Grsf1-Induced Translation of the SNARE Protein Use1 Is Required for Expansion of the Erythroid Compartment

Induction of cell proliferation requires a concomitant increase in the synthesis of glycosylated lipids and membrane proteins, which is dependent on ER-Golgi protein transport by CopII-coated vesicles. In this process, retrograde transport of ER resident proteins from the Golgi is crucial to maintain ER integrity, and allows for anterograde transport to continue. We previously showed that expression of the CopI specific SNARE protein Use1 (Unusual SNARE in the ER 1) is tightly regulated by eIF4E-dependent translation initiation of Use1 mRNA. Here we investigate the mechanism that controls Use1 mRNA translation. The 5′UTR of mouse Use1 contains a 156 nt alternatively spliced intron. The non-spliced form is the predominantly translated mRNA. The alternatively spliced sequence contains G-repeats that bind the RNA-binding protein G-rich sequence binding factor 1 (Grsf1) in RNA band shift assays. The presence of these G-repeats rendered translation of reporter constructs dependent on the Grsf1 concentration. Down regulation of either Grsf1 or Use1 abrogated expansion of erythroblasts. The 5′UTR of human Use1 lacks the splice donor site, but contains an additional upstream open reading frame in close proximity of the translation start site. Similar to mouse Use1, also the human 5′UTR contains G-repeats in front of the start codon. In conclusion, Grsf1 controls translation of the SNARE protein Use1, possibly by positioning the 40S ribosomal subunit and associated translation factors in front of the translation start site.


Introduction
Erythropoiesis, the daily generation of large numbers of erythrocytes, requires tight control of expansion and differentiation of the progenitor compartment. This depends on the cooperation of erythropoietin (Epo) with stem cell factor (SCF) and glucocorticoids. [4,12,22,55]. SCF does not exclusively act on erythroblasts, it is also important to maintain other progenitor compartments [13], and aberrant SCF signaling is observed in cancer [7,31,38,57].
SCF-induced expansion of the erythroblast compartment requires activation of mTOR (mammalian target of rapamycin) and subsequent release of eIF4E (eukaryote translation initiation factor 4E) [10,24]. The availability of eIF4E regulates the rate of mRNA translation in general, but transcripts with a long and structured 59UTR are hypersensitive to eIF4E levels, whereas transcripts with a short and simple 59UTR are only mildly affected by eIF4E levels [26,33]. We previously identified transcripts that are hypersensitive to eIF4E levels for their translation [24]. They included Use1 (unusual SNARE in the ER-1), the mammalian homologue of the yeast SNARE (Soluble N-ethylmaleimide-sensitive factor attachment protein (SNAP) receptor) Use1p [24,54].
Retrograde transport by CopI vesicles shuttles chaperones and other ER-resident proteins back from the Golgi to the ER, which is a requirement for anterograde transport of lipids and proteins to the Golgi by CopII vesicles [11,34]. Anterograde ER to Golgi transport depends, among others, on Sar1, member of the Arf family of GTPases, and on Sec23 as the Sar1-GAP protein [9]. Interestingly, mutations in Sec23B were identified as the genetic cause of congenital dyserythropoietic anemia type II (CDA-II) [25] and its specific erythroid phenotype is caused by cell type specific differential regulation of SEC family protein expression [48]. Obviously, the rate of anterograde transport to the Golgi has to increase in proliferating cells. SCF-induced, PI3K-dependent translation of Use1 is a novel mechanism rendering the Golgicoupled secretary pathway dependent on mitogenic signaling [24].
RNA-binding proteins can coordinate the translation of mRNAs encoding proteins that function in a specific pathway or in the developmental stage of a particular tissue. Misregulation of translation can cause severe diseases as exemplified by deregulation of Fragile X proteins [36]. Recently, a mitochondrial isoform of G-Rich Sequence binding Factor 1 (Grsf1) was shown to be important for the coordinated expression of mitochondrial proteins [2,29]. Grsf1 was initially shown to bind G-rich elements present in the 59UTRs of influenza virus and glutathione peroxidase 4 (Gpx4) to enhance translation of these transcripts [30,43,51]. In mitochondria, Grsf1 cooperates with RNaseP to stabilize and translate target transcripts [29]. Proteomic analysis indicated that Grsf1 is predominantly present as a monomeric protein [2].
To investigate the mechanisms by which mRNA translation contributes to adjust expression levels of Use1, we identified the full 59UTR and potential regulatory elements. A G-rich sequence located just upstream of the AUG start site is the docking site for Grsf1. Similar G-rich stretches are present just in front of the AUG start codon in human Use1. Modulation of Grsf1 expression directly correlated with Use1 protein expression, and both Use1 and Grsf1 appeared to be required to inhibit differentiation and sustain renewal divisions of erythroblasts.

Materials and Methods
Cell culture HEK-293T and NIH3T3 cells were cultured in DMEM (Invitrogen, Breda, The Netherlands) supplemented with 10% FCS, 100 IU/ml penicillin and 100 mg/ml streptomycin. Mouse liver was isolated from a C57/B6 mouse 8 weeks of age. The I/11 erythroblast line was cultured in StemPro-34 medium (Life Technologies) as described previously [55]. For expansion, the medium was supplemented with 0.5 U/ml Epo (Ortho-Biotech, Tilburg, The Netherlands), 100 ng/ml SCF (supernatant of CHO producer cells), and 10 26 M dexamethasone (Dex; Sigma-Aldrich). Differentiation was induced in medium containing 2 U/ml Epo, 0.5 mg/ml holo-transferrin. Cell numbers and cell size distribution were determined using an electronic cell counter (CASY-1; Roche). Cell morphology was analyzed in cytospines stained with histological dyes and neutral benzidine [28], using an OlympusBX40 microscope and OlympusDp50 CCD camera, and Viewfinder Lite 1.0 acquisition software.

RNA isolation, cDNA synthesis, RT-PCR and RACE
Total RNA from transiently transfected cells or from shRNA transduced I/11 cells was isolated using the Trizol reagent (Life Technologies) as recommended in the manufactures protocol. RNA from fractionated polyribosomes was isolated by proteinase K digestion, phenol-chloroform extraction and LiCl precipitation as described [20]. All RNA samples were treated with DNAse prior to cDNA synthesis according to manufacturer instructions. cDNA for real-time PCR (RT-PCR) was generated as described [28]. RT-PCR involved TaqMan technology (PE Applied Biosystems Model 7700 sequence detector), using the double stranded DNA-specific fluorescence dye SYBR green I to detect PCR product as previously described [32]. The amplification program consisted of 1 cycle of 50uC with 2 min hold (AmpErase UNG incubation), 1 cycle of 95uC with 10 min hold (AmpliTaq Gold Activation), 40 cycles of denaturation at 95uC for 15 s, annealing at 60uC for 30 s and extension at 72uC for 30 s. Gene specific primers were used (all primers are described in Table S1). For RACE experiment, 60 ng of purified poly(A)+ mRNA and GSP1 primer was used to synthesize cDNA at 60uC in accordance to manufacturer's protocol, using Transcriptor Reverse Transcriptase (Roche 03531317001) supplied together with the mRNA capture kit (Roche 117878960). First strand cDNA was purified using High pure purification columns (Roche 11732668001) and dA-tailed. The first PCR was performed using the GSP2 primer and the dT-linker forward primer supplied by the kit. Nested PCR was performed with the GSP3 primer and the forward primer supplied in the kit. Final products were cloned directly into pCR2.1 (Invitrogen) according to the instructions of the manufacturer. Nucleotide sequencing was carried out using a binding domain sequencing kit according to instructions from the provider (PE Biosystems). Sequencing was carried out on an ABI 310 automatic sequencer (PE Biosystems) using the M13 forward primer.

Constructs
The mouse and human Use1 UTR was amplified using the Expand High Fidelity PCR System (Roche), a forward primer harboring an XhoI site and T7 promoter, and a reverse primer containing a NcoI site. PCR products were ligated in pCR2.1 (Invitrogen) and pGL2-basic. The plasmids DSS, DGGGG and DAGGGCGGA were generated with the QuickChange Site-Direct Mutagenesis Kit (Stratagene). In the DSS the AAGAUGG sequence around the start codon was changed into a NcoI site: ACCAUGG. All modified vectors were confirmed by DNA sequencing. The constructs DIn, and 'b-globin intron' were generated by PCR cloning. The DIn was cloned as the spliced construct from mouse cDNA. The bglobin plasmid was constructed by replacing the Use1 intron of WT with the 80 bp b-globin intron sequence from bIVS II (kind gift from S. Philipsen, Erasmus MC). To obtain the b-globin plasmid three separate PCR reactions were performed with overlapping primers containing Use1 and b-globin intron sequence. PCR1 amplified Use1 59 flanking sequence, PCR2 the mini b-globin intron, and PCR3 the 39flanking sequence to the AUG start codon. The three PCR products were purified on agarose gel. PCR1 was fused to PCR2 product using forward primer PCR1 and reverse primer from PCR2, and PCR3 was fused using forward primer from PCR1 and reverse primer PCR3. The Grsf1 plasmid was cloned to pcDNA3.1(-) [52]. All indicated primers are shown in Table S1.

In vitro transcription and translation
Reporter plasmids (pGL2-basic) were linearized downstream of the luciferase coding sequence (Sal1). In vitro transcription was done with the Riboprobe in vitro Transcription Systems (Promega). In vitro translation assay were carried out with the Rabbit Reticulocyte Lysate System (Promega), according to the manufacture's protocol. After the reaction 5 ml was taken for the luciferase assay. Luciferase activity was measured using the Steady-Glo system (Promega).

Transfection, Western Blotting and Reporter Assay
For reporter assays, 10 7 Ba/F3 cells were electroporated (0.28 kV, capacitance 960 mFD) with maximum 20 mg of DNA. The HEK-293T cells were transfected using the transfection kit Mirus MT-1 (Mirus Bio Coorporation) in 24-well plates according to manufacture's protocol. Reporter plasmids (WT, DAGGGCGGA) and Grsf1 expression vector were co-transfected in NIH3T3 cells using the transfection kit Lipofectamine 2000 (Invitrogen) in 24-well plates. The reporter harboring the human Use1 59UTR, and the Grsf1 expression vector were co-transfected in 5610 5 HEK-293T cells/well seeded in 12-well dishes using calcium chloride. To exclude transcriptional effects due to different promoter doses, the total plasmid concentration was kept constant by adding appropriate amounts of empty expression vector pcDNA3.1(-). Transfected cells were harvested 24 hr posttransfection, or at the time indicated. Luciferase activity was measured in cell lysates using the Steady-Glo system (Promega). Transfection efficiency was determined by co-transfecting lacZ and analyzing b-galactosidase activity using the B-Gal kit (Promega). I/11 cells transduced with shRNA were harvested after the indicated time point by the addition of ice-cold PBS. Cell lysates, SDS-PAGE and Western blotting were performed as described previously [53]. To analyze Use1 and actin protein level, equal amounts of protein were loaded on a 12.5% polyacrylamide gel. The antibodies used: anti-p31 rabbit polyclonal antibody (Synaptic Systems) and anti-actin goat (Santa Cruz, CA), secondary antibody: Goat-anti-Rabbit 800 and Donkey-anti-Goat 800 (LI-COR Biosciences). Proteins were visualized using the Odyssey Infrared Imaging Machine (LI-COR Biosciences)

Virus production, cells transduction
The pLKO.1-puro lentiviral construct containing shRNA sequences against Use1, Grsf1 and control sequence were obtained from Sigma's MISSION TRC-Mm 1.0 (Mouse) shRNA library (Table S2). HEK-293T cells were transfected with 2.5 mg of pMD2.G DNA, 7.5 mg psPAX.2 DNA and 10 mg of lentiviral viral vector as described above. Medium was harvested 24 and 48 hours following transfection, and concentrated by ultracentrifugation at 60,000 g for 2 hours. The pellet was resuspended in 150 ml of 1% BSA/PBS and incubated for 2 hours at 4uC. Lentiviruses were then used directly for titration, snap-frozen and stored at 280uC. For titration, HEK-293T cells were seeded at 10 5 per well in 6-well plates and dilutions from 10 24 to 10 28 of lentivirus supernatant were used for transduction in presence of 8 mg/ml polybrene (Sigma Aldrich). Cells were selected for 8 days with puromycine, washed with PBS, fixed with 70% ethanol and stained with crystal violet for 1 hour. Colonies were counted by microscope at a 106magnification. Concentrated lentiviruses with titers around 500610 6 TU/ml were used to transduce 4610 6 of I/ 11 cells in presence of 8 mg/ml polybrene. Cells were selected by adding puromycine 24 hours post transduction.

RNA mobility shift assays
RNA probes were generated in the presence or absence of small amounts of Digoxigenin-11-UTP (Roche), using the T7 Megashortscript Kit (Ambion) and the primers listed in Table S1. As a template, WT or mutated DGGGG T7 promoter driven DNA constructs were used. In vitro transcription products were purified using the MEGAclear kit from Ambion. For RNA/proteinbinding studies, different amounts of protein (400 ng if not stated otherwise) were incubated with 0.5-1 pmol of digoxigenin-labeled RNA probe for 20 min at 30uC in binding buffer (10 mM HEPES-NaOH at pH 7.9, 25 mM KCl, 1.5 mM EDTA, 4% glycerol, 0.25 mM DTT, 7.5 mg/mL heparin, 25 ng/mL yeast tRNA) in a reaction volume of 15 mL. Afterwards the samples were analyzed on native 5% polyacrylamide gel electrophoresis. Protein/RNA complexes were transferred to a nylon membrane, and blots were visualized using the DIG Luminescent Detection Kit (Roche). The RNA binding affinity Kd was estimated as shown previously [35,46]. Briefly, following visualization bands (free RNA versus shifted RNA) were quantified using Biorad quantification software. Then logarithmic values of the ratio of shifted (complexed) versus free RNA were plotted as function of logarithmic values of the molar amount of protein present in the reaction. This revealed a linear correlation (y = mx+n), where m, n and r 2 were be determined in excel software. The logarithmic value of the Kd could be quantified from deducing the molar amount of protein when log (complexed RNA/free RNA) equals zero.

Circular dichroism (CD) spectroscopy
CD spectroscopy experiments were performed as described previously with some minor alterations [6]. A 40 nt nucleotide probe of the Use1 59UTR wt sequence (59-GUA-GCA-GGG-CGG-ACC-UCG-GAG-GGG-AAG-GAC-CUC-ACU-CAG-G-39) was used. Briefly, RNA stocks at 20 mM in Tris-HCl, pH 7.5, in the presence or absence of 100 mM of KCl or NaCl were heated to 70uC for 5 minutes and then slowly cooled down to 10uC at a rate of 0.01 K/s. RNAs were then diluted to 5 mM and CD spectra were monitored at a Jasco J720 spectropolarimeter at 10uC using a 200 ml quartz cuvette with a path length of 1 mm using a measurement range of 320 -220 nm, a response of 2 sec, data pitch of 0.1 nm, bandwidth of 1 nm and a scanning speed of 100 nm/min. Data were accumulated from ten scans. Presented data are representative for 3 independent experiments.

Hemoglobinisation and flow cytometry
Hemoglobin was measured as described before [3]. Propidium Iodide positive cells were measured by flow cytometry using the LSR II (Becton-Dickinson, San Jose, CA).

Identification of the 59UTR of Use1
To investigate polysome recruitment of Use1 we first identified the complete 59UTR of Use1. The reference cDNA NM_025917.4, representing Use1 version 1, contains only a short and probably incomplete 59UTR of 19 nt, possibly due to secondary structures in the 59UTR that prematurely stopped the reverse transcriptase. A longer Use1 isoform is encoded by reference cDNA NM_001145780.1 that includes an upstream exon with an alternative translation start site. For RACE (rapid amplification of cDNA ends) we used a recombinant RT-enzyme active at 60uC. Nested PCR was used in combination with primers located just downstream of the AUG start codon (figure 1A), which yielded two distinct RACE fragments, 420 and 230 nt long (figure 1B). The 420 nt fragment overlapped with EST BY097923 and represents a 59UTR of 387 nt. The shorter fragment extends as far upstream as the long fragment, but represents a transcript in which an intron of 156 nucleotides has been spliced out exactly as represented by NM_001145780.1 (figure 1A). The alternative intron is situated just in front of the Use1 AUG start codon.

Efficient Use 1 expression requires intronic sequences
Alternative splicing of the 59UTR may encode proteins with alternative start codons. In addition, splicing of the 59UTR may affect gene expression when regulatory sequences that determine transcript stability, localization, or translation are present in the alternatively spliced intron [15,58].
We isolated subpolysomal and polysome bound mRNA from erythroblasts and determined the presence of spliced and unspliced Use1 mRNA by RT-PCR in each fraction ( figure 2A,B). We found 65% of the unspliced form, and only 25% of the spliced form of Use1 transcript present in the actively translated fraction of polyribosomes, which suggested that the presence of the intron stimulated translation initiation ( figure 2A,B). Although the spliced Use1 transcript was far more abundant than the unspliced Use1 transcript in the 59RACE experiment (figure 1B), RT-PCR indicated near equal expression in total RNA (data not shown).
Next we examined protein expression. A start codon 44 nt upstream of the splice donor site gives rise to a short upstream ORF (uORF) in the unspliced transcript, and to a Use1 protein with a 14 amino acid extension at the N-terminus encoded by the spliced transcript (figure 2C). On Western blot we compared Use1 expression in erythroblasts, with Use1 expression in liver, an organ with high transcript expression. Only the p31 isoform of Use1, translated from the unspliced transcript, is detected in erythroblasts (figure 2D). A small amount of the longer isform is detectable in liver cells.
The role of the alternative intron in Use1 translation was further analyzed by reporter assays in which the expression of luciferase coupled to a wild type 59UTR (WT) was compared with constructs lacking the intron (DIn), or containing a mutated 39 splice site (DSS). First, identical molar amounts of in vitro synthesized mRNA were translated in reticulocyte lysate. The luciferase activity increased 3.5 fold when transcripts lacked the intron, which suggested that the intron contains sequences that inhibit in vitro translation ( figure 3A). Deletion of the 39 splice site did not affect in vitro translation ( figure 3A). Next, the same reporter constructs cloned into an expression vector (pGL2 basic) were expressed in HEK-293T cells. Quantitative PCR indicated that transcript expression was similar in all experiments (figure 3B). In contrast to translation in reticulocyte lysate, luciferase activity of the transcript lacking the intron was 4-fold decreased in these cells ( figure 3C). Mutation of the 39 splice site did not drastically alter luciferase activity. The splice site overlaps the Kozak sequence around the start codon. The small reduction of luciferase activity in the DSS mutant may be due to the alteration of the luciferase start codon from AAGAUGG to ACCAUGG (figure 2C). We also tested a construct in which the intron was replaced by the minimal

Grsf1 binds to Use1 59UTR and controls Use1 translation
Analysis of the alternative intron of Use1 revealed an A(G) 4 A element similar to the Grsf1-binding element found in the 59UTR of influenza virus and glutathione peroxidase 4 (Gpx4) that enhances translation of these transcripts ( figure 4A) [30,43,51].
The binding of Grsf1 to the G-rich elements in the 59UTR of Use1 was analyzed by RNA mobility shift assay. Increasing amounts of in vitro synthesized Grsf1 were incubated with a digoxigenin-labeled 40 nt probes containing the G-rich elements (figure 4B), or with a deletion of the A(G) 4 A motif (data not shown). The affinity of Grsf1 to both probes was calculated by plotting the quantified band shift signals against the Grsf1 concentration ( figure 4C). This algorithm revealed a linear correlation and the intercept with the X-axis indicated a Kd To test whether the G-rich motives are involved in Use1 translation regulation, reporter constructs containing the 59UTR of Use1 fused to the ORF of the luciferase gene were transfected into Ba/F3 cells. Luciferase activity and luciferase mRNA level were analyzed 24 h post transfection. Deletion of the A(G) 4 A motif reduced luciferase activity to 75%. Comparison of the Gpx4 and Use1 59UTR indicated that the A(G) 4 A is present in an area with multiple G-repeats, which may form to a secondary structure. This putative structure, as well as the A(G) 4 A sequence may contribute to control of mRNA translation. Therefore, we also deleted the adjacent G-rich element AGGGCGGA, which reduced the luciferase activity to 50% of wt levels (controlled for equal mRNA expression, figure 4D). Thus, G-rich repeats may bind Grsf1 to control mRNA translation of Use1 reporter constructs. Both a A(G) 4 A and an adjacent A(G) 3 C element may be involved in control of mRNA translation. The activity of the A(G) 3 C motif may explain why the Kd of Grsf1 dropped from 527 nM to 4.4 mM upon deletion of the A(G) 4 A motif, but not to insignificant binding.

Grsf1 stimulates Use1 translation
Next we examined whether expression of Grsf1 directly controls Use1 protein expression using over-and underexpression of Grsf1. We first expressed increasing amounts of Grsf1 together with luciferase reporter constructs harboring the WT 59UTR of Use1 or a 59UTR lacking the A(G) 3 CGGA motif (DAGGGCGGA) in NIH3T3 cells. Increasing levels of the Grsf1 expression plasmid, incremented expression of luciferase from the WT construct 2fold. Luciferase activity from a reporter construct lacking the A(G) 3 CGGA element was 2-fold reduced compared to the WT construct, and remained unaffected by increasing Grsf1 expression ( figure 5A).
Conversely, we also reduced Grsf1 expression using vectors expressing shRNA complementary to Grsf1, or a control shRNA. The mouse I/11 erythroblasts were transduced with the lentiviral constructs and harvested 48 h after puromycin selection. The efficiency of Grsf1 knock down was measured by quantitative RT-PCR and showed that 2 out of 5 shRNA vectors present in the TRC library [41] resulted in at least 65% knock down (figure 5B). Protein lysates isolated from the mouse erythroblasts were examined by Western blot. Grsf1 knock down clearly repressed expression of Use1 protein (figure 5C). In summary, increased Grsf1 levels induced expression of a reporter for Use1 translation, and Grsf1 knock down reduced Use1 expression in erythroblasts.

Homology between species
The splice-donor and -acceptor sites in mouse represent perfect consensus sites. Comparison between species, however, indicated that splicing is not conserved ( figure 6A). The splice acceptor site is absent in humans, the splice donor site is absent in rat. Also the branch point A-residue that is present in a perfect consensus in the alternative mouse intron is lacking in humans and rat (figure 6A). Together with the absence of spliced isoforms in EST databases of other species, this strongly suggested that the alternatively spliced intron of mouse could be regarded a bona fide part of the Use1 transcript. Surprisingly, the 59UTR of primates, man included, shows another unusual feature. The AUG start site of mouse Use1 is conserved, followed by a stop codon at the 8 th triplet, and the AUG start site of the main open reading frame of USE1 12 nt downstream of this stop codon (figure 6B). When we analyzed the presence of G-rich motifs in man and mouse, we found similar Grepeats. These repeats obtain scores of 19 (mouse) and 20 (human) as putative G-quadruplex structures using the QGRS-mapper (http://bioinformatics.ramapo.edu/QGRS/index.php; figure 6B). Interestingly, the predicted G-quadruplexes were not present in sequences conserved between man and mouse, but their position relative to the start codon of the Use1 ORF was similar.
To confirm the presence of G-quadruplexes in the Use1 59UTR, we monitored circular dichroism spectra of the 40 nt RNA construct that binds to Grsf1 ( figure 6C). The spectra are characterized by a positive peak at 265 nm and a negative peak at 240 nm, which is characteristic for a parallel quadruplex topology [6]. Spectra were assessed in the presence of KCl, NaCl or without salts. The presence of KCl coincided with the strongest (265 nm, 240 nm) peak intensities. CD spectra obtained in the presence of NaCl or without salts exhibited significant changes of the peak intensities. Thus, our data support the in silico data that indicate the presence of a G-quadruplex structure within the Use1 59UTR, and this structure is stabilized by the monovalent K + cation.
The 59UTR of human USE1 was cloned in front of a luciferase reporter gene to test whether translation of human USE1 is also  figure 6D). Thus, the presence of a G-rich motif and responsiveness to Grsf1 is conserved between man and mouse despite a different translation start site.

Grsf1 and Use1 are required for transient amplification of erythroblasts
Use1 is expressed under conditions that favor amplification of the erythroblast compartment. Upon induction of differentiation to mature erythrocytes, Use1 is rapidly down regulated [24]. If Use1 is required for expansion of the erythroblast compartment, its knock down and the knock down of its regulator Grsf1 are expected to abrogate amplification and to induce differentiation. Two lentiviral shRNA vectors able to repress Grsf1 expression (figure 5B), and two lentiviral shRNA vectors resulting in a 60% knock down of Use1 expression (figure 7A) were transduced to I/ 11 mouse erythroblasts. Cytospins were made 4 days following transduction, and revealed a marked increase in small hemoglobinised cells and in pycnotic cells (figure 7B). Cells were calculated and cumulative cell numbers were calculated. Cultures transduction with shRNA Grsf1-a and -b, or Use1-a continued to proliferate, although at reduced rate compared to cells transduced with control shRNA. Cell cultures transduced with shRNA Use1-b decreased in cell number and had to be excluded for further analysis of differentiation ( figure 7C). Upon induction of differentiation, control cells increased their hemoglobin content, but not upon knockdown of Use1 or Grsf1 ( figure 7D). Instead cells with suppressed expression of Use1 or Grsf1 accumulate dead cells that stain positive for propidium Iodide (PI) ( figure 7E). The data shown represent one of four experiments.

Discussion
Cell proliferation requires increased synthesis of glycosylated lipids and proteins which is dependent on ER-Golgi protein transport by CopII-coated vesicles. Retrograde Golgi to ER transport of chaperones, transport proteins, and other ER resident proteins in CopI vesicles is crucial to maintain ER integrity and allow for anterograde transport to continue. We identified the CopI specific SNARE protein Use1 as a gene under tight translational control in erythroblasts. Translation of Use1 Alternative splicing of the mouse Use1 59 UTR The 59RACE experiments indicated that a sequence of 155 nt immediately upstream of the AUG start codon is spliced out in Use1 transcripts. Reporter assays and polysome analysis indicated that the unspliced transcript is translated preferentially. Alternative splicing of the 59UTR as a mechanism to expose or remove translational control elements is increasingly recognized as a mechanism that controls protein expression [44,45]. However, the splice donor or acceptor site are lacking in all other species that were analyzed, suggesting different regulatory mechanisms in these species. In primates the sequence corresponding to the mouse AUG start codon and its surrounding Kozak consensus sequence is identical to the mouse transcript, but after 8 codons it is followed by a stop codon. The start codon of human USE1 is located 12 nt downstream of this stop codon (a CUG triplet in mouse). This suggests that translation in primates is not dependent on leaky splicing, but on leaky scanning.

Secondary structures in the 59UTR of Use1
Several studies identified transcripts that are hypersensitive to mTOR signaling and free eIF4E levels for their translation [24,26,33,49]. These transcripts mostly harbor a complex RNA structure in their 59UTR [23] and unfolding of such structures is thought to require a higher density of scanning complex loading, together with optimal activity of the eIF4A/B helicase complex. Comparison of luciferase reporter constructs that contain the spliced or unspliced 59UTR showed that intron sequences inhibit translation in reticulocyte lysate, but enhanced translation in cells. This indicates that secondary structure does not always inhibit translation in cells. In intact cells, secondary structures may have a positive role in translation, and they are most likely recognition sites for RNA binding proteins. The structural element in the alternatively spliced sequence appeared to bind Grsf1. Stable secondary structures with a positive effect on mRNA translation were also found in other transcripts identified as being translationally regulated in erythroblasts [24]. For instance, deletion of the secondary structures in Nme2 or Igbp1 enhanced translation in reticulocyte lysate, but impaired translation of reporter constructs in cells [Nieradka et al. unpublished data]. The regulatory domains in the 59UTR of Nme2 and Igbp1, however, did not bind Grsf1.
Grsf1 binds a secondary structure upstream the Use1 start codon The Grsf1 binding site in Influenza and in Gpx4 was determined to be an A(G) 3 U or A(G) 4 A element respectively whereas a larger structure was suggested to be involved in Grfs1 binding in Gpx4 [30,51,52]. Analysis of the G-rich sequences with the QGRS-mapper that predicts G-quadruplex structures suggests that the A(G) 4 A element is part of such a G-quadruplex structure both in Use1 and in Gpx4, and this prediction was supported by our in vitro data obtained by CD spectroscopy. The entire structure was present in the RNA gel shift probes, used to detect specific Grsf1 binding with a Kd intermediate between Gpx4 and the influenza NP 59UTR. Binding of Grsf1 to the larger structure, that may be a G-quadruplex, explains why deletion of the nearby A(G) 3 CGGA element affected translation of the reporter even more than deletion of the A(G) 4 A. Grsf1 potentially binds such structures similar to other hnRNP F/H proteins [51]. Detailed G-quadruplexes are increasingly recognized as elements that control translation [5,14]. Interestingly, also transcripts for TGFb2 and CyclinD3 contain G-quadruplex structure [1,56]. TGFb2 is a Epo-induced gene upregulated during erythroid differentiation [32]. When erythroblasts shift from a renewal cell cycle with size control, to differentiation-specific cell cycle control, they shift from CyclinD1 to CyclinD3 expression [40]. It needs to be investigated whether these G-quadruplexes also recruit Grsf1 for the translation of the corresponding mRNA, and whether reduced expression of TGFb2 and CyclinD3 contributes to impaired differentiation following Grsf1 knockdown. Grsf1 is involved in retrograde transport of ER resident proteins When we reported mEd2 to be translationally regulated [24] it was not yet known to be the human homologue of S. cerevisiae Use1p and part of the syntaxin 18 SNARE complex [54]. The essential role of Use1 in control of retrograde transport raises the question whether other regulatory proteins involved in ER-Golgi transport are tightly regulated by Grsf1. SCF controls translation of the ER export receptor Cnih1 (Cornichon) involved in transport of transmembrane proteins and EGF-like ligands (Gurken in Drosophila) [18,19,24]. The transcripts that required SCFinduced signal transduction to be translated also included the chaperone proteins Hspe1 (hsp10), hspcb (hsp90B) and hspa8 (hsp70 protein a8) [24]. Analysis of their 59UTR revealed putative secondary structures containing G-rich repeats that fulfill the criteria set for Grsf1 binding sites [30,51]. Thus, Grsf1 may be involved in translational control of critical ER-resident proteins that may be critical to cope with increased protein and lipid production in proliferating cells.

Grsf1 expression is transcriptionally and translationally controlled
Grsf1 is itself translationally regulated. Activation of mTOR results in translation of the RNA binding protein Dazl (Deleted in azoospermia-like), and Dazl subsequently binds the 39UTR of Grsf1 to enhance translation [27,47]. In addition, Grsf1 transcription is a downstream target of the Wnt/b-catenin pathway. Knock down of Grsf1 impairs embryonic development and recapitulates Wnt signaling defects [37]. Preliminary data indicate that the Wnt pathway sustains amplification of erythroblasts and induces Grsf1 expression in primary fetal liver derived erythroblast culture [Nieradka et al. unpublished data].