Drosophila Imp iCLIP identifies an RNA assemblage coordinating F-actin formation

Background Post-transcriptional RNA regulons ensure coordinated expression of monocistronic mRNAs encoding functionally related proteins. In this study, we employ a combination of RIP-seq and short- and long-wave individual-nucleotide resolution crosslinking and immunoprecipitation (iCLIP) technologies in Drosophila cells to identify transcripts associated with cytoplasmic ribonucleoproteins (RNPs) containing the RNA-binding protein Imp. Results We find extensive binding of Imp to 3′ UTRs of transcripts that are involved in F-actin formation. A common denominator of the RNA–protein interface is the presence of multiple motifs with a central UA-rich element flanked by CA-rich elements. Experiments in single cells and intact flies reveal compromised actin cytoskeletal dynamics associated with low Imp levels. The former shows reduced F-actin formation and the latter exhibits abnormal neuronal patterning. This demonstrates a physiological significance of the defined RNA regulon. Conclusions Our data imply that Drosophila Imp RNPs may function as cytoplasmic mRNA assemblages that encode proteins which participate in actin cytoskeletal remodeling. Thus, they may facilitate coordinated protein expression in sub-cytoplasmic locations such as growth cones. Electronic supplementary material The online version of this article (doi:10.1186/s13059-015-0687-0) contains supplementary material, which is available to authorized users.

In both iCLIP (254nm UV cross-linking) and PAR-iCLIP (365nm UV cross-linking), the isolation of RNA:Imp complexes was dependent on cross-linking (no signal is seen in the absence of cross-linking), dependent upon Imp antibody pulldown (no signal is seen when attempting pull-down with anti-mouse IgG coated Protein A Dynabeads), and it was sensitive to RNAse T1 treatment (a higher concentration of T1 results in trimming of the RNA:Imp complexes).
(B) Mapping statistics of all four CLIP libraries (C) Correlation analysis between the two iCLIP replicates, the two PAR-iCLIP replicates and iCLIP replicate 1 and PAR-iCLIP replicate 1. The scatter plot shows the natural logarithm of CLIP enrichment values per nucleotide in 3'UTRs for each dataset. There is a strong correlation both among the biological replicates and between protocols.
(D) Correlation analysis between CLIP tags per nucleotide and the length of the 3'UTR.
The scatterplots show a correlation between the length of the 3'UTR and the enrichment of CLIP tags per nucleotide. This phenomenon was seen for both iCLIP and PAR-iCLIP datasets. However, this correlation was not found for the RNA-binding protein HuR (HuR PAR-CLIP data obtained from [1]. (E) Top 10 most over-represented words from cWords motif analysis of top 25,000 variable sized sequences, covered by iCLIP and PAR-iCLIP clusters ranked by normalized CLIP enrichment.
(F) Relationship between co-occurrence frequency and motif spacing. The red line illustrates normalized frequency of co-occurring MACA motifs in highly enriched fixed sized sequences covered by iCLIP clusters in 3'UTRs. The blue line shows the frequency in randomly selected clusters of the same length as above from 3'UTRs. The boxed motif shows the highly enriched binding sequences and over-represented spacing distribution, with distances from 0 to 30 nt. In this plot, a horizontal line means that there is no dependence between MACA co-occurrences. We see a higher dependence when spacing between MACA occurrences is smaller than 30 nucleotides due to the presence of low-complexity regions, which is further supported by the strongly fluctuating pattern. Furthermore, we observe that in highly enriched fixed clusters co-occurrence of MACA motifs is around 4 times more frequent than in the background clusters.     (B) Knockdown efficiency of Imp in S2 cells treated with imp dsRNA or luciferase dsRNA S2 cells were treated with imp dsRNA or luciferase dsRNA. Western blot analysis of whole cell lysates from imp dsRNA-treated cells showed about 85% knockdown of Imp. The level of Imp was normalized to the TATA-binding protein (TBP) protein loading control.

C A A G C A U U A C A A C A A C A A C A A C A A C A A C A A C A A C A A C A A A U A U A A U A U A U A U U A U A U A U A A A C A C A C A C A A A A A A A C A C A C A C A C A C A C A C A C A C A G C A A C A C A C A C A C A C
(C) Western blot of luc dsRNA and imp dsRNA treated S2 cell lysate. Anti-Imp polyclonal antibody was used to detect the Imp protein. Anti-actin monoclonal antibody was used to detect the actin protein, while anti-Rps6 mouse monoclonal antibody was used to detect the presence of RpS6. RpS6 was used as a loading control and quantification of the intensities was performed in triplicates using ImageJ.

S5 Figure legend
Outline of the crosses that led to excision of the P-element.
Females are shown to the left. 1. cross: Imp G0072 heterozygous females were crossed to "jumpstarter" males that expressed a transposase that can induce the P-element to jump. Progeny possessing both the disrupted imp gene and the transposase gene was recognized by its phenotypic markers, red eyes from the P-element and stubble from the transposase. 2. cross: Female progeny from the first cross possessing a disrupted imp gene and a transposase was mated to FM7c balanced males. The transposase induced the P-element to jump, and progeny in which the P-element was excised could be recognized by their phenotypic markers, eye color and no bar from an FM7c allele.

Protein Putative function in neurons References Actin cytoskeleton dynamics R (Rap1)
A small GTPase-like protein, that has been found to stimulate F-actin polymerization upon chemoattractant stimulation in Dictyostelium[2] [2]. Rap1 controls inside-out signaling to integrins, with a model proposing that Rap1 contributes to the anchoring of integrins to the actin cytoskeleton[3] [3]. Rap1 localizes to spiny and aspiny neurons in cultured rat neurons.
[2]; [3]; [4] RhoGDI Rho-GDP disassociation inhibitor. Removes Rho family GTPases from membranes and solubilizes them in the cytosol. It also inhibits GTP hydrolyzing activities of Rho-proteins. [5] Rin (Rasputin) Rasputin encodes the RasGAP binding protein homolog G3BP, and has been shown to function in Ras-and Rhomediated signaling in Drosophila. [6] Cdc42 In primary Drosophila embryo cultures, this Rho GTPase is responsible for inducing large growth cones and long filopodia by promoting the polymerization of actin. [7,8] Rac1 In primary Drosophila embryo cultures, this Rho GTPase is responsible for inducing thick actin bundles. In mammalian cells, it is responsible for actin polymerization and the creation of lamellipodia. [7,8] Moe (Moesin) The sole Drosophila ERM protein, which is highly enriched in axons and other membrane protrusions, and promotes cortical actin assembly linking the membrane to the cytoskeleton.
[11] Zip (Zipper) Encodes the non-muscle myosin (myosin II) heavy chain, which is necessary for correct axon patterning and retrograde flow of the actin network in Aplysia californica bag cell neurons [12]; [13] Pp1-87B Protein phosphatase-1. Influences cytoskeleton dynamics in the growth cone of photoreceptor cells in Drosophila. [14]

Mys (Myospheroid)
Encodes an integrin-beta subunit. Integrins connect the extracellular matrix to the actin cytoskeleton. Is expressed in the CNS of Drosophila embryos, and loss of function mutants show axon guidance defects. [15] Msn (Misshapen) A Ste20-like Serine/Threonine Kinase present in Drosophila photoreceptor axons and growth cones. Is involved in the reorganization of actin cytoskeleton to control growth cone motility. [16]

Microtubule dynamics Mapmodulin
Stabilizing proteins important for microtubule dynamics. Interacts with MAP1A in mammalian cells resulting in axonal and dendritic budding. Differentially regulated in Drosophila mushroom bodies γ-neurons at the onset of axonal pruning.
[17]; [18] Ran A small Ras family GTPase that localizes to CNS during Drosophila development. In Drosophila primary neurons and mouse neurons, knockdown of Ran caused increased branch arborization, indicating a role for Ran in regulating neurite extension. Furthermore Ran is a regulator of microtubule dynamics.
[19]; [20] Sep1 Forms septin filaments in Drosophila, which are able to bind to microtubules in vitro. Some mammalian septins are localized to the tip of neurites where they are critical for neurite branching.
[21], [22] Pnut (Peanut) Forms septin filaments in Drosophila, which are able to bind to microtubules in vitro. Some mammalian Septins are localized to the tip of neurites where they are critical for neurite branching.
[21], [22] Vesicle dynamics Vsg (Visgun) Locates to endocytotic compartments and may contribute to endolysosomal biogenesis and trafficking. Has been shown to regulate cell proliferation in a Drosophila cell line. [23] Vha16-1, Vha26, Vha68-2, Vha55 Subunits of vacuolar ATPase proton pumps that localize to subcellular vesicles. V-ATPases are also present in the membrane of synaptic vesicles in the neurons of T. marmorata fish.
[24]; [25] SesB Mitochondrial ATP/ADP translocase activity is essential for neurotransmission, and loss of SesB is accompanied by a complete loss of synaptic transmission in the visual system in Drosophila. [26] Surf4 A cargo receptor involved in intracellular protein trafficking. In rat PC12 cells, it localizes to neurites of neuronal cells and is associated with proteins that induce neurite formation by controlling directional membrane trafficking. [27]

Sar1
Initiates formation of coated vesicles for forward trafficking from ER to Golgi. A genetic screen in Drosophila larval class IV dendritic arborization neurons showed defects in Sar1 expression resulted in decreased length of dendrites, due to dispersion of Golgi outposts in dendrites. [28]

Rab11
Rab11 endosomes regulate the transport of beta-integrins to the axonal growth cone. In rat PC12 neuronal cells, Rab11 is needed for normal neurite outgrowth. [29]

Rab5
Loss of the Rab5 endosomes result in a reduced number of dendritic branches in Drosophila dendritic arborization neurons, likely due to an inability to move proteins important for branching events along the microtubules. [30]

Arf51F
The mammalian homolog ARF6 transports integrins along the axon as well as in the growth cone and enhance neurite outgrowth in rat dorsal root ganglion axons. [31]

Membrane dynamics and signaling Sty (Sprouty)
Sprouty is an antagonist of the EGFR and FGFR signaling pathway and co-localizes with membrane ruffles and prevents the production of excessive number of branching events in neurons and glia cells in Drosophila. Mislocalization of Spry causes incorrect distribution of branching event.
[32]; [33] Arf79f A GTPase involved in the recruitment of factors responsible for polarization of actin filaments needed in the [34] 2012)

ruffles and prevents the production of excessive number of branching events in neurons and glia cells in Drosophila. Mislocalization of Spry causes incorrect distribution of branching event. Arf79f
A GTPase involved in the recruitment of factors responsible for polarization of actin filaments needed in the formation of lamellipodia in Drosophila neurons.
[34] 2012) Prominin-like A transmembrane protein that localizes in membrane protrusions. Knockdown of Prominin-like resulted in a disruption of primary neuron formation in Drosophila. [35]; [20] 14-3-3 epsilon Is highly expressed in the embryonic brain, CNS and motor axons in Drosophila. Binds to PKC and Raf, protein kinases with critical roles in neuronal signaling. [36] Fax Localizes to cellular membranes of CNS and PNS of Drosophila embryos. Deletions of Fax together with Abl (a tyrosin kinase that binds directly to F-actin) result in a disruption of the axonal architecture. [37] Eff (Effete) An E2 ubiquitin-conjugated enzyme. A key regulator of ubiquitin-proteasome mediated pruning of mushroom body γ-neurons and class IV dendritic arborization sensory neurons during metamorphosis in Drosophila. [38]; [39] PRL-1 Tyrosine phosphatase, Drosophila PRL is mainly enriched in developing CNS during embryogenesis.
[40] alpha-Spec Spectrins (α and β) link the cell adhesion proteins in membranes to the F-actin cytoskeleton.
Mutations have been shown to result in altered axon growth cone morphogenesis in Drosophila embryos. [41] Vap-33-1 In Drosophila, Vap is needed for axonal localization of the cell surface receptor Dscam, which is important for axon guidance. [42] Cnx99A Transmembrane protein, containing two Ca 2+ binding sites, that resides in ER and is necessary for correct function of the photoreceptor neuron in Drosophila. [43] Translational regulation Hrb27c Member of the hnRNP A/B family and binds to and regulates translation of oskar mRNA in Drosophila. Mutation of the Hrb27c gene caused a disruption of axonal projections in photoreceptor neurons in Drosophila.
[44]; [45] pAbp Cytoplasmic poly(A) binding protein, regulator of translational initiation. [46][46] Sqd (Squid) An hnRNP required for correct localization and translational regulation of gurken mRNA. Sqd interacts with Hrb27c and an RNAi screen conducted in Drosophila larval class IV dendritic arborization neurons showed that knockdown of Squid resulted in both branch loss and decreased length of dendrites.
[47]; [48] eIF-1A Knockdown of translational initiation factor eIF-1A leads to severe defects in dendrite morphogenesis, resulting in characteristic trees with truncated main branches and few and short higher order branches. [49]

S6 Figure legend
A subset of the 86 transcripts and their putative functions in neurons A literature search revealed that 40 out of the 86 transcripts, identified as key Imp targets by a combination of RIP-seq, iCLIP and PAR-iCLIP, can be associated with neuronal development, with a predominant involvement in axonal growth cone development or guidance. The protein names, as well as a short summary of their involvement in neuronal development, are listed in column 1 and 2, respectively, and references are listed in column 3. The proteins are divided into functional boxes based on their referenced functions: "Actin cytoskeleton dynamics", "Microtubule dynamics", "Vesicle dynamics", "Membrane dynamics and signaling", and "Translational regulation".

Immunostaining of S2 cells
Glass bottom dishes (GWSt-3522, Willco Wells) were treated with 40 µl 0.5 mg/ml concanavalin A (Sigma-Aldrich) [50]. 2x10 6 S2 cells were seeded in the petri-dish in 1 ml full medium and allowed to adhere to the bottom for 60 min. The medium was removed and cells were fixed by the addition of 1 ml fresh fixation solution [10% formaldehyde in PBS] and incubation for 10-15 min. Cells were washed briefly three times with PBS, and the cell membrane was permeabilized with 1 ml permeabilizing solution [PBS + 0.1% Triton X-100] for 3 min. Cells were washed briefly three times with PBS, and 500 µl blocking solution [1% normal goat serum (Sigma-Aldrich) in PBS] was added, followed by incubation for 1 h. 500 µl blocking solution containing a 1:10000 dilution of polyclonal antibody raised against Imp [51] was added, and the petri dish was incubated overnight at 4°C. Following three 5 min washes with PBS cells were incubated at room temperature for 1½ h with blocking solution containing 1:1000 Alexa Fluor® 568 goat anti-rabbit IgG (H+L)( Life Technologies). Three washes with PBS were completed in a dark container. To stain for cytoskeletal Factin, cells were incubated with 1:1000 Alexa Fluor® 488 phalloidin (Life Technologies) for 5 min followed by a wash with PBS.

RNA immunoprecipitation sequencing (RIP-seq)
The following procedure was performed in duplicates on separate days. The library preparation protocol was previously described by [52]. Primers and a schematic outline of the library preparation are seen in Figure SM1A and B, respectively. For the immunoprecipitation sample and the input RNA-seq sample, 4x10 7 S2 cells were pelleted and lyzed in 100 µl 50 mM Tris-HCl (pH 8.0), 100 mM NaCl, 1% NP40, 1.5 mM EDTA, complete EDTA-free protease inhibitor cocktail (Roche), 1 U/µl RiboLock (Thermo Scientific) lysis buffer. The lysates were cleared by centrifugation at 4°C and 22000 g for 20 min.
Total RNA was extracted for the input RNA-seq sample by adding Tri Reagent (Sigma-Aldrich) to the lysate according to the manufacturer's specifications, and subsequently 40 µg of input RNA was subjected to poly(A) RNA enrichment by using Poly(A)Purist MAG kit (Ambion) according to the manufacturer's specifications.
Immunoprecipitation was accomplished by first conjugating 40 µl polyclonal anti-Imp antibody (22053, made by BioGenes) to 400 µl Protein A Dynabeads (Life Technologies), according to the manufacturer's specifications. Before immunoprecipitation, 900 µl lysis buffer was added, and the total 1 ml of lysate was mixed with the beads, and Imp-RNP complexes were immunoprecipitated by rotation for 2 h at 4°C. Beads were washed three times in lysis buffer, before Tri Reagent (Sigma-Aldrich) was added to the beads, and immunoprecipitated RNA was isolated according to the manufacturer's specifications. For reverse transcription of the fragmented RNA samples, 1 µl RT_random_primer (100 pmol) and 1 µl 10 mM dNTP mix were added to 10 µl of the fragmented RNA, and the mixture was preheated for 5 min at 65°C before cooling. The RT mix [4.0 µl 5 x PrimeScript buffer (Takara), 0.5 µl RNAsin Plus (40 U/µl , Promega), 1.0 µl PrimeScript RT (200U) (Takara), 2.5 µl RNase free H 2 0] was added, and reverse transcription was performed with the following program; 25°C for 10 min, 42°C for 60 min, 70°C for 15 min and hold at 4°C.
The RNA was degraded by the addition of 0.75 µl RNAse H (5000 U/ml, New England Biolabs) and incubation for 20 min at 37°C. The cDNA was purified using RNAClean XP beads (Beckman Coulter) according to the manufacturer's specifications, and input cDNA was eluted from the beads using 20 µl H 2 O, while the immunoprecipitated cDNA was eluted using 10 µl H 2 O.
The PCR_forward primer anneals to the 3' adaptor sequence while the PCR_reverse_index primer, containing the unique sample identifier sequence, anneals to the 5' adaptor sequence. The PCR was performed as follows; 98°C for 3 min; (98°C for 80 sec, 64°C for 15 sec, 72° C for 30 sec)x4 and (98°C for 80 sec, 72°C for 45 sec)x24 followed by 72°C for 5 min. The PCR product was purified using AMPure XP beads (Beckman Coulter) according to the manufacturer's specifications and resuspended in 11 µl H 2 O.
The PCR samples were run in a 2% E-Gel SizeSelect agarose gel (Life Technologies) and the libraries of 210-500 bp were collected, purified using AMPure XP beads (Beckman Coulter) and resuspended in 11 µL H 2 O. The size and the concentration of the libraries were examined on a Bioanalyzer equipped with a DNA 1000 chip (Agilent Technologies), and molar equivalents of the different libraries were pooled and subjected to Illumina sequencing.

Figure SMM1
SMM1A: Oligonucleotides used in the preparation of the library [52] SMM1B: A schematic outline of the library preparation.   Electrophoretic mobility-shift assay DNA templates for in vitro transcription were synthesized from cDNAs with the primers: PABPCDST7-F + PABPCDS-R, PABP3UTRT7-F + PABP3UTR-R, SqdCDST7-F + SqdCDS-R and Sqd3UTRT7-F + Sqd3UTR-R. 32 Plabelled transcripts were generated by T7 RNA polymerase and purified by denaturing gel electrophoresis. The Imp coding region was amplified from EST SD7045 with the primers Imp-Bam-start and Imp-Xho-stop and inserted into the BamHI and XhoI sites of pET28a (Novagen). Recombinant protein was expressed in RNase E-deficient Escherichia coli cells (Invitrogen) containing plasmid-encoded tRNAs for rare Arg, Ile and Leu codons and subsequently purified on Ni-agarose beads (Sigma-Aldrich). Radiolabelled RNA and 100 ng Escherichia coli tRNA were incubated with recombinant Imp at concentrations of 12.5 nM, 25 nM, 50 nM or 100 nM for 30 min at 30°C in 10 μl 20 mM Tris-HCl (pH 7.8), 140 mM KCl, 2 mM MgCl2 and 0.1% Triton X-100. After addition of Ficoll (2%) and bromophenol blue, samples were applied directly to a 1 mm 5% polyacrylamide gel (19:1) in 90 mM Tris-borate (pH 8.3), and run at 80 V for 3 h.

Imp double-stranded RNA interference-mediated knockdown in S2 cells
The primers dsRNA Imp 1-F + dsRNA Imp 1&2-R and dsRNA Imp 2-F + dsRNA Imp 1&2-R, containing the T7 promoter sequence at the 5' end, were used to amplify a 343 and 479 bp fragment, respectively, of the Imp coding region using a pRmHa-4 Imp-FLAG plasmid as template. As a control dsRNA, primers dsRNA Luc 1-F + dsRNA Luc 1-R and dsRNA Luc 2-F + dsRNA Luc 2-R, containing the T7 promoter sequence at the 5' end, were used to amplify a 533 bp and 562 bp fragment, respectively, of the luciferase coding region using the pGL3 plasmid as template. Primer sequences are listed in Figure SMM4. The PCR product was purified using AMpure XP beads (Beckman Coulter) according to the manufacturer's specifications, and transcripts were generated by adding a reaction mix [10 µl 10x T7 RNA buffer (400 mM Tris-HCl, pH 7.5, 60 mM MgCl 2 , 50 mM DTT, 10 mM spermidin (Sigma-Aldrich)), 10 µl 7 mM rNTPs (Amersham), 1.5 µl T7 RNA polymerase and up to 100 µl H 2 O] to 2 µg of PCR template and incubating overnight at 37°C. The dsRNA was purified using AMpure XP beads (Beckman Coulter), and the concentration of the dsRNA was estimated by running an aliquot on a 1% agarose gel in the presence of an RNA dilution series of known concentration.
S2 cells were pelleted and resuspended in FBS-and antibiotics-free Schneider's Drosophila medium (Biowest) at a density of 1x10 6 cells/ml. One ml cell suspension was seeded in a 6-well tissue culture dish. Following adhesion, the medium was removed and 1 ml serum-free medium containing 40 µg/ml dsRNA Imp 1 or dsRNA Imp 2 was added to the well. The control cell culture was treated with 1 ml serum-free medium containing 40 µg/ml of dsRNA Luc 1 or dsRNA Luc 2. The dish was incubated at 26°C for 60 min, and 2 ml full medium (containing heat-inactivated FBS and antibiotics) was added to each well. Cells were incubated for 3 days to allow turnover of the targeted proteins. Each day, cells were spun down, medium removed, and a boost of serum-free medium containing 40 µg/ml dsRNA Imp 1/dsRNA Imp 2 or 40 µg/ml of dsRNA Luc 1/dsRNA Luc 2 was added.
Cells were washed twice in PBS, prior to addition of Tri Reagent (Sigma-Aldrich), and total RNA and protein extracts were isolated from the samples following the manufacturer's instructions, with the following modifications. Proteins were precipitated by adding 3 X acetone and incubating at RT for 10 min, followed by centrifugation at 12,000 g for 10 min at 4°C. Precipitated proteins were washed twice with protein wash 1 buffer [300 mM guanidine hydrochloride (Sigma-Aldrich) in 95% ethanol + 2.5% glycerol (v:v)] and twice with protein wash 2 buffer [ethanol containing 2.5% glycerol (v:v)]. The proteins were resuspended in 1% SDS with 100 mM DTT, and protein concentration was measured using the Qubit® Protein Assay Kit (Life Technologies) according to the manufacturer's specifications.
To test for efficient knockdown of Imp, western blot was performed with a SDS-PAGE electrophosesis system. Briefly, 2 µg total protein sample was resuspended in SDS-load buffer [50 mM Tris-HCl (pH 6.8), 100 mM DTT, 0.05% bromophenol blue, 10% glycerol] and electrophoresed in a 10% SDS-PAGE gel with SDSrunning buffer [25 mM Tris, 192 mM glycine, 0.1% SDS]. The proteins were blotted to a PVDF membrane (Amersham Hybond-P), and the upper half was probed with polyclonal anti-Imp antibody (Adolph, 2009), whereas the lower half was probed with monoclonal anti-TBP antibody (SC Biotechnology). Horseradish peroxidase-conjugated anti-rabbit IgG (Cell Signaling) or anti-mouse IgG (Cell signaling) were then added, respectively, and the secondary antibodies were detected using enhanced chemiluminescence (Chemiluminescent Substrate from Chemiluminescent Nucleic Acid Detection Module, Thermo Scientific). Chemiluminescence was detected using the luminescent image analyzer LAS-1000plus, and ImageJ was used to quantify the Imp/TBP ratio.

Figure SMM4
Western Blot of actin protein levels in Imp deficient S2 cells Knock-down of imp and western blot was performed as previously described. Following blotting of proteins to a PVDF membrane (Amersham Hybond-P), the upper half was probed with polyclonal anti-Imp antibody (Adolph, 2009), the middle part with polyclonal anti-actin antibody (Cytoskeleton) whereas the lower half was probed with monoclonal anti-S6 ribosomal protein antibody (Cell signaling). Horseradish peroxidaseconjugated anti-rabbit IgG (Cell Signaling) or anti-mouse IgG (Cell signaling) were then added, and the secondary antibodies were detected using enhanced chemiluminescence (Chemiluminescent Substrate from Chemiluminescent Nucleic Acid Detection Module, Thermo Scientific). Chemiluminescence was detected using an SLR camera (Khoury 2010) and ImageJ was used to quantification.

RNA-seq on Imp double-stranded RNA interference-mediated knockdown
The following experiment was carried out in biological triplicates on different days. Primers used for the library preparation are listed in Figure SMM5.
dsRNA treatment with dsRNA Imp 1/dsRNA Imp 2 and dsRNA Luc1 were performed on 5x10 6  The conversion of poly(A) RNA into cDNA containing 5' and 3' adaptors ready for PCR amplification was performed as previously described (RIP/RNAseq section) with the exception that the 3' ligation adaptor (LIGATION_ADAPTER_RB) in these experiments contain a 7-mer random barcode, making it possible to collapse PCR duplicates.
PCR was performed using the Phusion DNA polymerase (Thermo Scientific) as previously described (RIP-seq section) with 20 cycles of amplification, creating libraries with the index primers indicated in Figure SMM5. E-gel size selection and AMPure XP beads (Beckman Coulter) purification were performed as previously described (RIP-seq section). Molar equivalents, as assessed by a Bioanalyzer equipped with a DNA 1000 chip, were mixed of all the final libraries, and the mixture was subjected to Illumina flow cell sequencing.

RT-qPCR on pharate adult mutants
Hemizygous imp G0072 mutants or wild-type pharate adult males were removed from their pupae under an Olympus SZ40 microscope and approximately 20 males were pooled. Males were recognized by their sex combs and hemizygous imp G0072 mutants were recognized by their red eye-colour. RNA was extracted from the pharate adults using Trizol reagent (Invitrogen) according to the instructions of the manufacturer, including the optional centrifugation step. The quantification was therefore carried out on a collection of pupae and did not reveal whether there were fluctuations in the levels between individual flies. cDNA was synthesized from total RNA isolated from wild-type or hemizygous imp G0072 pharate adult males with Superscript III (Invitrogen) using an oligo(dT) primer. Quantitative RT-PCR was performed with the LightCycler-FastStart DNA Master SYBR Green I kit (Roche) according to the instructions of the manufacturer. The following primers were used: Imp-F: GGTTGAGTTGCTTGTGGCCA, Imp-R: CGGCGGATTTGTCGAAAGCA, RP49-F: ATCCGCCCAGCATACAGGCCC and RP49-R: GTTCTCTTGAGAACGCAGGCGA. The relative amount of imp mRNA was normalized to the level of rp49 mRNA in each sample.

Excision of P-element
In order to excise the P-element from the imp locus, imp G0072 females were crossed to jumpstarter males (y1 w*; ry506 Sb1 P{2-3}99B/TM6, stock 3664 from Bloomington Drosophila Stock Center). Among F1 progeny, females with red eyes and a stubble phenotype, possessing a disrupted imp allele and a transposase gene were collected and crossed to FM7c/Y males. Among the F2 progeny, 53 flies in which the P-element had been excised were recognized by their eye color (white if not containing the transposase gene or rosy if containing the transposase gene) ( Figure S5). 23 lines with the excised alleles were created by crossing excised females to FM7c/Y males. Genome DNA was extracted from males of these stocks and the region of the imp locus spanning the P-element insertion site was sequenced (with the primer Imp13208-F TGATTAACAAGTGAGCGCGT) to ensure that excision had taken place. DNA was also extracted from two flies that died a few days after eclosion, and the Imp genome region was sequenced. The sequence showed that both flies lacked part of the imp gene as a result of imprecise P-element excision.

Survival count
The fraction of imp G0072 , wild-type or FM7c/wt progeny able to proceed through embryogenesis, larval

Data treatment and bioinformatics analyses
Read mapping and processing The following mapping procedure applies to the PAR-iCLIP, iCLIP, RIP and RNA-seq controls. Reads were preprocessed with custom python scripts. First, we de-multiplexed the reads according to their fixed barcode allowing up to 1 mismatch. Next, we trimmed base calls with one of the 5 lowest quality scores from 3' ends and removed adapter sequences. Finally, we removed duplicates by collapsing all identical reads containing the same random barcode and trimmed the 5' end of reads to remove the random barcodes. After these steps, all reads longer than 17 nucleotides were further analyzed. The dmel3 version of the Drosophila melanogaster genome and ensembl72 annotation [57] were used to build the indexes. Reads were mapped to the dmel3 genome and an exon junction database using BWA-PSSM [58]. To map PAR-iCLIP reads, we used a custom matrix for scoring mismatches assuming a 12.5% T>C conversion rate. For further analysis, only reads mapped with a posterior probability (PP) > 0.99, referred to as confidently mapped reads, were used. Finally confidently mapped reads overlapping by one or more nucleotides were clustered. To perform motif discovery these large clusters were further divided in smaller regions with uniform coverage.
The RNA-seq datasets obtained from the Imp knock-down experiments used for the differential expression analyses were analyzed independently as they contained yeast RNA spike-in. In these datasets, we mapped the reads using BWA-PSSM as described above but we used a new genome index that included both Saccharomyces cerevisiae (sacCer3) and Drosophila (dmel3) genomes.
The human HuR PAR-CLIP data used in Figure S2C was obtained from [1] (GEO accession number GSE29780). The data was processed as described above. Reads were mapped also as described above using an index built from hg19 version of the human genome and an exon junction index extracted from ensembl70 annotation using the custom matrix to score T>C conversions. The RNA-seq used to normalize HuR PAR-CLIP data was a pool of RNA-seq datasets from [59] and [60] (GSM714684; GSM714685; GSM940576). These data were mapped to the same indexes as the HuR dataset.

Annotation
For this project the ensembl72 annotation of the Drosophila melanogaster was used. In all downstream analyses, one representative transcript was selected for each gene. This transcript was always the longest protein-coding transcript annotated for the gene.

3'UTR length versus coverage correlation analyses
In order to investigate the relationship between 3'UTR length and CLIP enrichment we pooled iCLIP and PAR-CLIP datasets and normalized them. The iCLIP replicates were normalized to their corresponding RNAseq datasets, whereas PAR-iCLIP replicates were normalized to combined RNA-seq iCLIP control replicates, normalization described by in the methods section (equation 1). For each transcript the sum of enrichment was normalized to the 3'UTR length and log2 transformed.

Differential expression analysis
For each transcript we counted the number of confidently mapped overlapping reads on the same strand. Only unambiguously assigned reads (i.e. that only overlap one gene) were considered. Differential expression was calculated using DESeq R package [61]. We estimated dispersion using pooled-CR method, which treats replicates as paired data. The rest of the parameters were default settings. To correct for multiple hypothesis testing, we used Benjamini-Hochberg adjusted p-values and considered genes differentially expressed with a corrected p-value (FDR) < 0.05.

Co-occurrence analysis
To explore over-representation of repeated CA-rich motif occurrences we selected the 3000 most enriched CLIP clusters, and for each of these we extracted the sequence around cluster starts, beginning 40 nucleotides upstream from the cluster start and ending 50 nucleotides downstream. In these clusters we counted the number of co-occurrences of MACA (M is IUPAC nomenclature for A or C) separated by a particular distance. This count was then normalized by the number of possible words of that particular length as shown in equation 2. (2) where n is the number of sequences, cc w is co-occurrence count at a particular distance, l w is the word length and s i is the spacing and f is the normalized word frequency at the i'th spacing. As a control for the co-occurrence analysis we used a set of 3000 clusters 90 nt long randomly sampled from 3'UTRs of the longest protein-coding transcripts associated with genes.

Positional enrichment analysis
Using the same clusters defined above, we counted the occurrences of all 4-mers in each position of the 90 nt long clusters. Next, for each word we calculated their normalized frequency by dividing the word count in position c wi by the number of sequences n as defined in equation 3 (3) This positional frequency was calculated independently for the iCLIP clusters and for the background sequences . In these clusters, we defined the cross-link region as position 40 to 57 in the cluster, and flanking regions as the regions upstream and downstream of the cross-link region. Figures 2E and S2H show a running mean over 5 nucleotides of the positional frequencies described above. A mean Z-score for each word in the cross-link and flanking regions was calculated as described in equation 4. These Z-scores were plotted in the scatter plots in Figure 2D and Supplementary Figure S2G. (4)

Standardized transcript profiles
To make standardized profiles, the longest protein-coding transcript of each gene with at least 30% RNAseq coverage in S2 cells was used. For these transcripts, each region (5'UTR, CDS, 3'UTR) was divided in a fixed number of equally sized bins as follows: 20 bins in 5'UTRs, 50 bins in CDS, and 50 bins in 3'UTRs. For iCLIP profiles, we calculated the mean enrichment of iCLIP reads relative to RNA-seq e for each bin b as ̅ ∑ ∑ where is the total number of genes, is the length of the bin which may differ by one if sequence length is not divisible by the number of bins. is the enrichment of iCLIP reads relative to RNA-seq calculated as described in the main text. For the standardized HMM profile, we calculated the match occurrence count for each bin b and inserted as the enrichment value in the equation above.