Translational reprogramming in tumour cells can generate oncoselectivity in viral therapies

Systemic treatment of cancer requires tumour-selective therapies that eliminate cancer cells yet preserve healthy tissues from undesired damage. Tumoral transformation is associated with profound effects in translational reprogramming of gene expression, such that tumour-specific translational regulation presents an attractive possibility for generating oncoselective therapies. We recently discovered that mRNA translational control by cytoplasmic polyadenylation element-binding proteins (CPEBs) is reactivated in cancer. Here we present a novel approach to restrict genetic-engineered therapies to malignant tissues based on CPEB translational regulation of target mRNAs. We demonstrate that tumour reprogramming of CPEB-mediated mRNA stability and translational regulation modulates tumour-specific expression of viral proteins. For oncolytic adenoviruses, insertion of CPE regulatory sequences in the 3′-untranslated region of the E1A gene provides oncoselectivity, with full potency in cancer cells but attenuated in normal tissues. Our results demonstrate the potential of this strategy to improve oncolytic virus design and provide a framework for exploiting CPE-regulated transgenes for therapy.

R egulation of transgene and viral protein expression is required to increase the safety and efficacy of gene and viral therapies. Delivery and expression of transgenes with anticancer activity, or the use of conditionally replicating viruses for cancer therapy, must be specific for tumours to avoid side effects to healthy tissues. Most efforts to achieve such selective control have been based on the use of tumour-specific promoters 1 and, more recently, by the engineering of target site-recognizing, tissue-specific microRNA (miRNA) [2][3][4][5][6] . Although both strategies contribute highly to tumour selectivity, it has become evident that the post-transcriptional regulation of specific mRNA subpopulations contributes substantially to the broad expression changes of genes responsible for the cancer phenotype 7 . Thus, the translational reprogramming of tumour cells has been proposed as a potential target for tumour-specific drugs 8 . These tumour-specific translational profiles could therefore be used to generate tumour specificity to transgene and viral protein expression.
One of the mechanisms to regulate the translation of specific subpopulations of mRNAs is through the presence of cis-acting elements in the 3 0 -untranslated region (3 0 -UTRs) of mRNAs, such as the cytoplasmic polyadenylation element (CPE; 5 0 -UUUUA 1-2 U-3 0 ). The CPE is bound by CPE-binding proteins (CPEBs), a family of four members conserved in their RNA-recognition domain but distinct in their regulatory motifs (reviewed in refs 9,10). All four CPEBs recognize RNA via tandem RNA recognition motifs located in their C-terminal halves that are followed by ZZ domains 11 . The RNA recognition motifs define two CPEB subfamilies (CPEB1 and CPEB2-4). Both CPEB subfamilies regulate overlapping populations of mRNAs and recognize similar cis-acting elements, albeit with different affinities [12][13][14][15] . The N-terminal domain is highly variable in both length and composition across various CPEB orthologues and paralogues 16 and contains all the identified regulatory motifs, including phosphorylations 17,18 , monoubiquitination 15 and sumoylation 19 . These post-translational modifications dictate whether CPEBs act as translational repressors or promote cytoplasmic polyadenylation and subsequent translational activation (reviewed in refs 9,10). In turn, the specific arrangement of CPEs, in number and in their relative position to other CPEs and the polyadenylation signal 20 , together with the presence of additional cis-acting elements such as AU-rich elements (AREs) 21 , quantitatively determine whether a particular mRNA is repressed or activated in a combinatorial code that responds to the activation of specific CPEBs 12 . Although CPEBs have been mainly characterized as regulators of maternal mRNAs during the transcriptionally silent germ cells, they also regulate cytoplasmic polyadenylation in somatic and tumoral cells 13,[22][23][24][25] . In fact, CPEBs are extremely conserved within vertebrates (96% for CPEB1 and 99% for CPEB4), the CPEs and the CPE-combinatorial code are conserved and functionally replaceable from frogs to mammals and promote translational repression and activation in rodents and humans 13,21,26,27 . CPEB levels and activities are differentially regulated in tumours 9,28 . CPEB1 levels are decreased in several human cancers 29,30 , and reduced levels of this protein have been associated with increased malignancy both in human 29 and mouse 23 . In contrast, high CPEB4 expression levels in pancreatic ductal adenocarcinoma and glioblastoma tumours correlated with stimulation of tumour growth, angiogenesis and decreased survival 14 . Furthermore, CPEB4 has been identified as one of the seven genes with an expression profile that is associated with colorectal cancer 31 .
Here we show the feasibility of using the tumoral reprogramming of CPE-mediated translational regulation to provide tumour selectivity to transgene expression. We have engineered a particular CPE arrangement that activates translation in tumour cells while promoting translational repression in non-transformed cells. We then generated a modified adenovirus in which the E1A protein expression is regulated by CPEBs, to obtain oncoselectivity and attenuated toxicity in non-transformed tissues. This novel targeting modality increases the therapeutic index of an oncolytic adenovirus and provides a new paradigm for its applicability to gene transfer-based therapeutic approaches.

Results
CPEs provide oncoselectivity to replicative adenoviruses. To identify the optimal sequences required for CPEB-dependent tumour selectivity, we tested chimeric mRNAs with three combinations of 3 0 -UTRs fused to the d2EGFP open reading frame. These mRNAs were expressed in a battery of normal (HPDE) and tumour (RWP-1, PANC-1 and MIA PaCa-2) pancreatic cells expressing variable levels of CPEB1 and CPEB4 (Fig. 1a). The first UTR was derived from Xenopus cyclin B1 (cB1) 3 0 -UTR mRNA and contained two consensus CPEs and one nonconsensus CPE. This CPE arrangement promotes both translational repression by unphosphorylated CPEB1 and translational activation by CPEB4 12,13,20 . The second UTR was synthetized by combining cB1 CPEs with an ARE sequence that opposes CPE-mediated polyadenylation and translational activation from the tumour-necrosis factor-a (TNF-a) 3 0 -UTR mRNA (TNF-a-cB1). The third UTR was generated from a fragment of the tissue plasminogen activator (tPA) 3 0 -UTR mRNA that contains two CPEs and two ARE sequences 14 . (Fig. 1b and Supplementary Table 1).
Lentiviruses expressing the d2EGFP with the indicated 3 0 -UTRs or a control wild-type (WT)-3 0 -UTR (without CPEs) and a lentivirus with a destabilized dRFP bearing a control 3 0 -UTR (WT) were used to coinfect the non-tumour HPDE cells and the pancreatic cancer cell lines ( Supplementary Fig. 1a). For each combination of UTR and cell line, the resulting d2EGFP-to-dRFP expression ratio indicated that cB1 was the only 3 0 -UTR promoting a differential translational control in normal cells compared to cancer cells. Thus, cB1-3 0 -UTR repressed the expression of d2EGFP in HPDE cells as compared to the dRFP control 3 0 -UTR, whereas it promoted activation in the cancer cells, with a stronger effect in PANC-1 (Fig. 1c). Interestingly, for the four cell lines, the d2EGFP/dRFP ratio was proportional to the CPEB4/CPEB1 ratio (Fig. 1a). As expected from the role that the poly(A) tail has on mRNA stability, reduced mRNA levels of d2EGFP-cB1-3 0 -UTR were detected in HPDE cells as compared to tumour cells ( Supplementary Fig. 1b). On the other hand, the two 3 0 -UTRs with ARE sequences led to reduced d2EGFP expression in all the cell lines, indicating that the destabilization effect of the ARE dominates over the effects of CPE regulation (Fig. 1c).
Next, we took advantage of the specific regulation promoted by the cB1-3 0 -UTR to generate an oncoselective replication-competent adenovirus Ad5. For this, we targeted viral replication by designing a virus in which the translation of E1A mRNA was regulated by CPEBs. The adenoviral immediate-early E1A gene was selected because it is the first gene transcribed after an adenoviral infection and thus acts as a master transcriptional regulator of further early viral genes and modifies several cell host functions required for viral DNA replication. We then replaced the WT-3 0 -UTR of the viral E1A coding sequence with the cB1-3 0 -UTR to give us AdCPE (Fig. 2a). Substitution of E1A WT-3 0 -UTR by cB1-3 0 -UTR had no effect on the transcription of this gene, as shown by the equal levels of pre-mRNA for both 3 0 -UTRs in normal and cancer cell lines (Fig. 2b). However, when the steady-state levels of mature transcripts were compared, we found E1A-cB1-3 0 -UTR mRNA to be significantly lower in HPDE cells as compared to E1A-WT-3 0 -UTR mRNA as well as reduced with respect to E1A-cB1-3 0 -UTR mRNA in tumour cells (Fig. 2b). This suggests a specific destabilization of the mRNA-containing CPEs in the non-tumour cells. Because the destabilization of the CPE-containing c-myc mRNA in non-transformed cells has been directly associated with its cytoplasmic deadenylation 25 , we measured the polyA tail length of the different E1A transcripts in the four cell lines by RNA ligation-coupled PCR with reverse transcription (RT-PCR) analysis (Fig. 2c). We found that the CPE-mediated destabilization of the E1A-cB1 mRNA in HPDE cells was associated with a shorter polyA tail (Fig. 2d), which resulted in reduced E1A protein expression from the cB1-3 0 -UTR mRNA as compared with the WT-3 0 -UTR (Fig. 2e). Accordingly, AdCPE-infected non-transformed cells (HPDE) showed a strong reduction in viral genome copy numbers (Fig. 2f) and decreased cytotoxicity (increased half-maximal inhibitory concentration (IC 50 ); Fig. 2g) as compared to the control virus Adwt. The lowest performance of AdCPE was also observed in human primary fibroblasts and the non-tumoral human kidney epithelial HK-2 cells (Fig. 2e,f and Supplementary  Fig. 2). However, in the tumour cells, cB1-3 0 -UTR was as efficient as the WT-3 0 -UTR in supporting optimum levels of viral-driven E1A expression (Fig. 2e), which in turn resulted in equal viral genome copy numbers and IC 50 from both 3 0 -UTRs ( Fig. 2f,g). CPEB-mediated control of E1A expression and viral release was also observed in AdCPE-infected colorectal and glioblastoma cells as well as in the patient-derived pancreatic tumour CP15 ( Supplementary Fig. 2). The similar effects of Adwt and AdCPE in cancer cells also highlight that the small size of the cB1-3 0 -UTR did not compromise virus fitness and packaging efficiency.
CPE-mediated oncoselectivity is provided by CPEB4. To determine whether the higher CPEB4 levels in tumour cells caused the specificity of cB1-3 0 -UTR regulation of mRNA translation, we first confirmed the specificity of CPEB4 binding to the CPEs in cB1-3 0 -UTR. As shown by RIP, CPEB4 bound cB1-3 0 -UTR WT in a CPE-dependent manner. In contrast, HuR, an ARE-binding protein that promotes mRNA stabilization, displayed almost negligible binding to cB1-3 0 -UTR WT that was not affected by the mutations that inactivate the CPE ( Supplementary Fig. 3). Then, we knocked down CPEB4 in tumoral RWP-1 cells (RWP-1-sh4) and compared it with a control non-target shRNA (RWP-1-shNT) (Fig. 3a). Depletion of CPEB4 (RWP-1-sh4) caused a significant reduction of E1A protein levels expressed from AdCPE (Fig. 3b), with the subsequent reduction of viral replication (Fig. 3c) and increased IC 50 values (Fig. 3d). On the other hand, Adwt was unaffected by a CPEB4 knockdown (Fig. 3b-d). Similar results were observed in HCT-116 colorectal and T98 glioblastoma cells ( Supplementary  Fig. 4). A reverse approach, of overexpressing CPEB4 in non-tumour cells (HPDE), partially rescued the expression of E1A from AdCPE (Fig. 3e) and viral replication (Fig. 3f), as compared to Adwt. Therefore, high levels of CPEB4 seem to be required for the oncoselective behaviour of the engineered virus and to prevent the CPE-mediated translational repression observed in non-tumour cells.
Oncoselectivity of AdCPE in vivo. To investigate whether AdCPE provides oncolytic specificity in vivo, nude mice carrying subcutaneous tumours were intravenously injected with Adwt or AdCPE viral particles, and tumour growth was monitored. In consonance with the results obtained in cell culture models, we found that AdCPE produced a significant inhibition of tumour growth that was similar or greater than that of Adwt ( Fig. 4a and Supplementary Fig. 5). Interestingly, at high viral doses at which Adwt compromises mice survival 32 , the antitumour effect of AdCPE was very strong (Supplementary  L iv e r P a n c r e a s K id n e y L iv e r P a n c r e a s K id n e y L iv e r P a n c r e a s K id n e y L iv e r P a n c r e a s K id n e y L iv e r P a n c r e a s K id n e y L iv e r P a n c r e a s K id n e y toxicity. To test the AdCPE effects on healthy tissues in vivo, we intravenously injected immunocompetent mice with Adwt and AdCPE, at the same viral dose used in the xenograft experiments. Three days later, we quantified viral proteins and viral genomes in the liver, pancreas and kidney. E1A protein levels were significantly reduced in the liver of AdCPE-treated mice, as compared to Adwt (Fig. 4b). Quantification of mRNA of E1A, hexon and fibre in the liver, pancreas and kidney revealed significantly reduced levels of the three mRNAs in AdCPEinjected mice as compared to Adwt (Fig. 4c). Consistent with in vitro data, E1A pre-mRNA levels in liver extracts from AdCPE and Adwt were similar, whereas the mature mRNA was reduced in AdCPE mice (Supplementary Fig. 7a). These results indicate that the CPE-dependent destabilization of viral mRNA is recapitulated in vivo. Despite the impairment of human Ad5 to productively replicate in mice, a low level of replication has been proposed to occur 33 , which is sufficient to detect differences between AdCPE and Adwt. Analysing viral genomes in the liver, pancreas and kidney showed, as expected, that most of the virus was retained in the liver, but that viral genomes were also present in the other organs. Interestingly, animals injected with AdCPE had a reduced viral content (of 18-fold in the liver, 7-fold in the pancreas and 6-fold in the kidney) as compared to Adwt (Fig. 4d). This effect was not associated with differences in viral titres, since the number of viral genomes that reached the liver at 4 h after injection, previous to any replication event, was similar between AdCPE-and Adwt-injected mice ( Supplementary  Fig. 7b). The attenuated activity of AdCPE in murine healthy tissues was further confirmed in human primary hepatocytes, a model fully permissive for adenoviral replication (Supplementary Fig. 8). Thus, AdCPE activity is highly impaired in normal tissues when compared to Adwt. Finally, to study liver damage-associated toxicity, which is one of the major side effects of Ad-expressing E1A, we analysed body weight, macroscopic liver appearance and serum parameters in mice after intravenous delivery of Adwt or AdCPE. Adwt caused a progressive loss of weight, whereas AdCPE led to a decrease in body weight by day 1, which then remained stable for the following days (Fig. 4e). Livers injected with Adwt showed a steatotic appearance that was not observed in AdCPE livers (Fig. 4f). Aspartate aminotransferase and alanine aminotransferase enzyme activities showed a remarkable increase in Adwt-injected animals with respect to the saline group, whereas a 4-fold lower induction was detected in AdCPE-injected mice (Fig. 4g). Altogether, these results indicate a reduced toxic profile of the AdCPE virus as compared to Adwt, while maintaining its oncolytic potential.

Discussion
Oncolytic viruses are advancing to clinical trials and are being envisioned as breakthrough agents in oncology in the near future. Accordingly, optimized engineered viruses are under development to maximize their anticancer effects 34 . However, as their potency increases, so do potentially associated toxicities, pointing to the need to develop highly tumour-specific viruses. We show that engineering adenoviruses with CPE regulatory elements, to control E1A expression post-transcriptionally, resulted in attenuated viral activity in normal cells while maintaining, or even increasing, potency in cancer cells. The specificity of the antitumour response is directly derived from the ectopic expression of CPEB4 in tumours 14 . Thus, depletion of CPEB4 attenuates viral activity in tumour cells, while its overexpression in non-transformed cells increases viral replication. In turn, the levels of CPE-mediated expression may be further increased by reduced levels of CPEB1. These results are consistent with a scenario in which CPEB1 in non-transformed cells (in which its levels are high) represses CPE-containing mRNAs 22 . In tumour cells, on the other hand, CPEB1 levels are reduced 9,28 , while CPEB4 levels are increased 14 , which promotes cytoplasmic polyadenylation and thereby increases mRNA stability and translation of CPE-regulated transcripts. Importantly, viral genome replication is dependent on CPEB4 activity and can only occur in tumour cells. This leads to the production of fully competent viruses in tumour cells but attenuated ones in healthy tissues. Accordingly, the CPE-regulated virus maintains its oncolytic capacity in tumour cells while significantly reducing its damage to non-tumour tissues, which is mainly in the liver as this is the target organ of adenovirus sequestration on intravascular delivery. This increased selectivity allows for an enhanced therapeutic index, since increased antitumour capacity was obtained after AdCPE treatment at viral doses that, with Adwt injection, compromised mice survival.
We demonstrate here a novel approach to fine-tuning protein expression in a tumour-selective manner by exploiting the post-transcriptional reprogramming of gene expression in tumour cells to control selectivity of an oncolytic virus. Our results have been validated in pancreatic cancer models. However, since CPEB4 is overexpressed in several tumours, such as gliomas and colorectal cancers 14,31 , this oncoselective strategy may be valid for many solid tumours. This approach may overcome some of the limitations associated with other commonly used postentry viral targeting approaches. Taking advantage of a tumourselective control by tumour-specific promoters is confronted with the fact that most of the genome is constantly transcribed at low levels, and is a strategy restricted to viruses that rely on the cellular transcription machinery but not for viruses that use virally encoded polymerases for replication in the cytoplasm, such as the measles virus and vaccinia virus 34 . The alternative postentry strategy to regulate viral replication is a negative targeting approach based on the miRNA expression in normal tissues to restrict viral replication of miRNA-target siteengineered viruses. Indeed, this is a very versatile approach for many different viruses and is highly efficient. However, the evolution of escape mutants in miRNA-targeted viruses, or the potential off-target effects on the host miRNA machinery, have been proposed as potential caveats 6 . Interestingly, no mutations were found in the CPE-regulated adenovirus under evolutionary pressure ( Supplementary Fig. 9). To achieve a tight restriction of replication to tumour cells, it should be possible to combine different targeting strategies such as regulation by more than one viral gene 3 or the introduction of transcriptional and posttranscriptional control in the same viral gene 35 . In this line, a double E1A-engineered virus, with the uPAR promoter and the CPE post-transcriptional control, displayed additivity in a non-tumoral context (Supplementary Fig. 10).
This novel tropism-modified strategy presented here provides a framework for exploiting CPE-regulated transgenes to achieve cancer cell-specific expression in therapeutic gene-transfer-based strategies, or to attenuate viruses for vaccine purposes. This technology could also extend to other viruses of interest in virotherapy that require attenuation in particular tissues, such as reoviruses in heart 36 or coxsackievirus A21 in the muscle 5 .
Altogether, our data provide a novel paradigm for the development of tumour-specific viruses and provide the proof of principle that CPEB-dependent regulation can be exploited to attenuate viral toxicity, by preventing the spread of the virus in normal tissues, without perturbation of the antitumour efficacy.
Lentiviral constructs with a modified 3 0 -UTR. The polyadenylation sequences (4,193-4,199 and 4,423-4,429) from the pLS-CG lentiviral vector (Addgene; no. 12161) were mutagenized with the help of the QuickChange II Site-Directed Mutagenesis Kit (Stratagene, Wilmington, NC, USA) according to the manufacturer's instructions (see Supplementary Table 1 for primers). Mutagenized sequences were verified by DNA sequencing.
d2EGFP and d2RFP genes were cloned by PCR with specific primers incorporating the AgeI and XhoI restriction sites. The 3 0 -UTR of tPA was PCR amplified using specific primers with sequences for the XhoI restriction enzyme at both ends. The 3 0 -UTR of cB1 was designed in the reverse primer of d2EGFP gene with a sequence for the XhoI restriction enzyme. The 3 0 -UTR with the TNF-a ARE sequences was generated by amplifying the previously generated cB1 3 0 -UTR with a primer containing the TNF-a main ARE sequence upstream of the CPE sites.
All primer sequences are listed in Supplementary Table 1. The indicated sequences were cloned into the pLS-CG lentiviral vector. The newly generated vectors were sequenced to confirm the DNA sequence.
Lentiviral production. Lentiviral particles were generated by cotransfection of plasmids pCMVAR8.91, pVSV-G and the pLS-CG-derived constructs, or a lenitiviral plasmid expressing CPEB4, in 293T cells by the calcium/phosphate DNA precipitation method (Clontech, Mountain View, CA, USA). Supernatants from 293T-transfected cells were collected at 24 and 48 h, filtered and processed for purification by ultracentrifugation for 2 h at 12°C at 20,200 r.p.m. After ultracentrifugation, the pellet was resuspended in PBS for 16 h at 4°C under constant agitation, aliquoted and frozen at -80°C for later use. Viral titration was performed by quantitative PCR (qPCR) with specific primers against the PBC-Psi region of the pLS-CG vector (listed in Supplementary Table 1). A standard curve was generated using serial dilutions of pLS-CG-d2EGFP-3 0 -UTR_tPA construct mixed with genomic DNA. The number of molecules was calculated using the formula: M ¼ (C Á 6.02 Â 10 23 )/(660 Á bp), where C is the concentration of the plasmid, bp the number of base pairs and M the number of vector molecules.
Flow cytometry assay. Flow cytometry assays were performed in HPDE, RWP-1, PANC-1 and MIA PaCa-2 cells. Forty thousand cells were plated per well in 12-multiwell plates and, 24 h postseeding, cells were transduced with the indicated lentiviruses at 12 multiplicity of infection per virus. Transduction was facilitated by spinning for 2 h at 12,000 r.p.m. Two days post-transduction, d2EGFP and dRFP levels were analysed by flow cytometry using BD LSR II (Becton Dickinson). Flow cytometry results were analysed using FlowJo 8.7 for Macintosh.
Adenovirus generation. The AdCPE genome was generated by the following steps: first, the E1A polyA sequence of the pEND-K plasmid was mutated from 5 0 -AATAAA-3 0 to 5 0 -ACTCGA-3 0 , generating a new XhoI restriction site with the primer 5 0 -GCTGAATGAGATTGATGTAAGTTTACTCGAGGGTGAGATA ATGTTTAACTTGC-3 0 using the QuickChange Multi Site-Directed Mutagenesis Kit (Stratagene, Wilmington, NC, USA). Second, the construct containing E1A and the 3 0 -UTR of cB1 with the three CPE sequences was generated directly by PCR, by amplifying Adwt E1A with the primers 5 0 -CCTTGGGTCCGGTTTCTA TGCC-3 0 and 5 0 -CGTCTCGAGGCTTTATTAAAACCAGTAAAACATTAA AAACACAATACACTATTTACAGAAGCACATGGTGCAACACTTAT GGCCTGGGGCGTTTACAGC-3 0 . Third, the E1A from the pEND-K with the mutated polyA sequence was replaced by the PCR product, using AgeI and XhoI restriction sites, to generate pEND-K-E1A-cB1. Finally, AdCPE was generated by homologous recombination of pEND-K-E1A-cB1 with the genome of the serotype 5 wild-type adenovirus in E. coli BJ5183 cells as described 40 .
Adwt was obtained from ATCC. AduPAR was described previously 40 . AdDUC genome was generated by incorporating the CPE containing 3 0 -UTR of E1A into the BoxI and AflII restriction sites of the pSH-DM-UPAR-E1A plasmid followed by recombination of the resulting plasmid with the Adwt genome in BJ5183 cells as described previously 40 .
Adwt, AdCPE, AduPAR and AdDUC were propagated in A549 cells and purified by cesium chloride banding. The concentration of viral particles (vp ml À 1 ) was determined by means of optical density, and infectious particles (pfu ml À 1 ) were determined by hexon immunostaining in HEK293 cells 41 . Both viruses presented equal vp per pfu ratio.
Western blot analysis. Protein extracts were obtained with lysis buffer (50 mM Tris-HCl at pH 6.8, 2% SDS) containing 1% Complete Mini Protease Inhibitor (Roche Diagnostics, Basel, Switzerland). BCA Protein Assay Kit (Pierce-Thermo Fisher Scientific, Waltham, MA, USA) was used to determine the protein concentration, and total proteins (35 mg) were resolved by electrophoresis on 7.5% gels and transferred to nitrocellulose membranes by standard methods. Membranes were immunoblotted with rabbit anti-adenovirus-2/5 E1A polyclonal antibody ( RNA-immunoprecipitation and RT-qPCR. RPE cells were transfected at 80% confluence with 25 mg of pCMV-lucRenilla plasmid (cB1-3 0 -UTR WT or cB1-3 0 -UTR Mut) 27 using Lipofectamine LTX and PLUS Reagent (Thermo Fisher Scientific, Waltham, MA, USA) for 48 h. After transfection, cells were crosslinked with 0.5% formaldehyde in free DMEM for 5 min at room temperature. Then, cells were lysed with RIPA buffer supplemented with EDTA-free protease inhibitor cocktail (Sigma-Aldrich, Saint Louis, MO, USA) and 200 U ml À 1 RiboLock RNAse Inhibitor (Thermo Fisher Scientific). Cell lysates were sonicated for 5 min at low intensity and centrifuged for 30 min at 13,200 r.p.m. and at 4°C. Eight hundred micrograms of lysates were precleared with 20 ml of Dynabeads protein A (Thermo Fisher Scientific) for 30 min at 4°C. Finally, they were incubated with 5 mg of anti-CPEB4 antibody (Abcam; ab83009) or anti-HuR antibody (Santa Cruz Biotechnology, sc-5261) or anti-immunoglobulin G antibody (Sigma-Aldrich) coupled to 50 ml of Dynabeads protein A overnight at 4°C. One-fourth of the immunoprecipitates were eluted with Laemmli sample buffer by heating at 65°C for 20 min, resolved in SDS-polyacrylamide gel electrophoresis and analysed by western blotting. Three-fourth of remaining immunoprecipitates were digested with proteinase K during 1 h at 65°C, and the RNA was isolated by phenol extraction. All the RNAs were treated with DNAse (Thermo Fisher Scientific) and retrotranscribed with random hexamers and Oligo d(T)20, using SuperScript IV Reverse Transcriptase (Thermo Fisher Scientific) and subjected to qPCR. The resulting cDNAs were used for gene-specific qPCR, using the following primers: pLucORF Fw, 5 0 -ACTGGGACGAAGACGAACAC-3 0 ; pLucORF Rv, 5 0 -GGCGACGTAATCCACGATCT-3 0 ; Renilla S, 5 0 -GATAACTGGTCCGC AGTGGT-3 0 ; Renilla AS, 5 0 -ACCAGATTTGCCTGATTTGC-3 0 . Fold enrichment of luciferase mRNA in the immunoprecipitated fraction was calculated after normalization with the gene expression from the inputs. SYBER Green I Master plus mix (Roche Diagnostics) and the primers Hexon Fw, 5 0 -GCCGCAGTGGTCTTACATGCACATC-3 0 and Hexon Rv, 5 0 -CAGCAC GCCGCGGATGTCAAAG-3 0 . Adenoviral copy number was relativized to the cellular DNA content using the albumin intron 12 primers of Fw, 5 0 -CTGTC ATCTCTTGTGGGCTGT-3 0 and Rv, 5 0 -GGCTATCCAAACTCATGGGAG-3 0 .
In vitro cell survival studies. Dose-response curves were constructed for all assessed cells transduced with doses ranging from 0.001 vp per cell to 10,000 vp per cell of Adwt or AdCPE. Cell viability was measured 3 days postinfection by a colorimetric assay following the manufacturer's instructions (MTT Ultrapure; USB, Cleveland, OH, USA).
Polyadenylation assays. Polyadenylation patterns were evaluated using a modified version of the RNA ligation-coupled RT-PCR presented previously 42 . One microgram of total RNA was ligated to 0.1 mg of P1 anchor primer (5 0 -P-GGTCACCTTGATCTCAAGC-NH2-3 0 ) in 10 ml reaction using T4 RNA ligase I (New England Biolabs, Ipswich, MA, USA) according to the manufacturer's instructions. Half of the reaction product was used in a 50 ml reverse transcription reaction with PrimeScript RT-PCR (Takara Bio, Mountain View, CA, USA) according to the manufacturer's instructions, using 0.1 mg of P1 0 as a reverse primer (5 0 -GCTTCAGATCAAGGTGACCTTTTT-3 0 ). An aliquot (2.5 ml) of this cDNA preparation was used for a first pre-PCR step with Fw-E1A-polyA primer (5 0 -GGTGTAAACCTGTGATTGCG-3 0 ). One microlitre of this reaction was later used in each 25 ml PCR with the primers Fw-E1A-polyA and P1 0 . Toxicity analysis. PBS or 2 Â 10 10 vp of Adwt or AdCPE were injected intravenously into the tail vein of 6-to 8-week-old male immunocompetent C57BL/6 mice. Animals were weighed and examined daily for any clinical signs of toxicity. Three days later, mice were killed, organs were isolated and blood samples were collected by intracardiac puncture. Serum aspartate aminotransferase and alanine aminotransferase were determined on an Olympus AU400 Analyser (Olympus, Tokyo, Japan) at the Clinical Biochemistry Service, School of Veterinary Medicine, Autonomous University of Barcelona. All animal procedures met the guidelines of European Community Directive 86/609/EEC and were approved by the ethical committee (CEEA-University of Barcelona) and by the local authorities of the Generalitat de Catalunya.
Mouse xenografts. RWP-1, MIA PaCa-2 and PANC-1 cells (2.5 Â 10 6 ), embedded in Matrigel 1:1 (BD Biosciences, San Jose, CA, USA), were subcutaneously injected into each flank of male, 7-to 8-week-old, athymic nu/nu mice (Harlan, Sant Feliu de Codines, Spain). Tumours were measured at least three times weekly, and their volumes were calculated using the formula V ¼ larger diameter Â (smaller diameter) 2 Â piÄ6. Mice were randomly assigned to either group for treatment. Virus was administered once tumours achieved a median volume of 100 mm 3 . The experimenter was blinded until the conclusion of the study.
Statistical analyses. The descriptive statistical analysis was performed on GraphPad Prism v5.0a (GraphPad Software, La Jolla, CA, USA). Statistical differences were evaluated using nonparametric two-tailed Mann-Whitney test.
Po0.05 was taken as the level of significance.
Statistical analyses for the d2EGFP/dRFP quantification were performed using R (version 2.10.0) and the ARM package, to perform the mixed model by REML and multcomp package for multiple comparison analysis of means by the Tukey-Kramer test.
Sample size calculation for animal studies took into consideration to have more than five mice in each group.
The in vivo tumour growth statistical analysis was evaluated using R v2.14.1 software (R: a language and environment for statistical computing; R Foundation for Statistical Computing, Vienna, Austria) with a linear mixed-effect model using the lme4 package. We associated a random-effects term with the day of measurement 43 . Statistical differences were evaluated using a multiple comparison of means by Tukey's contrasts.
Data availability. All relevant data are available from the authors on reasonable request.