 ml of extract (5 mg/ml total protein) was precleared with 40 μl streptavidin-agarose and 40 μl glutathione-agarose for 30 minutes. Precleared extract was incubated with 11 μg radiolabeled TRAP-ctrl or TRAP+2′-3X RNA for 15 minutes, then rotated with 200 μl streptavidin-agarose in a 1.2 ml RNase-free minicolumn (BioRad) for 2 hours. The resin was washed with 25 volumes of TPB and bound RNP complexes were eluted by incubation with 5 mM d-biotin in TPB for 1 hour, with a second elution for 30 minutes. The eluates were pooled and incubated with 100μl glutathione-agarose pre-bound with 1 mg GST-MCP for 2 hours with rotation. The resin was washed with 15 volumes of TPB and proteins were eluted by treatment with 2 μg RNase A in 200 μl TPB, 2× 30 minutes. Eluates were pooled, concentrated by TCA precipitation, then resuspended in boiling sample buffer (Gavis et al.,1996b). Samples were separated by SDS-PAGE, with the amount of each sample loaded adjusted for the amount of RNA captured, as determined by scintillation counting. Proteins were visualized by SYPRO Ruby (Invitrogen)staining, and the indicated ∼70 kDa band was excised and analyzed by mass spectrometry.

The full-length rump coding region was amplified by RT-PCR from 0-to 2-hour Oregon R embryonic poly(A⁺) RNA using the primers 5′-AAACTGCAGAGCATGGACGCTAGTAAC-3′ and 5′-CCCAAGCTTAAAAGTATGTTAC-3′, which introduce PstI and HindIII sites at the 5′ and 3′ ends, respectively. Following digestion with PstI and HindIII, the PCR fragment was inserted between the PstI and HindIII sites of pMAL-c2(New England Biolabs), fusing Rump to the C-terminus of MBP. MBP-Rump was expressed and purified on amylose-agarose(Kalifa et al., 2006),exchanged into Storage Buffer (100 mM KCl, 25 mM HEPES pH 7.9, 0.5 mM EDTA,10% glycerol, 1 mM DTT, 0.5 mM PMSF), and stored at -80°C. For monoclonal antibody production, the MBP tag was removed using Factor Xa (a gift from F. Hughson, Princeton University, Princeton, NJ).

Radiolabeled RNA probe synthesis and UV-crosslinking were performed as previously described using 27 μg MBP-Rump per reaction(Bergsten et al., 2001). The full-length +2′ element [nucleotides 101-186 of the nos3′UTR (Gavis et al.,1996a)] was inserted between the SmaI and BamHI sites of the pBluescript derivative pBS-SKΔKP(Kalifa et al., 2006). The+2′ element deletions (Δ1, nucleotides 101-128; Δ2, 101-142;Δ3, 120-162; Δ4, 128-152; Δ5, 144-186) were generated by PCR(Δ1-3, Δ5) or with annealed oligos (Δ4) and cloned into pBS-SKΔKP as above. The +2′(A)G:C, +2′(D)G:C and+2′(AD)G:C mutants were generated by PCR, introducing G-to-C point mutations into one or both of the two CGUU motifs (A, nos 3′UTR nucleotide 120; D, nucleotide 140] in the pBS-SKΔKPnos+2′ plasmid. Plasmids were linearized with BamHI for probe synthesis. For immunoprecipitation of crosslinked protein, each RNase-treated reaction was incubated for 1 hour at 4°C in 200 μl IP buffer (10 mM Tris-HCl pH 7.4,100 mM NaCl, 2.5 mM MgCl₂, 0.5% Triton X-100, 1 mM DTT, 1×EDTA-free Complete Protease Inhibitors) with 6 μl Protein-G Dynabeads(Invitrogen) coated with 30 μl antibody. Beads were washed twice with IP buffer, then resuspended directly in boiling sample buffer and separated by SDS-PAGE.

The rump¹ allele was generated by imprecise excision of P element P{SUPor-P} CG9373^KG02834(Bellen et al., 2004). Excisions in trans to Df(3R)by416, which deletes the rumplocus, were screened by single-fly PCR(Mansfield et al., 2002) using primer pairs to amplify the genomic regions immediately upstream and downstream of the P element insertion site. Genomic deletions were identified by the absence of a PCR product. The rump¹deletion endpoints were determined by PCR amplification with primers flanking the deletion and sequencing. The nos^BN allele(Wang et al., 1994) was recombined onto the rump¹ and Df(3R)by416chromosomes to generate rump¹ nos^BN and Df(rump) nos^BN chromosomes used for the analysis of nos-tub:nos+2 in Fig. 7.

The nos-tub:nos+2 transgene is described by Gavis et al.(Gavis et al., 1996a). The nos-tub:nos+2(A)G:C and nos-tub+2(D)G:C transgenes are identical to nos-tub:nos+2 except that a G-to-C mutation was introduced into the first (A) or second (D) CGUU motif within the +2′sequences, using PCR. The rump rescue transgene contains a 6.5 kb BanII genomic fragment from BACR19N07 (GenBank AC009350) containing rump inserted into pCaSpeR4. The rump transgene was introduced into y w^67c23 embryos by Pelement-mediated germline transformation(Spradling, 1986) and multiple independent lines were isolated.

Northern blotting and immunoblotting were performed as previously described(Kalifa et al., 2006). A ³²P-labeled probe for rump was generated by random-hexamer labeling of the full-length rump EST clone GH11495. For immunoblotting, anti-Rump 5G4 was used at 1:2000, anti-Snf (gift of P. Schedl,Princeton University, Princeton, NJ) at 1:20,000. Proteins were detected using Lumi-Light Western Blot Substrate (Roche).

Ovaries from well-fed females were dissected in Schneider's medium(Invitrogen), rinsed with PBS and frozen in liquid N₂. Approximately 15 mg ovaries or dechorionated embryos per sample were homogenized on ice in 750 μl RNase-free RNA co-immunoprecipitation buffer(RCB) [50 mM HEPES pH 7.4, 200 mM NaCl, 2.5 mM MgCl₂, 0.1% Triton X-100, 250 mM sucrose, 1 mM DTT, 1× EDTA-free Complete Protease Inhibitors, 0.4 mM Pefabloc (Roche)], supplemented with 300 U RNasin(Promega). Subsequent steps were carried out at 4°C. Soluble extract was obtained by two sequential 5-minute centrifugations at 16,000 g, then precleared with 40 μl Protein-G Dynabeads for 30 minutes at 4°C. After removal of samples for immunoblotting and quantitation of RNA input, Rump was immunoprecipitated from 450 μl precleared extract by incubation for 1 hour with 20 μl Protein-G Dynabeads bound with anti-Rump 5G4 antibody. Immunoprecipitates were washed eight times with RCB. Half of the IP was resuspended in RCB, treated with 8 units RQ1 RNase-free DNase (Promega) for 10 minutes at room temperature, then extracted with phenol:chloroform. RNA was ethanol precipitated with 30 μg tRNA and 5μg glycogen as carrier and resuspended in DEPC-treated distilled H₂O for RT-PCR. The remainder of the IP was resuspended in 20 μl boiling sample buffer for immunoblotting.

For RT-PCR analysis, RNA samples were denatured for 3 minutes at 80°C,then chilled on ice. Half the denatured RNA was reverse transcribed using SuperScript II (Invitrogen) and oligo(dT)₁₅ for 2 hours at 42°C. The remaining half was treated similarly, except that SuperScript II was omitted. Thirty cycles of PCR amplification with Taq polymerase were performed using nos primers (5′-GCGATCAAGGCGGAATCG-3′ and 5′-ATAGGATCCGAAAGTGTTCCTTGCTA-3′).

Embryonic cuticle preparation and in situ hybridization were performed according to Gavis and Lehmann (Gavis and Lehmann, 1992). The statistical significance of embryonic phenotypes was determined by the χ² test. Immunostaining of ovaries was performed as previously described(Shcherbata et al., 2004). Embryo immunostaining was performed as described(Duchow et al., 2005), except that for anti-Rump immunofluorescence, dechorionated embryos were heat-fixed in 68 mM NaCl/0.03% Triton X-100, transferred to PBS, and the vitelline membranes were manually removed using a 30-gauge needle. The following antibodies and dyes were used: 1:100 anti-Rump 10C3 (immunofluorescence),1:100 anti-Rump 5G4 (immunohistochemistry), 1:10,000 rabbit anti-Vas (gift of R. Lehmann, Skirball Institute, New York), 1:1000 Alexa Fluor 568 goat anti-mouse (Invitrogen/Molecular Probes), 1:1000 Alexa Fluor 488 goat anti-rabbit (Invitrogen/Molecular Probes), 1:1000 Oregon Green 488 phalloidin(Invitrogen/Molecular Probes), 1:1000 Hoechst (ovaries), 1:1000 DAPI(embryos). Images were obtained using standard Nomarksi optics or a Zeiss LSM510 confocal microscope.

To identify candidate nos regulatory factors, we used a two-step tandem RNA affinity purification (TRAP) method(Nelson et al., 2007) to isolate in vitro-assembled RNA-protein complexes containing the 88 nucleotide nos +2′ localization element. We focused on this element because previous mutational analysis suggested it as a target for multiple localization factors, and biochemical experiments identified a +2′element binding activity of ∼75 kDa (p75) in ovarian and embryonic extracts (Bergsten et al.,2001). In vitro-synthesized RNA containing the dual TRAP purification tags and three tandem copies of the +2′ element(TRAP+2′-3X; Fig. 1A) was allowed to form RNA-protein complexes with early (0- to 2-hour) embryonic protein extract. These complexes were isolated first on streptavidin resin,utilizing the specific interaction between streptavidin and the S1 RNA aptamer of the TRAP tag. Bound complexes were eluted by the addition of d-biotin and applied to glutathione resin coated with GST-MS2 coat protein (GST-MCP), which captures TRAP-tagged RNA through the interaction of GST-MCP with the six MCP-binding stem-loops present in the tag. This second purification step served to reduce non-specific background proteins that interact with the streptavidin-coated agarose beads. Subsequent RNase digestion allowed specific elution of the protein components of the recaptured RNP complexes and eluted proteins were separated by SDS-PAGE. To control for the isolation of non-specific RNA-binding proteins, an RNA of similar length containing naïve sequence in place of the nos regulatory sequences (TRAP-ctrl; Fig. 1A)was used in a parallel purification.

A band of ∼70 kDa was enriched by the TRAP+2′-3X RNA across several independent purifications (Fig. 1B and data not shown). In each case, this band was excised and the protein components were analyzed by mass spectrometry. Sixteen peptides were recovered spanning the 633 amino acids of a predicted 67 kDa protein encoded by the gene CG9373. The less intense band of ∼70 kDa present in the TRAP-ctrl sample corresponds to the E. coli Hsp70 chaperone, a contaminant of the GST-MCP protein preparation (data not shown).

CG9373 encodes the Drosophila homolog of the human hnRNP M family of nucleic acid-binding proteins and is orthologous to the 59 kDa splicing factor Hrp59 of Chironomus tentans(Kiesler et al., 2005). Like other hnRNP M family members, the Drosophila homolog contains three RNA recognition motifs (RRMs). Based on cross-reactivity of antibodies against the C. tentans protein with a protein in Drosophila S2 cells, the Drosophila protein has been previously referred to as Hrp59 (Kiesler et al., 2005). The phenotypes of a mutation we have generated in CG9373, described below,have led us to name this gene rumpelstiltskin (rump).

The TRAP method is capable of purifying proteins that interact directly with the nos +2′ element RNA as well as proteins associated with the purified RNA-protein complex. To determine whether Rump interacts directly with the +2′ element, we tested the ability of purified full-length recombinant MBP-Rump fusion protein to bind to radiolabeled+2′ element RNA using a UV-crosslinking assay. MBP-Rump bound to the+2′ element, but not to the adjacent nos +1 localization element, indicating that the interaction with the +2′ element is specific (Fig. 2A). Crosslinking of MBP-Rump to the radiolabeled +2′ element RNA probe was competed by an excess of unlabeled +2′ RNA, but not by unlabeled +1 element RNA, verifying the specificity of MBP-Rump for +2′ element sequences (Fig. 2B).

To identify the binding site for Rump, several overlapping fragments of the+2′ element were assayed by UV-crosslinking to MBP-Rump(Fig. 2C). Binding of MBP-Rump to the 5′ half of the +2′ element (Δ2) was indistinguishable from binding to the full-length +2′ element, whereas no interaction was detected with the 3′ half (Δ5; Fig. 2C). Two non-overlapping subregions (Δ1 and Δ4) were each sufficient for binding by MBP-Rump, although binding was reduced relative to the overlapping Δ2 region or the full +2′ element (Fig. 2C). This result indicates the presence of multiple binding sites for Rump in the 5′ half of the +2′ element.

The Δ1 and Δ4 regions each contain a CGUU motif and linker-scanning mutations that overlap these motifs were previously shown to disrupt p75 binding and +2′ element function(Bergsten et al., 2001). To determine whether this sequence is recognized by Rump, a single G-to-C point mutation was introduced into each CGUU, individually or in combination[+2′(A)G:C, (D)G:C and (AD)G:C]. Either point mutation significantly disrupted binding of MBP-Rump to the +2′ element, when assayed by UV-crosslinking (Fig. 2C). By contrast, a linker-scanning mutation of six residues in the 3′ half of the +2′ element [+2′(F)](Bergsten et al., 2001) had no detectable effect on MBP-Rump binding, consistent with the inability of theΔ5 fragment to bind to MBP-Rump (data not shown). Thus, binding of Rump to the wild-type +2′ element requires the integrity of both CGUU motifs. The context of these motifs is, however, important for Rump recognition, as little binding is detected using a synthetic RNA probe consisting solely of four CGUU repeats (data not shown).

To test whether Rump binding sites are required for localization, we introduced the (A)G:C and (D)G:C point mutations into the nos-tub:nos+2 transgene (Fig. 2D). The nos-tub:nos+2 transgene combines the +1 and+2′ elements, which act synergistically to confer substantial, albeit not wild-type, posterior localization(Gavis et al., 1996a). Because only two of the four nos localization elements are present in the nos-tub:nos+2 transgene, localization signal redundancy is reduced. Thus, using this transgene permits us to assay the effects of +2′element mutations on localization function. Both the (A)G:C and (D)G:C mutation, when tested individually or together, severely compromised posterior localization in 100% of embryos, consistent with their effect on Rump binding(Fig. 2E,F and data not shown). Together, these results suggest that Rump might function in vivo to regulate the posterior localization of nos mRNA through the +2′element.

For analysis of Rump function in vivo, we raised monoclonal antibodies to recombinant Rump protein. In addition, we generated a rump mutation, rump¹, by excision of a P element inserted in the 5′UTR of CG9373 (Bellen et al.,2004). Imprecise excision of this P element produced a deletion beginning 426 nucleotides upstream and extending 501 nucleotides downstream of the insertion site (Fig. 3A). This lesion removes the transcription start site, the entire 5′UTR and the initial 152 codons, and is thus predicted to be a null mutation of CG9373. Homozygous rump¹ flies are viable through adulthood, facilitating analysis of rump mutant ovaries and embryos. In contrast to wild-type ovaries, ovaries from rump¹ mutant females lack detectable rump mRNA(Fig. 3B). Similarly, anti-Rump monoclonal antibodies fail to detect any Rump protein in mutant ovaries(Fig. 3C), indicating that rump¹ is a null mutation.

Homozygous rump¹ females showed a variable maternal-effect defect in the percentage of progeny that complete embryonic development, ranging from 29% (n=1137) to 65% (n=886). This defect was rescued by a single copy of a genomic transgene containing only the rump transcription unit, with 80% (n=607) to 86%(n=1402) of the corresponding embryos hatching as larvae(P<0.001). Embryos displaying this and other maternal-effect defects will be referred to as rump mutant embryos. In addition to the maternal-effect defect, homozygous rump mutant males produced by parents heterozygous for rump¹ exhibited reduced fertility. Further investigation of this phenotype showed that only 10% of males were able to fertilize virgin females within a 10-day period(n=29). Fertility was restored to 100% (n=30) by the rump transgene.

The similarities in electrophoretic mobility of Rump and p75 and their recognition of the +2′ element CGUU motifs suggest that Rump corresponds to p75. To test whether the two are equivalent, we performed UV-crosslinking with a nos +2′ element probe and extract prepared from either wild-type embryos, wild-type ovaries or rump¹ ovaries. As predicted, p75 binding activity was present in wild-type extract, but could not be detected in the rump¹ extract(Fig. 3D). Furthermore, after UV-crosslinking, radiolabeled p75 could be immunoprecipitated with anti-Rump monoclonal antibody (Fig. 3D). These results show that Rump is the previously described p75 binding activity and confirm that the binding specificity exhibited by recombinant MBP-Rump mirrors that of the endogenous protein.

To determine whether Rump interacts with nos mRNA in vivo as well as in vitro, Rump was immunoprecipitated from wild-type or rumpmutant ovary extract and RNA isolated from the immunoprecipitates was analyzed by RT-PCR using primers for nos. Polyadenylated nos mRNA was reproducibly detected in immunoprecipitates from wild-type ovary extract, but not in immunoprecipitates from rump¹ ovary extract(Fig. 4A). In addition, a variety of control antibodies failed to immunoprecipitate nos mRNA(Fig. 4A and data not shown). To address whether the interaction between Rump and nos mRNA is maintained in the embryo, a similar co-immunoprecipitation experiment was performed using extract from 0- to 2-hour wild-type or rump¹ embryos. RT-PCR showed that nos mRNA is also immunoprecipitated from embryonic extract by anti-Rump antibody in a Rump-dependent manner (Fig. 4B).

Immunoblotting experiments detected Rump in the ovary and early embryo(Fig. 3C), consistent with a role in nos mRNA localization. To determine the specific distribution of Rump, ovaries and embryos were stained with monoclonal anti-Rump antibodies. From the germarium onward, Rump was detected in the nuclei of the germline nurse cells and oocyte, as well as in the somatic follicle cell nuclei (Fig. 5A,B and data not shown). This distribution is consistent with the previously described nuclear localization of Rump in Drosophila S2 cells and of its ortholog Hrp59 in the C. tentans Balbiani ring system(Hase et al., 2006; Kiesler et al., 2005). We were unable to detect cytoplasmic staining above background levels in the ovary. In the newly fertilized embryo, Rump was present throughout the cytoplasm(Fig. 5C,D). Just prior to cellularization, Rump could be detected in both the somatic blastoderm nuclei and their surrounding cytoplasm, as well as in the internal yolk nuclei(Fig. 5E,F). However, it was noticeably excluded from the nuclei of the germline precursors, the pole cells, which are marked by Vas (Fig. 5E′,F′). This nuclear exclusion was maintained during the subsequent migration of the pole cells to the interior of the embryo (data not shown).

The interaction of Rump with nos mRNA in vivo and the ability of Rump to recognize the nos +2′ element in vitro support a role for Rump as a nos mRNA localization factor. rump¹homozygous embryos exhibited a weak, though significant, segmentation defect that was rescued by the rump transgene, although this phenotype was variable in its penetrance (Fig. 6A). Such a mild defect is unsurprising given the redundancy of localization elements in the nos 3′UTR, as elimination of factors that recognize individual elements would be expected to produce only minor effects on nos localization. To devise a more sensitive assay for defects in nos localization, we reasoned that reducing the levels of nos mRNA would sensitize the embryo to defects in the noslocalization machinery. Indeed, reducing nos gene dosage to one copy in rump mutant embryos significantly increased both the frequency and severity of abdominal segmentation defects(Fig. 6A). The fourth and fifth abdominal segments, which are most sensitive to a reduction of nosactivity, were most frequently affected in this assay, consistent with a defect in nos localization. Importantly, loss of rump did not affect posterior localization of osk mRNA, Osk protein level, or posterior accumulation of Vas (see Fig. S1 in the supplementary material),supporting a direct effect on nos.

To complement the above analysis, we partially compromised the noslocalization machinery by using a null mutation in the germ plasm component tudor (tud^tux46)(Thomson and Lasko, 2004). In embryos heterozygous for tud^tux46, nos levels were unchanged, but mild abdominal patterning defects were observed in a small percentage (4%) of embryos (Fig. 6B and data not shown). Elimination of one or both copies of rump in this sensitized genetic background resulted in an enhancement of the abdominal segmentation defects in a dosage-dependant manner, with 44%of embryos showing segmental deletions and fusions in the absence of rump (Fig. 6B and data not shown). Similar results were observed using a second tud allele(tud^WC8, data not shown). Once again, the fourth and fifth abdominal segments were most often affected, consistent with a role for rump in the regulation of posterior nos activity.

Finally, we reduced nos localization signal redundancy by using the nos-tub:nos+2 transgene(Gavis et al., 1996a;described above), which confers substantial, albeit not wild-type, posterior localization. Since this transgene includes sequences required for nos translational control, all of the nos activity that is produced results from the translation of localized nos-tub:nos+2 mRNA(Gavis et al., 1996b). A single copy of the nos-tub:nos+2 transgene partially rescued the abdominal defects of nos mutant embryos, such that the majority developed four or more abdominal segments and over 30% were nearly wild-type,with seven to eight segments (Fig. 7A). In the absence of rump, the ability of the nos-tub:nos+2 transgene to rescue abdominal segmentation was severely compromised and more than 90% of embryos developed fewer than four abdominal segments (Fig. 7A). As a molecular measure of abdominal segmentation, we monitored expression of the pair-rule gene even-skipped (eve), which is expressed in seven stripes corresponding to even-numbered segments(Macdonald et al., 1986). nos mutant embryos expressing nos-tub:nos+2 had four to seven eve stripes, with over 60% of embryos displaying the complete set of seven stripes (Fig. 7B). In the absence of rump, this distribution was shifted, such that the majority of embryos (66%) had fewer than seven stripes(Fig. 7B). Eve stripes four to seven, which designate the abdomen, were most sensitive to loss of rump (Fig. 7C),consistent with the abdominal segmentation defects of these larvae.

We hypothesized that the reduction in posterior nos activity in nos-tub:nos+2 embryos lacking rump results from decreased posterior localization of nos-tub:nos+2 mRNA. We therefore carried out in situ hybridization to nos-tub:nos+2 mRNA in the presence or absence of rump. Consistent with the partial rescue exhibited by the nos-tub:nos+2 transgene, nos-tub:nos+2 mRNA showed variable posterior localization, although the majority of embryos (77%) had qualitatively `substantial' (++) localization and only 1% of embryos showed no detectable (-) posterior localization (Fig. 7D,E). However, in the absence of rump, only 42% of embryos showed `substantial' posterior localization of nos-tub:nos+2RNA, whereas 38% showed weak (+) or diffuse localization and 20% of embryos had no detectable localization (Fig. 7D,E). Thus, the abdominal segmentation defects observed with loss of rump can be explained by loss of nos mRNA localization. Moreover, these results demonstrate a role for rump in nosmRNA localization.

Using a biochemical approach, we have identified Rump, the Drosophila hnRNP M homolog, as a bona fide nos mRNA localization factor. Rump corresponds to the p75 binding activity that was previously shown to recognize the nos +2′ localization element(Bergsten et al., 2001). Recombinant and endogenous Rump recognize two CGUU motifs in the +2′element. A point mutation within either of these motifs is sufficient to disrupt both interaction with Rump and mRNA localization function. The demonstration that sequences required for Rump binding are also required for posterior localization provides strong evidence that Rump plays a direct role in mediating nos mRNA localization through its interaction with the nos 3′UTR.

Rump was previously identified in Drosophila S2 cells through its cross-reactivity with an antibody to the orthologous C. tetansprotein, Hrp59 (Kiesler et al.,2005). In our experiments, Drosophila ovarian, embryonic and S2-cell Rump protein migrate with an apparent mobility of ∼70 kDa,consistent with its calculated molecular weight of 67 kDa. Mammalian hnRNP M proteins have been implicated in splicing through their association with pre-mRNA at an early step in spliceosome assembly(Kafasla et al., 2002). Studies in Chironomus have shown that Hrp59 binds to pre-mRNA co-transcriptionally and remains associated with the RNA until it reaches the nuclear envelope, and recent analysis of the Drosophila S2 cell protein shows that it regulates alternative splicing of its own mRNA(Hase et al., 2006; Kiesler et al., 2005).

Our results indicate that Rump has multiple functions at various stages of the Drosophila life cycle. In addition to defects in abdominal segmentation, many embryos produced by rump mutant females arrest development prematurely and adult rump mutant males are largely sterile. Neither the maternal-effect developmental arrest nor the zygotic sterility phenotypes is characteristic of nos mutants, indicating that Rump regulates both nos-dependent and nos-independent processes. This functional pleiotropy is consistent with the isolation of numerous Rump target mRNAs from S2 cells, the steady state levels of which are decreased by RNAi knockdown of rump(Kiesler et al., 2005). Since nos is not expressed in S2 cells, its behavior in this assay cannot be determined. However, we detect no alteration in the abundance or splicing of nos mRNA in rump mutant ovaries or embryos. Instead, our data demonstrate a specific and previously unidentified role for Rump as an mRNA localization factor.

The different regulatory functions of Rump might be distinguished by distinct recognition motifs in Rump target mRNAs. Rump protein isolated from S2 cells by immunoprecipitation with anti-Hrp59 antibody binds preferentially to a purine-rich motif with similarity to exonic splicing enhancers. This motif, with GGAGG at its core, is enriched among Rump target mRNAs in S2 cells identified by whole-genome analysis(Kiesler et al., 2005). The GGAGG core sequence does not exist within the nos +2′ element,nor within the entire nos 3′UTR. The ability to recognize different motifs might be conferred by different RRMs within Rump. Similarity between RRM3 and the human splicing regulator ASF (also known as SF2 and SFRS1- Human Gene Nomenclature Database) RRM suggests that RRM3 might be specifically responsible for interaction with splicing targets(Kiesler et al., 2005),whereas RRMs 1 and/or 2 might recognize a different set of Rump targets. In addition, binding of Rump to nos might be cooperative, as mutation of one or both CGUU motifs is equally deleterious. Consistent with this idea,Rump self-interaction was detected in a yeast two-hybrid screen(Giot et al., 2003),suggesting that Rump might function as a dimer.

Multifunctionality appears to be a common feature of proteins that regulate mRNA localization and localized translation. For example, DrosophilaHrp48 (also known as Hrb27C - FlyBase), an hnRNP A/B family member that participates in localization and translational regulation of osk and gurken mRNAs (Goodrich et al.,2004; Huynh et al.,2004; Norvell et al.,2005; Norvell et al.,1999; Yano et al.,2004), also functions as a regulator of alternative splicing(Hammond et al., 1997) and mammalian homologs of Xenopus Vg1RBP60, an hnRNP I protein involved in Vg1 mRNA localization (Cote et al., 1999), play various roles in nuclear RNA biogenesis(Valcarcel and Gebauer, 1997). Similarly to Rump, activities of two other nos regulators, Glorund(Glo) and Smaug (Smg), may be distinguished through multiple, distinct binding specificities. Glo, an hnRNP F/H homolog, represses translation of unlocalized nos during oogenesis through its interaction with an AU-rich double-stranded region within the nos 3′UTR(Kalifa et al., 2006), whereas its mammalian counterparts regulate alternative splicing through their interaction with a G-rich motif (Caputi and Zahler, 2001; Matunis et al.,1994). Smg represses translation of unlocalized nos in the embryo through its interaction with Smaug Recognition Elements (SREs) in the nos 3′UTR (Crucs et al.,2000; Dahanukar et al.,1999; Smibert et al.,1999), but its function in degradation of maternal mRNAs is SRE-independent (Semotok et al.,2005).

Elimination of rump function has a more variable effect on localization of nos-tub:nos+2 RNA than does mutation of Rump binding sites. Thus, another ovarian protein with a similar recognition motif might be able to compensate for the loss of Rump. Alternatively, mutation of the Rump binding site(s) might have a more global effect on the structure of the localization element, disrupting the function of other localization factors. The CGUU motifs recognized by Rump flank one of the two SREs recognized by Smg within the nos 3′UTR and both the (A)G:C and (D)G:C mutations disrupt the function of the SRE, consistent with the ability of these mutations to cause pleiotropic effects. By contrast, mutation of the SRE had only a small effect on p75 binding to the +2′ element(Bergsten et al., 2001),suggesting that Rump binding is less sensitive to local perturbations. Finally, several linker-scanning mutations in the +2′ element disrupt the localization function of this element but do not disrupt Rump binding in vitro (Bergsten et al., 2001)(our results), suggesting that at least one additional, as yet unknown, factor mediates localization via the +2′ element.

During early and mid-oogenesis, Rump accumulates in the nurse cell nuclei where nos is synthesized, suggesting that it might first associate with nos mRNA in the nucleus. In this regard, Rump may resemble Xenopus Vg1RBP60/hnRNP I, which associates with Vg1 and VegT mRNAs in the nucleus and travels with the RNAs to the cytoplasm where the RNP is remodeled (Kress et al.,2004). Owing to the impenetrability of the late-stage oocyte to antibodies, we have not been able to determine the distribution of Rump in late oocytes when nos becomes posteriorly localized. However, Rump appears cytoplasmic in newly fertilized embryos and is not enriched at the posterior of the embryo, suggesting that it does not maintain a stable association with nos once anchored at the germ plasm. Biochemical evidence for a continued association of Rump with nos during embryogenesis, together with the uniform distribution of Rump protein,indicate that in the early embryo Rump is most likely to be associated with unlocalized, translationally repressed nos mRNA. The close apposition of Rump and Smg binding sites raises the possibility that binding by these factors is mutually stabilizing.

Taken together, data presented here suggest a model whereby Rump associates with nos mRNA during oogenesis, possibly in the nucleus, where it participates in the generation or stabilization of localization-competent nos RNP complexes. A small fraction of nos RNPs are entrapped by the germ plasm and this association results in reorganization of the RNP and release of Rump. The majority of nos mRNA remains unlocalized and, consequently, bound to Rump. Whether Rump plays a role in coordinating translational control with nos mRNA localization remains to be determined.

We thank S. Chatterjee for help with protein purification, fly work and,with M. Klosinska, for help with monoclonal antibody screening; S. Kyin(Princeton Mass Spectrometry Facility); C. DeCoste and J. Levorse (Princeton Monoclonal Facility); J. Goodhouse (Princeton Microscopy Facility); C. Schwartz and H. Krause for the TRAP vector and advice on purification; T. Thomson and P. Lasko for the tud^tux46 allele; P. Schedl,R. Lehmann and A. Ephrussi for antibodies; and T. Weil and K. Sinsimer for comments on the manuscript. This work was supported by an NSF Predoctoral Fellowship (R.A.J.) and a grant from the National Institutes of Health(GM067758).