The UPR sensor IRE1α and the adenovirus E3-19K glycoprotein sustain persistent and lytic infections

Persistent viruses cause chronic disease, and threaten the lives of immunosuppressed individuals. Here, we elucidate a mechanism supporting the persistence of human adenovirus (AdV), a virus that can kill immunosuppressed patients. Cell biological analyses, genetics and chemical interference demonstrate that one of five AdV membrane proteins, the E3-19K glycoprotein specifically triggers the unfolded protein response (UPR) sensor IRE1α in the endoplasmic reticulum (ER), but not other UPR sensors, such as protein kinase R-like ER kinase (PERK) and activating transcription factor 6 (ATF6). The E3-19K lumenal domain activates the IRE1α nuclease, which initiates mRNA splicing of X-box binding protein-1 (XBP1). XBP1s binds to the viral E1A-enhancer/promoter sequence, and boosts E1A transcription, E3-19K levels and lytic infection. Inhibition of IRE1α nuclease interrupts the five components feedforward loop, E1A, E3-19K, IRE1α, XBP1s, E1A enhancer/promoter. This loop sustains persistent infection in the presence of the immune activator interferon, and lytic infection in the absence of interferon. Adenovirus (AdV) can cause persistent infections, but underlying mechanisms are poorly understood. Here, Prasad et al. show that the AdV glycoprotein E3-19K activates the unfolded protein response sensor IRE1α, and that this triggers a feedforward loop that sustains persistent infection in the presence of interferon.

infection by reducing the recruitment of the positive transcription regulator GABPα/β, and enhancing the E2F/Rb repressor complex on the E1A promoter sequence 36 . Removal of IFN leads to virus lytic release, akin to acutely immunosuppressed patients 17 .
Here, we show that the AdV glycoprotein 19K selectively activates IRE1α but not PERK and ATF6. This gives rise to a transcriptional feedforward loop, including five components-19K, IRE1α, XBP1s, the E1A enhancer/promoter (e/p), and E1A protein. This loop maintains long-term viral persistence in the presence of IFN, and boosts lytic infection in the absence of IFN.

Results
IRE1α-mediated XBP1 splicing enhances AdV infection. Mammalian cells express two homologs of the yeast Ire1p, IRE1α (encoded by the ERN1 gene) and IRE1β (ERN2), the latter in a tissue-specific manner, for example in the digestive tract 37,38 . To explore the role of IRE1α in AdV infection, we used CRISPR/ Cas9 to generate IRE1α-knockout HeLa cells (HeLa I-KO). A guide RNA targeting exon 2 yielded a KO phenotype affecting all three allelic copies of IRE1α (Fig. 1a, Supplementary Fig. 1a, b). Two IRE1α alleles were edited by frameshift mutations and one had a 15-nt in-frame deletion. HeLa I-KO was significantly less susceptible to AdV-C5 infection than HeLa or HeLa I-KO ectopically expressing IRE1α from a lentivirus, as shown by AdV late protein VI expression and virus production (Fig. 1a). Infection inhibition was not due to reduced virus entry, as the incoming vDNA was effectively delivered into the nucleus 2.5 hpi 39 (Supplementary Fig. 1c). Akin to IRE1α KO, the IRE1α nuclease inhibitor 4µ8C 40 reduced viral replication, as shown by quantitative (q)-PCR (Fig. 1a).
To test if AdV infection enhanced the IRE1α nuclease activity, we measured the levels of XBP1s by quantitative reverse transcriptase (qrt) PCR in extracts of infected HeLa and HeLa I-KO cells 41 . AdV-C5 infection consistently increased the XBP1s mRNA levels at 24, 48, or 72 hpi (Fig. 1b). Note that the band denoted with asterisk (*) is a background product and can be removed by EndoT digest (see Supplementary Fig. 1d). XBP1s (denoted as 1S) was not enhanced in HeLa I-KO cells, but restored by ectopic expression of IRE1α (Fig. 1b). Importantly, XBP1s induction occurred in AdV-C2 or C5-infected HeLa and human corneal epithelial (HCE) cells, or in diploid human fibroblasts WI-38, but not in cells infected with AdV-C5 lacking the E1 region ( Fig. 1b, Supplementary Fig. 1e-g). E1 encodes the E1A immediate early transactivator protein, the antiapoptotic E1B-19K and E1B-55K proteins 24,42 . Collectively, the data show that AdV infection of both transformed and nontransformed non-transformed human cells induces XBP1s depending on IRE1α and the early viral genes E1A-E1B.
Phosphorylation of IRE1α and not PERK in AdV infection. We next tested if XBP1s induction was conserved in murine cells. AdV-C5 infection for 24-48 h readily induced XBP1s in mouse embryonic fibroblasts (MEFs) expressing Flag-IRE1α, but not in cells lacking IRE1α (I-KO) 43 , akin to the short-term chemical stimulation of ER stress by thapsigargin (5 h), which inhibits the ER calcium pump and depletes calcium ions from ER stores 44 (Fig. 2a). Both I-KO and normal MEFs were readily infected with AdV-C2, as indicated by the expression of the early 19K glycoprotein.
To test if AdV-C5 activated IRE1α, we analyzed the migration of the endogenous IRE1α protein in SDS-PAGE containing Phos-Tag. Such gels retard the migration of phosphoproteins relative to the nonphosphoproteins. IRE1α migration was retarded upon AdV-C5 infection at 16 or 24 hpi, and retardation blunted by treatment of the cell lysates with phosphatase (Fig. 2b). In contrast, PERK did not show an upward shift upon AdV-C5 infection, unlike treatment with the reducing agent dithiothreitol (DTT), a known activator of PERK 45 (Fig. 2c). We found no evidence that AdV-C5 induced the regulated IRE1α-dependent decay (RIDD) pathway, unlike DTT, as assayed with Bloc1S1 mRNA (see Supplementary Fig. 2a).
BiP dissociates from IRE1α in AdV infection independent of 19K. In UPR, the association of the ER chaperone BiP/Grp78 (BiP) with IRE1α decreases upon initial ER stress, and restores under persistent UPR stimulation, when BiP and co-chaperones are transcriptionally induced 46,47 . We assessed the levels of BiP-IRE1α by immunoprecipitation experiments of Flag-hIRE1α expressed in IRE1α −/− MEFs at near-endogenous levels 43 . The coimmunoprecipitation data, western blots, and rt-PCR measurements of E1A mRNA showed that BiP dissociated from IRE1α at 7 hpi, when only low amounts of E1A protein were present, and neither 19K protein nor XBP1 splicing were detectable (Fig. 2d). This was in contrast to conventional UPR induced by DTT where BiP-IRE1α dissociation was rapidly followed by XBP1 splicing (Supplementary Fig. 2b). Similar results were obtained in HeLa I-KO cells transduced with lentivirus encoding Flag-hIRE1α, where BiP dissociated from IRE1α at 4 h post AdV-C2 infection (Supplementary Fig. 2c, d). In MEFs, AdV-C5 lacking E1A and E1B  ORFs did not dissociate BiP from IRE1α, in contrast to a mutant lacking the 19K ORF (Fig. 2e). Notably, AdV-C5 infection did not increase the levels of BiP and IRE1α, unlike treatment with DTT, which massively increased the BiP levels ( Fig. 2f). Accordingly, the ectopic expression of BiP did not attenuate the induction of XBP1s by AdV-C2, unlike thapsigargin treatment ( Supplementary Fig. 2e, f), consistent with earlier reports 46 . Collectively, the data indicate that AdV activates IRE1α in a non-canonical manner, whereby BiP dissociates long before XBP1 splicing, and independent of 19K. f a AdV-C2 (h) 24 48 Tg ( An early AdV gene product activates IRE1α. E1A activates all early viral promoters, and together with E1B enhances transformation and replication 48,49 . We assessed if viral replication was required for the IRE1α/XBP1s induction by treating HeLa cells with the nucleoside analog cytosine arabinoside (AraC), which blocks viral replication and late protein expression past the immediate early phase of E1A induction 50,51 . AraC neither affected the E1A expression nor the levels of XBP1s (Supplementary Fig. 3a). In contrast, an inhibitor of the positive transcription elongation factor, flavopiridol, blocked E1A expression 24 hpi as expected 52 , and wiped out the induction of XBP1s ( Supplementary Fig. 3b). The data reinforce the notion that E1A is required for XBP1s induction.
The lumenal domain of C2/5-19K activates IRE1α. We used a range of E3 mutant viruses to test for induction of XBP1s (Fig. 3a). AdV-C5-dl327 lacks 7.1K, 19K, ADP, and RIDα/β 53 , dl309 lacks RIDα/β and 14.7K 54 , and AdV-C2-dE3B has a deletion of RIDα/β 55 . The deletions in dl327 and dl309 were validated by PCR, and in addition, all three viruses were found to express E1A and 19K, except dl327 that lacked 19K ( Supplementary Fig. 3c). Importantly, dl327 did not activate IRE1α unlike dl309 and AdV-C2-dE3b (Fig. 3a). We investigated the role of 19K in the activation of IRE1α in HDF-TERT cells. Phos-tag SDS-PAGE indicated that a mutant lacking the 19K ORF (AdV-C5-Δ19K) poorly induced IRE1α phosphorylation, and barely induced XBP1s at 24 hpi unlike AdV-C5 (Fig. 3b). Similar results were obtained in time-course studies in HeLa cells ( Supplementary Fig. 3d). Remarkably, the AdV-C5-Δ19K-infected cells showed reduced E1A expression levels compared with AdV-C5. This finding was in accordance with the RNAi data against 19K, where a pool of more than a dozen synthetic dsRNAs targeting 19K strongly reduced the induction of XBP1s, and also reduced the expression of E1A, while control RNAi pools against E4-Orf4 had no effect on E1A ( Supplementary Fig. 3e).
The abrogation of individual ORFs demonstrated that expression of 19K was necessary for XBP1s induction (Fig. 3c, Supplementary Fig. 3f). The expression of 19K alone was sufficient to induce XBP1s, but not PERK phosphorylation, unlike DTT (Fig. 3c).
The 19K glycoprotein has an N-terminal cleavable signal sequence, a transmembrane segment, and a small cytosolic tail with a di-lysine ER retention signal 56 . To identify the domain that triggered IRE1α activation, we expressed 19K chimeras with transmembrane and cytoplasmic domains from murine MHC-I H-2K d in HeLa cells 35 . All three chimeras had comparable expression levels and induced XBP1s, but 19K lacking the LD only weakly induced XBP1s (Fig. 3d, and Supplementary Fig. 3g). The human MHC class-I antigen binding defective M87A 19K mutant 35 strongly induced XBP1s, indicating that MHC-I binding was not required for XBP1s induction (Fig. 3d). The expression of the C2 19K-LD alone with or without a C-terminal HDEL motif activated IRE1α to comparable levels as the fulllength protein, and much more effectively than the D8 19K-LD or full-length 19K (Fig. 3e, Supplementary Fig. 3h, i). The AdV-C2/ 5-19K glycoproteins have a highly conserved LD of 122 and 123 amino acids, 92% of which are identical, whereas the D8-LD of 119 amino acids is 30% identical with the C2/5-LDs (Supplementary Fig. 3j). This underscores that 19K alone or in context of AdV-C infection induces the phosphorylation of endogenous IRE1α followed by XBP1 splicing. This suggests the formation of IRE1α oligomers triggering trans-autophosphorylation and allosteric activation of the endonuclease domain 9 .
The C2-19K lumenal domain interacts with IRE1α but not PERK. To test if 19K interacted with IRE1α, we stably expressed C2 19K in human embryonic kidney (HEK) 293 cells, yielding a reticular ER-like pattern in the cytoplasm ( Supplementary Fig. 4a). Immunoprecipitation of 19K by the monoclonal IgG antibody 3A9 significantly enriched IRE1α compared with pulldowns with a control IgG, suggesting a complex of IRE1α and 19K. We next tested if the lumenal domains of C2-19 K and IRE1α interacted in the ER. Tripartite split-GFP fluorescence complementation assays were used in cells expressing the 19K-LD and IRE1α or PERK-LD. The Flag-IRE1α containing the 20-amino-acid domain GFP10 at the C terminus and the C2 19K-LD with the C-terminal 18 amino acids of GFP11 gave rise to reticular ER-like green fluorescence signals in transfected cells expressing signal sequence containing GFP1-9 with a HDEL ER retention signal (Fig. 4a, b). Green fluorescence signals colocalized with the anti-19K and anti-Flag immunostainings validating the GFP complementation signals. In contrast, no green fluorescence was obtained with Flag-tagged PERK-GFP10-LD, although both 19K and PERK localized in reticular cytoplasmic Three independent experiments gave similar results. Source data are provided as a Source Data file. b AdV induces phosphorylation of IRE1α. Flag-IRE1α expressing IRE1α-KO MEFs at 7, 16, and 24 hpi with AdV-C5 (MOI 300) immunoblotted with anti-IRE1α antibody. Lysates were resolved on a 6% SDS-PAGE gel containing 25 µM Phos-tag. Samples were treated with or without alkaline phosphatase. Phosphorylated (p) and hypophosphorylated (o) forms of IRE1α are indicated by the dashed lines, and the percentage of IRE1α phosphorylated was calculated as indicated. Lysates are the same as in panel D demonstrating β-tubulin loading. Three independent experiments gave similar results. Source data are provided as a Source Data file. c AdV-C5 infection of HeLa cells (MOI 200) does not activate PERK, unlike treatment of cells with the reducing agent DTT. Activated phosphorylated PERK is indicated by p, and the inactive form by o. Two independent experiments gave similar results. d BiP displacement from IRE1α occurs before XBP1 splicing in Flag-IRE1α-expressing IRE1α-KO MEFs infected with AdV-C5 (MOI 300). Cells were lysed and BiP-IRE1α complexes immunoprecipitated (IP) with anti-Flag antibody, and a western blot with anti-IRE1α, anti-BiP, and anti-β-tubulin antibodies was performed. A separate non-reducing immunoblot probed with anti-19K Tw1.3 antibodies revealed monomeric and dimer forms of 19K indicated as mo and di, respectively. Input lysates were 1% of the immunoprecipitated samples. The corresponding samples were also analyzed for XBP1 splicing and E1A mRNA levels by rt-PCR (reverse transcription polymerase chain reaction), as indicated. Three independent experiments gave similar results. Source data are provided as a Source Data file. e BiP-IRE1α dissociation requires E1, not 19K. Co-immunoprecipitation of Flag-IRE1α and BiP was performed as described in d with AdV mutants lacking E1 (AdV-C5-ΔE1) and 19K (AdV-C5-Δ19K) at MOI 300 each. Two independent experiments gave similar results. Source data are provided as a Source Data file. f AdV-C5 infection does not increase BiP/Grp78 and IRE1α levels given in arbitrary units (a.u.), unlike the canonical UPR triggered by DTT (2 mM). The bar graph shows the normalized levels of IRE1α and BiP from three independent experiments. Data show the means ± SD from three independent experiments. Source data are provided as a Source Data file.
XBP1s boosts E1A and 19K, IFN-γ inhibits E1A expression. We next asked if XBP1s directly enhanced AdV infection. Lentivirus-mediated transduction of human XBP1s gave a dosedependent increase of XBP1s in HeLa I-KO cells, and increased AdV-C2 infection (Fig. 5a). This was confirmed by qrt-PCR measurements of E1A mRNA and 19K protein levels (Fig. 5a). The XBP1s enhancement of E1A was dependent on the activation domain of XBP1s, as demonstrated by expression of mutant murine   24h pi: Wilcoxon non-parametric XBP1s lacking the activation domain due to a premature stop codon after the leucine zipper domain ( Supplementary Fig. 5a). Both wild-type and mutant XBP1s mRNAs were expressed similarly, although we could only detect wild-type XBP1s in western blots with an antibody against the C-terminal region containing the transactivation domain. The ectopic expression of XBP1s enhanced the E1A protein levels under the viral E1A-e/p in HeLa and HDF-TERT cells (Fig. 5b). The application of the IRE1 nuclease inhibitor 4µ8C reduced E1A expression, indicating that both XBP1s and IRE1α boost E1A expression in the absence of other AdV gene products (Fig. 5b). In contrast to XBP1s and IRE1α, IFN-γ reduced E1A expression. Notably, the XBP1s-binding sites on the E1A-e/p are separated from the E2F co-repressor-binding sites controlled by IFN-γ 36 . The enhancement of E1A expression by IRE1α and XBP1s was further confirmed by small-interfering RNAs targeting IRE1α and XBP1, which reduced the lentivirus-based E1A expression in both untreated and IFN-γ-treated HeLa or HDF-TERT cells (Fig. 5b). The treatment of HDF-TERT cells with IFN-γ also suppressed AdV-C5 long-term infection, as indicated by strongly reduced expression of the E4-GFP-Orf4 fusion protein (Supplementary Fig. 5c), in agreement with the literature 36 .
Since reduced XBP1s levels may affect cell proliferation 57,58 , we checked if XBP1s levels affected the growth of HDF-TERT cells. Label-free, real-time impedance measurements (xCELLigence) reporting on cell numbers, cell adhesion, and cell-cell interactions 59 indicated that both 4µ8C and RNAi against XBP1 reduced cell numbers upon 2-3 days of treatment. The washout of 4µ8C 5 days post treatment restored cell proliferation, indicating that 4µ8C affected cell growth but not viability, whereas XBP1s RNAi led to cell death 4-5 days post treatment ( Supplementary Fig. 5d). We conclude that IRE1α KO cells exhibit reduced levels of basal XBP1s, and do not increase XBP1s upon ER stress induction. XBP1 on the other hand is essential for cell viability. Ectopic XBP1s enhances E1A expression, even if E1A is expressed in the absence of other viral genes.
XBP1s binds to the E1A-e/p and promotes E1A expression. XBP1s is an integral element of the UPR, and binds to the promoters of genes restoring homeostasis upon ER stress 60,61 . We identified multiple elements of the ACGT and the CCACGbinding box motifs in the E1A and the E4 promoters, and one of each in the major late (ML) promoter, conserved in several AdV species, including B, C, D, and F (Fig. 6a). To test if XBP1s bound to the E1A-, E4-, and ML promoters, ChIP analysis of AdV-C2infected cells was performed. XBP1s antibodies enriched the E1A, MLP, and E4-e/p regions fivefold to sevenfold over isotype control antibodies (Fig. 6b). The deletion of the four E1A XBP1sbinding sites in E1A significantly reduced E1A expression in HDF-TERT cells from dl309_Δ63-95 compared with dl309, most prominently, if XBP1s was overexpressed (Fig. 6c). This result was supported by site-specific mutagenesis of the five XBP1sbinding sites (ACGT, CCACG, and CACG boxes) yielding AdV-C5-XBP1s-mut, which showed strongly attenuated E1A expression compared with AdV-C5 (Fig. 6c). In accordance, the vDNA copy numbers were four logs reduced in dl309_Δ63-195-infected HDF-TERT cells compared with dl309 at 216 hpi ( Supplementary  Fig. 6a). Similar result was obtained with AdV-C5-XBP1s-mut compared with AdV-C5. The reduced cytopathic effects of dl309_Δ63-195 and AdV-C5-XBP1s-mut compared with dl309 and AdV-C5 were confirmed by impedance measurements (Supplementary Fig. 6b). We conclude that XBP1s binds to the E1A-e/p, transactivates E1A transcription, and drives lytic AdV infection.
The IRE1α-XBP1s axis enhances persistence. We next explored if IRE1α-XBP1s controlled persistent AdV infections of HDF-TERT cells. Short-term (72 h) persistent infection in the presence of IFN-γ reduced the expression of protein VI about tenfold, which was further reduced by RNA interference against either IRE1α or XBP1, also in the presence of IFN-γ (Fig. 7a). In longterm infections (22 days), 4µ8C reduced both lytic (without IFNγ) and persistent (in the presence of IFN-γ) infections, as determined by E1A expression and vDNA copy numbers (Fig. 7b).
To test if persistence depended on the E1A-e/p, we used impedance measurements and q-PCR to monitor infections with dl309_Δ63-195 and AdV-C5-XBP1s-mut lacking the major XBP1s-binding sites in the E1A-e/p due to deletion or point mutations, respectively. The E1A-wild-type viruses dl309 and AdV-C5 were used as controls. HDF-TERT cells were seeded onto xCELLigence plates, treated with IFN-γ, infected 36 h later, and incubated in the presence of IFN-γ until 22.5 days post seeding, followed by IFN-γ washout and incubation for another 13.5 days. Multiplicities of infection (MOI) of both viruses was Fig. 3 The lumenal domain of the 19K glycoprotein activates IRE1α. a Schematic drawing showing the deletions in the E3 region of AdV mutants with yellow boxes indicating the deletions (left panel). Infection was carried out with a median of 150 particles bound per cell. Asterisk denotes a background product. At least three independent experiments gave similar results. Source data are provided as a Source Data file. b AdV-C5 19K enhances IRE1α phosphorylation, XBP1s splicing, and E1A levels. Phosphorylation of IRE1α in AdV-C5-and AdV-C5-d19K-infected HDF-TERT cells (MOI 75000, 24 hpi) was analyzed in lysates treated with or without alkaline phosphatase, and fractionated by SDS-PAGE (6%, containing 25 µM Phos-tag, first panel). XBP1 splicing in AdV-C5-and AdV-C5-d19K-infected HDF-TERT cells 24 hpi (MOI 75,000, second panel). Immunofluorescence data of E1A are shown as a scatterplot using n = 14,000 cells randomly chosen per condition, 24 hpi (MOI 75000). Central line of the box plot indicates median with first and third quartiles, and whiskers are shown as boxes and lines, respectively. Statistics were performed using the Wilcoxon two-sided nonparametric test with *p < 0.0001 (third panel). Three independent experiments gave similar results. Source data are provided as a Source Data file. chosen such that similar cytopathic effects occurred without IFNγ treatment at 120 h post seeding (Fig. 7c, right panel, see also E1A expression data in Supplementary Fig. 7b). While dl309 and AdV-C5 infections led to cell death after 12.5 and 11 days, respectively, in the presence of intermittent amounts of IFN-γ, the dl309_Δ63-195 and the AdV-C5-XBP1s-mut-infected cells remained viable throughout the course of the experiment to about 36 days (870-h time point), akin to uninfected cells (Fig. 7c). Under IFN-γ, the titers of dl309_Δ63-195 and AdV-C5-XBP1smut did not rise (middle panel), and the cells remained viable as shown by phase-contrast microscopy and impedance measurements using xCELLigence, indicative of only low levels of persistence, whereas titers of dl309 or AdV-C5 increased by 10 2 −10 3 -fold, indicating high persistence of E1A normal viruses (Fig. 7c). Importantly, upon removal of IFN-γ 22.5 days post seeding, the dl309_Δ63-195 and AdV-C5-XBP1s-mut titers increased several hundred to a thousand-fold 36 days post seeding, while the dl309 or AdV-C5 infections lead to rapid cell death. These results demonstrate that the XBP1s-binding sites in the E1A-e/p support the persistence under IFN-γ. The data show  4 19K forms a complex with IRE1α in the ER. a Schematic representation of tripartite split-GFP lumenal domain (LD) constructs used in ER-lumenal GFP complementations. Green fluorescence is restored when proteins containing the GFP10 and GFP11 domains are in close proximity together with the core GFP1-9 targeted to the ER lumen. b Interaction of C2 19K-LD with IRE1α-LD but not PERK-LD. HeLa cells co-transfected with C2 19K-LD-GFP11, SS-(signal sequence)-GFP1-9-HDEL, and Flag-IRE1α-GFP10 or PERK-LD-GFP10 were fixed and stained with anti-19K (3A9) and anti-Flag antibodies and DAPI (nuclei). Confocal images were segmented with CellProfiler using DAPI as a nuclear mask, and the reticular ER signal was measured in a ten-pixel area around the nuclei. Arrows indicate high-intensity split-GFP complementation signals in IRE1α-LD-transfected cells, and arrowheads low-intensity split-GFP complementation. Zoomed in representative immunofluorescence micrographs of split-GFP complementation in SS-GFP1-9-HDEL, IRE1α-LD-GFP10, and C2 19K-LD-GFP11-transfected cells. Cells were imaged with Leica SP8 microscope using Nyquist x-y-z sampling with ×63 objective. Following acquisition, images were deconvolved using SVI Huygens using theoretical point-spread function (PSF) automatically calculated from the imaging parameters. Split-GFP complementation signals appear in the region of ER tubules where 19K and IRE1α colocalize. infected HeLa I-KO cells (n = 3627 and 2317 for lenti-empty and lenti-hXBP1s, respectively, fourth panel). Data show the medians and first and third quartiles, and whiskers as boxes and lines, respectively. Statistical analyses were done by two-tailed Wilcoxon nonparametric tests with significance *p < 0.001. Two independent experiments gave similar results. b E1A expression from the E1A-e/p can be increased by ectopic XBP1s, or reduced by the IRE1α inhibitor 4µ8C in HeLa cells (left panel). A similar experiment was carried out in HDF-TERT cells including IFN-γ to suppress the E1A promoter/enhancer expressed from a lentivirus vector (middle panel). E1A expression in HDF-TERT cells is reduced by RNA interference against IRE1α and XBP1, but not by nontargeting siRNA (siNeg1, 20 nM siPools, right panel). Error bars represent standard deviations. For all the graphs, data are presented as mean from two technical replicates, and two independent experiments gave similar results. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-15844-2 ARTICLE that noncanonical activation of IRE1α maintains a feedforward loop between the ER and the nucleus, gives rise to E1A and 19K expression, maintains AdV persistence, and boosts lytic infection (Fig. 7d).

Discussion
AdVs encode several immune-modulatory membrane proteins. We showed that the E3-19K glycoprotein of species C2/5 but not of the divergent D8 AdV is necessary and sufficient to selectively activate IRE1α, but not PERK, ATF6, and RIDD, see also ref. 59 . Activations of PERK and RIDD normally restore homeostasis upon ER stress by phosphorylation of the eukaryotic initiation factor 2 GTPase inhibiting global protein synthesis, and degrading mRNA and microRNAs, respectively 45,62 . This is in accordance with unabated protein production at the time of full IRE1α activation, that is no loss in Flag-hIRE1α, BiP, or β-tubulin, indicating that AdV does not activate the UPR to inhibit protein synthesis. The ensuing 19K expression and association of the 19K-LD with the IRE1α-LD induced IRE1α phosphorylation and XBP1 mRNA splicing yielding XBP1s. These results are compatible with proteomics data showing that XBP1s levels are increased in AdV-infected cells 63 .
Remarkably, AdV-infected cells dissociated BiP from IRE1α before XBP1 splicing, and independent of 19K expression, but dependent on E1A, indicating that the E1A protein, its mRNA, or an E1A-controlled gene product dissociates BiP from IRE1α. This may be akin to a subset of ER-targeted cellular mRNAs, signal recognition particle RNA, ribosomal RNAs, or transfer RNAs, which can directly activate IRE1α 64   triggers the transition of IRE1α from a closed to an open conformation, and leads to IRE1α oligomerization 65 . Consequences of IRE1α activation comprise inflammatory reactions in conjunction with activation of pattern recognition receptors 66 . The LD of C2 19K and IRE1α was in close association based on immunoprecipitations and split-GFP complementation assays. This mode of IRE1α activation contrasts with a recent lipid stress activation model of IRE1α, where BiP binding to IRE1α is not affected 43,67 . Importantly, IRE1α activation by C2 19K was independent of 19K interaction with MHC-I in the ER lumen, The initiation of XBP1 mRNA splicing is a key output from activated IRE1α reviewed in refs. 9,68 . XBP1s is central in boosting lytic and persistent AdV-C2/5 infections. Lytic infection could be rescued by overexpression of XBP1s in IRE1α KO cells. More specifically, XBP1s was found in a complex with the E1A-e/p in infected cells, and boosted E1A (and also 19K) expression, and lytic infection. The XBP1s-binding sites on the E1A-e/p are distantly located from the E2F-Rb repressor-binding sites 36 , a configuration that allows for coregulation by repressors and stimulators, which we show here is crucial in AdV persistently infected cells in the presence of IFN-γ. The concept of coregulation by activators and repressors is found with herpes viruses and HIV, where transcriptional coregulation occurs during persistence 69,70 . In accordance, the IRE1α nuclease inhibitor 4µ8C, which reduces XBP1s, reduced AdV-C5 persistence in HDF-TERT cells under IFN-γ.
The C2/5 19K glycoprotein of the E3 region together with the transcriptional activator E1A is important to maintain the levels of XBP1s, as indicated by infection of HDF-TERT cells with AdV-C5-Δ19K. 19K activates IRE1α, which increases XBP1s, and XBP1s enhances E1A transcription. This supports both lytic and persistent infection of HDF-TERT cells, as indicated by AdV-C5 mutants lacking functional XBP1s-binding sites in the E1A-e/p. In nonlymphoid cells, the E3 promoter, which lacks XBP1sbinding sites is controlled by E1A, and in lymphoid cells, it is E1A-independent but NF-κB dependent 71 . This may lead to cell death. For example, TNFα exposure induces ER stress, IRE1α, and NF-κB activation, and exacerbates death signaling through the TNF receptor 1, involving c-JUN N-terminal kinase 72 . If and how AdV antagonizes this signaling pathway is unknown. Alternatively, other host transcription factors could be involved in E3 regulation. For example, SP1, which is induced by XBP1s 60 has 29 predicted binding sites in the E3 promoter, and could enhance E3 transcription. Regardless of the nature of the E3 promoter regulation, the deletion of the 7.1K and 19K region in AdV-C5 reduced viral persistence in Syrian hamsters 73 . Our data highlight the therapeutic potential of the IRE1α branch of the UPR, and converge on the notion that in Zika virus-infected human and mouse embryos, the UPR was observed in the cerebral cortex of postmortem fetuses, and the IRE1α nuclease inhibitor 4µ8C reduced the microcephaly frequency 74 . In conclusion, the selective activation of IRE1α by 19K provides an environment conducive for AdV persistence, and supports the lytic cycle under conditions of impaired immunity.

Methods
Cells and viruses. A549 and HeLa-ATCC cells were obtained from American Type Cell Culture (ATCC). HCE cells were obtained from Dr Niklas Arnberg (UMEA University, Sweden). Human diploid fibroblasts immortalized with telomerase (HDF-TERT) 75 and other cells were grown at 37°C in 5% CO 2 environment in DMEM (Sigma) supplemented with 10% fetal calf serum (FCS). For generation of the AdV-C2_dE3B-mCherry virus, the GFP ORF in AdV-C2_dE3B_GFP 55 was replaced with the mCherry ORF. The virus genome was cloned into pKSB2 76 , followed by two homologous recombination steps according to the recombineering protocol numbers 1 and 3 77 . Note that the MOI varied between different experiments and viruses depending on the susceptibility of the cells, the nature of the experiment (persistence/lytic/early/late infection readout), and the infection protocol (warm or cold-synchronized infection). For generation of AdV-C5-E3-Δ19K and AdV-C5-XBP1s-mut viruses, AdV-C5 (wt300) DNA 78 was inserted into pKSB2 76 followed by two homologous recombination steps according to the recombineering protocol numbers 1 and 3 77 . Insertion of the sequence 5′-ATTTATTGTC AGCTTTTTAAACGCTGGGGTCGCCACCCAAGATGATTTACTAAGTTACAA AGCTAATGTCACCACTAACTGCTTTACTCG-3′ in the second step introduced a complete deletion of the 19K ORF except the first 4 nt, allowing to keep the overlapping stop of the upstream 6.7K CR1a gene (AdV-C5-dE3-19K). Similarly, insertion of 5′-CCTTAATTAAGGGCGGCCGCATTTAAATTAATTAACATCATCAATAAT  ATACCTTATTTTGGATTGAAGCCAATATGATAATGAGGGGGTGGAGTTTGT  GCATGTTCGCGGGGCGTGGGAACGGGGCGGGTGCATGAGTAGTGTGGCGG  AAGTGTGATGTTGCAAGTGTGGCGGAACACATGTAAGCGACGGATGTGG  CAAAAGTGCATGTTTTGGTGTGCGCCGGTGTACACAGGAAGTGACAATT  TTCGCGCGGTTTTAGGCGGATGTTGTAGTAAATTTGGGCGTAACCGAGTA  AGATTTGGCCATTTTCGCGGGAAAACTGAATAAGAGGAAGTGAAATCTGA  ATAATTTTGTGTTACTCATAGCGCGTAATATTTGTCTAGGGCCGCGGGGA  CTTTGACCGTTTCATGGGAGACTCGCCCAGGTGTTTTTCTCAGGTGTTT  TCCGCGTTCCGGGTCAAAGTTGGCGTTTTATTATTATAGTCAGCTTCAT  GGTAGTGTATTTATACCCGGTGAGTTCCTCAAGAGGCCACTCTTGAGT  GCCAGCGAGTAGAGTTTTCTCCTCCGAGCCGCTCCGACACCGGGACTGAA AATGAGACATATTATCTGCCACGGAGGTGTTATTACCGAAGAAAT-3′ in the E1A-e/p region introduced XBP1s-binding sites mutagenized using A<>C; G<>T substitution rule (AdV-C5-XBP1s-mut, Fig. 6c). All the indicated MOI were calculated as the estimated number of viruses per cell where virus titer was determined by q-PCR based described in the section "Virus titer estimation." For specific reagents, see Supplementary Table 1.
Cas9 knockout of IRE1α in HeLa-ATCC cells. HeLa-ATCC and 293T cells were transduced with Lentiviral vector expressing Cas9 (Addgene, 49535) and a guide RNA against exon 2 of mammalian ERN1 (IRE1α, chromosome 11) 79 . A polyclonal population of both cell lines expressing Cas9 was selected with puromycin (2 µg/ml), and single clones were isolated from HeLa-ATCC. For selection of clonal population, a polyclonal population was seeded at high dilution in a 15-cm dish and allowed to grow into several sparse clones. Using circular rings, small populations from single clones were trypsinized, collected, and seeded on 96-well clearbottom plates for expansion. After several weeks, DNA from the growing clonal population was extracted, and using primers flanking the target site of Cas9 on exon 2 of IRE1α, PCR was performed, and fragments were ligated into plasmid vector (Bluescript). Positive clones were sequenced, aligned to the wild-type Fig. 7 IRE1α and XBP1s facilitate persistent and lytic AdV infections. a Reduction in late AdV-C5 protein VI expression (MOI 180) of HDF-TERT cells upon RNA interference against IRE1α and XBP1 in the presence or absence of IFN-γ, including nontargeting siRNA (siNeg1). Data show the means ± SD from three independent experiments (n = 3). b E1A expression of AdV-C5-infected HDF-TERT cells (MOI 200, 37°C, 1 h) with or without 500 IU IFN-γ, or IRE1α RNase inhibitor 4µ8C (100 µM) at 13 days pi was analyzed by immunofluorescence showing representative images (left) and a scatterplot with 15,000 cells per condition (middle). Data show the median, first and third quartiles, and whiskers as boxes and lines, respectively. Significance was assessed with two-tailed Wilcoxon nonparametric test (middle panels). Virus titers (q-PCR, right panel) were determined after 5 days of incubation with the drug (18 d pi). Q-PCR data show the means from two technical replicates. Two independent experiments gave similar results. Scale bar, 100 µm. c xCELLigence impedance plots showing HDF-TERT cell viability upon dl309 or dl309-63/195 and AdV-C5 or AdV-C5-XBP1s-mut infections (MOI 200). Cells were seeded on xCELLigence E-16 plate, and impedance readout for cell viability was measured live at 15-min intervals. Data show the means ± SD from three technical replicates. Two independent experiments gave similar results (left row). Experimental conditions and virus amount were as in panel b Genome copy numbers of virions released to the supernatant from the same experiment. Data show the means from two technical replicates. Two independent experiments gave similar results (middle row). Representative phase-contrast images of parallel samples imaged live for dl309 or dl309-63/ 195 infections (images on the right side, scale bar, 200 µm). d Schematic model depicting AdV infection under the control of a five-component feedforward loop.
(1) The immediate early E1A protein transactivates early promoters. including the E3 and E4, giving rise to the 19K glycoprotein (2). Activation of IRE1α by 19 K increases XBP1s mRNA, and XBP1s protein (3), which translocates into the nucleus, and binds to the E1A enhancer/promoter (e/p) of the episomal viral genome (4). Binding of XBP1s to the E1A-e/p increases the E1A levels (5), which enhances output from the E3 promoter, enhances the 19K levels, and maintains a feedforward loop supporting viral persistence and lytic infection.
mammalian IRE1α locus (NCBI identifier NC_000017.11 region 64039142-64132469), and the resulting mutations were identified. As IRE1α is triploid in HeLa cells 80 , sequencing of the clone used in this study revealed three distinct mutations, namely insertion of A, frameshift deletion of 23 nt, and in-frame deletion of 15 nt (Supplementary Fig. 1b).
Lentivirus production, cloning, and transduction. The cDNA inserts for human IRE1α 43 , human XBP1s (Addgene 63680, NCBI Gene ID 7494), mouse XBP1s (NCBI Gene ID 22433), AdV-C5 E1A enhancer, promoter, and E1A expression cassette (1-1701 genomic region from AdV-C5 genome NCBI AC_000008.1) were cloned into PLVX-IRES-Puro (Clontech 632183). Lentiviral vectors were produced in HEK293T cells by transfecting lentiviral construct, pCMVR8.91-Gag-Pol and pVSV-G (Clontech) 81 . Relative titers of different lentiviruses were determined using dose-response of puromycin 72 h post transduction. For transduction of cells, predetermined amounts of lentivectors that gave good expression of the target gene were added to cells at the time of seeding, and transduction was performed for 48 h. Cells were either collected for analyses of the transgene expression levels by western blotting or for chromatin immunoprecipitation or infection assays, as indicated.
For the ectopic expression of AdV-C2 and D8 E3-19K (Genbank AB448767.1) in the absence of other viral factors, an expression system was developed where codon-optimized AdV-C2 E3-19K construct after the N-terminal calreticulin signal sequence (Css) was synthesized by GeneArt/Life Technologies and inserted into pSG5 expression vector 82 (pSG5-19K-CO Css). For insertion of AdV-D8 E3-19K in the same construct, D8 E3-19K ORF after the wild-type signal sequence was amplified by PCR from purified vDNA and substituted with the C2 E3-19K sequence in pSG5-19K-CO Css plasmid after the Css sequence using NEBuilder ® HiFi DNA assembly cloning kit (New England Biolabs). Lumenal domain (LD) 19K-LD-Flag-HDEL mutants of C2 and D8 19K were made by the deletion of the transmembrane and cytoplasmic domains of E3-19K and insertion of Flag-HDEL domain by site-directed mutagenesis using the Q5 ® Site-Directed Mutagenesis Kit (New England Biolabs). The correctness of all the constructs was verified by sequencing.
Transfection and infection. For knockdown experiments, siPools TM oligos (siTools Biotech GmbH, Martinsried, Germany) were used. siPools have minimal off-target effects due to subthreshold concentrations of individual dsRNAs 83 . siRNAs were mixed with 9.8 µl of Opti-MEM medium (Invitrogen) and Lipofectamine RNAiMax reagent (0.2 µl/well, Invitrogen), and incubated at room temperature for 5 min in a 96-well plate (Greiner). Ten thousand HeLa-ATCC or HDF-TERT cells were added in 90 µl per well of DMEM medium supplemented with 10% FCS of the 96-well plate and incubated at 37°C in a 5% CO 2 environment for 48 h. After this, continuous infection with AdV-C5 or AdV-C2-dE3B_mCherry was carried out for 24, 48, or 72 h. Multiplicity of infection (MOI) was calculated from the absorbance of purified particles at 260 nm in the case of AdV-C5_dl309 or dl309_Δ63-195, by the infectious titer determined in A549 cells, or as indicated.
For HeLa cells, the final concentration of 10 nM, and for HDF-TERT cells, 20 nM of siPools were transfected, based on knockdown efficiency. For the rescue of infection in knockout cells, cells were transduced with lentivectors expressing human XBP1s or IRE1α for 48 h, incubated with indicated viruses for 24 h followed by fixation and staining with protein VI or E1A (M58). Control cells were transduced with lentivectors devoid of the transducing gene. Further analysis was done with high-throughput microscopy, western blotting, or quantitative PCR as described below.
For transfection of split-GFP constructs in HEK293T cells, 100 ng of each construct per well of the 96-well plate was mixed with 9.8 µl of Opti-MEM medium (Invitrogen) and Lipofectamine 2000 reagent (0.2 µl/well, Invitrogen), and incubated at room temperature for 5 min after directly spotting on a 96-well plate (Greiner). Ten thousand HEK293T cells were added in 90 µl of DMEM medium supplemented with 10% FCS per well of the 96-well plate and incubated at 37°C at 5% CO 2 for 48 h. Cells were fixed 48 h post transfection and stained with anti-E3-19K (3A9) and anti-Flag antibodies followed by imaging and analysis as described in the section called high-throughput imaging and measurement of infection. Representative images of split-GFP constructs for Fig. 4b were taken with a confocal SP5 microscope (Leica) and maximal projection images are shown.
For Neon transfection (Thermo Fisher Scientific) of E3-19K constructs in HeLa-ATCC cells, 5 µg of the plasmid DNA was mixed with Resuspension buffer (Neon) at a density of 0.5 × 10 7 cells/ml and filled in the Neon Tip (100 µl of plasmid plus buffer) attached to the Neon Pipette. Neon Tube was filled with electrolytic buffer and attached to the Neon Pipette Station. The cells were electroporated at 1005-V pulse voltage, at 35-ms pulse width, and two pulses. Following this, the cells were resuspended in antibiotics-free DMEM medium supplemented with serum (FCS) and seeded on the 6-well plate for RNA extraction or western blotting or 96-well plate for immunofluorescence staining. Twenty-four hours after transfection, cells were either collected for RNA extraction or fixed and DAPI stained for the calculation of the percentage E3-19K-expressing cells.
XBP1-splicing assay. Total RNA from HeLa-ATCC and HDF-TERT cells was lysed in Trizol Reagent (Invitrogen) and extracted using the Direct-zol RNA MicroPrep kit (Zymo Research). The cDNA synthesis was carried out with MMLV RT (Promega, M170) using Oligo dT 15 mer primers (Promega, C1101). PCR amplification for XBP1 gene was performed, and products were digested with PstI and treated with T7 Endonuclease (NEB, E3321) to get rid of the hybrid XBP1s-XBP1u bands ( Supplementary Fig. 1f). The DNA bands were resolved in 2% agarose gel and images were taken with GeneSnap (Syngene).
IRE1α and PERK mobility shift assay. Activation of PERK signaling arm of UPR was checked using the electrophoretic mobility shift assay. Hyperphosphorylated forms of PERK were detected by sufficiently resolving the 1-h DTT-(10 mM) treated cell lysates in 4-20% gradient gels (Biorad), and detecting the protein using anti-PERK antibody (Cell Signaling). For separating the phosphorylated from hypophosphorylated forms of IRE1α, 6% polyacrylamide gels containing 25 µM Phos-tag (Fujifilm WAKO) and ZnCl 2 was used. All buffers were prepared as described by the manufacturer (https://www.igz.ch/downloads/16079/phos-tagtm_sds-page_guidebook_11.pdf), and Phos-tag-bound proteins resolved on the polyacrylamide gels were incubated with 10 mM EDTA for 10 min prior to transferring them to PVDF membranes to reduce the adverse effects of Zn 2+ on protein transfer efficiency.
Immunoprecipitation and immunoblotting. HeLa-ATCC, Flag-IRE1α-MEFs, or 293T cells were seeded on 10-cm dishes. Twenty-four hours later, cells were infected with AdV-C2 (MOI 100 for HeLa-ATCC or twofold higher for Flag-IRE1α-MEFs) at 37°C, 5% CO 2 for 24 h. Following this, the medium was aspirated and washed with ice-cold PBS several times. Cells were scraped in 600 µl of DMEM and collected in 1.5-ml tubes. Cells were pelleted and resuspended in 600 µl of icecold IP lysis buffer (Tris-HCl, pH 8.5, NaCl 150 mM, MgCl 2 1 mM, EDTA 1 mM, Nonidet P-40 0.5%, and protease inhibitor cocktail) under agitation at 4°C for 20 min. The cells were centrifuged at 16,000 × g at 4°C for 10 min, and the supernatant was collected. Fifty microliters of the cell lysates were kept for an input control sample and the rest was precleared with protein-A sepharose beads (Abcam, ab193256). Precleared lysates were incubated with anti-3A9 84 , anti-Flag (Sigma, F7425), or an equivalent amount of IgG control overnight at 4°C. The next day, 30 µl of Protein-A sepharose beads were added per sample and incubated at 4°C for 1 h with agitation. Beads were spun down, the supernatant removed, and beads washed in IP lysis buffer several times. After this, the beads were mixed with 2× SDS lysis buffer (10% w/v), boiled at 95°C for 5 min, and the supernatant was collected. Samples were resolved with polyacrylamide gel electrophoresis after addition of DTT (50 mM) for anti-Flag and without DTT for anti-3A9 immunoblotting. For immunoprecipitations of Flag-IRE1α from Flag-IRE1α-expressing MEFs and lumenal domain of IRE1α-expressing HeLa cells, the ratio of BiP to IRE1α in the IP fraction was normalized to their respective levels in the lysate.
Tripartite split-GFP protein interaction assay. The tripartite split-GFP complementation assay was modified based on earlier protocols 85,86 . For the construction of tripartite split-GFP plasmids, lumenal domains of E3-19K from AdV-C2 or D8, and Flag-tagged human IRE1α or PERK was PCR amplified (Supplementary Table 2) and inserted into pcDNA 3.1 plasmid having the GFP11 and 10 domains, respectively. The rest of the GFP1-9 and GFP1-10 domain was inserted after Css at the N terminus and ER retention signal HDEL at the C terminus and cloned in pcDNA 3.1. For full-length split-GFP constructs shown in Supplementary Fig. 4b, IRE1α-GFP11 and 19K-GFP10 pair was chosen due to the better signal:noise ratio of this pair. For IRE1α, GFP11 domain was inserted at the linker region after the transmembrane domain of IRE1α. The rest of the GFP domains 1-9 were cloned in pcDNA 3.1. One hundred nanograms of all the plasmids were transfected in HeLa or 293T cells using Lipofectamine transfection method. Forty-eight hours later, cells were fixed with 3% paraformaldehyde (PFA) and stained with anti-Flag and anti-E3-19K (3A9) antibodies and imaged with high-throughput and confocal microscopes to measure the expression levels. The number of cells was determined with DAPI staining. Total tripartite split-GFP puncta were identified using custom-made CellProfiler script (Supplementary  Table 4) and plotted using JMP 13.
Virus titer estimation. Long-term persistent AdV-C5-infected HDF-TERT cells were washed with PBS and collected after scraping of the cells from the plate. DNA extraction was performed with Qiagen blood and tissue kit (Qiagen, #69504), and viral genome copy numbers were determined using quantitative PCR with primers on the AdV E1A enhancer region (Fwd 5′-GGTGGAGTTTGTGACGTGG-3′ and Rev 5′-CGCGCGAAAATTGTCACTTC-3′). First the relative abundance of E1A templates was calculated using ΔΔCt method and normalized to the cellular PDK1 gene (see Supplementary Table 2). Afterward, absolute viral gene copy numbers of one of the infected samples were calculated with an E1A standard curve using a plasmid DNA containing the E1A promoter and enhancer region followed by relating it to the relative fold difference derived from the ΔΔCt method above. Details of the quantitative PCR are mentioned in the section about quantitative PCR.
Establishment of persistent AdV infection. HDF-TERT cells were incubated with IFN-γ (500 IU/ml) for 24 h prior to infection. Infection with AdV-C5 (MOI 25) was performed in the presence of IFN-γ for 14 h; cells were washed and incubated with IFN-γ. For establishment of long-term persistence, the medium and fresh IFN-γ were replaced every 5 days until the end of the experiment 36 .
Promoter activity assays. The AdV-C5 E1A gene under the E1A enhancer and promoter elements (genomic region 1-1701 of AC_000008.1) were cloned in pLVX-IRES-Puro lentivector (Clonetech) and used to transduce HeLa-ATCC and HDF-TERT cells along with empty and human XBP1s lentivectors. Forty-eight hours post transduction, 4µ8C (100 µM) or IFN-γ (500 IU) were added to the cells. Seventy-two hours post transduction, cells were fixed with PFA and stained with anti-E1A (M58). The percentage of E1A-expressing cells was determined using custom-made CellProfiler script and data analyzed with Knime software.
High-throughput imaging and measurement of infection. After fixation and staining, cells were imaged with high-throughput wide-field microscope (Molecular Devices IXM-XL). DAPI staining was used to segment the nuclear mask for the cells, and the intensity of E1A (M58 and 73) and protein VI was measured over this mask using CellProfiler 87 . For the reticular staining of split-GFP and E3-19K staining, nuclear mask was expanded by ten pixels, and E3-19K intensity was measured over this ring-shaped object. The separation of mCherry, E1A, protein VI, or E3-19K-expressing and nonexpressing cells was calculated with KNIME, and graphs were plotted either in JMP version 13 or GraphPad Prism version 8.
Western blotting. At the time of anlaysis, cell lysates were collected in lysis buffer (0.2 ml of 200 mM Tris, pH 8.8, 20% glycerol, 5 mM EDTA, 50 mM DTT, 5% SDS, and 0.02% bromophenol blue) by boiling at 95°C and shearing through G21 needles (Sterican). Proteins were resolved in polyacrylamide gels and transferred to a polyvinyldifluoride membrane (Amersham). Blocking of unspecific proteinbinding sites of the membrane was done with 5% milk powder. Primary antibody incubation was performed at 4°C, followed by horseradish peroxidase (HRP)tagged secondary antibody incubation at room temperature. Protein bands were visualized by incubation with chemiluminescent reagent (Amersham) and the Amersham Imager 600.
Quantitative rt-PCR. Cells were lysed in Trizol Reagent (Invitrogen) and total RNA extracted using the direct-zol RNA MicroPrep kit (Zymo Research). cDNA synthesis was carried out with MMLV RT enzyme (Promega, M170) using Oligo dT 15 mer primers (Promega, C1101). The amplification was done using SYBR Green JumpStart Taq ReadyMix in Applied Biosystems Quant Studio 3 Real-Time PCR System with primers listed in Supplementary Table 2. For ChiP experiments, the relative quantification procedure of the Pfaffl method was used to convert the average Ct values for each sample to relative fold-change information 88 .
xCELLigence impedance measurements. For measuring cell toxicity of AdV and chemicals, the xCELLigence impedance measurement (Roche Applied Science and ACEA Biosciences) was used 59,52 . For AdV toxicity in lytic or persistent infections, cells were seeded on E-16 plates for 48 h, followed by incubation with IFN-γ (persistent infection) for 24 h and incubation with AdV at 37°C for 1 h. Cells were either further incubated with IFN-γ (persistent infection), or with normal medium (lytic infection), and cell index measurements were recorded live at intervals of 15 min. For assessing the effects of 4µ8C on cell viability, cells were seeded and incubated with 4µ8C for 7 days followed by drug washout.
Click chemistry and visualization of single viral DNA dots. HeLa wild-type and I-KO cells were seeded on alcian-blue-coated coverslips and infected with EdClabeled AdV-C5 virus at 37°C for 1 h, followed by washout of unbound virus and further incubation at 37°C for 1.5 h. Cells were fixed with formaldehyde and analyzed for capsid-free vDNA by copper-catalyzed click chemistry 39 . Viral capsids were stained with antihexon antibody (9C12) and samples imaged with an SP8 confocal microscope (Leica) 52 . Maximal projections of signals were analyzed with CellProfiler where DAPI staining was used to identify a nuclear mask together with overexposed hexon signals to segment the cell boundaries. Viral capsids, vDNA over the nucleus and in the cytoplasmic area was counted, and the distribution of capsid-free vDNA over nuclei and the cytoplasm calculated.
Prediction of SP1-binding sites in the E3 promoter. An overall consensus around the GC-box core element GGGCGG of SP1-binding sites can be predicted at http:// tfbind.hgc.jp/ 89 . Accordingly, up to 34 putative binding sites around the GC box were predicted in the E3 promoter, and 29 of them have a predicted consensus match for SP1 binding.
Quantifications and statistical analysis. Statistical analyses were performed using JMP version 13. For statistical analysis used in Figs. 5a and 6c, pairwise Wilcoxon nonparametric tests were performed.
We declare that no data have been excluded from experimental replicates. Random sampling in Figs. 6c and 7b as well as S4 d was done to keep an even number of cells analyzed between samples.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data supporting the findings of this study are available within the article and its Supplementary Information files, or are available from the authors upon request. The source data underlying Figs. 1-3 and 5a, and Supplementary Figs. 1e, 2b-e, 3a, b, d-i, 4a, and 5a are provided as Source Data files.