Dicistrovirus type 6 IRES diversity

To explore the diversity of Type 6 IRES in dicistroviruses, we compiled dicistrovirus genomes from the Earth’s Virome Project (Serratus.io), NCBI-annotated virus genomes, and metagenomic studies [38,41]. We used the RNA-dependent RNA polymerase (RdRP) protein sequences from known dicistroviruses (International Committee on Taxonomy of Viruses:ICTV) as a taxonomic barcode to scan these datasets in initially selecting viromes with an RdRP identity of at least >50% at the amino acid level compared to known dicistrovirus RdRP using Palmscan (Fig 1A). In total, we identified 64,245 transcripts containing dicistrovirus-like RdRP sequences. To eliminate redundant and short sequences, we clustered sequences at a 98% nucleotide identity and excluded sequences below 6,000 bases in length and with at least one ORF > 1000 nucleotides (i.e., > 333 codons), thus maximizing the identity of full-length dicistrovirus genomes, considering that dicistrovirus genomes are ~ 9,000 nt in length. This curation resulted in a final list of 4,704 dicistrovirus-like genomes used for subsequent analyses (S1 Data).

To identify genomes with Type 6 IRES-like structures (Fig 1B), covariance models of the six known Type 6 IRES subtypes were generated and used to scan the dicistrovirus genome database using the covariance model algorithm INFERNAL [42] (Fig 1A). A total of 3257 Type 6 IRES-like structures were identified (~33% of genomes), among these, 1912 IRESs were unique in their nucleotide sequence, while 1345 IRESs were found in multiple genomes (Fig 1C and S1 Data). Moreover, the bioinformatics pipeline identified previously-described subtype 6a and 6b IRESs including IGR IRESs from the genera Cripavirus (CrPV, Drosophila C virus (DCV), Nilaparvata lugens C virus (NLCV), Rhopalosiphum padi virus (RhPV), and aphid lethal paralysis virus (ALPV)), Triatovirus (Triatoma virus (TrV), Himetobi P virus (HiPV), Plautia stali intestine virus (PSIV), and black queen cell virus (BQCV), and Aparavirus (acute bee paralysis virus (ABPV), Kashmir bee virus (KBV), IAPV, Formica exsecta virus 1 (FEV1), TSV, Macrobrachium rosenbergii Taihu virus (MrTV), mud crab virus (MCV)) thus providing further confidence and validating this bioinformatics approach (Fig 1D and S1 Data). Of the 1912 unique IRES-like structures identified, the majority were IRESs of subtypes 6a (849, 44.4%), 6b (206, 10.7%) and 6d (452, 23.6%), whereas a smaller fraction of subtypes 6c (3, 0.1%), 6e (83, 4.3%), and 6f (98, 5.1%) IRES structures were identified. Interestingly, a fraction of the identified IRES structures was recognized by more than one covariance model subtype, 113 (5.9%) IRES structures were classified as both subtype 6a and 6b IRESs, 107 (5.6%) as both subtype 6a and 6d IRES structures, and 1 (0.05%) as both subtype 6d and 6f IRES structures (Fig 1C and 1D). In general, these mixed subtype IRES-like structures have low INFERNAL scores, containing conserved elements such as the SLIV, SLV, and PKI and deviate in other regions.

While the bioinformatics pipeline identified IRES-like structures in a third of the genomes, IRES-like structures were not identified in the other genomes. The absence of IRES-like RNA structures in the majority of dicistrovirus-like genomes could be attributed to incomplete or truncated genomes. Most detected IRESs were present in genomes larger than 6 kb (Fig 1E and S1 Data). In this subset, 2051/4294 (50.1%) of the genomes contained candidate IRES-like structures. This fraction remained constant in genomes >7,000 nucleotides in length (1948/3754, 51.9%), or >8,000 nucleotides in length (1662/3140, 52.9%). Similarly, the majority of dicistrovirus genomes that do not contain an Type 6 IRES-like structure within the intergenic region were >6 kb in length.

To gain insights into the evolution of Type 6 IRESs, we constructed an RdRP-based phylogenetic tree of the dicistrovirus genomes (<6 kb) and plotted the predicted IRES-like structure subtypes in this tree (Fig 1F and S1 Data). Genomes with subtype 6a IRES structures were distributed in multiple clades throughout the tree, suggesting this IRES structure may be closer to the ancestral Type 6 IRES and/or that distant viral genomes obtained IRESs via recombination events. Genomes from viruses of the Cripavirus genus, that have been described to contain 6a IRESs, were separated into a basal clade (containing RhPV and ALPV and a derived clade including DCV, NLCV and CrPV), confirming that this is paraphyletic [43]. In contrast, the Triatovirus genus was in a single monophyletic group. Genomes with subtype 6b IRES structures were mostly clustered in two clades; a large basal clade (containing Solenopsis invicta virus-1 (SINV-1), ABPV, KBV, FEV1, and IAPV) and a smaller derived clade (containing TSV, MrTV, and MCV). This distribution confirmed that this is likely a paraphyletic group [43]. Type 6c IRES structures were the rarest, present in only two closely related genomes. Most genomes with a subtype 6d IRES structure clustered in a monophyletic clade. Genomes with subtype 6e were closely related in a monophyletic clade thus suggesting that this IRES subtype may be present only in specialized dicistrovirus genomes. Genomes with subtype 6f IRES structures were in a monophyletic clade, clustered within the subtype 6d IRES structure clade, hinting that the 6f subtype may have derived from subtype 6d IRESs. IRES-like structures that were identified by multiple covariance models (6a-6b, 6a-6d, and 6d-6f) did not show clustering into monophyletic or nearly monophyletic clades, suggesting that these may be recently evolved IRES structures that have not been positively selected or that the chimeric IRES structures occurred via recombination.

IRES consensus models

To examine conserved elements within the predicted Type 6 IRES-like structures, we aligned unique high-scoring IRES subtypes through multiple sequence alignments to construct consensus models (Fig 2, top 50% INFERNAL scores of each subtype, S2 Data). Type 6a IRES structures, as exemplified by the CrPV IRES, (Fig 2) showed high conservation in the internal loop 1.1 (L1.1), stem-loop IV (SLIV), and the stem-loop V (SLV), which was not unexpected as these domains are essential for the Type 6a IRESs to interact with ribosomal subunits; the L1.1 interacts with the L1 stalk from the 60S subunit [26,29,44], the SLIV interact with the rps7 and the SLV with the rps25 from the 40S subunit [26,27,30]. Conservation was also observed within the PKI domain; VLR, PKI base pairing and L3.1 mediate specific steps of IRES translation including frame selection and translocation [45,46]. The Type 6a IRES consensus alignment further identified conserved regions within adenines in the L1.2 close to the PKII base pairing, UA-rich single-stranded regions (S2.2 and S2.3) upstream of the SLV, residues in the helical P1.1 region, and a CG pairing within PKII. From structural analysis of the CrPV IRES and IRES:ribosome complexes, the adenines within L1.2 are involved in A-minor interactions with the minor groove of the P2.2 helix [26,29].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Consensus models of Type 6 IRES.

Secondary structure models of indicated Type 6 IRESs are shown with colour-coded heat map of nucleotides representing percent conservation from alignment of the top 50% scoring sequences for each subtype based on INFERNAL output (S2 Data). The range of base pairing are also denoted and the number of sequences used for alignment of each sub-type are indicated. Pie charts show the percentages of the predicted codon adjacent to the pseudoknot I (PKI) IRES sequences. The resulting multiple sequence alignments were manually curated to remove gaps.

https://doi.org/10.1371/journal.ppat.1013255.g002

Type 6b IRESs (Fig 2), as exemplified by the IAPV and TSV IRESs, adopt a structure similar to 6a IRESs, containing three PKs and uniquely an additional stem-loop close to the PKI (SLIII). The predicted Type 6b IRES structural model showed high levels of conservation in SLIV and SLV, which have similar interactions with the 40S ribosomal subunits as in Type 6a IRES:40S complexes [27]. Surprisingly, within the L1.1 region of 6b IRES model, only a few nucleotides in the distal part of the internal loop were strongly conserved, despite the role of this region for ribosome assembly on the IRES [28,47]. Nucleotides surrounding the SLIV showed moderate levels of conservation (S2.2, S2.3, and P2.2) whereas L1.2, L3.2, VLR, the linker S2.5, and the junction of L3.1, L3.2, and L3.3 showed a higher level of conservation. The L3.1, L.3.2 and L3.3 junction is important for +1 frame translation in a subset of 6b IRESs [48]. L3.3 and SLIII showed low levels of conservation despite their reported role in promoting IRES-mediated translation.

For predicted Subtype 6d IRES structures, conservation was observed within SLIV and L1.1, domains that interact with both 40S and 60S subunits (Fig 2). Other regions included two adenosines within L1.2a. For predicted subtypes 6e and 6f IRES structures, conservation was observed in L1.1 and P2.2, and within PKI. Subtype 6f IRES-like structures showed some conservation within the three nucleotides within S2.4. Subtypes 6d-6f IRESs have been shown to recruit 80S ribosomes directly, similar to subtype 6c IRESs [35–37]. In general, these analyses highlighted the length of the base-paired regions and key nucleotide identity conservation that may be important for IRES function.

A unique feature of IGR IRES mechanism is that the IRES initiates translation from a non-AUG codon adjacent to the PKI domain. We examined the putative start codon usage of the predicted Type 6 IRES-like structures (Fig 2 and S2 Data). For subtypes 6a to 6d, the dominant predicted start codon was an alanine codon with GCU being the most prevalent start codon. There was a strong preference for a G at the first nucleotide of the starting codon. Structural studies of IGR IRES:ribosome complexes have shown that this G base adopts intrastrand stacking interactions within the PKI loop, thus contributing to the anticodon-codon mimicry of PKI [26,32]. For predicted subtype 6f IRES structures, the most prevalent putative start codons were ACA, CCU and ACU, from highest to lowest (Fig 2). These analyses suggested a strong selection of putative start codons for IGR IRES-mediated viral protein synthesis.

Functional validation of predicted IRES structures

We next examined whether the bioinformatically-predicted IGR IRES structures are functional. To address this, we cloned a library of 1111 of the 1912 predicted Type 6 IRESs within the intergenic region of a lentivirus dual fluorescent protein reporter containing an upstream eGFP and downstream mRuby3 (Fig 3A and S3 Data). eGFP and mRuby3 are driven by cap-dependent and IRES-mediated translation mechanisms, respectively. The cloned sequences were selected by clustering the 1912 Type 6 IRESs at 90% nucleotide identity to prevent overrepresentation of highly similar sequences (S3 Data). Lentivirus-transduced HEK293T cells were then treated with the ER stress inducer, thapsigargin, which is a known treatment known to increase dicistrovirus IGR IRES translation [49,50]. Cells were subjected to flow cytometry analysis (FACS) and cell-sorted for both eGFP and mRuby3 positive cells. Under basal conditions, approximately ~1.3% of the cells displayed both high eGFP and mRuby3 fluorescence suggesting that relatively few IRESs were active (Fig 3B). By contrast, in Thapsigargin-treated cells, the number of cells displaying both high eGFP and mRuby3 increased to ~4.1% of cells, which is consistent with increased IRES translation under ER stress [50]. Cells displaying high and low mRuby3 (i.e. IRESs that were active in both basal and Thapsigargin-treated) were isolated and genomic DNA was purified, from which the IRES region was PCR amplified and identified by next-generation sequencing. A total of 209 IRES-like structures were identified, the majority of which were enriched of subtype 6a and 6b IRESs (Fig 3C and S3 Data). In general, there was an enrichment of IRES structures (i.e., more read counts) in thapsigargin-treated cells over untreated cells (Figs 3C and S1C and S3 Data). Plotting the fold-change (mRuby3 positive/mRuby3 negative) enrichment of IRES structures in untreated (x-axis) vs thapsigargin-treated cells (y-axis) revealed a negative correlation (S1A Fig), indicating that IRESs that were not translated under basal conditions were relatively more actively translated under ER stress.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Functional analysis of IRES-like RNAs.

(A) Schematic of IRES screening pipeline in mammalian cells. A library of IRESs were cloned within the intergenic region of a lentivirus eGFP-mRuby3 dual reporter construct. HEK293 cells were transduced with lentivirus and fluorescent cells were analyzed by FACS analysis and sorted. Cells were either untreated or treated with Thapsigargin (1 μM) for 18 hours. (B) FACS plots of cells eGFP (y-axis) and mRuby3 (x-axis) fluorescence. The percent of mRuby3 positive cells are shown under specific cell treatments. IRESs identified in untreated and Thapsigargin treated cells are listed in S3 Data. (C) Sequencing read counts for IRESs in mRuby3 positive cells transduced with lentiviral reporter library in untreated (red) and Thapsigargin treated (blue) conditions. The counts are ranked from highest to lowest for IRESs enriched in untreated conditions. All raw read counts for unfiltered and filtered IRESs are in S3 Data. (D) In vitro translation. Select Type 6a and 6b IRESs were tested in (left graph) rabbit reticulocyte lysates (1 hour, 37°C) or (right graph) in Sf21 insect lysates (2 hours, 30°C) using in vitro transcribed dual luciferase reporter RNAs. IRESs selected are listed in S4 Data. Shown are averages ± s.d. from at least three independent experiments of FLuc/RLuc luciferase activity ratios normalized to that of bicistronic RNAs containing the wild-type (WT) CrPV IGR IRES. mPKI represents mutant CrPV IRES containing mutations that disrupt PKI base pairing. INFERNAL scores are shown below the indicated IRESs. Shown are averages ± s.d. from at least three independent experiments.

https://doi.org/10.1371/journal.ppat.1013255.g003

To further confirm that the predicted IRESs are functional, we selected candidate Type 6a and 6b IRES structures from the FACS-sorted fluorescent cell screen and the bioinformatics pipeline (i.e., high INFERNAL scores) (S1, S3 and S4 Data). We chose a range of subtype IRES-like structures and cloned them within the intergenic region of the dual luciferase reporter constructs (Fig 3D and S4 Data). For simplicity, we re-labeled the IRES subtypes 6a-1 to 6a-11 and 6b-1 to 6b-9 (S4 Data). The IRES sequences, the genomes from which the IRESs were identified, the INFERNAL predicted score and the corresponding predicted secondary structure models are shown in S4 Data and S2 Fig. In vitro transcribed RNAs were incubated in rabbit reticulocyte lysate (RRL) or insect Sf21 extracts and luciferase activities were measured. As expected, the reporter RNA containing the wild-type CrPV IGR IRES led to Firefly luciferase activity (shown as Firefly:Renilla (FLuc/RLuc) ratio) indicative of active IRES translation, whereas a mutant CrPV IRES containing mutations that disrupted PKI base-pairing did not (Fig 3D). Normalizing the translational activities to that of the wild-type CrPV IRES, the select Type 6a and 6b IRESs displayed a range of translational activities (Fig 3D). Approximately half of the IRES structures (6a-14 and 6b-15) showed IRES activities equal to or ~two-fold higher as compared to the CrPV IRES in RRL (Fig 3D). Although the majority tested showed IRES translation in both RRL and Sf21 extracts, the level of FLuc/RLuc ratios were not as high in Sf21 extract as those observed in RRL. Furthermore, some IRESs displayed opposite translational activities in RRL versus Sf21 extracts (Fig 3D; see 6a-1, 6b-1, 6b-2 IRESs as examples), suggesting that there may be species-specific differences that govern IRES translation in vitro. These results conclusively validated the bioinformatics approach in identifying bona fide dicistrovirus IRES elements.

To determine whether the predicted RNA structures are important for IRES function, we generated mutations that disrupt PKI base-pairing within select active Type 6a and 6b IRESs and tested them using the dual luciferase reporter assay in vitro in RRL (Fig 4A). As predicted, disruption of the predicted PKI base-pairing abolished IRES translation whereas compensatory mutations that restored PKI base-pairing partially rescued IRES activity (Fig 4A and 4B - representative 6b-2 IRES model; S2 Fig for models of each IRES), similar to that observed with other Type 6 IRESs [51,52]. These results demonstrated that the integrity of the predicted PKI structure is important for IRES translation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Integrity of IRES structure for translation and ribosome binding.

(A) Integrity of PKI for IRES function. Bicistronic luciferase reporter RNAs containing select IRESs containing either mutations that disrupt PKI base pairing (mPKI) or compensatory mutations that restore PKI base pairing (cPKI) were incubated in RRL for 1 hour, 37°C. Shown are averages of FLuc/RLuc luciferase activity ratios normalized to that of bicistronic RNAs containing the wild-type (WT) CrPV IGR IRES. mPKI represents mutant CrPV IRES containing mutations that disrupt PKI base pairing. (B) Secondary structure model of the Type 6b-2 IGR IRES. Mutations that disrupt the PKI base pairing and compensatory mutations that restore the base pairing are indicated in rectangular boxes. (C) 80S ribosome binding. Filter binding assay of purified, salt washed human 40S and 60S incubated with select [³²P]-labeled RNAs (0.5 nM) and the fraction of IRES:ribosome complexes were measured by phosphorimager analysis. mPKII/PKIII and mPKI/mPKII/mPKIII represent mutant IGR IRESs whereby PKI, PKII and PKIII base pairing are disrupted. (D) Apparent dissociation constant (K_D) values of 80S:IRES complexes. (E) Relative luciferase activity of Type 6a IRES subtype in S2 cells. S2 cells were transfected with reporter RNAs containing the indicated wild-type 6a-11 IRES or CrPV IRES (WT), mPKI (mutation in IRES that disrupts PKI base pairing). Shown are averages of Firefly:Renilla (Fluc:RLuc) luciferase activities at 6 hours post-transfection. All data are averages ± s.d. from at least three independent experiments.

https://doi.org/10.1371/journal.ppat.1013255.g004

We next examined whether the IRESs can bind to ribosomes directly, which is a key unique property of Type 6 IRESs [49]. Select [³²P]-end labeled Type 6a and 6b IRES RNAs were incubated with increasing concentrations of purified salt-washed 40S and 60S subunits. Of note, we chose IRESs that were active and inactive in in vitro translation extracts (Fig 3D) and were of relatively high INFERNAL scores. All IRESs tested bound directly to 80S ribosomes with an apparent K_D ranging from 2.5-12 nM (Fig 4C and 4E), which is within the range of known IGR IRES:ribosome binding affinities [25,28]. We examined the Type 6a-11 IRES structure in more detail and investigated if mutations that disrupt PKII and/or PKIII base pairing reduced IRES binding to the ribosome (Fig 4C, left bottom). As expected, PKII and PKIII mutant versions of this IRES impaired binding to the 80S ribosome, indicating that the integrity of the predicted IRES structure directed ribosome binding. To examine if the IRES structure is important for translation in a more biologically relevant context, we tested the IRES activity in insect S2 cells by transfection of 5′ cap bicistronic reporter RNA (Fig 4D). The wild-type IRES supported translation in S2 cells, whereas the PKI mutant (mPKI) showed significantly reduced translational activity (Fig 4D). In summary, we have demonstrated that the bioinformatics pipeline approach identified bona fide IRES-like structures that can assemble ribosomes directly and initiate IRES translation.

Novel dicistrovirus genomic architectures

We next analyzed the location of the predicted IRES structures within the viral genome. We examined viral genomes that were >6 kb in length in order to capture near complete genomes (S1 and S5 Data). Of the 805 genomes that contained predicted unique IRES structures, the majority of genomes (532/805) resembled architectures of known dicistroviruses harboring a single IRES within the intergenic region (Figs 5 and S3). Of these genomes, the IRES structures can be grouped based on the location within the intergenic region: (i) the majority of IRES structures (824) are adjacent to or close to the ORF1 stop codon (intergenic region is < 230 nucleotides including the predicted IRES structure), whereas 275 genomes ((ii) and (iii)) contained IRES structures downstream of the ORF1 stop codon that are within an intergenic region of 230–420 nucleotides or >420 nucleotides (Figs 5 and S3). These subgroups of IRES-containing IGRs can be further divided into specific IRES sub-types: most are of Type 6a and 6b IRESs (S3 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Genome architectures of dicistrovirus-like genomes.

Genome organization of dicistrovirus genomes was obtained by predicting ORFs using ORffinder, RdRps with Palmscan, and IRESs with INFERNAL. ORF2 > 1 kb in length were predicted using any possible non-stop codon as a start codon. The number of genomes within specific sub-types and the total numbers are shown to the right. Genomes are categorized with predicted IRESs located within the 5′ UTR or a single long ORF, or genomes with multiple predicted IRESs. Genomes containing predicted IRES with more than two ORFs are shown. Predicated helicase (Hel) and 3C-like protease (Pro) and RNA-dependent RNA polymerase (RdRP) are shown. Genomes are listed in S5 Data.

https://doi.org/10.1371/journal.ppat.1013255.g005

Interestingly, a fraction of genomes ((iv), (61 genomes) had the stop codon of ORF1 within the predicted IRES structure (Figs 5 and S3). This genomic arrangement is reminiscent of the honeybee dicistrovirus genomes, such as IAPV and ABPV, that contain a stop codon within a 5’ stem-loop that promotes IRES translation [51,53]. Within this subgroup, subtype 6a and 6d IRES structures were predominant. Unexpectedly, there was one genome which has an extended ORF1 whereby the stop codon is downstream of the IRES structure and putative start site (v) (Figs 5 and S3).

Unlike IRES-bicistronic genome architectures described above, we also identified other atypical IRES genome arrangements (Figs 5 and S3). We identified 30 genomes (vi) that contained a predicted Type 6 IRES structure within the 5′ UTR, of which they were similar to Type 6a and 6b subtype IRESs. Moreover, we also identified a small subset of genomes (19 genomes, (vii)) with predicted Type 6 IGR IRES structures within both the 5′ UTR and the intergenic region. Most of these genomes contained subtype 6a IRES structures in both the 5′ UTR and IGR, whereas a few genomes contained mixtures of subtype IRES-like structures (S3 Fig). Remarkably, we identified 24 genomes containing two predicted IRES structures within the intergenic region (viii). In these cases, the upstream IRES structure is predicted to drive translation of a small ORF ranging between 93–286 amino acids in length, whereas the downstream IRES structure directs translation of a downstream ORF. Finally, and surprisingly, there was a small subset of genomes that were monocistronic and contained a predicted embedded IRES structure overlapping within the coding region (14 genomes, (ix)). In these genomes, the organization and order of non-structural and structural proteins were similar to other dicistrovirus genomes [54]. The putative embedded IRES is downstream of the RdRP and is predicted to drive translation of the viral structural proteins.

Functional characterization of IRESs within atypical dicistrovirus genomes

The location of the predicted IRES structures in atypical dicistrovirus genome architectures suggests that the IRESs may be utilized in distinct ways to regulate the stoichiometry of viral protein expression during infection. We chose two distinct IRES/genome architectures for follow-up experiments: the genome with the ORF1 stop codon downstream of the predicted IRES structure (v) and genomes that are monocistronic and contain an overlapping IRES (ix). The location of the IRES structure embedded within the viral open reading frame poses the question of how translation of ORF1 or the monocistronic main ORF affects IRES-mediated translation.

The Type 6a-2 IRES is functional (Fig 3D) and is located within the genome whereby the stop codon (UGA) of ORF1 is directly downstream and adjacent to the putative PKI domain (Fig 6A). The RdRP domain is located upstream of the IRES whereas the translated protein sequence encoded within the IRES did not yield any obvious known structure or sequence via Blast or Alphafold3. We hypothesized that ribosomes that translate ORF1 unwind the IRES and thus, inactivate IRES-mediated translation. To address this, we generated three dual bicistronic reporters in which the 6a-2 IRES is cloned between RLuc and FLuc luciferase. In the first reporter (Fig 6B, WT), the 6a-2 IRES is situated to mimic the natural genome arrangement whereby the RLuc ORF extends into and past the IRES and the stop codon is located downstream of the IRES structure. Here, scanning-dependent translation of the RLuc and IRES-driven FLuc would express ~47 kDa and ~65 kDa proteins, respectively (Fig 6B). The second reporter (RS; RLuc stop codon) is similar to the first reporter but contains a stop codon before the IRES, thus generating a bicistronic reporter. The third reporter contains a strong hairpin structure (HP) within the 5′ UTR and thus blocks scanning-dependent translation. We tested these reporters in in vitro translation extracts monitoring expression of RLuc and FLuc by incorporation of [³⁵S]-methionine. As shown previously, bicistronic RNAs containing the wild-type but not mutant PKI CrPV IGR IRES, resulted in expression of FLuc (Fig 6C). Bicistronic RNAs containing the 6a-2 IRES in its natural context resulted in translation of both the ~ 47 kDa and ~65 kDa proteins. Similarly, FLuc expression was also detected in bicistronic RNAs containing a stop codon downstream of RLuc. Insertion of a strong hairpin within the 5′ UTR abolished scanning-dependent RLuc translation as expected but resulted in higher IRES-mediated FLuc translation (~1.5 fold) compared to the other two WT and RS bicistronic reporters (Fig 6C and S4 Data). These results indicated that the IRES is functional though it is located within the ORF.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Functional analysis of IRES within an atypical dicistrovirus genome.

(A) Schematic of the Type 6a-2 IRES (type 6a-2) within the atypical dicistrovirus genome. Predicted secondary structure model of the Type 6a-2 IRES. (B) Schematic of reporter RNAs mirroring their locations in atypical dicistrovirus genome. “RS” denotes a stop codon at the end of the Renilla luciferase ORF (stop codon is represented by a red octagon). “HP” denotes a strong hairpin stem-loop inserted within the 5′ UTR. The predicted translated protein masses are shown above. (C) SDS-PAGE analysis of in vitro translation reactions (top) and quantification of proteins (bottom). In vitro-transcribed bicistronic reporter RNA (600 ng) containing the indicated wild-type or mutant IRESs were incubated in RRL (1 hour, 37°C) with [³⁵S]-methionine. (D) IRES-mediated translation in mock and CrPV-infected (MOI 10) S2 cells. S2 cells were transfected with reporter RNAs containing the indicated wild-type 6a-2 IRES or CrPV IRES (WT), mPKI (mutation in IRES that disrupts PKI base pairing), RS (Renilla with a stop codon) or HP (hairpin within the 5′ UTR). Shown are averages of Firefly (FLuc) and Renilla (RLuc) luciferase activities ± s.d. from at least three independent experiments at 6 hours post-transfection.

https://doi.org/10.1371/journal.ppat.1013255.g006

To validate these results, we monitored translation in S2 cells by transfecting in vitro transcribed capped-bicistronic mRNAs. In these experiments, we infected S2 cells with the dicistrovirus, CrPV, to simulate virus infection which activates IRES activity. Cells transfected with the bicistronic RNA containing the stop codon after RLuc (RS) resulted in higher IRES-mediated translation as compared to that of the reporter RNA containing the 6a-2 IRES in its natural context (WT) (Fig 6D). These results are consistent with the model that ribosomes may translate through the overlapping IRES and reduce IRES translation, possibly by unwinding and disrupting IRES structure. However, these results cannot rule out that there is competition between cap- and IRES-dependent translation within the bicistronic reporter.

Embedded IRES-like structure in the viral ORF

An unexpected finding from the bioinformatics search was the discovery of embedded Type 6 IRES structures within the viral monocistronic open frame (Fig 5ix and S5 Data). These overlapping IRES structures (herein called contiguous or embedded IRES) are located downstream of the RdRP and are predicted to drive translation of viral structural proteins in ORF2 (Fig 7). There were sixteen genomes containing embedded IRES structures, which were all of either subtype 6a or 6b IRESs (Figs 5 and S2). We cloned these IRES structures into the dual luciferase reporter construct whereby the RLuc and FLuc were fused in frame with the contiguous IRES structure (Fig 7B, schematic of bicistronic reporters). Thus, translation of the full-length RLuc-IRES-FLuc would produce a 106 kDa protein (Fig 7B). Reporter RNAs with contiguous IRES structures led to translation of the FLuc 64 kDa protein in vitro (Fig 7C) with varying degrees of expression (ranging from 5 to 100% as compared to that of the wild-type CrPV IRES), thus demonstrating that most of these contiguous IRES structures are functional within the context of an ORF. Interestingly, scanning-dependent translation of the full-length polyprotein (RLuc-IRES-FLuc, 106 kDa protein) was also observed for the majority of contiguous IRES-containing reporter RNAs in vitro, suggesting that both scanning-dependent and IRES-mediated translation can occur (Fig 7C). Moreover, a reporter RNA with contiguous IRES 6a-9, resulted in expression of a truncated RLuc (36 kDa protein) which may suggest that ribosomes may be dropping off upon reaching the IRES structure (Fig 7C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Translation of embedded contiguous IRESs.

(A) Genomic location (top schematic) of a representative embedded IRES (type 6a-11) and its predicted secondary structure (bottom). P2A – denotes 2A peptide from porcine tescovirus-1. Scar – denotes sequence left over after RNA circularization. (B) Schematic of the luciferase reporter RNAs containing the IRES embedded within the ORF mirroring the location of the IRES in the native monocistronic genome. Predicted MW of RLuc (36.0 kDa), FLuc (61.7 kDa) and fusion RLuc-IRES-FLuc (106 kDa) are shown. Stop codon is indicated as a red octagon. (C) Translation of luciferase reporters containing select embedded IRESs (listed in S4 Data). In vitro transcribed bicistronic reporter RNAs containing the indicated IRESs were incubated in RRL (1 hour, 37°C) with [³⁵S]-methionine. Reactions were loaded on 12% SDS-PAGE gels and imaged by phosphorimager analysis. Representative gels (top panel) and quantification of relative Firefly luciferase band intensities (bottom graphs) are shown. (D) Mutational analysis of Type 6a-11 IRES. Representative SDS-PAGE gel of [³⁵S]-methionine in vitro translation reactions of reporter RNAs in RRL. mPKI-III denotes mutations that disrupt PKI, PKII and PKIII base pairing, m4SL denotes mutations within SLIV, SLV and L1.1 collectively. “Stop ORF1” denotes a luciferase reporter wherein a stop codon is inserted at the end of Renilla luciferase as shown in (B). (E) Schematic of circular RNAs and predicted circRNA protein products. (F) Representative SDS-PAGE gel of in vitro translation (RRL) reactions containing [³⁵S]-methionine and the indicated circRNA reporter. “SubP2A” denotes P2A sequence lacking the N-terminal GSG. 3mPK denotes mutations that disrupts all three pseudoknots within the IRES. (G) Translational activity of IRES circRNA in S2 cells. S2 cells were transfected with circular RNAs containing the embedded IRES. At 8 hours post-transfection, cells were harvested and lysed and luciferase activities were measured. Shown are averages ± s.d. from at least three independent experiments.

https://doi.org/10.1371/journal.ppat.1013255.g007

Using the subtype 6a-11 embedded IRES structure as a model (Fig 7A), we used several approaches to further confirm that this embedded IRES structure is functional. First, introducing a stop codon directly upstream of the IRES (and downstream of the RLuc) still led to translation of FLuc (Figs 3C and 7D, compare lanes 1 and 2 (Ren-Stop)). Second, mutations that disrupt PKI base pairing (mPKI), all three PK base pairing (mPKI-III) or mutations within SLIV, SLV and L1.1 (mSL4) all abolished FLuc expression (Fig 7D, lanes 3–6).

Third, to conclusively confirm IRES translation, we tested whether the contiguous 6a-11 IRES can function in a circular RNA (circRNA). Because the contiguous IRES is devoid of in-frame stop codons in the native polyprotein ORF, we reasoned that the IRES could drive continuous translation (i.e., a circRNA without a stop codon) [55]. Towards this, we generated an IRES-NLuc circRNA that does not contain an in-frame stop codon, and thus ribosomes would continuously drive translation of multiple NLuc proteins (WT-NLuc-circ) (Fig 7E, schematic of circRNA reporters and expected protein products). Further, we generated reporters whereby two porcine teschovirus 2A (P2A) “stop-go” sequences flank NLuc in order to release individual NLuc proteins (WT-P2A-NLuc-circ). To monitor translation, we tested RNAs in RRL with [³⁵S]-methionine to follow protein synthesis. In RRL, the WT-NLuc-circ RNA led to the expression of a smear of high molecular weight proteins (>80 kDa), consistent with rolling circle translation (Fig 7F, lane 3). In contrast, the insertion of flanking P2As of NLuc or addition of a stop codon downstream of NLuc (and thus, preventing circRNA translation) resulted in the expression of ~20 kDa NLuc (Fig 7F, lane 4 and 5). To confirm that rolling circle translation was occurring, we generated circRNA that contains flanking suboptimal P2A sequences that lack the N-terminal GSG sequence of P2A (WT-subP2A-NLuc-circ) [56]. This circular RNA WT-subP2A-NLuc-circ resulted in two predominant bands at ~20kDa and ~40 kDa, which is consistent with inefficient 2A peptide activity during rolling circle translation (Fig 7F, lane 2). Mutations that disrupt PKI (mPKI) or all three PK (3mPK) base-pairing inhibited translation, albeit there was still minor expression in reactions containing the mPKI mutant, suggesting that an intact PKII and PKIII can still recruit the ribosome and direct circRNA translation in vitro (Fig 7F, lanes 5 and 6). In summary, we showed that the 6a-11 IRES can direct rolling circle translation.

We next examined rolling circle translation by the contiguous 6a-11 IRES in transfected S2 cells. Transfection of IRES-containing circular RNAs resulted in higher NLuc activity as compared to the linear versions (Fig 7G). Of note, the circular RNAs that permitted rolling circle translation displayed higher NLuc than that of the WT-NLuc circRNA which contains a stop codon (WT-Nluc-stop-circ) (Fig 7G). The circRNA displaying the highest NLuc activity was the WT-P2A-Nluc-circ, followed by the WT-NLuc-circ and WT-sub2A-NLuc-circ. These results indicated that protein processing by the 2A peptide affects the NLuc translation and enzymatic activity. The mutant IRES containing disruptions of all three PKs, abolished translation (Fig 7G, 3mPK). By contrast, mutations that disrupt PKI base pairing (mPKI), which inhibits ribosome positioning on the IRES but not ribosome binding [25], still displayed NLuc activity (Fig 7G), suggesting that ribosome recruitment by the IRES with intact PKII and PKIII basepairing can still direct rolling circle translation. In summary, the contiguous IRES can direct internal ribosome entry translation initiation in the context of a reporter circRNA.

Factorless IRESs beyond family Dicistroviridae

Given that the bioinformatics pipeline identified bona fide factorless IRESs, we next asked whether IRES-like structures can be identified in other RNA viral genomes. Searching the RNA virome from the NCBI database and Serratus using INFERNAL, we identified eight IRES-like elements in positive-sense single-stranded genomes of Picornaviridae, Marniviridae, Narnaviridae and Tombusviridae, and of these, six were Type 6a and two were Type 6b IRESs (S2 Fig for models and S4 Data). Of note, the predicted start sites of these IRESs were in-frame of the downstream viral ORF, thus strongly suggesting that these putative IRESs direct translation in these non-dicistrovirus genomes. For instance, in the picornavirus and marnavirus genomes, the IRES-like structures are located within the intergenic region between two main ORFs, presumably to direct the translation of the downstream structural proteins (Fig 8A). In the narnavirus genome, the IRES is located within the 5′ UTR upstream of the single polyprotein ORF (Fig 8A). The IRES-like structures in the tombusvirus genomes were all located within the intergenic region upstream of the ORF encoding the capsid (Fig 8A). During the replication cycle of tombusvirus, a subgenomic RNA containing the capsid ORF is generated, which is typically translated by a 3′ UTR translational element called a CITE (cap-independent translation enhancer) [57]. Given the location of the IRES-like structures within the tombuvirus-like genomes, they would also be predicted to drive the translation of the tombusvirus capsid ORF.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Functional IGR IRES in non-dicistrovirus viral genomes.

(A) Schematic of non-dicistrovirus genomes and the location of predicted Type 6 IRES. (B) Non-dicistrovirus-like IRES translation. In vitro transcribed dual luciferase reporter RNAs containing the indicated non-dicistrovirus predicted IRESs within the intergenic region were incubated in RRL (1 hour, 37°C) or Sf21 extracts (2 hours, 30°C). Luciferase activities were measured and the ratio of FLuc/RLuc was calculated and normalized to that of CrPV IRES containing reporter RNAs. (C) Genome organization of Tombusvirus-like non-DV1 genome with a predicted IGR IRES (top) and secondary structure model of the putative IRES (bottom). Mutations used for subsequent binding experiments are denoted in the boxes. (D) Predicted non-DV1 RNA structure model from SAXS data (light blue) overlapped with atomistic structure, represented in ribbons. (E) Sucrose gradient centrifugation analysis (top) showing percent total radioactive counts (y-axis) in fractions (top to bottom) of reactions containing [³²P] labeled IRES incubated with purified salt-washed 40S and 60S. Wild-type (black lines) and mutant (grey lines) of CrPV IRES and non-DV-1 RNA are shown. Mutant IRES denotes mutations that disrupt both PKII and PKIII base pairing. Filter binding assays (bottom) of purified human 80S with indicated wild-type and mutant IRES. ³²P labeled IRESs (0.5 nM) were incubated with increasing concentrations of purified, salt-washed human 40S and 60S subunits and the fraction of IRES:ribosome complexes were measured by phosphorimager analysis. Shown are averages ± s.d. from at least three independent experiments.

https://doi.org/10.1371/journal.ppat.1013255.g008

Using the bicistronic reporters, we tested whether these non-dicistrovirus IRES-like structures can drive translation in RRL and insect Sf21 extracts (non-DV-1 to -8) (Fig 8B). Four of the eight IRESs directed FLuc expression and notably two of them, one from picornavirus and one from narnavirus, displayed higher IRES activity than that of the CrPV IGR IRES (Fig 8B). Despite the non-dicistrovirus IRES-like structures having relatively high INFERNAL scores, it was intriguing that the four putative IRESs from tombusvirus-like genomes did not support IRES activity (non-DV-1–4 IRES-like RNAs). We reasoned that given that tombusvirus typically infects plants, IRES-like RNAs may be functional in a plant extract or that the RNAs may be directing translation in alternative reading frames as a subset of dicistrovirus IRESs can drive alternative reading frame translation [51]. However, the non-DV-1 and non-DV-2 RNAs from tombusvirus-like genomes did not support translation in wheat germ extracts, nor could direct translation in all three frames (S4 Fig).

We next addressed whether the non-DV IRES-like RNA adopts an RNA structure as predicted by INFERNAL. To follow-up, we performed SAXS and SHAPE-seq analysis of the non-DV-1 IRES-like RNA (S5 Fig and S6 Data, see details of analysis). Briefly, fifty models of the non-DV-1 RNA were calculated with a Χ2 value of approximately 1.07, indicating a good fit of the experimentally collected scattering data with the ab initio envelopes. These models were then averaged and filtered to derive a single representative model for each RNA using DAMAVER, which exhibits an NSD value close to 1 (S5E and S5F Fig), implying that the average model closely resembles all generated models. We used the SHAPE reactivity profiling to gain insights into the secondary structure profile of non-DV-1 RNA. The SHAPE reactivities of each nucleotide were mapped and represented as a folded structure in S5G Fig. Overall, nucleotides throughout the RNA structure showed high reactivities (> 0.4). The secondary structure constraint provided by SHAPE, combined with the three-dimensional SAXS envelope, provided insights into the high-resolution structural organization of non-DV-1 RNA. Fig 8D shows the fitting of the SAXS envelope (surface structure in light blue) with the SHAPE-derived computational structure (dark cyan). These structures were overlapped using DAMSUP and presented an NSD of 1.07 (Figs 8D and S5E), indicating a good fit. Additionally, we used the “fit in map” function of ChimeraX to calculate the correlation coefficient between these two structures, which is 0.7749. This statistical parameter confirms the good fit, indicating a strong and positive correlation between the structures. Based on these structural analyses, the non-DV-1 RNA adopts an RNA structure that is reminiscent of a Type 6a IRES, as predicted by the INFERNAL pipeline.

Given that non-DV-1 adopts a Type 6a IRES structure, we next asked whether the non-DV-1 RNA binds directly to ribosomes. Using sucrose gradient centrifugation, we fractionated reactions containing [³²-P]-labeled RNA and purified salt-washed ribosomes. As shown previously, the wild-type but not mutant (TM, triple PKI/PKII/PKIII mutation) CrPV IRES was detected in the 80S fraction (Fig 8E, fractions 23–24), indicating direct ribosome binding the IRES. Similarly, the non-DV-1 IRES-like RNA was present in the 80S fraction. Mutating PKII and PKIII base pairing of non-DV-1 IRES-like RNAs abolished 80S binding (Fig 8E). To further validate these findings, we monitored 80S binding to the non-DV-1 RNA by filter-binding analysis. The non-DV-1 RNA bound to purified 80S ribosomes with an apparent K_D 23.45 ± 4.16 nM whereas mutating PKII and PKIII base-pairing impaired 80S binding (Fig 8E). These results indicated that the non-DV-1 IRES-like RNA structure can mediate direct ribosome binding. Overall, these data revealed IGR-IRES like RNA elements in viral genomes other than dicistroviruses and that IRES-driven mechanism may be more widespread in virosphere.

Plasmids

Dual and monocistronic luciferase reporter constructs were generated as described [77]. Briefly, IRESs were synthesized (Twist Biosciences) and cloned within the intergenic region of bicistronic construct or within the 5′ UTR of monocistronic construct. The library of predicted IRESs were synthesized (Twist Biosciences) and cloned into the intergenic region of a lentivirus reporter expression plasmid containing eGFP and mRuby3.

Tissue culture cells

HEK293T cells transduced with lentivirus were maintained in DMEM supplemented with 10% fetal bovine serum (FBS) at 37°C. Drosophila Schneider S2 cell line (S2) cells were maintained in Shields and Sang M2 insect medium (Sigma-Aldrich, St. Louis, MI, USA) supplemented with 10% (FBS) at 25°C.

In vitro transcription

Bicistronic and monocistronic DNA were linearized with BamHI and NcoI (NEB) restriction enzymes respectively, and RNAs were in vitro transcribed in reactions with T7 RNA polymerase as described [77]. Integrity of the RNA was monitored by gel analysis and quantitated by Nanodrop spectrophotometer. In vitro transcribed RNA was 5′ capped post-transcriptionally (Cellscript).

For SAXS analysis, the plasmid constructs were purified and then digested using XbaI (NEB, Canada). In vitro transcription (IVT) reactions were carried out under the control of the T7 RNA polymerase promoter, and the synthesized RNA was subsequently purified using size exclusion chromatography (SEC), following previously published protocols [63,78]. The purified fractions were eluted using an RNA buffer (10 mM Bis-tris pH 6.0, 100 mM NaCl, 15 mM KCl, 5 mM MgCl2, and 5% glycerol) and concentrated through overnight ethanol precipitation. The resulting pellets were then resuspended in RNA buffer to achieve a concentration of approximately 1.5 mg/mL for downstream applications. The integrity of the RNA was assessed using nondenaturing agarose gel electrophoresis.

In vitro translation assays

In vitro transcribed RNAs were incubated in rabbit reticulocyte lysates (RRL) or Sf21 insect lysates (Promega) for 1 hour at 37°C or 30°C in the presence of [³⁵S]-methionine as described previously [78]. Reactions were either resolved on 12% SDS-PAGE gel and analyzed by phosphorimager analysis (Typhoon, Amersham, Chicago, IL, USA) or analyzed for enzymatic luciferase activity (Promega) using a dual luciferase kit (Promega) in Tecan Spark-10M Multimode plate reader.

Transfection

S2 cells (2.5 × 10⁶ cells) were transfected with bicistronic RNA (2 μg) using Lipofectamine 2000 reagent (Thermo Fisher Scientific, Waltham, MA, USA). For monitoring reporter RNA translation under CrPV infection, S2 cells were transfected with bicistronic RNA for 1 h, followed by infection with CrPV (MOI = 10) for 4.5 h. Cells were harvested at the indicated time and lysed in 1X passive lysis buffer (Promega). Lysates were cleared and protein concentration was measured by Bradford assay (Bio-Rad, Hercules, CA, USA). Equal amounts of total protein were used to measure luciferase enzymatic activity according to the manufacturer’s protocol (Promega).

Purification of the 40S and 60S subunits

HeLa cell pellets (Cell Culture Company) were used to purify ribosomes as described [25]. Briefly, HeLa cells were lysed in a lysis buffer (15mM Tris-HCL (pH 7.5), 300 mM NaCl, 6mM MgCl2, 1% (v/v) Triton X-100, 1mg/mL heparin). Debris was removed by centrifuging at 23,000 ~ g and the supernatant was layered with 30% (w/w) cushion of sucrose in 0.5M KCl and centrifuged at 100,000 ~ g to pellet crude ribosomes. Pelleted ribosomes were gently resuspended in Buffer B (20 mM Tris-HCL (pH 7.5), 6 mM magnesium acetate, 150mM KCL, 6.8% (w/v) sucrose, 1 mM DTT) at 4°C. Next, the ribosomes were treated with puromycin (final concentration 2.3 mM) to release mRNA and KCl (final concentration 500 mM) was added to wash and dissociate 80S ribosomes into 40S and 60S. The dissociated ribosomes were then separated on a 10%-30% (w/w) sucrose gradient. Absorbance at 260 nm was measured to detect the 40S and 60S peaks and corresponding fractions were pooled and concentrated using Amicon Ultra spin concentrators (Millipore Sigma) in buffer C (20mM Tris-HCl (pH 7.5), 0.2 mM ED buy TA, 10mM KCL, 1mM MgCl2, 6.8% sucrose). The conversions 1 A260 nm = 50 nM for 40S subunits, and 1 A260 nm = 25 nM for 60S subunits were used to determine the concentration of 40S and 60S subunits by spectrophotometry.

Assembly and analysis of ribosomal complexes by sucrose density gradient centrifugation

10 pmol [³²P]-labeled IGR IRES mRNA were incubated with 30 pmol 40S subunits and 40 pmol 60S subunits in buffer E (final concentration: 20 mM Tris pH 7.5, 100 mM KCl, 2.5 mM MgOAc, 0.25 mM Spermidine and 2 mM DTT) for 20 min at 30°C to form ribosomal complexes, and then resolved by centrifugation through 10%-30% sucrose density gradients in buffer E in Beckman SW55 rotor at 55,000 rpm for 90 min at 4°C. Optical density of fractionated gradients was measured at 260nm, and the presence of radioactivity was monitored by Cherenkov counting in a scintillation counter (Beckman LS 6000IC).

Ribosome filter binding assays

Ribosome:IRES complexes were monitored by filter binding assays as described [33]. Briefly, in vitro transcribed RNAs (final 0.5 nM) were preheated at 65 °C for 3 min and 1X buffer E was added and slow cooled in a preheated water bath for 20 min. Pre-folded RNAs, along with non-competitor RNA were incubated with an increasing concentration of 40S and 60s subunits at room temperature for 15 minutes. Reactions were then loaded onto a Bio-Dot filtration apparatus (Bio-Rad) with a double membrane of nitrocellulose and nylon pre-washed with buffer E. The membranes were then washed thrice with buffer E, dried, and the radioactivity was imaged and quantified by phosphorimager (Amersham) analysis. The dissociation constant was calculated by the formula

where [A] is the concentration of RNAs, [B] is the concentration of ribosomes, [AB] is the concentration of RNAs bound to the ribosomes, fmax is the saturation point, and K_D is the dissociation constant.

SHAPE RNA structural probing

For SHAPE-MaP analysis of the minimal IRES, in vitro transcribed RNA for non-DV1 IRES was used as described [79]. Briefly, 500 ng of RNA was heated to 95° for 3 min, followed by the addition of Buffer E (final concentration of 20-mM Tris, pH 7.5, 0.1-M KCl, pH 7.0, 2.5-mM MgOAc, 0.25-mM spermidine and 2-mM dithiothreitol (DTT) and incubated at 30°C for 20 min. Folded RNA was modified by adding N-methylisatoic anhydride (NMIA) dissolved in DMSO (final concentration 5mM) and incubated for 45 minutes at 30°C. Control reactions containing only DMSO (no NMIA), as well as an additional denaturing control (DC) reactions were performed in parallel. The RNA was recovered by ethanol precipitation. RT Primer extension of modified and control RNAs was performed with primer 5’ -GTGTTGTTGTTTTGTTGTTTTCGTTG-3’ using Superscript III Reverse Transcriptase (SSIII, Thermo Fisher Scientific), with Mn2 + ; followed by second strand cDNA synthesis (NEB #E6111). The cDNA was then used for subsequent library preparation and sequencing. Libraries were sequenced on an Illumina MiSeq platform following the manufacturer’s standard cluster generation and sequencing protocols. SHAPE-Mapper 2.0 was used to analyze the sequencing data as described [80]. RNAstructure [81] was used to predict and model the secondary structure using the SHAPE-MaP data.

Small-angle X-ray scattering (SAXS)

SAXS data was collected at the B21 beamline in Diamond Light Source, located in Didcot, Oxfordshire, UK. The collection was carried out using an Agilent 1200 high-performance liquid chromatography (HPLC) system with a specialized flow cell, as previously outlined [82]. A Shodex KW403-4F SEC column was employed at a flow rate of 0.160 mL/min with RNA buffer. X-ray exposure lasted for 3 seconds over 600 frames. Analysis of the data was conducted using the ATSAS suite [83]. CHROMIXS [84] was initially utilized on each raw data set to account for buffer contribution. Guinier analysis was employed to assess the radius of gyration (Rg) and the homogeneity of the samples (q2 vs. ln(I(q))) [85]. The folding state of the RNA was determined using dimensionless Kratky analysis (qRg vs. qRg2*I(q)/I(0)) [86]. Paired distance distribution (P(r)) plots were generated to establish the real space Rg and the maximum dimension (Dmax) of the RNAs [87]. DAMMIN [88] was used to produce twenty models for non-DV-1 RNA based on the P(r) data. Subsequently, DAMAVER and DAMFILT [89] were employed to generate an average and filtered model for each RNA using these models. SHAPE-Mapper 2.0 was used to analyze the sequencing data as described [80]. RNAstructure [81] was used to predict and model the secondary structure using the SHAPE-MaP data.

Atomistic RNA structure determination

Reactivity data from SHAPE-MaP was employed to determine the secondary structure. The SHAPE sequencing map’s reactivity profile was analyzed using RNAstructure, and a dot-bracket secondary structure profile was visualized through StructureEditor [81]. The lowest free energy structure from the SHAPE analysis was chosen and used as secondary structure constraints in SimRNA v3.20 [90] to generate three-dimensional tertiary models for each RNA. A total of 20000 models for non-DV-1 RNA were produced. DAMSUP was utilized to align the tertiary models of each RNA with the SAXS models and calculate the normalized spatial discrepancy (NSD). Finally, the UCSF ChimeraX-1.8 [91] fit-in map tool was employed to determine the correlation coefficient between computational models and the SAXS envelope. The best representative models were filtered and chosen based on NSD and correlation coefficient values and visual inspection of structural fitting. The models were visually depicted using ChimeraX.

IRES lentiviral screen

The IRES library cloned into the lentiviral dual reporter was transformed into competent bacteria until a 10X coverage of colonies to inserts was achieved. Colonies were pooled and plasmids extracted for lentiviral infection. Cells were subjected to FACS under basal conditions and after 18h of exposure to 1 μM of Thapsigargin, selecting both eGFP and mRuby3 positive cells.

GFP and mRuby3 (putative IRES activity) and GFP-only positive cells (putative inactive IRES) were subjected to DNA extraction. Inserts were PCR amplified using flanking primers and 15 PCR cycles. The resulting PCR product was purified and sequenced on an Illumina MiSeq with 150 bp paired-end reads. The resulting sequencing files were quality-cleaned using TrimmGalore with a minimum Phred score of 20 and adapter trimming. Quality control was assessed with FastQC and MultiQC. Sequences were mapped with STAR to the initial IRES library sequences, and sequences with less than 88% overall mapping coverage and fewer than 2 raw counts were excluded from downstream analyses. We performed exploratory data analysis to assess the distribution of reads and identify potential biases in representation across samples (Gini coefficient [0.98], Spearman correlation [ρ ≈ 0.58], and top-10% IRES contribution), finding that a small subset of RNAs dominated the total read counts. Based on these observations, normalization was performed to improve comparability between conditions. Read counts were normalized using the trimmed mean of M-values (TMM) method in EdgeR to adjust for differences in sequencing depth and RNA composition. Normalized counts were then used to calculate log₂ fold changes (log₂FC) between GFP and mRuby3 versus GFP-only populations to identify IRESs enriched under basal and thapsigargin-treated conditions. All sequencing data are deposited in Jan, Eric; Chapagain, Subash (2025), “Discovery of functional factorless internal ribosome entry site-like structures through virome mining”, Mendeley Data, V1, https://doi.org/10.17632/3rwdwcrfhw.1.

Phylogenetic analysis

Phylogenetic relationships among dicistrovirus genomes were analyzed using maximum likelihood methods implemented in IQ-TREE (version 2.0.7). RdRP sequences from genomes >6 kb were aligned using MUSCLE v3.8 with the parameters -maxiters 1 -diags1 -sv. The optimal substitution model used was Blosum62 + I + G4 as determined by the ModelFinder in IQ-TREE. A phylogenetic tree was constructed with 1,000 ultrafast bootstrap replicates to provide branch support. RdRp sequences from the Hepatitis A virus (AWK22878.1), Heterosigma akashiwo RNA virus (NP_944776.1), and Nilaparvata lugens honeydew virus-1 (YP_009505599.1) were used as outgroups based on published RdRp-bases phylogenetic trees.

Dicistrovirus genomes selection and filtering

Viral genomes were selected from multiple sources, including the Serratus database, NCBI Viral genomes, and published metatranscriptomes. Dicistrovirus-like genomes were identified by screening RNA-dependent RNA polymerase (RdRP) sequences from known ICTV-classified dicistroviruses using Palmscan. Initially, 64,245 dicistrovirus-like genomes were identified. To minimize redundancy, genomes were clustered at 98% nucleotide identity, and sequences shorter than 2,000 bp were excluded. This filtering resulted in a final list of 9,151 genomes, which were then analyzed for IRES identification.

IRES search

RNA secondary structure models were generated to search for Type IRESs in a set of dicistrovirus genomes using INFERNAL v1.1.3. The INFERNAL search was performed using default parameters. 6a subtype IRES covariance models were generated from IGR IRESs of Drosophila C virus (DCV - AF014388), Homalodisca coagulata virus 1 (HoCV1 - DQ288865), aphid lethal paralysis virus (ALPV - AF536531), Rhopalosiphum padi virus (RhPV - AF022937), Triatoma virus (TrV - AF178440), Cricket paralysis virus (CrPV - AF218039), Black queen cell virus (BQCV - AF183905), Himetobi P virus (HiPV - AB017037), and Plautia stali intestine virus (PSIV – AB006531). 6b subtype IRES covariance models were generated from IRESs of the Acute bee paralysis virus (ABPV – AF150629), Israeli acute paralysis virus (IAPV – EF219380), Kashmir bee virus (KBV AY275710), Taura syndrome virus (TSV – AF277675), and Solenopsis invicta virus 1 (SINV1 – AY634314). 6c subtype IRES covariance models were generated from IRESs the Kuiper virus (KX657785.1), Changjiang picorna-like virus 14 (NC_032773.1), Halastavi árva RNA virus (NC_016418.1), and Shahe arthropod virus 1 strains SHWC01c3692 (KX883988). 6d subtype IRES covariance models were generated from IRESs of the Antarctic picorna-like virus 1 (APLV1 - KM259869.1), Beihai picorna-like virus 78 (KX883307.1), Changjiang picorna-like virus 9 (KX884541.1), Gingko biloba dicistrovirus strain pt112-dic-11 (MN729613.1), nedicistrovirus (NediV - JQ898341.1), Sanxia picorna-like virus 12 (KX883723.1), s64-k141_ 2464283 (MZ678982.1), Picornavirales sp. isolate R35-k141_316374(MZ678988.1), and Pallasea cancelloides TSA sequence (GEQX01016691.1). 6e subtype IRES covariance models were generated from IRESs of the Caledonia beadlet anemone dicistro-like virus 1 (MF189971.1), Beihai picorna-like virus 85 (KX883346), bivalve RNA virus G5 (NC_032115), and Proasellus solanasi 2 (HAFJ01050538.1). 6f subtype IRES covariance models were generated from the Changjiang picorna-like virus 9 (KX884560.1), Changjiang picorna-like virus 6 (KX884548.1), Wenzhou picorna-like virus 31 (KX884341.1), and Lactuca sativa dicistroviridae strain pt151 (MN722414.1). The identified IRESs were manually curated to remove alignment gaps, and the sequences were classified into subtypes based on scores and structural similarity. All datasets are shown in S1–S5 Data.