A Novel Double-stranded RNA-binding Protein, Disco Interacting Protein 1 (DIP1), Contributes to Cell Fate Decisions during Drosophila Development*

We report the identification of the Disco Interacting Protein 1 (DIP1) gene isolated in a yeast interaction trap screen using the zinc finger protein disconnected (disco) as a bait. DIP1 encodes a protein containing two double-stranded RNA binding domains (dsRBD). Consistent with the presence of dsRBD, DIP1 binds dsRNA or structured RNAs in Northwestern assays. DIP1 is found in nuclear subdomains resembling speckles known to accumulate transcription and splicing factors. In early embryos, nuclear localization of DIP1 protein coincides with the onset of zygotic gene expression. Later in development DIP1 expression is decreased in dividing cells in different tissues. Overexpression of DIP1 in the eye-antennal imaginal disc, early in embryonic and larval development, causes the formation of supernumerary structures in the head capsule. A role for DIP1 in epigenetic mechanisms that lead to the establishment and/or maintenance of cell fate specification is discussed.

The fate and function of a large number of different RNA molecules are dependent upon their interactions with specific RNA-binding proteins. RNA-binding proteins compose a diverse group of proteins that participate in a wide range of biochemical functions (1). A subset of these proteins preferentially recognizes and binds dsRNA. 1 These dsRNA-binding pro-teins (DSRBP) feature the presence of one or more doublestranded RNA binding domains (dsRBD). These structural motifs of 65-68 amino acids in length can be divided into two groups according to their similarity to the consensus sequence and their ability to bind dsRNA. Type A dsRBDs generally bind dsRNA with high affinity and show a high degree of sequence similarity with the entire consensus sequence. Type B dsRBDs bind dsRNA less efficiently and are similar at their C termini to the consensus sequence but not at their N termini (2).
DSRBPs are involved in a variety of cellular functions that range from RNA localization to RNA editing and more recently post-transcriptional gene silencing (3)(4)(5)(6). In the Drosophila genome there are 12 dsRBD-containing proteins, and of these only 5 have been genetically characterized (7)(8)(9)(10)(11). The Staufen gene, first identified as mutations that disrupt early embryonic pattern formation, is required for RNA localization and translational control. Mutations in the maleless (mle) gene cause disruption in X chromosome dosage compensation and thus male-specific lethality. The Mle protein is part of the malespecific lethal complex that associates with the X chromosome in males and leads to a 2-fold increase in the expression of X-linked genes presumably through an increase in the level of histone acetylation (reviewed in Ref. 12). Three of the Drosophila DSRBPs have been reported to function in RNA interference and editing (9,10). The remaining six dsRBD-containing proteins include one protein highly similar to TAR-binding protein and five others of unknown function (7). Therefore Drosophila DSRBPs compose a finite gene family whose function is likely to represent the range of biochemical functions fulfilled by this class of proteins for the development and homeostasis of an organism.
Here we report the identification and functional characterization of a novel dsRBD-containing gene. This gene, called DIP1, for Disco Interacting Protein 1, was isolated in a yeast two-hybrid screen to identify potential regulators of disconnected (disco) gene function. The disco gene encodes a nuclear protein with two Kruppel-like zinc fingers expressed in a developmentally regulated fashion and displays a tissue-specific autoregulatory feedback loop (13,14). Mutations in the disco gene cause a variety of phenotypes that range from disruption in the connectivity of the larval visual system to failure in the differentiation of the circadian pacemakers in the central nervous system (13,15,16).
The DIP1 gene is widely expressed in all stages of development. In early embryos, nuclear localization of DIP1 coincides with the onset of zygotic gene expression. Additionally, the level of DIP1 protein appears reduced in dividing cells in different tissues during embryonic and postembryonic development. Interestingly, DIP1 is found mostly in nuclear subdo-mains resembling speckles shown to accumulate transcription and splicing factors. DIP1 overexpression in the eye-antennal imaginal disc causes cell fate transformations and the formation of supernumerary structures in the head capsule. Taken together our results suggest that DIP1 may mediate epigenetic mechanisms required for the establishment and/or maintenance of cell fate specification.
Yeast Interaction Trap Screening and Cloning of DIP1-A Drosophila 0 -12-h embryonic cDNA library, constructed in the galactose-inducible yeast expression vector pJG4-5, was kindly provided by Dr. Roger Brent. The library was screened according to Ref. 19. The baits used for the interaction trap assay were constructed by cloning 5Ј (corresponding to 186 N-terminal amino acids) or 3Ј (corresponding to 153 Cterminal amino acids) portions of the disco cDNA in-frame with the LexA coding sequence contained in the pEG202 plasmid. EGY48 was used as the host yeast cell. Yeast cell transformations were done according to Ref. 20. A total of 2 ϫ 10 6 transformants were screened by plating on Ura Ϫ , His Ϫ , Trp Ϫ , and Leu Ϫ plates containing galactose each time. The Disco N-terminal bait yielded 150 colonies after 3 days at 30°C, 18 of which showed galactose-dependent blue color in 5-bromo-4-chloro-3-indolyl-␤-galactopyranoside medium (X-gal, Sigma). Sixteen of these clones contained a 1.1-kb 3Ј-end fragment of DIP1 cDNA. Additional full-length DIP1 cDNAs were isolated by conventional screening of a 0 -12-h cDNA library (DIP1-a) (21) and from an ovarian cDNA library (DIP1-c) (22). The Berkeley Drosophila Genome Project (BDGP, EST LD14381) kindly provided the DIP1-b cDNA. The DIP1 genomic sequence was obtained from cosmids isolated from a genomic library kindly provided by J. Tamkun (University of California, Santa Cruz).
Disco Binding Studies-DIP1 protein was synthesized in vitro using the expression system from MBI Fermentas. Briefly, the DIP1-a cDNA was inserted in the vector pSPUTK and transcribed in vitro from the SP6 promoter following the manufacturer's protocol (MBI Fermentas). DIP1 protein was labeled by [ 35 S]methionine incorporation during disco translation. Both wild type and mutated N-terminal Disco-GST fusion constructs as well as GST alone were expressed from the pGEX vector in the protease-negative Escherichia coli strain BL21-DE3 (Stratagene) following the manufacturer's protocol (Amersham Biosciences). Binding protocol is essentially as described in Ref. 23.
DIP1 Protein Purification-The DIP1-a cDNA was cloned at NdeI and XhoI sites of pET29b(ϩ) expression vector (Novagen). The ATG in the NdeI site was used as the translation initiation site, and the termination codon of pET29b(ϩ) was used to tag the protein with a His 6 . This construct was then transformed into E. coli BL21-(DE3) cells. Bacterial cultures (24) grown for large scale protein production were processed for protein purification under denaturing conditions (8 M urea) according to the Qiagen protocol with 15 mM imidazole present in the lysis and wash buffers. The DIP1 His-tagged protein was purified from the bacterial lysate by elution from nickel-nitrilotriacetic acid Superflow slurry (Qiagen) following the manufacturer's protocol for batch purification. For injection purposes the eluted protein was precipitated in 8 volumes of 80% acetone and lyophilized. For antibody purification, the protein was stored in the elution buffer at 4°C until needed.
Generation of DIP1-c and -b Deletion Constructs and dsRBD Constructs-Constructs expressing His-tagged DIP1-b and DIP1-c were obtained by ligating the BamHI-HindIII digest of the pET21b(ϩ) vector (Novagen) with the BamHI-HindIII-restricted amplification products of the corresponding cDNAs. The following PCR primers were used: 5Ј-C-GCGGATCCGCGATGAAGCGAAATCGTCGTGC and 3Ј-CCCAAGCT-TGGGAGTGGTGTCGCTGTAGGTG.
The double-stranded RNA binding domains (dsRBD) were cloned by PCR amplification using Pfu DNA polymerase (Stratagene) of the corresponding dsRBD regions from DIP1-c using primers containing NdeI and XhoI restriction sites, respectively, to allow for in-frame cloning with histidine tag (His 6 tag). The dsRBDs were cloned into pET29b(ϩ) and transformed into E. coli BL21 (DE3). The primers for dsRBD1 are  5Ј-TGCGCAAGCATATGCTCCCCAAGAAC and 3Ј-CTTGGGGGTCTC-GAGTGCAATAATAAAATC, and the primers for dsRBD2 are 5Ј-GCT-GGGAGCATATGCACCCGGCGA and 3Ј-GTAGGTGAACTCGAGGCC-GAACAGAGA. A construct expressing DIP1-c with an N-terminal T7 tag was made by PCR amplification of DIP1-c from pET21b(ϩ)-DIP1-c with 5Ј-ATGAAGAAGCTTCGTCGTGCATTTGCTGGA containing HindIII restriction site in-frame with the T7 tag (T7 viral epitope tag) and 3Ј-GTGGTGCTCGAGTTAAGTGGTGTCGCTGTAGGTGAA having XhoI restriction site and a stop codon. DIP1-c deletion constructs were made by PCR amplification using Taq DNA polymerase (Invitrogen) of pET21b(ϩ)-DIP1-c using primers going in opposite directions flanking dsRBDs. The primers contained NcoI restriction sites, and the PCR product was ligated and transformed into E. coli BL21(DE3). Primers for dsRBD1 deletion are 5Ј-TTGGGGAGCCATGGCTTGCGCAGCTGG-CGGTT and 3Ј-ATTTTATTCCATGGAAAATGACCCCCAAGCCG. Primers for dsRBD2 deletion are 5Ј-GGGTGCATCCATGGCCAGCCA-GAGGGTAGCTC and 3Ј-GCAACTCTCCATGGGGCACAACTT-CACCTAC.
Northwestern Blots-Proteins used in Northwestern blots were purified in the same manner as the one used for immunization. The Northwestern protocol is essentially as described previously (2). dsRNA probes were prepared by phosphorylating poly(I)-poly(C) dsRNA (Amersham Biosciences). Structured RNA probes were prepared by transcribing from pT7VA1 (kindly provided by M. B. Mathews) with T7 polymerase (Roche Applied Science), and 50 Ci of [␣-32 P]CTP 5 ϫ 10 5 cpm/ml was used for the binding assay.
Generation of DIP1 Antibody-Purified DIP1 protein was sent to Pocono Rabbit Farm (Canadensis, PA) for injection into two rabbits following the facility's protocol for popliteal immunization. The antibody was affinity-purified using DIP1 protein (isoform a or c) immobilized on nitrocellulose filters essentially as described (25). Hemagglutinin (HA)-tagged DIP1 protein expressed in Drosophila Schneider (S2) cells was purified by immunoprecipitation with anti-HA monoclonal antibody (Roche Diagnostics). The immunoprecipitated protein was recognized by DIP1 antibody in Western blots (data not shown). Likewise the DIP1 antibody was able to immunoprecipitate HA-tagged DIP1 protein expressed in S2 cells (data not shown). The specificity of the DIP1 antibody was confirmed by the following experiments. DIP1 antibody labeling was abolished when the antibody was preincubated with recombinant DIP1 protein in all specimens reported here (data not shown). Ectopic expression of the DIP1 gene via the GAL4 system was recognized by this antibody as an ectopic nuclear signal (data not shown). DIP1 gene expression as detected by this antibody was entirely consistent with that detected by RNA in situ (data not shown). Finally, antibody affinity-purified using two different DIP1 isoforms from two different rabbits detected an indistinguishable pattern of expression not detected by the preimmune sera.
Immunohistochemistry of Ovaries-Ovaries from 2-to 3-day-old mated Oregon-R females were prepared essentially as described previously (26). Dissected ovaries were incubated with rabbit anti-DIP1 antibody (1:15 dilution) overnight at 4°C followed by incubation with goat anti-rabbit secondary antibody conjugated to the fluorochrome Alexa 488 (dilution 1:200; Molecular Probes, catalogue number A11008) for 2 h at room temperature. The whole mount ovaries were mounted in 70% glycerol in phosphate-buffered saline containing trace amounts of p-phenylenediamine (Sigma P-6001). For propidium iodide staining, ovaries were fixed and permeabilized similarly except that the specimens were treated with RNase A (50 g/ml) for 2 h prior to incubation in propidium iodide (ICN number 195458) at 100 g/ml for 1 h at room temperature. Ovaries were subsequently treated for immunolabeling as described previously (26). The preparation of ovaries for phalloidin labeling was carried out according to a previously published procedure (26). DIP1 immunolabeling was completed prior to incubation in phalloidin-conjugated rhodamine (Molecular Probes R-415) for 1 h at room temperature.
Immunohistochemistry of Embryos-Oregon-R embryos were collected at 22°C on apple juice agar plates and staged as defined previously (27). Specimen preparation was carried out as described previously (28). The anti-DIP1 antibody was used at a dilution of 1:15. Secondary antibodies used were goat anti-rabbit conjugated to Alexa 488 (1:200) or goat anti-rabbit conjugated to horseradish peroxidase (1:100) (Jackson ImmunoResearch 111-035-003). For fluorescent secondary antibodies, embryos were treated and mounted as specified for ovaries in Ref. 26, whereas horseradish peroxidase-stained embryos were processed for diaminobenzidine and mounted as described in Ref.
28. Phalloidin labeling of embryos was performed as outlined previously (28) except that 80% cold ethanol replaced methanol. Following fixation, embryos were incubated with rhodamine-conjugated phalloidin (1:20) for 1 h at room temperature, rinsed several times in 1ϫ PBT (0.5% Triton X-10), and processed for antibody labeling (26,28). For detection of mitotic spindles, embryos were prepared as specified previously (29) and labeled with anti-tubulin antibody (E6 Iowa Hybridoma Bank, dilution 1:50). Bound antibody was detected using a goat anti-mouse conjugated to Texas Red secondary antibody (Jackson Im-munoResearch 115-075-147) at a dilution of 1:150. Specimens were viewed under Normarski optics in a Zeiss Axioscope. Confocal microscopy was performed on a Bio-Rad MRC 600 krypton/argon laser confocal microscope. Photomicrographs were taken on Fujichrome 64 film and panels constructed using Adobe Photoshop 5.0.
Immunohistology of Eye-Antennal Discs-Larval brains and eye-antenna imaginal discs were dissected and stained with X-gal (Sigma) for homothorax-, spalt-, and distalless-lacZ reporter lines as described previously (30). Samples were visualized as using Nomarski optics.

Identification and Cloning of the Disco Interacting Protein 1
Gene (DIP1)-In order to begin addressing the mechanisms underlying disco tissue-specific autoregulatory function, we undertook a yeast two-hybrid screen to isolate putative partners of the disco gene product. We used an interaction trap assay developed as a modification of the yeast two-hybrid system. This approach employs baits fused to the LexA DNA binding domain, the LEU2 and lacZ reporters under the control of the LexA operator, and a cDNA library fused to the E. coli B42 activation domain (19,24). The library screened was a Drosophila 0 -12-h embryonic cDNA library constructed and kindly provided by Dr. R. Brent (Harvard Medical School).
Two portions of the disco gene were used as bait: a C-terminal fragment of 154 amino acids (residues 415-568 (32)) and a 186-amino acid N-terminal fragment containing the two zinc fingers (residues 1-186). Due to the ability of the N-terminal fragment to activate LexA reporter gene expression in the absence of a heterologous activation domain, a mutated version of this fragment was used in which Cys-127 was substituted by a Ser. This mutation, which is the same found in the disco 1 mutation, greatly reduces the N-terminal fragment transactivating capability allowing it to be used as bait in the interaction trap screen (data not shown and see Ref. 14).
The Disco C-terminal bait did not yield any positive clones in two rounds of the yeast two-hybrid screen (10 6 clones each). The screen using the Disco N-terminal bait yielded 18 positive interactors by galactose-dependent activation of two reporter genes, lacZ and LEU2 (Fig. 1A). Sixteen of these colonies contained the same cDNA as determined by restriction analysis of PCR-amplified DNA. Plasmid DNA from 4 of these 16 positive colonies was sequenced alongside additional full-length clones isolated through conventional methods from other cDNA libraries.
Physical interaction between the Disco and DIP1 proteins was verified in a disco GST pull-down experiment. Purified N-terminal GST-Disco fusion protein bound to glutathioneagarose beads was allowed to bind to in vitro translated DIP1 protein labeled with [ 35 S]methionine. The GST-bound protein was eluted and analyzed on denaturing SDS-polyacrylamide gel. DIP1 protein bound only to GST-Disco fusion and not to GST alone (Fig. 1B).
The DIP1 Gene Encodes a dsRNA-binding Protein That Contains Two dsRBDs-The DIP1 gene corresponds to the transcription unit that distally flanks the not-yet-sequenced repeatrich region in polytene band 20A at the base of the X chromosome (33). The genomic organization of the DIP1 gene and the features of three independently isolated DIP1 cDNAs are shown in Fig. 2 (33). Differences in the sequence of these clones exist at the 5Ј-and 3Ј-ends. Sequence heterogeneity at the 5Ј-end of the DIP1 cDNA clones can be accounted for by the inclusion of different segments of exons 1 and 2. The boundaries of these segments include consensus donor and acceptor splicing sites suggesting that alternative use of these sites is responsible for the sequence heterogeneity at the 5Ј-end. The existence of a putative start site (ATG) in DIP1-b and DIP1-c but not in DIP1-a indicates that proteins with a different N terminus may be synthesized by the corresponding mRNAs. The absence of this ATG in clone DIP1-a would lead to the synthesis of a truncated protein from a second initiator site located in exon 2 (Fig. 2B). Thus, assuming that the first start site is used preferentially, the predicted size of the DIP1-a, -b, and -c proteins is 34, 44, and 46 kDa, respectively. The DIP1 gene contains a putative bipartite nuclear localization signal sequence (NLS) that spans exons 2 and 3 (31). The 3Ј-untranslated region (3Ј-UTR) includes variable number of repeats of ϳ120 bp represented 8 times in DIP1-a, 7 times in DIP1-b, and 5 times in DIP1-c (Fig. 2). The number of repeats present has been shown to be strain-specific (33).
Sequence analysis using BLAST tool (34) revealed that the predicted DIP1 protein contains two regions with similarity to double-stranded RNA binding domains (dsRBD) (Figs. 2 and 3). A consensus for dsRBDs has been derived from in vitro binding experiments, mutational analysis and sequence comparison of several dsRNA-binding proteins (8,35).
The defined dsRBD consensus sequence is 65-68 amino acids long (2). Two types of dsRBD, a long (type A) and short (type B), have been recognized according to the degree and region of similarity to the full-length consensus sequence (2). The first dsRBD (dsRBD1) is of the type A dsRBD, which shows strong similarity to the entire length of the consensus sequence (Fig.  3, DIP1-1). The second dsRBD (dsRBD2) is of the type B dsRBD and matches well the C-terminal portion (or core motif) but shows poor similarity to the N-terminal portion (Fig. 3, DIP1-2). The amino acid sequence of DIP1 dsRBDs shares the highest degree of amino acid similarity (50%) with those present in rat and human glutamate receptor RNA editase (rRED1 and hRED1) (Fig. 3). This similarity extends beyond the dsRBDs to include all of the DIP1 protein sequence. The alignment of the predicted DIP1 protein sequence with the N-terminal 301 amino acids of rRED1 and hRED1, which includes the two dsRBDs but not the catalytic domain, shows 43% amino acid similarity.
The ability of DIP1-c, the longest isoform, to bind RNA was investigated in Northwestern blots probed with doublestranded (poly(I)-poly(C)) and single-stranded (poly(C)) homopolymers. The VA1 RNA of adenovirus, known to form extensive secondary structure, was also used to probe DIP1-c Northwestern blots. DIP1-c did not bind single-stranded homopolymers although it showed strong binding to either double-stranded RNA homopolymer or adenovirus VA1 RNA (Fig.  4), demonstrating that in vitro it behaves as a dsRNA-binding protein.
The functional significance of the dsRBDs for dsRNA binding in Northwestern assays was further investigated. His-tagged constructs of DIP1-c were created in which either the dsRBD1 (DIP1-c⌬dsRBD1) or the dsRBD2 (DIP1-c⌬dsRBD2) was deleted (Fig. 5A). Additionally, similar constructs were made for the expression of dsRBD1 or dsRBD2 (Fig. 5A). Northwestern analysis was carried out as described above to evaluate the dsRNA binding ability of these recombinant proteins (Fig. 5B).
Our results indicate that the dsRBD2 by itself was sufficient to bind dsRNA homopolymers (Fig. 5B, 1st lane). In contrast, dsRBD1 did not show any significant binding in these conditions (Fig. 5B, 2nd lane). Consistent with these results is the observation that a deletion of dsRBD2 (DIP1-c⌬dsRBD2) causes a significant decrease in binding, whereas deletion of dsRBD1 (DIP1-c⌬dsRBD1) does not significantly affect DIP1-c binding to the dsRNA homopolymer (Fig. 5B, 3rd and 4th lanes,  respectively). Taken together these results suggest that the dsRBDs may play relatively different roles in the ability of DIP1-c to bind dsRNA.
The DIP1 Gene Product Is Found in the Nucleus in a Variety of Tissues throughout Development-The DIP1 gene is widely expressed during embryonic development as seen by immunohistochemistry using an anti-DIP1 antibody (Fig. 6). During later stages of embryogenesis, DIP1 expression appears to be somewhat higher in the central nervous system (Fig. 6, E and  F). The overall distribution of DIP1 protein is similar to that of the DIP1 mRNA as seen by in situ hybridization to whole mount embryos (data not shown). Thus DIP1 expression overlaps with that of the disco gene in a variety of locations including the optic lobe primordium where disco shows an autoregulatory function (Fig. 6C, arrow).
DIP1 gene is expressed postembryonically in a several tissues as seen by DIP1 antibody labeling of adult ovaries, all imaginal discs, and the central nervous system of the third instar larvae (Fig. 7, A and B, Fig. 9C, and data not shown). In all specimens analyzed so far DIP1 gene expression is nuclear, except in ovaries where significant expression is also found in the cytoplasm of nurse cells (Fig. 7A, star). Expression of DIP1 in the cytoplasm of nurse cells appears to be localized to the ring canals suggesting that it is transported from the nurse cells to the developing oocyte (Fig. 7A, arrow  and inset).
The nuclear expression of the DIP1 gene is not uniform in the specimens analyzed. In the ovaries, DIP1 expression is distinct in nurse cells, follicle cells, and in the developing oocyte. In nurse cells DIP1 expression is found in large, apparently interconnected, irregular domains ( Fig. 7B and green in Fig. 7C). In these cells, regions devoid of DIP1 are apparently filled with DNA as seen by propidium iodide staining (Fig. 7C). In follicle cells, DIP1 is found concentrated in granules superimposed onto a diffuse apparently less abundant background (Fig. 7, A  and D). In the developing oocyte DIP1 is found in the pronucleus and in the nucleolus (Fig. 7E, arrowhead and arrow,  respectively).

Presence of DIP1 in Subnuclear Domains during Early Embryogenesis Correlates with the Onset of Zygotic Transcription-In
Drosophila the first 13 nuclear divisions are not accompanied by cytokinesis, and the embryo consists of a syncytium of 6,000 nuclei in a monolayer in the cortex. During this period, development is directed mostly from maternally contributed gene products with only a few genes being transcribed. Cellularization and cessation of synchronized nuclear division signals the transition from maternal to zygotic gene expression (reviewed in Ref. 36). Subsequent cell division is restricted to stereotypic domains (37).
In order to begin an investigation into the relationship between DIP1 function and nuclear gene expression, we investigated the subcellular localization of the DIP1 protein during early embryogenesis prior to and soon after the maternal to zygotic transition. Whole mount embryos were double-labeled with anti-DIP1 and anti-tubulin antibodies to visualize mitotic spindles, with phalloidin to visualize the cytoplasmic membrane outline and thus cellularization, and with propidium iodide to visualize DNA. DIP1 protein is found in low level in early embryos (stage 3) prior to the onset of the bulk of zygotic transcription, suggesting that it is maternally contributed (Fig. 8A). In these embryos, . 3rd lane shows no binding of DIP1-c to ssRNA homopolymer poly(C). Each gel was loaded with 5 g of total protein, and the same protein preparation was used for all three RNA binding assays. DIP1 protein is present throughout the embryo surrounding mitotic spindles (Fig. 8A). Upon cellularization, DIP1 expression appears to increase and is found concentrated in the nucleus partially overlapping with DNA (Fig. 8, B-D, yellow dots on red background). In the nucleus DIP1 is found concentrated in irregular subnuclear domains (Fig. 8C, inset). The presence of DIP1 in the nucleus coincides with the onset of zygotic transcription.

DIP1 Protein Is Decreased during Cell Division-
The results above show that during early embryogenesis, in the syncytial blastoderm stage, the level of DIP1 is low. At that stage the embryo consists of a syncytium of nuclei undergoing synchronized DNA replication and nuclear division. Development is directed mostly from maternally contributed gene products with only a few genes being transcribed. In order to determine whether this observation could be generalized, we investigated the expression of DIP1 during periods when mitotically active cells can be seen adjacent to non-dividing cells. To that end the localization of DIP1 was analyzed in embryos after stage 5 when stereotypical mitotic domains can be identified by double FIG. 7. DIP1 expression in a stage 10 egg chamber. A and B show a sagittal view of a whole-mount egg chamber at a superficial and a more medial focal plane, respectively. Star in A points to the presence of DIP1 in the cytoplasm. The arrow points to a ring canal, although the inset in A shows a higher magnification of a confocal image of a ring canal. C shows a higher magnification view of a nurse cell nucleus from a chamber, and D shows the follicle cells covering the oocyte both of which have been double-labeled with propidium iodide (red) and with anti-DIP1 (green). In the nuclei of nurse and follicle cells, DIP1 is found in irregularly shaped domains. In nurse cells regions devoid of DIP1 are filled with DNA (C). E shows the presence of DIP1 in the pronucleus (arrowhead) and nucleolus (arrow) of a developing oocyte. DIP1 expression is consistently reduced in mitotic domains present in embryos (Fig. 9). Fig. 9, A and B, shows a lateral view of stage 10 embryo double-labeled with anti-DIP1 and anti-tubulin where a mitotic domain can be seen. In these dividing cells, the DIP1 level is apparently reduced and located more superficially in comparison to neighboring non-dividing cells. In the third instar larva, DIP1 is nearly absent in the proliferating centers of the developing optic lobes and in dividing neuroblasts of the central brain. This observation is in contrast to the high level of expression in postmitotic neurons (Fig. 9C, star and arrowhead, respectively).

Ectopic Expression of DIP1 Causes Organismal Lethality and Cell Fate Transformations in the Eye-Antennal Imaginal Disc-
The pattern of expression of DIP1 during development and its subcellular localization suggest that the DIP1 gene is involved in the control of gene expression. In order to test this hypothesis, we up-regulated the expression of the of DIP1-b and -c isoforms using the UAS-GAL4 system (18). A variety of GAL4 drivers was used to target the expression of the DIP1 gene to different stages of development and cell types. Ubiquitous overexpression during embryogenesis, using the hsp70-GAL4 driver, caused organismal lethality. Similar results were obtained when the ubiquitously expressed tubulin or armadillo-GAL4 elements were used (data not shown). These observations suggest that proper expression of the DIP1 gene is required for viability of the organism.
In order to address the function of the DIP1 gene in the differentiation of specific cell types, we overexpressed both DIP1 isoforms using the GMR-GAL4 and the eyeless (ey)-GAL4 drivers. These drivers reflect the expression of the glass (gl) gene and the eyeless (ey) genes, respectively, and are expressed in the third instar larva in non-overlapping portions of the eye-antennal imaginal disc. The GMR-GAL4 is expressed posteriorly to the morphogenetic furrow in differentiating photoreceptor cells, whereas the ey-GAL4 expression in the third instar is restricted to the non-differentiated portion of the eye imaginal disc, ahead of the furrow (38,39). During the first instar larval stage, ey expression encompasses all of the eyeantennal imaginal disc but is not present in the antennal portion in the second instar larval stage and thereafter (39). Additionally, ey-GAL4 is also expressed in subsets of neurons in the central nervous system during embryogenesis and larval development (40).
Overexpression of the DIP1 gene in differentiating retinular cells using the GMR-GAL4 driver disrupted development as seen by the fusion of ommatidial units and duplication of bristles (data not shown). Targeting of DIP1 isoforms to ey-expressing cells caused lethality in ϳ81% of the progeny. Interestingly, in the escapers a variety of phenotypes was observed. These included duplication, deletion, and formation of ectopic structures. These phenotypes were never observed in either paren- FIG. 8. DIP1 expression prior to and during the maternal to zygotic transition. A shows a superficial view of a syncytial blastoderm embryo (stage 3) double-labeled for tubulin (red) and anti-DIP1 (green). B and C show a superficial view of a cellular blastoderm embryo (stage 5) double-labeled with either propidium iodide (red) and anti-DIP1 (green) in B or phalloidin (red) and anti-DIP1 (green) in C. Inset in C shows a higher magnification of the same nuclei illustrating the speckled domains. D shows a lateral view of a stage 6 embryo at a higher magnification. During the nuclear divisions of the syncytial blastoderm, DIP1 protein is found throughout the embryo surrounding the mitotic spindles (A). Upon cellularization DIP1 protein level increases, and it is found restricted to the nucleus.

FIG. 9. DIP1 expression is down-regulated in dividing cells.
A and B show a sagittal view of different focal planes of the same mitotic domain of a stage 10 embryo double-labeled for DIP1 (green) and for tubulin (red). A represents a more superficial focal plane of the same area outlined in B showing the corresponding mitotic spindles (arrowhead). B (green channel only) shows an area with reduced DIP1 expression (outline). Dividing cells overlap with the area of reduced DIP1 expression. In cells undergoing mitosis, DIP1 is located between the mitotic spindles (arrowhead). Non-dividing cells (arrow) show the typical punctate expression of DIP1. C shows a ventrolateral view of a third instar larval brain hemisphere labeled with anti-DIP1 antibody. The optic lobe proliferation center (asterisk) and a portion of the central brain region (arrowhead) are depicted. DIP1 expression is reduced in the developing optic lobe (asterisk) and in the neuroblasts of the central nervous system (arrowhead). tal strain. The penetrance, but not the kind of phenotypes observed, varied according to the temperature at which the cultures were maintained during development, the isoform used, and the particular insertion line. When flies were maintained at 28°C, the cell fate transformations observed were more frequent and extensive due to the higher activity of the GAL4 transcription factor at this temperature. For instance, at 28°C the highest UAS-DIP1-b-expressing line yielded 92% (n ϭ 101) of the viable progeny displaying a variety of phenotypes as exemplified in Fig. 10.
The phenotypes observed can be divided in two groups. In one group reduction of eye tissue was always seen accompanied by the appearance of unidentified anterior ectopic structures or by an apparent enlargement of the dorsal portion of the head capsule (Fig. 10, A and B, C and D, and arrowhead in F). In the second group, duplication of antennal or mouth parts were found in the absence of disruption in eye development (e.g. antennal duplication Fig. 10F, arrow). These supernumerary structures were found in the anterior-ventral portion of the head near the location of the normal appendages.
The presence of both dsRBDs is required for the observed phenotypes. Strains carrying deletion constructs of UAS-DIP1-b and -c missing either the first dsRBD (dsRBD⌬1) or the second dsRBD (dsRBD⌬2) were created. Targeted expression of the deleted constructs was carried out using the same ey-GAL4 element described above. Deletion of either dsRBD1 or dsRBD2 in the DIP1-b or DIP1-c isoforms was sufficient to prevent the phenotypes observed (data not shown). This suggests that these domains are equally required for the hypothesized function of DIP1 in the specification of eye tissues.
In order to begin understanding the mechanism underlying the phenotypes observed, we examined the expression of genes known to mark different fields within the eye-antennal imaginal disc and to play a role in the patterning of head structures (41,42). We chose to examine the expression of lacZ reporter constructs of homothorax (hth), spalt major (salm) and distalless (dll) genes. hth is expressed throughout the antennal primordium for most of larval development (43). Elimination of hth expression in the antennal primordium causes antenna to leg transformation (41). In the eye, down-regulation of hth by decapentaplegic, ahead of the morphogenetic furrow, is required for the commitment of the eye primordium and possibly to control the ratio of eye to head tissue (44). salm expression in the second segment of the antennal primordium is regulated by hth and dll, and its function is required for proper joint formation between antennal segments 2 and 3 (45). In the eye primordium, salm gene function is essential for the terminal differentiation of R7 and R8 photoreceptors (46). dll expression in the antenna marks the distal-most portion of the antenna and is required to repress leg cell fate in the antenna (42).  F). The ectopic expression of head capsule tissue in presumptive eye tissue (arrowhead) can also be seen. Fig. 11, A, C, and E, shows the expression of hth, salm, and dll-lacZ reporter genes, respectively, in a wild type background. Overexpression of DIP1-b caused a variety of changes in the expression of these reporter constructs. As with the external phenotypes described above, these phenotypes were never observed in either parental strain. An apparent expansion of hth expression ahead of the morphogenetic furrow at the expense of eye tissue behind the furrow was observed (Fig.  11B). salm-lacZ expression in the second antennal segment was sometimes missing (Fig. 11D). In contrast, salm expression was not affected in the eye portion of the imaginal disc. Finally, we observed an apparent duplication of dll expression in the eye disc (arrowhead in Fig. 11F). The disruption in the pattern of expression of these reporter constructs described above was less frequent than that found for the external phenotypes found in adult flies (hth-lacZ 6%, salm-lacZ 25%; dll-lacZ 12%). The relatively low frequency of these disruptions may reflect the wide range of phenotypes observed and the overall lethality associated with the overexpression of DIP1 under the regulation of the ey-GAL4 driver.
These results suggest that overexpression of DIP1 disrupts the allocation of cells into different fates within the eye-antennal disc. The observation that these phenotypes include a wide range of cell fates seems to indicate that overexpression of DIP1 may simultaneously disrupt several developmental pathways. Thus DIP1 function during development may not be instructive. The observed changes in the expression salm and dll in the third instar antenna were due to a pulse of DIP1 expression in this primordium during the first instar stage. The antennal primordium from the first instar until the third instar stage undergoes 7 rounds of cell division (47). Changes in cell fate caused by the overexpression of DIP1 in the first instar stage therefore were maintained after DIP1 expression had ceased. Taken together these results support the notion that DIP1 overexpression may alter epigenetic mechanisms required for the establishment and/or maintenance of cell fate specification. DISCUSSION We report the identification and characterization of a novel gene, DIP1, isolated by virtue of its association with the zinc finger-containing disco gene in a yeast interaction trap screen. Besides the interaction detected in yeast, we have demonstrated binding of DIP1 to Disco in GST pull-down experiments. The experiments described here do not address whether these interactions are relevant for the function of disco in vivo. Rather they investigate the function of DIP1 gene in development.
DIP1 Pattern of Expression Correlates with Period of Active Gene Expression-DIP1 localization in subnuclear domains in the early embryo coincides with the onset of zygotic transcription. Several transcription and splicing factors have been found concentrated in similar granules or speckles in tight association with the nuclear matrix (48 -50). Recent reports support the notion that speckles represent sites of storage for transcription and splicing factors rather than sites of active gene expression (51)(52)(53). During periods of transcriptional activity, these proteins appear dispersed throughout the nucleus with abundant sites of preferred localization, which have been described as irregularly shaped speckles. Transcription inhibition brings about a recruitment of these proteins into larger, more accentuated speckles with a concomitant decrease in overall dispersal of these proteins (51).
In Drosophila the organization of the chromatin in the salivary gland polytene chromosomes is stereotypical and easily visualized. DIP1 expression is found primarily associated with interbands known to represent sites of higher gene expression.
Moreover, subsequent to heat shock, DIP1 protein is found associated with heat shock puffs where the highly transcribed heat shock genes are located (data not shown). These observations are analogous to those reported for the Xenopus RNAediting enzyme ADAR1 (54). This enzyme is found associated with transcriptionally active lampbrush chromosome loops, suggesting that it acts co-transcriptionally (54). Interestingly, chromosomal staining is abolished by removal of both dsRBDs which may be potentially due to loss of RNA or protein interactions. Chimeric proteins carrying only one or multiple copies of a specific dsRBD were found associated with specific chromosomal loops (55).
Significance of the dsRBD-Several proteins known to bind double-stranded RNA or highly structured RNA substrates contain one or more regions with significant similarity to the dsRBD consensus sequence (2,56). A number of these proteins contain in addition shorter versions of the dsRBD, the so-called short C-terminal core domain, whereas two have been reported to contain only one copy of the short domain (2). In several instances both types of dsRBDs have been shown to be sufficient and necessary for RNA binding (2,57). In the case of the human immunodeficiency virus type 1 TAR RNA-binding protein, extensive deletion analysis led to identification of a 15amino acid subpeptide within the conserved dsRBD core motif, as being sufficient for high affinity binding to TAR (58,59).
In addition to the traditional role in dsRNA binding, dsRBD have also been shown to mediate protein-protein interaction in the absence of dsRNA binding (60). In the human RNA-editing enzyme ADAR1, the third dsRBD is required for nuclear localization of this protein (61). Interestingly, this function is modulated by dsRNA binding, which has been postulated to mask this NLS and thus lead to cytoplasmic accumulation of this protein (62).
Our results suggest that the two dsRBDs of DIP1 behave differently regarding the dsRNA homopolymer binding in Northwestern assays. The first domain, which corresponds to a type A dsRBD, does not bind dsRNA homopolymer, and deletion of this domain in the full-length DIP1-c isoform does not significantly impair RNA binding in this assay. By contrast, the second dsRBD, most similar to a type B, is sufficient and required for dsRNA homopolymer binding of DIP1-c in Northwestern assays. When human immunodeficiency virus TAR RNA was used to probe the same deleted proteins in a Northwestern assay the results were the opposite (see Supplemental Material, and Ref. 63). The first dsRBD was required for TAR binding, whereas the second dsRBD was not. These results indicate that the inability of DIP1-c lacking the second dsRBD to bind dsRNA homopolymers is not likely to be due to disruption in the conformation of the protein. These experiments do not adequately address the relative dsRNA binding ability or specificity of these two domains. Nevertheless, the results indicate that these domains differ functionally as expected from their sequence. We conclude that, as has been shown for a number of dsRNA-binding proteins (55), individual dsRBDs within the DIP1 protein may recognize different RNAs and/or may perform a function other than RNA binding.
What Is the Function of the DIP1 Gene?-Proteins containing dsRBDs are involved in a wide variety of biochemical functions that range from the control of transcription (64), RNA degradation and localization (3), translational control (4, 5), posttranslational modifications (6), and most recently post-transcriptional gene silencing (reviewed in Ref. 55). In most instances the dsRBD is required to perform the activity in question. For instance a subunit of the Xenopus GATA-2 transcription factor, p122, has been identified that contains dsRBDs similar to that of DIP1 (65). Additionally, similar to what was found for DIP1 in Drosophila, p122 translocates from the cytoplasm to the nucleus at the mid-blastula transition (65). The 90-kDa subunit of the nuclear factor of activated T cells, a bona fide transcription factor required for the expression of interleukin II, contains two copies of the core dsRBD motif (66). The NF90 dsRBDs have been shown to enhance the inhibitory and stimulatory transcriptional effects on reporter gene expression (67). Interestingly, NF90 has been independently identified in a screen for cellular proteins that associate with the adenovirus VA1 RNA (6). It is possible that NF90 performs a dual role in the control of gene expression: one is its well defined role as a transcription factor, and the other a yet undefined biochemical function that involves interaction with dsRNA.
Overexpression of DIP1 in eye-antennal disc causes an apparent cell fate transformation. The range of the different phenotypes exhibited by flies overexpressing DIP1-b and DIP1-c excludes a specific instructive role for DIP1 in developmental decisions as exemplified by the role of Hox genes (reviewed in Ref. 68). This conclusion is also supported by the observation that DIP1 expression is not restricted to any one particular cell type but is widespread during development. Although our studies did not address the critical period in development during which elevated DIP1 expression led to the transformations described, it is clear that continued expression is not required. The ey-GAL4 driver is expressed in the antennal primordium transiently, during embryogenesis, and early first instar. Nevertheless, this expression is sufficient to induce and maintain changes in gene expression seen in the late third instar eye-antennal imaginal disc.
Taken together these observations support the notion that DIP1 plays a role in the control of gene expression. DIP1 may function in conjunction with specific transcription factors such as disco. Alternatively, DIP1 may function in some aspect of RNA processing, and its association with disco indicates an overlap between transcription and RNA processing. Several recent reports (17, 51, 69 -71) document physical association between components of splicing and transcription machinery with the nuclear matrix. These findings form the underpinning of the current view of the nuclear matrix as a scaffold where the dynamic interaction between components of transcription and RNA processing machineries determine gene expression (48). The dynamic association of transcription and splicing factors with the nuclear matrix is essential for accurate and maximal transcriptional activity and thus for the implementation and maintenance of developmental decisions (49,50,56).