Transcriptional Regulation in Ebola Virus: Effects of Gene Border Structure and Regulatory Elements on Gene Expression and Polymerase Scanning Behavior

ABSTRACT

The highly pathogenic Ebola virus (EBOV) has a nonsegmented negative-strand (NNS) RNA genome containing seven genes. The viral genes either are separated by intergenic regions (IRs) of variable length or overlap. The structure of the EBOV gene overlaps is conserved throughout all filovirus genomes and is distinct from that of the overlaps found in other NNS RNA viruses. Here, we analyzed how diverse gene borders and noncoding regions surrounding the gene borders influence transcript levels and govern polymerase behavior during viral transcription. Transcription of overlapping genes in EBOV bicistronic minigenomes followed the stop-start mechanism, similar to that followed by IR-containing gene borders. When the gene overlaps were extended, the EBOV polymerase was able to scan the template in an upstream direction. This polymerase feature seems to be generally conserved among NNS RNA virus polymerases. Analysis of IR-containing gene borders showed that the IR sequence plays only a minor role in transcription regulation. Changes in IR length were generally well tolerated, but specific IR lengths led to a strong decrease in downstream gene expression. Correlation analysis revealed that these effects were largely independent of the surrounding gene borders. Each EBOV gene contains exceptionally long untranslated regions (UTRs) flanking the open reading frame. Our data suggest that the UTRs adjacent to the gene borders are the main regulators of transcript levels. A highly complex interplay between the different cis-acting elements to modulate transcription was revealed for specific combinations of IRs and UTRs, emphasizing the importance of the noncoding regions in EBOV gene expression control.

IMPORTANCE Our data extend those from previous analyses investigating the implication of noncoding regions at the EBOV gene borders for gene expression control. We show that EBOV transcription is regulated in a highly complex yet not easily predictable manner by a set of interacting cis-active elements. These findings are important not only for the design of recombinant filoviruses but also for the design of other replicon systems widely used as surrogate systems to study the filovirus replication cycle under low biosafety levels. Insights into the complex regulation of EBOV transcription conveyed by noncoding sequences will also help to interpret the importance of mutations that have been detected within these regions, including in isolates of the current outbreak.

INTRODUCTION

Ebola virus (EBOV) is a member of the filovirus family and causes a severe febrile disease in humans with high case fatality rates. The nonsegmented negative-sense (NNS) RNA genome of EBOV is 19 kb in length and contains seven genes (Fig. 1A). The transcription of all NNS RNA viruses follows a common mechanism: the viral polymerase gains access to the genome exclusively at the 3′ promoter region and successively transcribes the viral genes following a stop-start mechanism at each gene boundary, which results in the production of discrete mRNAs from each gene (reviewed in references 1 and 2). While moving along the genome template toward the 5′ end, the polymerase scans for the gene start (GS) and gene end (GE) signals of each gene. At the gene junctions, following termination of a transcript, the polymerase is thought either to progress to the GS of the next gene to reinitiate mRNA synthesis or to dissociate from the template. As it can reenter the genome only at the 3′ promoter, this process inevitably leads to the more frequent synthesis of genes located proximal to the promoter and a gradient in transcript abundance. It is presumed that filoviruses follow this stop-start model, as transcription gradients have been observed in both EBOV- and Marburg virus-infected cells (3–5).

FIG 1.

Scheme of EBOV gene borders, bicistronic minigenomes, and transcribed mRNAs. (A) EBOV genome showing the diverse gene borders. Gray boxes, ORFs; white boxes, UTRs flanking the ORFs; white triangles, GS signals; black bars, GE signals; le, leader; tr, trailer; gray lines, IRs (IR lengths are shown above each IR); overlap, overlapping gene borders; oval dotted lines, the gene borders examined in this study. (B) Sequences and structures of Bicis containing the gene borders between the NP and VP35 genes (Bici NP/VP35-wt IR 5nt; top) and the VP35 and VP40 genes (Bici VP35/VP40-wt overlap; bottom). Regulatory regions are illustrated as described in the legend to panel A. The conserved pentamer shared by the overlapping VP40 GS and VP35 GE signals is highlighted in bold. Bicis and sequences are shown in negative-sense orientation (not to scale). The Bicis differ only in the gene border regions and the 3′ and 5′ UTRs flanking the gene borders; thus, only these regions are shown for Bici VP35/VP40-wt overlap. The three different mRNAs transcribed from each Bici template as well as their calculated length [excluding the lengths of the poly(A) tails] are shown underneath each Bici. rth-mRNA, readthrough mRNA.

The GS and GE signals that flank each EBOV gene are highly conserved (6–10). GS signals direct the viral polymerase to initiate mRNA synthesis and to cotranscriptionally cap and methylate the mRNA's 5′ end, while the GE signals mediate polyadenylation and the termination of transcripts, as best characterized for vesicular stomatitis virus (VSV) and Sendai virus (11–16). A stretch of uridine (U) residues present in all GE signals of NNS RNA viruses is reiteratively copied by a stuttering polymerase, which results in the synthesis of the poly(A) tail (17). In the EBOV genome, 6 of the 7 GE signals are highly conserved and differ mainly in the number of uridine residues, ranging from 5 to 6 nucleotides (nt) (7, 8, 10).

In contrast to the GE and GS signals, the EBOV gene borders are highly variable. There are two 5-nt-long intergenic regions (IRs), one exceptionally long IR of 144 nt, two overlapping gene borders without an IR, and one gene border that contains both a 4-nt-long IR and a gene overlap due to the presence of two GE signals (Fig. 1A). Within the NNS RNA viruses, a broad spectrum of gene border configurations has evolved. These configurations range from gene borders with variable IRs, as found in the respiratory syncytial virus (RSV) and rabies virus genomes, to gene borders with highly conserved, short IRs strictly of 2 or 3 nucleotides in length, as in the VSV and Sendai virus genomes. For the latter, it was shown that sequence elements of the GE signals and IRs are essential not only for termination but also for efficient reinitiation at the downstream GS signal. This suggests a functional overlap of the signals that guide the polymerase during transcription at highly conserved gene borders (18–21). Consequently, changes to the IR sequence or length have dramatic effects on gene expression in these viruses (reviewed in reference 1). In contrast, previous analyses with EBOV showed that the IRs are not essential for transcription of the adjacent genes (6, 22). However, specific truncations of the 144-nt-long IR to 10, 20, or 30 nucleotides significantly inhibited downstream mRNA synthesis, while shorter or longer IRs did not result in substantial attenuation (6). This finding is in contrast to reports for other viruses with variable IRs showing that either IR length did not influence downstream gene expression (23–25) or elongation of the IR proportionally reduced reinitiation efficiency (26). As the previous results with EBOV were solely based on an analysis of the VP30/VP24 gene border containing an exceptionally long IR, the question as to whether IR length-dependent inhibition of downstream gene expression extended to gene borders with naturally short IRs arose.

Adding to the complexity of cis-acting signals at the gene borders, the EBOV genome contains three overlapping genes in which the GE signal of the upstream gene directly overlaps with the GS signal of the downstream gene (Fig. 1A, overlap). Intriguingly, all filoviral overlaps show exactly the same structure (Fig. 1B, bottom). The GS of the downstream gene precedes the GE of the upstream one, and the overlap is restricted to a pentameric sequence motif, 3′-UAAUU, which is part of both the GE and GS signals (6, 7, 22). This makes filoviruses exceptional, as, to our knowledge, other NNS RNA viruses contain at most one overlapping gene border in which the much longer overlapping region is not limited to the transcription signals (27–31). It is unclear if the direct overlaps unique to filovirus are recognized differently from the way in which IR-containing gene borders are recognized.

Another characteristic feature of EBOV genes is unusually long untranslated regions (UTRs) flanking the open reading frames (ORFs) at both the 3′ and 5′ ends (Fig. 1A). For clarity, we point out that the orientation of the UTRs is indicated in the negative-sense orientation and not in the mRNA orientation throughout this article. The EBOV 3′ UTRs are, on average, about 10 times longer than their counterparts in VSV, the prototypical NNS virus, and the 5′ UTRs are about 5 times as long. The function of these long UTRs remains largely elusive, although there is some evidence that they play a role in transcription regulation and translational control (3, 6, 22).

In this study, we analyzed EBOV gene borders of divergent structure using bicistronic minigenomes (Bicis) containing gene borders with overlapping genes (the VP35/VP40 gene border), a short IR (the NP/VP35 gene border), and an exceptionally long IR (the VP30/VP24 gene border). We aimed to disentangle the different levels of regulation mediated by the IRs, the UTRs, and the GE signals and shed light on the modulation of gene expression at overlapping gene junctions. Our data show that the filovirus gene borders containing conserved, short overlaps are recognized according to the stop-start mechanism, similar to the mechanism of recognition of IR-containing gene borders. We further provide evidence for multiple levels of gene expression control in EBOV, with the UTRs being crucial regulators of transcript levels. Gene expression is further modulated by the upstream IR length through a gene border-independent mechanism. Additionally, the length of the GE uridine stretch fine-tunes transcription activity. Our data also suggest a complex interplay between the different cis-acting regulatory elements for specific combinations of IR length and surrounding UTR sequences.

MATERIALS AND METHODS

Cell lines and viruses.

VeroE6 cells and the human hepatoma cell lines Huh7 and Huh-T7 were grown in Dulbecco's modified Eagle medium (DMEM) containing penicillin (50 U/ml) and streptomycin (50 μg/ml) (P/S) supplemented with 10% fetal bovine serum (FBS). The Huh-T7 cell line (kindly provided by V. Gaussmüller, University of Lübeck, Lübeck, Germany) constitutively expresses the T7 RNA polymerase. To maintain expression of the T7 RNA polymerase, the culture medium for Huh-T7 cells was additionally supplemented with 1 mg/ml of Geneticin (32). BSR-T7/5 cells constitutively expressing the T7 RNA polymerase (kindly provided by K. K. Conzelmann, Max von Pettenkofer Institute and Gene Center, Munich, Germany) were cultured as described in reference 33.

Ebola virus, species Zaire ebolavirus, Mayinga variant (GenBank accession number AF086833), was propagated in VeroE6 cells. Virus titers were determined by a 50% tissue culture infectious dose (TCID₅₀) assay. All work with infectious virus was performed under biosafety level 4 conditions at the Institute of Virology, Philipps University of Marburg, Marburg, Germany.

RNA isolation and Northern blot analysis.

Total RNA of transfected or infected cells was isolated using an RNeasy kit (Qiagen). Polyadenylated RNA was purified using oligo(dT) cellulose as described in reference 34 or using a MicroPoly(A)Purist kit (Ambion). Purified mRNA was analyzed by Northern hybridization using digoxigenin-labeled, negative-sense riboprobes directed against the chloramphenicol acetyltransferase (CAT) gene, as described previously (34). For the analysis whose results are presented in Fig. 5D, the protocol was adjusted as follows. The RNA samples were transferred onto a Biodyne B membrane and hybridized with a digoxigenin-labeled riboprobe targeting the CAT gene. For detection, the blots were incubated with a biotinylated antidigoxigenin antibody (Abcam) followed by Streptavidin-IRDye 800 CW (Rockland Immunochemicals) and scanned with an Odyssey imager (LI-COR). All data shown for a given experiment were taken from the same blot and exposure time, while lanes not relevant for this study were excised. For RNA quantification, purified mRNA was measured using a spectrophotometer, and equal RNA amounts were loaded onto the gel prior to Northern hybridization. The mRNA bands of the scanned blots were quantified with Quantity One software (Bio-Rad Laboratories).

FIG 5.

Effects of IR length and sequence changes on reinitiation efficiency at the NP/VP35 gene border. (A) Scheme of wild-type and mutated gene borders in Bici NP/VP35. Sequences are shown in negative-sense orientation (3′ to 5′). Substituted or inserted nucleotides in the IRs of the respective mutants are underlined. Deleted nucleotides are marked by a hyphen. (B to H and J) The Bicis shown in panel A were analyzed as described in the legend to Fig. 2C. Representative results of at least three independent experiments are shown. Minigenome activity was determined by Northern hybridization (B, D, F, H) and quantitative CAT assay (C, E, G, J). CAT activity (+SD) was normalized to that of the wild-type Bici (wt IR 5nt; black column). The CAT expression level of the mutant Bicis was compared against the value for the wild type using a one-sample t test (***, P < 0.001; **, P < 0.01; *, P < 0.05). (B and C) Deletion of the IR; (D and E) substitution of the IR; (F and G) truncation of the IR; (H and J) elongation of the IR. (I) Densitometric quantitation of the mRNA 2/mRNA 1 ratio from Northern blot analyses. The mRNA ratios (+SD) for the elongation mutants (a representative blot is shown in panel H) were normalized to the ratio for wild-type Bici (wt IR 5nt; black column). Expression levels from each Bici were compared against the values from the wild type using a one-sample t test (***, P < 0.001; **, P < 0.01; *, P < 0.05). Rel., relative.

Construction of Bicis.

Bici NP/VP35-wt IR 5nt, containing the wild-type (wt) NP/VP35 gene border (nt 2728 to 3128; Ebola virus, species Zaire ebolavirus, Mayinga variant; GenBank accession number AF086833) with a 5-nt IR, is identical to minigenome E-bici-1,2 described in reference 35. To clone the Bici VP35/VP40-wt overlap, the VP35/VP40 gene border (nt 4152 to 4478) was amplified by reverse transcription-PCR with primers flanked by BglII and BamHI restriction sites. The PCR fragment was cloned into the construct 3E-5E-ΔCAT (35), containing the first 300 nt of the CAT gene and the full-length CAT gene separated by a BglII restriction site that was used for insertion of the PCR fragment. Mutant Bicis were created by standard cloning techniques, including QuikChange site-directed mutagenesis (Stratagene) and fusion PCR (36). To move the VP35 GE signal downstream of the VP40 GS signal in Bici VP35/VP40, we inserted stretches of nucleotides including the sequence of the short NP/VP35 IR (3′-GAUUA) in the mutant with an 8-nt insertion (the ins 8nt mutant) as well as one (the ins 11nt mutant), two (the ins 21nt and ins 39nt mutants), or three (the ins 31nt mutant) XhoI restriction sites (CTCGAG), used for cloning. To reach the desired length of the insertions, random nucleotides were added (see Fig. 4A). In all mutants, the inserted IRs start with 3′-GA and end with UA-5′, which is consistent with the terminal nucleotides of the VP30/VP24 and NP/VP35 IRs.

FIG 4.

Transcription initiation at an upstream-located GS signal. (A) Schematic of wild-type and mutated gene border sequences in Bici VP35/VP40-wt overlap. Inserted nucleotides are underlined. To preserve both transcription signals in the insertion mutants, the overlapping pentameric sequence was duplicated (inserted nucleotides are in bold and underlined). Due to the insertions, the VP35 GE signal was moved downstream, creating overlaps from 8 to 39 nt in length and total overlap lengths ranging from 18 to 62 nt (left). (B) Northern blot analysis of Bici VP35/VP40 insertion mutants. Shown is a representative result of at least three independent experiments in BSR-T7/5 cells, performed as described in the legend to Fig. 2C. Lanes not relevant to this study (indicated by the dashed line) were excised. (C) 5′RACE analysis of mRNA 2 synthesized from Bici VP35/VP40. The sequence of the 5′ end of mRNA 2 was determined by 5′ RLM-RACE for Bici VP35/VP40-wt (top) and Bici VP35/VP40-ins 8nt (bottom) and is shown in negative-sense orientation. Red box, the VP35 GE signal, green box, the VP40 GS signal; black box, the inserted nucleotides; gray box, the sequence of the RNA adapter ligated to the 5′ end of the mRNA. The inner primer of the RLM-RACE kit used for this experiment likely misannealed at the VP40 GS, resulting in a mixture of fragments differing in their 3′ sequences. The sequence of the inner primer is shown in gray and underlined, and its putative annealing sites are indicated. (D) Transcripts synthesized at the VP35/VP40 gene border. Diagram showing a putative mRNA 2_short transcribed from mutated Bicis in which VP35 GE is separated from VP40 GS by more than 21 nt. The gene overlap created by insertion mutagenesis is shown in gray.

Transfection and infection of cells.

BSR-T7/5 and Huh-T7 cells grown to 60 to 70% confluence in 6-well plates were transfected using the FuGENE 6 (Roche Molecular Applied Science) or TransIT-LT1 (Mirus) transfection reagent following the manufacturers' recommendations. To analyze minigenome activity in EBOV-infected cells, Huh-T7 cells were transfected with 2 μg plasmid DNA of each Bici or the monocistronic minigenome 3E-5E (37), along with 1 μg of pCAGGS-T7 expressing the T7 RNA polymerase (kindly provided by T. Takimoto, St. Jude Children's Research Hospital, Memphis, TN, and Y. Kawaoka, University of Wisconsin, Madison, WI). At 12 h posttransfection, the transfected cells were infected with EBOV at a multiplicity of infection (MOI) of 5 TCID₅₀ units/cell and subjected to Northern blot analysis at 48 h or 72 h postinfection (p.i.).

To analyze the transcriptional activity of Bicis in minigenome assays, BSR-T7/5 or Huh-T7 cells were transfected with 1.0 μg pT/L_EBO, 0.5 μg pT/NP_EBO, 0.5 μg pT/VP35_EBO, or 0.1 μg pT/VP30_EBO, along with 1.5 μg of the respective Bici constructs (37). When Huh-T7 cells were used, 0.5 μg of pCAGGS-T7 was added. For analysis of CAT reporter activity, 0.3 μg of the pGL3-Control plasmid (Promega) expressing firefly luciferase was added as a transfection control. All transfections were adjusted to the same amount of DNA using the empty pTM1 vector backbone. As negative controls, plasmid pT/L_EBO was replaced by the same amount of pTM1 (indicated by −L in the appropriate figures).

Passaging of Bici-containing supernatants.

Supernatants of Bici-transfected and EBOV-infected Huh-T7 cells were harvested at 48 h or 72 h p.i. and cleared of cell debris by low-speed centrifugation. Approximately 2 × 10⁵ Huh-T7 cells were infected with 250 μl supernatant mixed with 250 μl DMEM. Following virus adsorption, 2.5 ml of DMEM plus P/S containing 2.5% FBS was added. The cells were harvested at 48 h p.i. and subjected to Northern blot analysis.

Quantitative CAT assay.

Approximately 6 × 10⁵ to 8 × 10⁵ BSR-T7/5 cells grown in 6-well plates were transfected as described above. At 2 days posttransfection, the cells were lysed and luciferase activity (representing transfection efficiency) was determined using a luciferase assay system (Promega). Quantitative CAT assays were performed with normalized amounts of cell lysates following the protocol of the Fast CAT Green (deoxy) chloramphenicol acetyltransferase assay kit (Molecular Probes) as described before (6).

RLM-RACE.

For a minigenome assay, BSRT7/5 cells were transfected with the Bici of interest, as described above. Total RNA was harvested with the RNeasy kit (Qiagen) for the detection of mRNA 2 or with the TRIzol reagent (Invitrogen) for the detection of a short, attenuating mRNA 2 transcript (mRNA 2_short). Poly(A) mRNA was purified with the MicroPoly(A)Purist kit (Ambion). Poly(A) mRNA (250 ng) was used to perform 5′ RNA ligase-mediated (RLM) rapid amplification of cDNA ends (RACE) using a FirstChoice RLM-RACE kit (Ambion). For the outer 5′ RLM-RACE PCR, gene-specific and 5′ RACE outer primers were used. Two microliters of the outer PCR product was used to amplify a nested PCR fragment with gene-specific and 5′ RACE inner primers. The PCR fragment was isolated via gel extraction (with a QIAquick gel extraction kit) and sequenced using the gene-specific inner primer. To detect mRNA 2_short, cDNA was synthesized using either random decamers or oligo(dT) as primers. From both cDNAs, we successfully amplified a fragment of mRNA 1 using the gene-specific primers designed for the amplification of mRNA 2_short. Since the mRNA 2_short-specific primers bind within the overlapping region, they also detect mRNA 1 (see Fig. 4A). This confirmed that mRNA isolation, reverse transcription, and PCR were successful and that the primers bound to the target sequence. However, despite considerable efforts, we were unable to amplify mRNA 2_short.

Statistical analysis.

For quantitative analyses of CAT activity or mRNA amounts, a one-sample t test or one-way analysis of variance (ANOVA) was performed using GraphPad Prism (version 5) software. Correlation analysis was performed using the same software. This analysis computes the value of the Pearson correlation coefficient (r). r values can range from −1 to +1, with positive values indicating a linear relationship between variables (e.g., variable x increases as variable y does).

RESULTS AND DISCUSSION

Structural differences in the gene borders do not significantly affect transcription profiles.

To analyze the effect of structural differences of the gene borders on transcriptional activity, we compared the mRNA synthesis at the NP/VP35 gene border containing an IR of 5 nt to that at the overlapping VP35/VP40 gene border. The gene borders of interest and parts of the adjacent UTRs were cloned into Bicis to allow a comparative analysis independently of their position in the viral genome (Fig. 1B). Within the Bicis, expression of two reporter genes, a truncated and, therefore, inactive form of the chloramphenicol acetyltransferase (CAT) gene and an intact version of this gene, is controlled by cis-acting elements located in the different gene borders (6, 35). Transcription of the Bicis by the EBOV polymerase results in the synthesis of three different transcripts, mRNA 1, mRNA 2, and readthrough mRNA, all of which can be detected simultaneously by Northern hybridization with a single CAT-specific probe (Fig. 1B). Use of a single probe to detect all mRNA species allows the direct comparison of mRNA amounts (38, 39). Readthrough mRNAs are bicistronic transcripts that are synthesized when the polymerase fails to terminate transcription at the GE signal of the gene border.

We previously showed that Bici NP/VP35 is able to be replicated and transcribed in EBOV-infected cells (6). To ensure that Bici VP35/VP40 was also accepted as a template for transcription and replication by the authentic EBOV polymerase complex, Huh-T7 cells were transfected with either Bici NP/VP35 or Bici VP35/VP40 and subsequently infected with EBOV. Both Bicis were recognized as a template for transcription, as shown by the synthesis of all three expected mRNA products (Fig. 2A). In addition, Bici VP35/VP40 was as efficiently replicated, packaged into virions, and passaged onto fresh cells as Bici NP/VP35, as indicated by the presence of Bici mRNAs in cells infected with Bici-containing supernatants from transfected and infected cells (Fig. 2B). The two analyzed gene borders showed similar transcription profiles overall, irrespective of whether the transcription signals were separated by a short IR or overlapped. This was also consistently the case when the Bicis were tested in an EBOV minigenome system (Fig. 2C). Readthrough transcription took place irrespective of the gene border structure, in line with previous observations in EBOV-infected cells (4, 6). This confirms that the close proximity of the GS and GE signals in the overlaps does not prevent polymerase readthrough. Although the overall low levels of readthrough mRNA in comparison to the levels of the much more abundant mRNA 1 and mRNA 2 did not allow reliable quantitative comparative analyses, Northern blot analyses consistently showed that the levels of readthrough mRNA produced at the VP35/VP40 gene overlap were less than the already low levels produced at the NP/VP35 gene border. However, the data presented below (see Fig. 6) suggest that the structure of the NP GE signal rather than the IR is responsible for the more abundant production of readthrough mRNA (see Fig. 6B). Slight differences in the relative abundance of the three mRNAs in the minigenome system compared to that in virus-infected cells were observed (Fig. 2A and C). This is not surprising, given that all virus proteins are present in EBOV-infected cells, while only those viral proteins required for replication and transcription are expressed in the minigenome system. Despite these differences, the mRNA expression pattern of the various analyzed Bicis was consistently similar across the systems used. Therefore, we are confident that the transcriptional effects observed in the minigenome system relate well to the regulatory mechanisms at play in infected cells.

FIG 2.

Transcriptional activity of Bicis containing an overlap or a short IR. (A) Transcripts synthesized from Bicis in EBOV-infected cells. Huh-T7 cells were transfected with plasmid DNA encoding the indicated Bicis along with an expression plasmid encoding the T7 RNA polymerase. At 12 h posttransfection, cells were infected with EBOV at an MOI of 5 or left uninfected. Poly(A)-positive mRNA was harvested at 48 h p.i. and analyzed by Northern hybridization using CAT-specific riboprobes. As a negative control, nontransfected cells infected with EBOV were used (lane 5). Shown is a representative result of three independent experiments. A short exposure (top) and a long exposure (bottom) of the X-ray film revealing readthrough mRNA (rth) are shown. mRNA products are labeled according to the labeling used in Fig. 1B. Lanes not relevant to this study (indicated by dashed lines) were excised. (B) Transcripts synthesized in cells infected with Bici-containing viral supernatants. Supernatants of cells transfected and infected as described in the legend to panel A were collected at 72 h p.i. and used to infect Huh-T7 cells. At 48 h p.i., poly(A)-positive mRNA was purified and analyzed as described in the legend to panel A. The experiment was repeated twice with similar results each time. (C) Transcriptional activity of Bicis in an EBOV minigenome system. Huh-T7 or BSR-T7/5 cells were transfected with plasmids carrying the Bici construct along with expression plasmids carrying the T7 RNA polymerase and the EBOV polymerase complex. Cells were harvested at 48 h posttransfection and subjected to Northern hybridization. Shown is a representative result obtained in Huh-T7 cells. The experiment was performed twice in Huh-T7 cells and twice in BSR-T7/5 cells. 1 and 2 to the left of the gels, mRNA 1 and mRNA 2, respectively; numbers to the right of the gels, RNA size markers (in thousands [k] of nucleotides).

FIG 6.

Swapping of IRs between gene borders VP30/VP24 and NP/VP35. (A to C) Swapping of the IRs of Bici NP/VP35-wt (IR 5nt) and Bici VP30/VP24-wt (IR 144nt). (A) (Top) Schematic showing the location of the analyzed gene borders within the EBOV genome; (bottom) gene borders in wild-type and mutant Bicis. White bars, UTRs of Bici NP/VP35; hatched bars, UTRs of Bici VP30/VP24. The NP GE signal is highlighted in bold. (B) Wild-type and mutant Bicis were analyzed by Northern hybridization in BSR-T7/5 cells as described in the legend to Fig. 2C, and a representative result of four experiments is shown. (C) Densitometric quantitation of mRNA amounts. The mRNA 2/mRNA 1 ratios of the intensities (+SD) measured for mRNAs for which the results are presented in panel B are shown. Values were normalized to the value for Bici NP/VP35-wt IR 5nt (NP/VP35 wt). White columns, values obtained for Bicis containing the NP/VP35 gene border; hatched columns, values obtained for Bicis containing the VP30/VP24 gene border. The mRNA 2 expression level from Bici NP/VP35 IR 144nt was compared against the level from Bici NP/VP35 wt using a one-sample t test (***, P < 0.001; **, P < 0.01; n.s., no significant difference). Differences between all other samples were analyzed for statistical significance by one-way ANOVA.

Transcription at EBOV gene overlaps follows the general stop-start mechanism.

We further analyzed recognition of the transcription signals at the VP35/VP40 gene overlap in the EBOV minigenome system (Fig. 3A). To examine whether the overlapping transcription signals are recognized individually by the viral polymerase or rather as a contiguous regulatory unit, we first disrupted the GS signal of the second gene (VP40 GS) by substitution mutagenesis, leaving the GE signal of the first gene (VP35 GE) intact (Fig. 3A, start mut). Northern hybridization revealed a complete loss of mRNA 2 due to the lack of reinitiation at the mutated VP40 GS signal (Fig. 3B, third lane). Since mRNA 1 was efficiently transcribed, it can be concluded that transcription termination occurs independently of transcription reinitiation at the gene overlaps. We also observed an increase in mRNA 1 levels relative to those for the wild type, suggesting that the polymerase initiated transcription of the first gene more frequently when reinitiation at the second gene was abrogated (compare the second and third lanes in Fig. 3B). In a second construct, we mutated the GE signal of the first gene (VP35 GE) without compromising the GS signal of the second gene (VP40 GS) (Fig. 3A, stop mut). This mutation led to the exclusive production of readthrough mRNA (Fig. 3B, fourth lane), indicating that transcription termination is a prerequisite for reinitiation at the overlapping gene border, as mRNA 2 synthesis could not be initiated, despite the presence of an intact GS signal when the overlapping GE signal was not functional. Our observations are in line with results described for IR-containing gene borders (11, 24). We thus conclude that the filovirus-specific gene overlaps are transcribed by the EBOV polymerase following the stop-start model postulated for IR-containing gene borders. This is also consistent with transcription at the overlapping M2-L gene border in the RSV genome. In agreement with our results, transcription initiation at the L GS was dependent on a functional M2 GE signal (40).

FIG 3.

Analysis of overlapping transcription signals in Bici VP35/VP40-wt overlap. (A) Scheme of wild-type and mutated gene border sequences in Bici VP35/VP40-wt overlap. The distance from the end of the VP35 GE signal to the beginning of the VP40 GS signal is shown as total overlap. Substituted nucleotides are underlined. Bici VP35/VP40 mutants contain either an inactive GS signal (start mut) or an inactive GE signal (stop mut). (B) The Bicis shown in panel A were tested in the EBOV minigenome system in BSR-T7/5 cells as described in the legend to Fig. 2C. Shown is a representative result of the Northern blot analysis from three independent experiments.

In the overlapping region of the RSV L and M2 genes, the GS is located substantially upstream of the GE. The L and M2 genes overlap for a total region of 68 nt, with 45 nt separating the 12-nt-long M2 GE signal and the 11-nt-long L GS signal (27). To reinitiate transcription, the viral polymerase has the ability to scan for upstream-located GS signals (40). In contrast, all filovirus overlaps are restricted to the conserved 3′-UAAUU pentamer that is shared by the overlapping GS and GE signals with a total overlap length of only 18 nt (Fig. 3A, total overlap). Because of the close proximity of the GS and GE signals, it is conceivable that the EBOV polymerase recognizes the GS signal while still sitting atop the GE signal. We therefore investigated next whether the EBOV polymerase is able to scan for upstream-located GS signals.

The EBOV polymerase possesses upstream template-scanning capabilities.

In a previous study, the pentamer shared by the GS and GE signals in the EBOV VP35/VP40 gene border was duplicated, thereby moving the VP35 GE signal directly downstream of the VP40 GS signal. The reporter activity of either gene in this construct was not substantially different from that in wild-type Bici (22). We extended this analysis and moved the VP35 GE signal further downstream of the VP40 GS signal by inserting increasing numbers (8 to 39 nt) of nucleotides (Fig. 4A). In these constructs, the EBOV polymerase has to scan backwards after terminating the transcription of mRNA 1 to access the upstream GS signal and initiate mRNA 2 synthesis. Northern blot analysis of the synthesized mRNAs showed the transcription of both mRNA 1 and mRNA 2 from Bicis in which the GS and GE signals were separated by 8, 11, 15, or 21 nt (Fig. 4B). This suggests that the polymerase has the ability to recognize GS signals located upstream of the GE. At an overlap of 21 nt, however (representing a total overlap of 44 nt), mRNA 2 levels seemed to be substantially reduced compared to those for the wild-type Bici. When the transcription signals were further separated by 31 nt or 39 nt (total overlap of 54 and 62 nt, respectively), mRNA 2 was not detected, even after prolonged exposure times (Fig. 4B). Note that the apparent increase of readthrough mRNA with mutants containing 15 inserted nucleotides or more was not consistent throughout the experiments.

To confirm that transcription of mRNA 2 was initiated at the translocated GS signal rather than at a suboptimal GS signal located downstream of the GE signal (18, 20, 41), we sequenced the 5′ end of mRNA 2 synthesized from Bici VP35/VP40-ins 8nt and the wild-type Bici. The 5′ RACE analysis revealed that mRNA 2 was initiated at the VP40 GS even when it was located upstream of the VP35 GE (Fig. 4C). In a previous study, we showed that the EBOV polymerase also has a pronounced ability to scan downstream (6). Together, our data demonstrate that the EBOV polymerase is able to scan the template bidirectionally at the gene borders to access the next GS signal. Upstream template scanning has also been observed for other NNS RNA viruses, including VSV, a virus with a genome that does not contain any gene overlaps, indicating that bidirectional scanning for GS signals is a common feature of NNS RNA virus polymerases (18, 40).

A possible explanation for the lack of mRNA 2 observed for gene overlaps exceeding 44 nt could be that the EBOV polymerase is not able to access a GS signal that is located more than 21 nt upstream of the GE signal. Although we cannot rule out a possible effect of the inserted sequences on the scanning activity of the polymerase, a sequence-specific effect seems unlikely, as scanning through IR sequences by the EBOV polymerase was largely unaffected by sequence alterations (Fig. 5D and E) (6).

Studies with RSV and VSV showed successful upstream scanning at gene overlaps spanning 90 and 200 nt, respectively (18, 40). Moreover, the VSV polymerase was able to scan equally well for GS signals moved as far as 200 nt up- or downstream of the corresponding GE signal (18). Given that the EBOV polymerase is able to access GS signals that lie more than 200 nt downstream of the GE signal (6) and its ability to move bidirectionally along the template RNA, it is puzzling that an upstream-located GS signal must be rather close to the GE signal to be recognized. Another possible explanation for the lack of mRNA 2 could be that the EBOV polymerase reinitiates transcription at the VP40 GS and prematurely terminates at the translocated VP35 GE signal, leading to the synthesis of a short, attenuating mRNA 2 transcript, mRNA 2_short (Fig. 4D). Due to its small size, mRNA 2_short would not be detected by our Northern blot analysis. Support for the latter hypothesis comes from observations with other NNS viruses, where shorter transcripts, similar to mRNA 2_short in Fig. 4D, were detected with gene overlaps of 68 nt (for RSV) or more than 66 nt (for VSV) (18, 27). Efficient termination of VSV transcription does not occur at a transcript length of less than about 51 nt, indicating that a minimal distance between the GS and GE signal is required (42). As for EBOV, the reduced level of mRNA 2 transcription from a template containing a 44-nt overlap (Bici VP35/VP40-ins 21nt) and the lack of transcription of mRNA 2 from templates with 54- and 62-nt-long overlaps (Bicis VP35/VP40-ins 31nt and -ins 39nt) suggest a similar length requirement for minimal transcription units. To screen for the transcription of short mRNA products from Bici VP35/VP40-ins 39nt, we used 5′ RACE analysis but were unable to detect mRNA 2_short transcripts. Failure to detect mRNA 2_short, however, does not rule out the possibility that mRNA 2_short is transcribed.

In conclusion, our data confirm that template scanning in an upstream direction is a conserved feature of NNS RNA virus polymerases. This ability is independent of the structure of their genomes and the requirement of this activity during transcription.

Recognition of the GS signal at the NP/VP35 gene border is inhibited by specific IR lengths.

Next, we investigated transcription reinitiation at EBOV gene borders in which the GE and GS signals are separated by a short IR. We recently showed that the reinitiation efficiency at the VP30/VP24 gene border containing the single long IR (144 nt) in the EBOV genome is regulated by specific IR lengths (6). It is not clear, however, whether the inhibitory effect of distinct IRs is exclusive for the VP30/VP24 gene border or reflects a more general mechanism that can also be observed for other gene borders. To examine the effect of IR length variation in a gene border naturally containing a short IR, we chose the NP/VP35 gene border, which has an IR of 5 nt. Several mutant Bicis bearing a shortened, substituted, or elongated IR were generated (Fig. 5A) and analyzed by Northern hybridization. In addition, expression of mRNA 2 was determined by a quantitative CAT reporter assay, which is considered a confirmatory assay because reporter gene expression relies on mRNA translation and, therefore, does not directly reflect transcriptional activity.

Deletion of the IR did not affect mRNA 1 levels, while we observed a strong decrease in mRNA 2 expression and CAT reporter activity (Fig. 5B and C). This confirms previous results indicating that neither the short IRs nor the long IRs are essential for EBOV transcriptional activity but regulate reinitiation rates (6, 22). Partial or complete substitution of the short IR had only a minor effect on reinitiation (Fig. 5D and E). This indicates that the IR sequence does not strongly affect downstream gene expression, consistent with our previous observations that the IR sequence is of minor importance for reinitiation efficiency (6). To investigate the impact of IR length on transcriptional activity, the IR was either shortened to 3 nt or gradually elongated to up to 50 nt by inserting sequences derived from the long IR in the EBOV genome (Fig. 5A). While truncation was well tolerated (Fig. 5F and G), elongation of the IR generally led to a decrease in mRNA reinitiation (Fig. 5H to J). As observed before with gene border VP30/VP24 (6), there was no linear correlation between IR length and reinitiation rates, as stepwise changes of the IR length did not result in concomitant alterations of reinitiation rates. Rather, some IR lengths were better tolerated than others. Consistent with our results from gene border VP30/VP24, IR lengths of 10 and 30 nt caused a significant reduction of the level of mRNA 2 reinitiation to about 14% of the wild-type level and a corresponding decrease in CAT activity, while IRs comprising 35 to 50 nt did not reduce mRNA 2 amounts and CAT activity to the same extent. Inhibition of transcription reinitiation by IRs of distinct lengths has not been observed for any other NNS RNA virus but seems to be a common theme for EBOV gene borders. Possible explanations for this phenomenon include changes in the NP encapsidation phase that move the GS signal to an unfavorable position where it is overlooked by the scanning polymerase or steric constraints that prevent recognition of the GS by the polymerase (6). In contrast to earlier results with gene border VP30/VP24, where an IR of 20 nt inhibited reinitiation as efficiently as a 10- or 30-nt-long IR, the inhibitory effect of the 20-nt-long IR with the same sequence was less pronounced at gene border NP/VP35 (Fig. 5H to J). This suggests differences in the regulation of gene expression between the two EBOV gene borders that do not depend only on IR length.

Surrounding sequences at the gene borders additionally influence the regulation of transcript levels.

We previously observed that Bici VP30/VP24 containing a 144-nt-long IR expressed less mRNA 2 than Bici NP/VP35 containing a 5-nt short IR (6). Surprisingly, truncation of the 144-nt-long VP30/24 IR to 5 nt did not increase the mRNA 2 levels, indicating that factors other than IR length play a regulatory role in reinitiation. The NP/VP35 and VP30/VP24 gene borders differ not only in the length of the IRs; they are also flanked by different UTRs and show slight differences in the GE signals (Fig. 6A; compare NP/VP35 wt with VP30/VP24 wt). To examine the effect of these additional regulatory regions on downstream gene expression, the IRs of Bici NP/VP35 wt (IR 5nt) and Bici VP30/VP24 wt (IR 144nt) were swapped, placing each in the context of the other gene border (Fig. 6A, NP/VP35 IR 144nt and VP30/VP24 IR 5nt). The two constructs were analyzed in the EBOV minigenome system and compared to their wild-type counterparts (Fig. 6B and C). As observed previously, placing the 5-nt-long NP/VP35 IR into gene border VP30/VP24 resulted in mRNA 2 levels that were not significantly different from wild-type mRNA 2 levels (Fig. 6C, third and fourth columns). The complementary experiment, i.e., insertion of the 144-nt-long IR into Bici NP/VP35, resulted in a moderate, albeit significant decrease in mRNA 2 levels compared to the mRNA 2 levels in wild-type Bici NP/VP35 (Fig. 6C, first and second columns). Intriguingly, both constructs containing the NP/VP35 gene border promoted high mRNA 2 levels, while the VP30/VP24 gene border resulted in lower levels of mRNA 2, largely irrespective of IR length. This indicates that the regions flanking the IRs, comprising the GS and GE signals and UTRs, play a crucial role in the regulation of EBOV gene expression.

While the GS signal sequences of the NP/VP35 and VP30/VP24 gene borders are identical, the GE signals differ by 1 nucleotide. The NP GE signal ends with a stretch of six uridine (U) residues, whereas the VP30 GE signal contains only five U residues (Fig. 6A). Since it was shown for other NNS RNA viruses, including RSV and parainfluenza virus 5, that the length of the U tract can influence transcriptional activity at the gene borders (43–46), we elongated the U stretch in the GE signal of Bici VP30/VP24 from 5 to 6 U residues (Fig. 6A, VP30/VP24 NP GE). While mRNA 2 expression levels were not significantly affected by the additional U residue (Fig. 6C; compare the third and fifth columns), the amount of readthrough mRNA was slightly elevated in all constructs containing the NP GE signal (Fig. 6B, first, second, and fifth lanes), suggesting that the additional U residue in the NP GE signal impairs transcription termination, consistent with our previous observations (Fig. 2).

In line with previous results, our data indicate that multiple genetic elements in the EBOV genome, including IRs, UTRs, and GE signal length, are involved in the regulation of EBOV gene expression (22).

Transcription regulation by IR length is conserved across different gene borders but is complemented by a complex interplay of the neighboring regulatory regions.

To untangle the regulatory effects of IR lengths and flanking UTRs, we compared the reinitiation efficiency at the NP/VP35 and VP30/VP24 gene borders containing identical or very similar IRs (Fig. 7A; IRs 50 and 51 nt and IRs 46 and 45 nt differ by a single nucleotide). The IR lengths in these Bici pairs ranged from 5 to 144 nt. For the VP30/VP24 gene border, we used previously determined mRNA 2/mRNA 1 ratios obtained from Bici mutants that varied in their IR lengths (see Fig. 5G in reference 6). These data were compared to the mRNA 2/mRNA 1 ratios from NP/VP35 Bicis containing the IRs indicated in Fig. 7A (Fig. 5I and 7B). In most cases, the relative mRNA 2 levels for the NP/VP35 Bicis were consistently about 2.5 times higher than those for the VP30/VP24 Bicis. To visualize regulatory effects that occurred independently of the UTRs bordering the IRs, the NP/VP35 and VP30/VP24 data sets were plotted on a graph with two differentially scaled y axes, adjusting the 2.5-fold difference in reinitiation efficiency observed with the two sets of constructs (Fig. 7B). Interestingly, comparison of the relative differences in reinitiation rates across the two gene borders revealed that the effect of IR length on reinitiation was remarkably similar for both gene borders. There were two notable exceptions. First, the wild-type minigenome Bici NP/VP35 wt IR 5nt seemed to contain a combination of elements promoting the most efficient reinitiation of all constructs (Fig. 7B, second column). Second, an IR of 20 nt did not lead to an equally strong reduction of mRNA 2 expression in Bici NP/VP35 compared to that in Bici VP30/VP24 (Fig. 7B, fifth and sixth columns). This indicates that, in addition to IR length, the flanking regulatory regions (GE signals and/or UTRs) play an important role in reinitiation, suggesting a complex interplay between the IRs and adjacent sequences during viral transcription.

FIG 7.

Surrounding sequences at the gene borders regulate transcriptional activity in conjunction with the intergenic regions. (A) Scheme of wild-type and mutant VP30/VP24 (top) and NP/VP35 (bottom) Bicis. White bars, UTRs of Bici NP/VP35; hatched bars, UTRs of Bici VP30/VP24. The sequence of the wild-type 144-nt-long VP30/VP24 IR is shown. Horizontal lines beneath the sequence indicate which IR sequences are present in the respective truncation mutants of Bici VP30/VP24 or in the elongation mutants of Bici NP/VP35. (B) Comparison of relative changes in mRNA 2/mRNA 1 ratios for the Bicis depicted in panel A. The absolute mRNA ratios of NP/VP35 Bicis (Fig. 5I) are shown in comparison to previous results obtained with Bicis containing the VP30/VP24 gene border (GB) with various IR lengths. The best match between mRNA levels from the two gene borders containing similar IRs was achieved when the y axis for gene border VP30/VP24 was rescaled by a factor of 2.5. Black columns, wild-type values. (Republished from reference 6.) (C) Correlation analysis of relative mRNA 2 expression levels (±SD) from Bicis containing similar IR lengths in the NP/VP35 and the VP30/VP24 gene borders. (D) Shortening of the NP GE U stretch in Bici NP/VP35 does not substantially influence reinitiation efficiency. Sequence variations of the GE signal in mutant Bicis NP/VP35 with an IR of 20 nt are shown. The last nucleotide of the NP GE signal was either deleted (NP/VP35 IR 20nt stop del) or replaced by an A residue (NP/VP35 IR 20nt stop mut). Substituted or deleted nucleotides are underlined. Sequences are shown in the negative-sense orientation. The deleted nucleotide is indicated by a hyphen. (E) Northern blot analysis of the Bicis shown in panel D. The experiment was performed using BSR-T7/5 cells as described in the legend to Fig. 2C. Shown is a representative result from three independent experiments. (F) Densitometric quantitation of mRNA amounts. The mRNA 2/mRNA 1 ratios of densitometric intensities (+SD) measured for the mRNAs whose results are shown in panel E and normalized to the ratio for the wild-type gene border (wt IR 5 nt; black column) are shown. mRNA 2 expression levels from each Bici were compared against the values for the wild type using a one-sample t test (***, P < 0.001; **, P < 0.01).

To rule out the possibility that the additional U residue in the GE signal of the NP/VP35 gene border accounts for the differences observed for the Bici pair containing an IR of 20 nt, the U residue was either removed or replaced by an A residue in construct Bici NP/VP35 IR 20nt (Fig. 7D). These mutations did not substantially alter the mRNA 2/mRNA 1 ratios compared to the ratio for Bici NP/VP35 IR 20nt (Fig. 7E and F).

Both the effect of IR length on reinitiation and the deviant behavior of wild-type Bici NP/VP35 and Bici NP/VP35 IR 20nt were quantitatively confirmed by correlation analysis. When we excluded the last two constructs from the analysis, a strong correlation between the reinitiation frequencies (mRNA 2/mRNA 1 ratios) of constructs with equal IR lengths at both gene borders was detected (r = 0.987, P = 0.0002; Fig. 7C). This indicates that IRs of similar lengths mediate equal reinitiation rates at both gene borders but that this effect is superimposed on the basic transcript level set by the specific UTRs surrounding the individual gene border. When the two deviant constructs were included in the analysis, we did not find a significant correlation (r = 0.659, P = 0.0755).

In conclusion, our data provide supporting evidence for at least four layers of gene expression control in EBOV, in addition to the transcript gradient. These are mediated by (i) the UTRs, (ii) the IRs, (iii) the transcription signals, and (iv) in some cases, such as the wild-type Bici NP/VP35 and Bici NP/VP35 IR 20nt, an interplay of these regulatory regions. The impact of these genetic elements on gene expression ranges from only the subtle effects observed for the construct with the additional U residue in the GE signal to the moderate effects observed for specific IR lengths and the strong effects observed for the UTRs. Since we used minigenomes encoding reporter genes for our analyses, we cannot rule out the possibility that EBOV transcription regulation is even more complex and also involves the coding regions of the viral genes.

In most cases and in accordance with previous results (6), IR length regulated downstream gene expression regardless of the gene border analyzed (Fig. 7B and C), while changes to the IR sequence did not substantially affect transcription of the downstream gene (Fig. 5D and E). Recent whole-genome sequencing analyses of the current EBOV outbreak variant Makona revealed that the noncoding regions are prone to mutations (47, 48). While the mutations were primarily mapped to the UTRs, those identified within the IRs exclusively led to sequence substitutions rather than to deletions or insertions, indicating a strong selection pressure to preserve IR length (47; UCSC Ebola genome browser at https://genome.ucsc.edu/ebolaPortal/). This supports our observation that the IR length is more important than the IR sequence.

The presence of numerous mutations within the UTRs is intriguing, in light of the regulatory role that UTRs play in the control of EBOV gene expression. Our results are in accordance with those of a previous study in which deletion of either the up- or downstream UTR in reporter Bicis containing the NP/VP35 or VP35/VP40 gene border with neighboring sequences resulted in dramatic changes of reporter gene expression (22). Importantly, deletion of the VP35 upstream UTR in a Bici reporter construct containing the NP/VP35 gene border not only resulted in a decrease of reporter activity from the transcription unit containing the deleted UTR but also enhanced expression of the reporter gene located upstream, even though the deleted UTR was not part of this transcription unit (22). This indicates an interaction of this UTR with cis-acting elements at the preceding gene border during transcription, similar to our observations with wild-type Bici NP/VP35 and Bici NP/VP35 IR 20nt, and suggests a complex interplay of the regulatory regions during EBOV transcription.

The regulatory role of the UTRs and IRs in combination with sequence variations in these regions, as detected in outbreak isolates, advise vigilance not only to focus on changes in protein-coding sequences but also to pay attention to changes in noncoding regions that potentially affect gene expression and, as a consequence, might influence virus pathogenicity. Our observations will be instrumental to inspire future research aimed at further unraveling the sophisticated strategies by which gene expression is regulated within the highly pathogenic filoviruses.