Main

Alternative 3′ UTR isoform abundance was shown to be highly cell-type-specific and can change upon proliferation, differentiation and transformation1,2,3. Alternative 3′ UTR isoforms produce the same protein, but the long 3′ UTRs contain additional regulatory elements that can regulate mRNA localization and protein abundance1,2,10. We have discovered a new function of 3′ UTRs: they can regulate protein localization independently of RNA localization.

CD47 is best known as a ubiquitous cell surface molecule that acts as a marker of self and protects cells from phagocytosis by macrophages11,12. We found CD47 protein expressed on the cell surface, as well as intracellularly (Fig. 1a and Extended Data Fig. 1a–c). The CD47 gene produces alternative 3′ UTRs as determined by the 3′-end sequencing method 3′-seq and confirmed by northern blot analysis (Fig. 1b, c)3. Exclusive knockdown of the longer 3′ UTR isoform by short hairpin RNAs (shRNAs) decreased CD47 surface expression without changing intracellular expression (Fig. 1d and Extended Data Fig. 1d–g). This suggests that the long 3′ UTR isoform facilitates cell surface localization of CD47 protein.

Figure 1: The long 3′ UTR of CD47 localizes GFP-TM protein to the plasma membrane, whereas the short 3′ UTR localizes it to the endoplasmic reticulum.
figure 1

a, Fluorescence confocal microscopy of endogenous CD47 protein in non-permeabilized (top) and permeabilized (bottom) cells. IC, intracellular. b, 3′-seq analysis of naive B cells shows two 3′ UTR isoforms of CD47 mRNA (short 3′ UTR (SU) and long 3′ UTR (LU)). Shown is the last exon of the gene model. Isoform abundance shown in transcripts per million (TPM). c, Northern blot analysis of human cell lines confirming CD47 mRNA isoforms from b. The corresponding ethidium-bromide-stained RNA gel is shown as loading control. d, Staining of U2OS cells as in a after transfection of a control shRNA (sh Co) or an shRNA against the long CD47 3′ UTR isoform. e, CD47 protein contains an N-terminal signal peptide (S; green), ECD (blue), five TMDs (grey) and a cytoplasmic C terminus (C; red). In both constructs, the ECD was replaced with GFP and either fused with the long (GFP-TM-LU) or the short CD47 3′ UTR (GFP-TM-SU). Constructs are drawn to scale. f, Fluorescence confocal microscopy of fixed U2OS cells after transfection of GFP-TM-LU or GFP-TM-SU. Bottom, with additional staining of the endoplasmic reticulum with anti-calnexin. g, FACS analysis of GFP expression in transfected U2OS cells with (dark blue lines, detection of total expression) and without permeabilization (light blue lines, detection of surface (surf.) expression). Values for mean fluorescence intensity (MFI) are shown in parentheses. Unstained cells are shown in grey. Representative image from more than 20 experiments. h, RNA-fluorescence in situ hybridization (FISH) (red) against GFP in permeabilized U2OS cells after transfection of GFP-TM-LU or GFP-TM-SU. Bottom panel also shows GPF-TM protein. a, d, f and h are representative images from hundreds of cells. Scale bars, 10 µm.

PowerPoint slide

Full size image

To test this hypothesis, we asked whether green fluorescent protein (GFP) encoded by an mRNA containing the long (with a mutated proximal polyadenylation signal; Extended Data Fig. 1h) or the short 3′ UTR of CD47 would localize differently. To allow GFP to enter the secretory pathway, we replaced the extracellular domain (ECD) of CD47 with GFP, while preserving the CD47 signal peptide, transmembrane domains (TMDs) and carboxy terminus, which we refer to as GFP-TM (Fig. 1e). We observed that GFP-TM encoded by an mRNA containing the long 3′ UTR of CD47 (GFP-TM-LU) localizes primarily to the cell surface whereas GFP-TM encoded by an mRNA with the short 3′ UTR of CD47 (GFP-TM-SU) localizes predominantly to the endoplasmic reticulum (Fig. 1f and Extended Data Fig. 1i). The localization results were confirmed by fluorescence-activated cell sorting (FACS) analysis, using an anti-GFP antibody on permeabilized and non-permeabilized cells to measure total and surface GFP levels, respectively (Fig. 1g). The localization step occurs at the protein level, as both the LU- and SU-containing GFP transcripts show a similar distribution near the perinuclear endoplasmic reticulum (Fig. 1h). Thus the LU isoform of CD47 encodes information that is necessary for cell surface expression of GFP-TM protein, in a manner independent of RNA localization.

To address the mechanism of 3′ UTR-dependent protein localization (UDPL; Fig. 2a), we reasoned that there must be an RNA-binding protein (RBP) that binds to the long, but not the short, 3′ UTR of CD47. The long 3′ UTR of CD47 contains many uridine-rich elements (see later), which are potentially bound by HuR6,7,8,9. HuR is known for its role in mRNA stabilization and translation activation7,13,14. However, HuR knockdown by shRNAs did not affect CD47 mRNA abundance or isoform levels, nor did it affect total CD47 protein levels (Fig. 2b, bottom, and Extended Data Figs 1d, 2a–c). But, strikingly, knockdown of HuR reduced CD47 surface expression (Fig. 2b, top, and Extended Data Fig. 2c). This suggests that for CD47, HuR mediates protein localization post-translationally.

Figure 2: 3′ UTR-dependent protein localization (UDPL) depends on HuR, SET and RAC1, and mediates surface localization of membrane proteins.
figure 2

a, Model of UDPL. HuR binds to the long 3′ UTR and recruits SET. During translation of CD47 mRNA, this protein complex is targeted to the endoplasmic reticulum (ER) surface where SET binds to the newly translated cytoplasmic domains of CD47. This step probably requires energy. SET interacts with RAC1 and active RAC1 translocates SET and CD47 to the plasma membrane. b, FACS analysis of endogenous CD47 protein expression in HEK293 cells. Left panel is after transfection of control shRNA (sh Co) or shRNAs against HuR (shown are all GFP+ cells). Middle and right panels depict cells stably expressing the indicated shRNAs. Surface CD47 (top) and total CD47 protein (bottom) were measured. c, 3′-seq analysis for CD44 in HEK293 cells and for TNFRSF13C in B-LCL cells, as shown in Fig. 1b. FACS analysis of endogenous CD44 protein in HEK293 cells (left) and endogenous BAFFR protein in SHSY-5Y cells (right) shown as in b (left). d, Left, FACS analysis of GFP after transfection of constructs containing a signal peptide and GFP fused to the TMD and C terminus of CD44 and either the long 3′ UTR (dark blue line) or the short 3′ UTR (light blue line) in U251 cells. Right, as in left panel, but for BAFFR with transfection into HeLa cells. e, FACS analysis of GFP expression shows that HuR-BS is sufficient for surface localization of GFP-TM (dark blue line). Deletion of the binding sites from the HuR-BS construct (HuR-BSΔ) abrogates GFP surface expression (light blue line). For b, c, d and e, surface and total protein expression were determined and shown as in Fig. 1g. Representative images from three (sh2 HuR, n = 5) biological replicates.

PowerPoint slide

Full size image

Beyond the role of HuR as an RBP6,7,8,9, HuR interacts through protein–protein interactions with SET, ANP32A and ANP32B4. Nuclear SET binds to histone tails and prevents acetylation15, but phosphorylated SET localizes to the cytoplasm and the surface of the endoplasmic reticulum5,16. Also, SET interacts with RAC1, and active RAC1 translocates SET to the plasma membrane5. In our model of UDPL (Fig. 2a), HuR binds to the long 3′ UTR of CD47 and recruits SET. Upon targeting of the mRNA to the endoplasmic reticulum surface, the scaffold function of the 3′ UTR results in local recruitment of SET to the site of translation. After translation of CD47 mRNA, the ECD is located in the endoplasmic reticulum lumen, whereas its C terminus is cytoplasmic. This allows SET to interact with the newly translated C terminus and cytoplasmic domains of CD47 and to translocate CD47 to the plasma membrane via active RAC1. Transfer of SET from CD47 mRNA to CD47 protein probably requires energy input, as has been shown for transfer of the signal peptide from the signal recognition particle to the translocation channel17. In this model of UDPL, surface expression of CD47-LU depends on SET and active RAC1. And indeed, knockdown of SET or RAC1 by shRNAs reduced surface expression of CD47 without affecting overall CD47 levels (Fig. 2b and Extended Data Fig. 2c–e).

To determine if UDPL is a more widespread phenomenon, we examined the localization of four additional transmembrane proteins that are derived from mRNAs with 3′ UTR isoforms that can be bound by HuR6,7,8 (Extended Data Figs 3 and 4). TSPAN13 has only one 3′ UTR isoform, whereas the other three genes (CD44, ITGA1 and TNFRSF13C (encoding the BAFF receptor, BAFFR) use ApA to generate alternative 3′ UTR isoforms (Fig. 2c and Extended Data Fig. 2f)3. As was the case for CD47, knockdown of HuR decreased surface expression of all four proteins without changing total protein levels (Fig. 2c and Extended Data Fig. 2a, f). As was done for CD47, we generated GFP-fused LU and SU constructs for CD44, ITGA1 and TNFRSF13C with their respective TMDs, C termini and 3′ UTRs. For all tested genes, the longer 3′ UTR increased surface localization of GFP-TM (Figs 1f–h, 2d and Extended Data Fig. 2g). This demonstrates that UDPL has the potential to be a widespread phenomenon.

The 3′ UTR of CD47 contains over 30 putative HuR-binding sites (Extended Data Fig. 3). We tested whether a 3′ UTR with a few HuR-binding sites (HuR-BS) (Extended Data Fig. 3) is enough to mediate surface localization. Indeed, the uridine-rich sequence was necessary and sufficient for surface localization of GFP-TM, although it was less potent than the full-length 3′ UTR of CD47 (Fig. 2e).

Next, we investigated each step of our UDPL model in more detail. We demonstrated by RNA-immunoprecipitation that HuR binds to the HuR-BS and to the LU but not to the SU isoform of CD47 (Fig. 3a, left). SET also associates with the long 3′ UTR of CD47, which is dependent on HuR (Fig. 3a, right). SET or HuR overexpression was insufficient to localize GFP-TM-SU to the cell surface (Extended Data Fig. 5a, b). However, tethering of SET or HuR to the short 3′ UTR isoform of CD47 was sufficient to redirect GFP-TM localization from the endoplasmic reticulum to the plasma membrane (Fig. 3b and Extended Data Fig. 5c, d; see Extended Data Fig. 5 for experimental details). This indicates that local recruitment of SET to the site of translation, mediated by the scaffold function of the long 3′ UTR, is required for UDPL.

Figure 3: Mechanism of UDPL.
figure 3

a, Left, RNA-immunoprecipitation after transfection of the indicated constructs into HEK293 cells. Protein–RNA complexes were pulled-down with anti-HuR antibody and GFP abundance was normalized to GAPDH and is shown as fraction of input. Right, RNA co-immunoprecipitation after transfection of the indicated shRNAs. Protein–RNA complexes were pulled down with anti-SET antibody and the abundance of CD47-LU was normalized to GAPDH and is shown as fraction of input. Shown is mean ± standard deviation (s.d.), n = 3 biological replicates. ***P < 0.0003, **P < 0.002, NS, not significant (P > 0.05), two-sided t-test for independent samples. b, MS2-binding sites (see Extended Data Fig. 5) were added to GFP-TM-SU and co-transfected with MS2-mC (left), MS2-mC-HuR (middle) or MS2-mC-SET (right). mC, mCherry. Fluorescence confocal microscopy of HEK293 cells after transfection of indicated constructs shows that recruitment of HuR or SET to the short 3′ UTR redirects localization of GFP protein from the endoplasmic reticulum to the cell surface. Representative images from hundreds of cells. Scale bars, 10 µm. c, Co-immunoprecipitation of endogenous SET using anti-GFP antibody after transfection of CD47-SU or CD47-LU in HEK293 cells (for constructs, see Fig. 4a). Two per cent of input was loaded. WB, western blot. d, FACS analysis of surface GFP expression after transfection of GFP-TM-LU (dark blue line), GFP-TM-LU with a C-terminal deletion (ΔC; light blue line; left) or with destruction of both SET-binding sites (ΔC and K163A, K166A, K175A; ΔCL; light blue line, right). Shown as in Fig. 1g. Representative image from more than four experiments. e, Co-immunoprecipitation of endogenous SET using anti-GFP antibody after transfection of the indicated constructs. Two per cent of input was loaded. Asterisk indicates unspecific band.

PowerPoint slide

Full size image

Furthermore, by co-immunoprecipitation we demonstrated that endogenous SET only interacts with CD47-LU protein, but not with CD47-SU protein (Fig. 3c). Since SET binds to lysine residues18, we mutated single lysines in the C terminus of CD47, which decreased GFP-TM surface localization by up to 37% (Extended Data Fig. 6a). Mutation of 2/5 lysines decreased it by more than 50% and deletion of the entire C terminus decreased it by 80% (GFP-TM-LUΔC; Fig. 3d and Extended Data Fig. 6b). Additional mutation of three lysines in the first cytoplasmic loop abolished surface GFP expression (GFP-TM-LUΔCL; Fig. 3d and Extended Data Fig. 6b). The reduction in surface expression is probably due to partial or complete loss of SET binding to CD47, as demonstrated by co-immunoprecipitation (Fig. 3e).

To test if the difference in surface localization has phenotypic consequences, we added the ECD of CD47 to the GFP constructs (called CD47-LU or CD47-SU; Fig. 4a). Both constructs resulted in comparable overall CD47 protein levels (Extended Data Fig. 7a). CD47-LU efficiently localized to the cell surface via UDPL mediated by active RAC1 (Fig. 4b, c and Extended Data Fig. 7b). Whereas GFP expressed from the GFP-TM-SU construct nearly completely localized to the endoplasmic reticulum (Fig. 1f–h), CD47-SU primarily localizes to the endoplasmic reticulum, but also localizes partially to the cell surface, but independently of active RAC1 (Fig. 4b, c).

Figure 4: CD47 protein has different functions depending on whether it was generated by the SU or LU isoform.
figure 4

a, To generate GFP–CD47, GFP was inserted in frame between the signal peptide and the rest of the CD47 open reading frame. GFP–CD47 was fused with either the long or short CD47 3′ UTR, called CD47-LU and CD47-SU, respectively. b, FACS analysis of surface (surf.; light blue) and total (dark blue) GFP–CD47 expression in transfected JinB8 cells. Shown as in Fig. 1g. Representative images from four experiments. c, FACS analysis of GFP expression after transfection of CD47-LU or CD47-SU with or without co-transfection of dominant-negative RAC1 (N17RAC1). Shown as in Fig. 1g. Representative images from n = 7 (LU) and n = 2 (SU) experiments. d, Fraction of Mitomycin-C-treated cells that survived at day (d)3 after co-culture with macrophages is displayed for Jurkat, JinB8 (CD47−/− ) and the GFP+ JinB8 cells after nucleofection of CD47-LU or CD47-SU. Shown is mean ± s.d., n = 3 biological replicates. **P < 0.005, *P < 0.02, NS, not significant (P > 0.05), two-sided t-test for independent samples. e, The fraction of surviving cells (TO-PRO3 negative) measured by FACS analysis at day 3 after γ-irradiation is shown for the same populations as in d. Shown is mean ± s.d., n = 3 biological replicates of the 20% of cells with the highest GFP expression. Gy, Gray. f, Fluorescence confocal microscopy of permeabilized U251 cells after transfection of CD47-LU or CD47-SU co-stained with anti-RAC1 antibody. Yellow indicates co-localization. Representative images from hundreds of cells. Scale bars, 10 µm. g, Immunoprecipitation of endogenous RAC1–GTP (active RAC1) in HEK293 cells after transfection of CD47-LU, CD47-SU, or empty vector. Total RAC1 and GFP–CD47 were measured from input. n = 3 biological replicates.

PowerPoint slide

Full size image

We added the respective ECDs to CD44 and BAFFR, which also increased surface expression of their SU isoforms compared with their GFP-TM isoforms, but to a lesser extent than was observed for CD47 (Extended Data Fig. 7c, d). Di- or multimerization of cell surface receptor subunits, which often occurs through their ECDs, is a common strategy for overcoming endoplasmic reticulum retention, because it results in masking of endoplasmic reticulum retention signals19. We speculate that CD47-SU, CD44-SU and BAFFR-SU might use such a mechanism (although the multimerization partners are unknown) for their partial surface expression. In the case of BAFFR the ECD only increased surface expression by 1.2-fold, indicating that BAFFR strongly depends on UDPL for surface expression (Extended Data Fig. 7d). This is supported by the absence of BAFFR on B cells in Rac-deficient mice20. Taken together, our data suggest that membrane proteins rely on UDPL for surface expression to varying degrees.

Cells with high CD47 surface levels are protected from phagocytosis by macrophages (Extended Data Fig. 7e)12. To examine if the difference in surface expression of CD47-LU and CD47-SU protects cells to a different extent from phagocytosis, we used CD47-deficient Jurkat cells (called JinB8 cells21) and expressed similar total amounts of CD47-LU or CD47-SU (Extended Data Fig. 7a). Co-culture of these cells with macrophages demonstrated that CD47-LU fully protected the cells, whereas CD47-SU only partially protected the cells from phagocytosis (Fig. 4d).

CD47 also functions in the regulation of apoptosis22, as JinB8 cells or tissues from Cd47-knockout mice23 fail to undergo apoptosis after γ-irradiation24,25. Interestingly, expression of CD47-SU in JinB8 cells restored apoptosis, but expression of CD47-LU did not affect the loss-of-apoptosis phenotype (Fig. 4e and Extended Data Fig. 7f, g). Thus, when a cell requires increased CD47 surface expression, transcriptional upregulation alone would be non-optimal as it would confer increased susceptibility to apoptosis. ApA-generated 3′ UTR isoforms allow independent regulation of differentially localized and functionally distinct CD47 protein.

As the surface localization of CD47-SU is RAC1-independent, it also does not co-localize with RAC1 at the plasma membrane (Fig. 4f). In contrast, CD47-LU shows strong co-localization with RAC1 at lamellipodia (Fig. 4f). Both activated RAC1 and CD47 are necessary for efficient cell migration and activated RAC1 localizes to the leading edge of migrating cells5,23,26. We show that only the expression of CD47-LU resulted in changes in cell morphology with the generation of lamellipodia at the leading edge of cells. Furthermore, CD47-LU, but not CD47-SU, resulted in increased active RAC1 (Fig. 4g), which suggests that CD47-LU may cooperate with RAC1 during cell migration. Thus, CD47 protein localized to the same cellular compartment, but produced by the SU or LU mRNA isoforms, can exert different functions. It is currently not known if other surface proteins derived from their LU or SU isoforms also have distinct biological roles.

We propose that UDPL is a widespread mechanism for surface expression of membrane proteins. All currently known components of the pathway (HuR, SET and RAC1) are ubiquitously and highly expressed (Extended Data Fig. 8a)3. UDPL requires the presence of HuR-binding sites in the 3′ UTR and SET-binding sites in the cytoplasmic domains of membrane proteins (Extended Data Figs 3, 4 and 9). All candidates tested so far that met both requirements used UDPL for surface expression (Figs 1f–h, 2d, 4b and Extended Data Figs 2g, 7c). HuR-binding sites are highly abundant as HuR binds to thousands of mRNAs6,7,8,9, with a third of them being membrane proteins (Extended Data Fig. 8b). Although the SET-binding motif is currently unknown, we (Fig. 3d, e and Extended Data Fig. 6) and others have shown that SET binds to positively charged amino acids in histone tails or cytoplasmic domains of membrane proteins18,27. According to the positive-inside rule for integral membrane proteins, the cytoplasmic domains of membrane proteins are enriched in positively charged amino acids for topological reasons28. Therefore, potential SET-binding sites in cytoplasmic domains of membrane proteins are very widespread.

So far, efforts to determine the consequences of alternative 3′ UTRs have largely focused on mRNA stability and translation1,2,29. We expand the known functions of 3′ UTRs and show that they can act as scaffolds for RBPs that serve as adaptors to recruit effector proteins to the site of translation, which determines subcellular protein localization and function. With respect to CD47, CD44, ITGA1, TSPAN13 and BAFFR, the adaptor protein is HuR and the effector protein is SET. We further speculate that the scaffold function of 3′ UTRs may extend beyond the regulation of membrane proteins. RBPs could recruit other effector proteins, for example, enzymes that post-translationally modify proteins, as was shown for long non-coding RNAs30. We showed here that CD47 produced by alternative 3′ UTR isoforms localizes to different cellular compartments and has independent and sometimes opposite functions with respect to cell survival and cell migration. Thus, through the generation of alternative 3′ UTR isoforms, ApA contributes to functional diversity of the proteome without changing the amino acid sequence.

Methods

Cell lines

MCF7 (breast cancer), HeLa (cervical cancer), HEK293 (embryonic kidney), Caov-3 (ovarian carcinoma), NTERA2 (embryonic carcinoma) and THP-1 cells (monocytic leukaemia) were purchased from ATCC. B-LCL are Epstein Barr virus (EBV)-immortalized human B cells described earlier3. U2OS cells (sarcoma) were a gift from T. Brummelkamp, Toledo (B-cell lymphoma) cells were a gift from M. Mueschen, U251 (glioblastoma) cells were a gift from I. Mellinghoff, SHSY-5Y (neuroblastoma) cells were a gift from T. Tuschl and Jurkat (T-cell leukaemia) and JinB8 (CD47-negative Jurkat) cells were a gift from W. Frazier.

Constructs

For some of the shRNA knockdown experiments, pSUPERretropuro was modified by cloning IRES::GFP (derived from pMSCVpig)2 downstream of puromycin to obtain pSUPERretropuro containing enhanced (e)GFP (shRNA-GFP). The following DNA oligonucleotides served as shRNA precursors and were cloned into pSUPER-GFP or pSUPER. CD47-LU2F: 5′-GATCTCCCAGCTGTGTTACCGTTAAATTCAAGAGATTTAACGGTAACACAGCTGTTTTTC-3′; CD47-LU3F: 5′-GATCCCCCAGCTGTGTTACCGTTAAATTCAAGAGATTTAACGGTAACACAGCTGTTTTTC-3′; HuR2F: 5′-GATCTCCGATCAGACTACAGGTTTGTTTCAAGAGAACAAACCTGTAGTCTGATCTTTTTC-3′; HuR3F: 5′- GATCCCCGAGGCAATTACCAGTTTCATTCAAGAGATGAAACTGGTAATTGCCTCTTTTTC-3′; HuR4F: 5′-GATCCCCTCTTAAGTTTCGTAAGTTATTCAAGAGATAACTTACGAAACTTAAGATTTTTC-3′; SET1F: 5′-GATCCCCTGAAATAGACAGACTTAATTTCAAGAGAATTAAGTCTGTCTATTTCATTTTTC-3′; SET2F: 5′- GATCTCCCTGGTTTACTGACCATTCTTTCAAGAGAAGAATGGTCAGTAAACCAGTTTTTC-3′; RAC2F: 5′-GATCCCCCTTGCCTACTGATCAGTTATTCAAGAGATAACTGATCAGTAGGCAAGTTTTTC-3′; RAC3F: 5′-GATCCCCGTCCCTTGGAACCTTTGTATTCAAGAGATACAAAGGTTCCAAGGGACTTTTTC-3′; ControlF: 5′-GATCTCCTTCTCCGAACGTGTCACGTTTCAAGAGAACGTGACACGTTCGGAGAATTTTTC-3′.

The sequence of shRNA1 SET and of shRNA Control were published earlier31,32. The GFP fusion constructs were generated in pcDNA3.1 expression vector (Life Technologies). The short and long 3′ UTRs of CD47 were PCR-amplified from genomic DNA using Q5 High Fidelity DNA polymerase (NEB) and the primers listed later and inserted between the NotI and XbaI sites. To obtain expression of only the long 3′ UTR isoform of CD47, the proximal polyadenylation site was mutated from AAUAAA to ACUCAA using the QuikChange Multi Site Directed Mutagenesis Kit (Agilent). The resulting plasmids were used to test qPCR primers for accuracy in measuring short to long 3′ UTR isoform ratios (see later). The short 3′ UTR of CD47 used in the MS2-BS construct was cloned from the plasmid containing the long 3′ UTR with the mutated proximal polyadenylation site. eGFP was PCR-amplified from pMSCV-pig and inserted upstream of each CD47 3′ UTR (BamHI, NotI). The signal peptide of CD47 was generated by annealing two DNA oligonucleotides that were inserted into the BamHI site. To generate the GFP-TM constructs, the sequence of the TMDs and C-terminal tail of CD47 (the longest isoform, isoform 4; ref. 33) were cloned from Toledo cDNA and inserted downstream of eGFP (BsrgI, NotI). Isoform 4 was chosen as it is the most abundant isoform in Jurkat cells (data not shown). The two nucleotides of eGFP that occur after the BsrgI site were included in the forward primer. To generate the GFP-CD47 constructs, the ECD, TMDs and C terminus of CD47 were PCR-amplified from Toledo cDNA using the TM reverse primer and the CDS forward primer and inserted downstream of eGFP (BsrgI, NotI). The sequence of HuR-BS and HuR-BSΔ are shown in Extended Data Fig. 3 and replaced the long 3′ UTR in GFP-TM-LU. The GFP constructs containing the TMDs and C termini fused to either the long or short 3′ UTRs of CD44, ITGA1 and TNFRSF13C were generated as follows. TMDs, C termini and short 3′ UTRs of CD44, ITGA1 and TNFRSF13C were cloned from Toledo, SHSY-5Y and B-LCL cDNA respectively and inserted downstream of eGFP (BsrgI, XbaI). Short 3′ UTRs consisted of the first 122 nucleotides of CD44, the first 45 nucleotides of ITGA1 and the first 337 nucleotides of TNFRSF13C. Long 3′ UTRs: 3,068 nucleotides after the stop codon of CD44 and 3,996 nucleotides after the stop codon of ITGA1 were used. For TNFRSF13C, 1,221 nucleotides after the stop codon together with the last 600 nucleotides of the 3′ UTR (nucleotides 2,712–3,311 after the stop codon containing the majority of HuR-binding sites) were used as long 3′ UTR and were cloned from genomic DNA. To generate BAFFR-SU and -LU, the complete open reading frame of TNFRSF13C (without the start codon) was amplified from human B-cell cDNA and cloned using Gibson Assembly Cloning (NEB) into pcDNA3.1 vector used above downstream of eGFP. To generate CD44-SU and -LU, the open reading frame of CD44 (without the start codon) was amplified from cDNA of the human breast cancer cell line MDA-MB231 and cloned using Gibson Assembly Cloning (NEB) into pcDNA3.1 vector used above downstream of eGFP. To generate the GFP-TM-LUΔC construct the sequence of just the TMDs of CD47 was cloned from Toledo cDNA using the TM forward primer and TMΔC reverse primer and inserted downstream of eGFP (BsrgI, NotI). The 24 MS2-binding sites were cloned from a plasmid obtained from J. Gerst34 using the primers listed later (XbaI, ApaI). The constructs in which K163, K166, K175, K290, K297 and K304 were mutated to alanines were generated using the QuikChange Multi Site Directed Mutagenesis Kit (Agilent).

CD47UTRF: 5′-ATGCGCGGCCGCAGTGAAGTGATGGACTCCGATT-3′; CD47shortUTRR: 5′-ATGCTCTAGATGGGCAAACAACATAGATCA-3′; CD47longUTRR: 5′-ATGCTCTAGAAACACATTGGACTGATTTAAAACTT-3′; GFPF: 5′-ATGCGGATCCATGGTGAGCAAGGGCGA-3′; GFPR: 5′-ATGCGCGGCCGCTTACTTGTACAGCTCGTCCATG-3′; SPCD47F: 5′-GATCCATGTGGCCCCTGGTAGCGGCGCTGTTGCTGGGCTCGGCGTGCTGCGGATCAGCTG-3′; SPCD47R: 5′-GATCCAGCTGATCCGCAGCACGCCGAGCCCAGCAACAGCGCCGCTACCAGGGGCCACATG-3′; CD47TMF: 5′-ATGCTGTACAAGATTCTTATTGTTATTTTCCCAATT-3′; CD47TMR: 5′-ATGCGCGGCCGCTTATTCATCATTCATCATTCCTT-3′; CD47CDSF: 5′-ATGCTGTACAAGCAGCTACTATTTAATAAAACAA-3′; TMΔCR: 5′-ATGCGCGGCCGCTTATTTCATATAAACTAGTCCAAGTAA-3′; MS2-BSF: 5′-ATGCTCTAGAGGGCCCTATATATCGATCCTAAG-3′; MS2-BSR: 5′-ATGCGGGCCCTTTATTATGCTTGGTACCGAGCTCG-3′.

To create the MS2 fusion constructs35, the pUG34-MS2-GFP-SBP plasmid was obtained from J. Gerst34. SBP was replaced by either HuR, SET or a stop codon. HuR and SET were PCR-amplified from U2OS cDNA using the primers listed below (BsrgI, XbaI). The stop codon was generated by annealing two DNA oligonucleotides (BsrgI, XbaI). mCherry was PCR-amplified and replaced GFP (BamHI, BsrgI). After cloning was complete, all plasmids were sequenced to assure fidelity of the sequences.

HuRF: 5′-ATGCCGTACGAGTCTAATGGTTATGAAGACCACA-3′; HuRR: 5′-ATGCTCTAGATTATTTGTGGGACTTGTTGGT-3′; SETF: 5′-ATGCTGTACAAGTCGGCGCCGGCGGCCAAA-3′; SETR: 5′-ATGCTCTAGATTAGTCATCTTCTCCTTCATCCTC-3′; StopF: 5′-GTACAAGTAATAATAAT-3′; StopR: 5′-CTAGATTATTATTACTT-3′; mCherryF: 5′-ATGCGGATCCGTGAGCAAGGGCGAGGAG-3′; mCherryR: 5′-TTACTTGTACAGCTCGTCCATGC-3′.

The N17RAC1 construct was provided by A. Hall36.

Transfections

For transfections into U2OS, HEK293, HeLa and U251 cells Lipofectamine 2000 (Invitrogen) and for SHSY-5Y cells Xtreme reagent (Roche) was used. JinB8 and Jurkat cells were transfected using the Neon Transfection System (Invitrogen) according to the manufacturer’s protocol for transfecting Jurkat cells. To account for differences in the sizes of transfected plasmids the same molar amounts were transfected. When RNA was to be extracted, the cells were grown in the presence of high amounts of puromycin (4 mg ml−1) for 3 days and FACS analysis was performed to ensure that >90% of the cells were GFP+.

Generation of cell lines with stable expression of shRNAs

Stable cell lines were generated as described previously2.

FACS analysis

For surface FACS (in order to detect surface protein expression), cells were incubated with mouse anti-CD47-PerCy5.5 (BD Biosciences, 561261), mouse anti-CD44-PE (BD Biosciences, 561858), chicken anti-GFP (Abcam, ab13970), mouse anti-ITGA1-PE (BD Biosciences, 555749), mouse anti-BAFFR-PE (BD Biosciences, 554680) or rabbit anti-TSPAN13 (Genetex, GTX52155) in FACS buffer A (0.5% FBS in PBS) for 30 min at 4 °C, and then washed twice in FACS Buffer A. For detection of GFP and TSPAN13, cells were then incubated with goat anti-chicken Alexa Fluor 568 or 633 (Invitrogen, A11041 or A21103) and goat anti-rabbit Alexa Fluor 680 (Invitrogen, A-21076), respectively, for 30 min at 4 °C, and then washed twice in FACS Buffer A. At least 30,000 cells were analysed on a BD FACSCalibur cell analyser (BD Biosciences) and FACS data were computed using the FlowJo VX software.

For intracellular FACS (in order to detect total protein expression), cells were fixed for 15 min at room temperature in fixation buffer (4% PFA, 0.02% sodium azide, and 0.1% Tween 20 in PBS), washed in FACS buffer B (0.02% sodium azide, 0.1% Tween 20 in PBS), permeabilized for 10 min at 4 °C in permeabilization buffer (0.02% sodium azide, 0.1% Tween 20 and 10% dimethyl sulfoxide in PBS), washed, re-fixed for 5 min at room temperature in fixation buffer, and washed again. Cells were incubated with the same primary and secondary antibodies as for surface FACS in FACS buffer B for 30 min at 4 °C, and then washed twice in FACS Buffer B. For live/dead analysis by FACS, cells were incubated with TO-PRO3 in FACS buffer A (0.5% FBS in PBS) for 10 min at 4 °C, and then washed twice in FACS Buffer A. Cells were analysed is the same manner as for surface FACS.

For all the GFP-expressing plasmids the 20% of cells with the highest GFP expression are shown.

Immunocytochemistry

For surface staining of CD47, cells were fixed for 15 min at room temperature in fixation buffer A (4% PFA and 0.02% sodium azide in PBS), washed with PBS, blocked for 15 min at 4 °C in 5% Normal Goat Serum (Invitrogen, PCN5000) in PBS and then incubated with mouse anti-human CD47 (Santa Cruz, sc-59079) primary antibody for 1 h at 4 °C in PBS. After washing in PBS, donkey anti-mouse Alexa Fluor 594 (Invitrogen, A-21203) secondary antibody was incubated for 1 h at 4 °C in PBS, and during the last 10 min of incubation 4′,6-diamidino-2-phenylindole (DAPI; Invitrogen, D1306) was added, followed by three washes of PBS. Mowiol mounting media (Sloan Kettering Institute, Molecular Cytology Core Facility) was used to mount the slides. Imaging was performed at the Sloan Kettering Institute Molecular Cytology Core Facility, on a Leica TCS SP5 confocal microscope, using a ×63, 1.4 numerical aperture oil objective.

For intracellular staining of CD47, cells were fixed for 15 min at room temperature in fixation buffer B (4% PFA, 0.02% sodium azide, and 0.1% Tween 20 in PBS), washed in wash buffer (0.02% sodium azide, 0.1% Tween 20 in PBS), permeabilized for 10 min at 4 °C in permeabilization buffer (0.02% sodium azide, 0.1% Tween 20 and 10% dimethyl sulfoxide in PBS), washed, re-fixed for 5 min at room temperature in fixation buffer B, washed and blocked for 15 min at 4 °C in 5% Normal Goat Serum in wash buffer. Mouse anti-CD47 (Santa Cruz, sc-59079) primary antibody was incubated overnight at 4 °C in wash buffer. The strong permeabilization and overnight staining were necessary to visualize intracellular CD47, as the CD47 antibody recognizes an epitope in the ECD of CD47, which is located in the lumen of the endoplasmic reticulum. Owing to the extended treatment with a buffer containing Tween 20 the plasma membrane could no longer be visualized in these cells.

For co-staining of GFP with calnexin or RAC1, the cells were fixed for 15 min at room temperature in fixation buffer B (4% PFA, 0.02% sodium azide, and 0.1% Tween 20 in PBS), washed with wash buffer B, blocked for 15 min at 4 °C in 5% Normal Goat Serum (Invitrogen, PCN5000) in wash buffer and then incubated with rabbit anti-calnexin (Santa Cruz, sc-11397) or mouse anti-RAC1 (Abcam, ab12048) primary antibodies for 1 h at 4 °C in wash buffer. After washing, goat anti-rabbit Alexa Fluor 680 (Invitrogen, A-21076) or donkey anti-mouse Alexa Fluor 594 (Invitrogen, A-21203) secondary antibodies were incubated for 1 h at 4 °C in wash buffer, and during the last 10 min of incubation DAPI (Invitrogen, D1306) was added, followed by three washes. Mounting and imaging was performed as for the surface staining. Owing to the lack of permeabilization and short period in Tween 20, the plasma membrane was still visible. When calnexin was co-stained with endogenous CD47, the protocol for intracellular staining of CD47 was used and the plasma membrane was again not visible.

Alexa Fluor 594 and 680 were pseudo-coloured red and endogenous GFP and mCherry were imaged as they appear, without any antibody.

RNA-FISH

Custom Stellaris FISH Probes (Biosearch Technologies) were designed for the open reading frame of eGFP using the Stellaris Probe designer website, and with the assistance of Biosearch Technologies staff. The probes were conjugated to the Quasar 670 fluorochrome. Staining was carried out according to the manufacturer’s protocols. Briefly, 24 h after transfection of the GFP-TM constructs cells were trypsinized and plated on Millicell EZ glass slides (Millipore) and allowed to grow overnight. Cells were washed in PBS, fixed in 4% PFA at room temperature for 10 min and permeabilized in 70% ethanol at 4 °C for 2 h. After washing, the probes were hybridized at 37 °C for 4 h, washed, incubated with DAPI at 37 °C for 30 min, washed and mounted in Mowiol mounting media. Imaging was performed as for the surface immunostaining. RNA was pseudo-coloured red, and GFP was imaged as it appears, without any antibody.

3′-seq

3′-seq reads of naive B cells, B-LCL and HEK293 cells were analysed and visualized as described previously3.

Northern blot analysis

Northern blots were performed as previously described2.

CD47 probe F: 5′-TTGATGGAGCTCTAAACAAGTCC-3′; CD47 probe R: 5′-GAATAACCAATATGGCAATGACG-3′; GFP probe F: 5′-TAAACGGCCACAAGTTCAGC-3′; GFP probe R: 5′-CTTGTACAGCTCGTCCATGC-3′.

Quantitative PCR

RNA was extracted using TRI Reagent (Ambion) according to the manufacturer’s protocol. cDNA was synthesized using random hexamers and the TaqMan Reverse Transcription Kit (Applied Biosystems). qRT–PCR was performed using the Power SYBR Green master mix (Applied Biosystems) on a 7500 HT Fast Real-Time PCR System (Applied Biosystems). Each reaction was performed in triplicate. The experiments were performed at least three times to obtain at least three biological replicates. The following primers were used to quantify total CD47 mRNA, the long 3′ UTR isoform of CD47 and GAPDH for normalization.

CD47TotalF: 5′-AGTGATGGACTCCGATTTGG-3′; CD47TotalR: 5′-GGGTCTCATAGGTGACAACCA-3′; CD47LongF: 5′-AAGAGAACTCCAGTGTTGCT-3′; CD47LongR: 5′-ACGGTAACACAGCTGTAAAACA-3′; GAPDHF: 5′-ACAACTTTGGTATCGTGGAAGG-3′; GAPDHR: 5′-TATTTGGCAGGTTTTTCTAGACG-3′.

To measure 3′ UTR isoform expression by qRT–PCR and to take into account different affinities of primers, we generated a standard using plasmids that contained either the short or the long 3′ UTR of CD47 (see earlier). We mixed together known quantities of the two plasmids ranging from 3:1 to 1:3 of short to long 3′ UTR and performed qPCR on these mixtures to test the accuracy of our primer sets. The fraction of the long 3′ UTR isoform was calculated by subtracting the CT value of the long isoform from the CT value for total CD47 expression. The fraction of the long 3′ UTR isoform corresponds to 2CT difference. The fraction of the short 3′ UTR isoform was obtained by subtracting the fraction of the long 3′ UTR isoform from the total CD47 mRNA.

Immunoprecipitation of RNA complexes and RT–PCR

GFP-TM-LU, GFP-TM-SU, GFP-TM-HuR-BS and GFP-TM-HuR-BSΔ were transfected into HEK293 cells and immunoprecipitation of protein–RNA complexes was carried out as previously described37. RNA-immunoprecipitations were performed with crosslinking to prevent re-association of HuR with mRNA after lysis38. Briefly, cells were harvested and washed twice in cold PBS. Formaldehyde was added to a final concentration of 1% (v/v) and the cells were incubated at room temperature for 10 min. The reaction was quenched by addition of glycine to a final concentration of 0.25 M and incubated at room temperature for 5 min. After centrifugation the cell pellet was washed twice in cold PBS and resuspended in RIPA buffer (25 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% NP-40, 1% Na-deoxycholate, 0.1% SDS, 1 mM EDTA, protease inhibitor cocktail (Roche, 04693124001)). The mRNPs were solubilized by three rounds of sonication for 15 s each in a Misonix Ultrasonic Processor S-4000 at an output of 8–9 W. Insoluble material was removed by centrifugation. Lysates were pre-cleared by addition of magnetic protein A beads (Millipore, LSKMAGA10) and incubation for 30 min at 4 °C with constant mixing. The pre-cleared lysate was divided into three parts. One portion was retained for the input control and anti-HuR (Millipore, 07-1735) or IgG (Santa Cruz, sc-2025) were added to the two other portions and incubated at 4 °C for 2 h. Magnetic protein A beads (Millipore, LSKMAGA10) were added and incubated at 4 °C for 1 h. The beads were washed five times with high-stringency RIPA buffer (50 mM Tris-HCl (pH 7.4), 1 M NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS and 1 mM EDTA). The crosslinking was reversed by resuspending the beads in 50 mM Tris-Cl (pH 7.0), 5 mM EDTA, 10 mM dithiothreitol (DTT) and 1% SDS, followed by incubation at 70 °C for 45 min. RNA was extracted from the beads and buffer using TRI Reagent (Ambion) according to the manufacturer’s protocol, and cDNA was synthesized as previously described. qRT–PCR was carried out using primers for GAPDH (see earlier) and GFP, so as not to amplify endogenous CD47. The primers used were as follows: GFPF: 5′-TAAACGGCCACAAGTCAGC-3′; GFPR: 5′-AAGTCGTGCTGCTTCATGTG-3′.

HEK293 cells transiently transfected with sh2 HuR or sh Co were used to assess the presence of SET on endogenous CD47 mRNA and the requirement of HuR for this association. Cells were transfected and treated with high dose puromycin for 3 days. FACS analysis was used to ensure greater than 90% of surviving cells expressed the shRNA. RNA-immunoprecipitation was carried out as described earlier, but using anti-SET (Abcam, ab181990) or IgG (Santa Cruz, sc-2025). The following primers were used for qRT–PCR of CD47-LU: CD47F: 5′-AAGAGAACTCCAGTGTTGCT-3′; CD47R: 5′-ACGGTAACACAGCTGTAAAACA-3′.

The obtained CT values were first normalized to GAPDH and then the fraction of input mRNA was plotted.

Western blotting

HEK293 cells stably expressing sh3 HuR, sh4 HuR, sh1 SET, sh2 SET, sh2 Rac1, sh3 Rac1 or sh Co, as well as HEK293 cells transiently transfected with sh2 HuR, CD47-SU, CD47-LU, CD47-LU+N17Rac1, CD47-LU2Km, CD47-LUΔC or CD47-LUΔCL, were lysed in Laemmli buffer (Sigma, S3401), boiled for 7 min and then cooled on ice. Lysates were run on NuPAGE Novex 4–12% Bis-Tris Gel (Invitrogen, NP0322BOX) and transferred to PVDF membrane (Bio-rad, 162-0177). After blocking for 1 h at room temperature in Odyssey Blocking Buffer (Li-Cor, 927-40000) the following primary antibodies were used: rabbit anti-HuR (Millipore, 07-1735), rabbit anti-SET (Abcam, ab181990), mouse anti-RAC1 (Cell Signaling Technology, 8631S), mouse anti-ACTB (Sigma, A4700), rabbit anti-ACTB (Sigma, A2066), chicken anti-GFP (Abcam, ab13970), mouse anti-CD47 (Santa Cruz, sc-59079), mouse anti-CD44 (BD Bioscience, 561858), and rabbit anti-TSPAN13 (GeneTex, GTX52155). The antibodies were diluted in Odyssey Blocking Buffer containing 0.1% Tween 20 and the blots were incubated overnight at 4 °C. After four washes in PBST (PBS plus 0.1% Tween 20) the blots were incubated for 1 h at room temperature in Odyssey Blocking Buffer containing 0.1% Tween 20 and 0.01% SDS and the following secondary antibodies: donkey anti-mouse IRDye 700 (Rockland Immunochemicals, 610-730-002), donkey anti-rabbit IRDye 680 (Li-Cor Biosciences, 926-68073), donkey anti-rabbit IRDye 800 (Li-Cor Biosciences, 926-32213), donkey anti-mouse IRDye 800 (Li-Cor Biosciences, 926-32212) and rabbit anti-chicken IRDye 800 (Rockland Immunochemicals, 603-432-002). The blots were washed four times in PBST and then thoroughly soaked in PBS before imaging. Imaging was performed on an Odyssey CLx imaging system (Li-Cor). Quantification of western blots was performed using Image J.

Co-immunoprecipitation

CD47-SU, CD47-LU, CD47-LUΔC and CD47-LUΔCL were transfected into HEK293 cells and the cells were lysed in ice cold RIPA buffer ((25 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% NP-40, 1% Na-deoxycholate, 0.1% SDS, 1 mM EDTA, protease inhibitor cocktail (Roche, 04693124001)) for 5 min on ice. After the cells were spun down at 20,000g for 20 min the supernatant was pre-cleared as described earlier. The lysate was divided in two equal parts (a small portion was removed to be used as the input control). Anti-GFP (Invitrogen, A-6455) or IgG (Santa Cruz, sc-2025) was added to the lysates and a 1 h rocking incubation at 4 °C was performed, followed by addition of magnetic protein A beads (Millipore, LSKMAGA10) and another 1 h rocking incubation at 4 °C. After seven washes in RIPA buffer the beads were then boiled in Laemmli buffer (Sigma, S3401) for 7 min and then cooled on ice. Western blotting was carried out as described earlier. Chicken anti-GFP (Abcam, ab13970) antibody was used to confirm immunoprecipitation of GFP constructs and rabbit anti-SET (Abcam, ab181990) antibody was used to assess co-immunoprecipitation of SET protein.

Immunoprecipitation of RAC1–GTP

CD47-SU, CD47-LU and pcDNA 3.1 vector alone were transfected into HEK293 cells. The levels of RAC1–GTP were assessed using an Active Rac1 Detection Kit (Cell Signaling Technology, 8815) following the manufacturer’s protocol. Briefly, cells were lysed in ice cold lysis buffer and centrifuged at 16,000g for 15 min at 4 °C. The supernatant was added to GST-PAK1-PBD and glutathione resin and incubated with rocking for 1 h at 4 °C (a small portion was removed to be used as the input control). The resin was washed three times with lysis buffer and then SDS sample buffer was added to elute the bound proteins. Western blotting was carried out as described earlier. Mouse anti-RAC1 (Cell signaling Technology, 8631S) antibody was used to detect RAC1–GTP as well as to assess total RAC1 in the samples.

Irradiation and cell survival assay

Jurkat, JinB8, and transfected JinB8 cells (24 h post-transfection) were irradiated for a total of 0, 5, 10, 20 or 40 Gy using a Shepherd Mark-1 caesium irradiator. For JinB8 or Jurkat cells transfected with CD47-LU, CD47-SU or with sh2 CD47-LU the percentage of GFP+ cells was determined by FACS before irradiation. Three days after γ-irradiation, cells were stained with TO-PRO3 (Sloan Kettering Institute, Flow Cytometry Core Facility) in FACS buffer A (0.5% FBS in PBS) for 10 min at 4 °C and analysed using a BD FACSCalibur cell analyser to evaluate cell survival. Per cent survival was calculated as the TO-PRO3-negative cells divided by the total number of GFP+ cells. Shown are mean and standard deviation of three biological replicates.

Phagocytosis assay

Human macrophages were obtained by differentiation of THP-1 cells with 25 ng ml−1 phorbol 12-myristate 13-acetate (PMA; Sigma) for 3 days. On day 3, Jurkat, JinB8, or transfected JinB8 cells were treated with 10 µg ml−1 Mitomycin C for 2.5 h at 37 °C and washed three times in media. Mitocycin C treatment halts cell division and allows for a more accurate assessment of the percentage of cells that are phagocytosed. For JinB8 cells transfected with CD47-LU or CD47-SU the percentage of GFP+ cells was determined by FACS before co-culture. The cells were either cultured alone or co-cultured with fully differentiated macrophages. After 3 days, cells were counted and the fraction of GFP+ cells was determined by FACS analysis. The fraction of surviving cells after co-culture with macrophages was normalized by the number of surviving cells without co-culture and is show as mean and standard deviation of three independent experiments.

To demonstrate that the cells were phagocytosed, during the last 10 min of Mitomycin C treatment the cells were also labelled with carboxyfluorescein succinimidyl ester (CFSE; Invitrogen). Washing after Mitomycin C and CFSE treatment was carried out in cold media according to the manufacturer’s protocol. After co-culture, CFSE uptake by macrophages was measured by FACS analysis to demonstrate that a decrease in the number of surviving cells is due to phagocytosis by macrophages.

Fraction of membrane proteins among HuR target genes

The list of HuR target genes was obtained from previous publications7,9. The union of genes from both publications was analysed using gene ontology analysis39 and all genes with the tag “membrane” were considered membrane proteins. This number is consistent with the number of membrane proteins obtained in yeast40. Fisher’s exact test was used to test for significance.

Statistical analysis

To test for significant differences between samples a two-sided t-test for independent samples was performed using SPSS.