De novo assembly.

In the task “Assemble, Blast and Parse” (iii), retained RNA sequences are subjected to de novo assembly. For short reads (<50 nt), we tested several rounds of de novo assembly by Velvet [7] using the Oases software package [22] (Fig 1) and k-mer lengths ranging from 15 to 35 (S1 Table). For reads between 50 nt and 100 nt, we also used the Oases with k-mer lengths ranging from 13 to 69. Finally in Use Case 3–3, we used the Trinity assembly software which is available as a Galaxy tool and was reported to performs well with long reads [23]. Trinity as well as SPAdes [24] assembly softwares were also tested as alternate option to Oases in the Use Case 2–2 (S1 Table), giving similar outputs. It is noteworthy that users can adapt a Metavisitor workflow using any assembly software available in the Galaxy tool shed.

Blast.

Next, de novo assembled contigs are aligned to both nucleotide and protein vir1 BLAST databases built from the viral reference sequences (Fig 1) using the blastn or blastx Galaxy tools [18]. These tools search nucleotide or protein databases using nucleotide or translated nucleotide queries, respectively [25]. The default parameters are adjusted in order to report only the 5 best alignments per contig (Maximum hits option is set to 5) and to generate a tabular blast output that includes the 12 standard columns plus a column containing the length of the aligned subject sequences (extended columns option, “slen” checked). Note that this additional column in the blast output is required for subsequent parsing of the blast output by the “Parse blast output and compile hits” tool.

Parsing.

Tabular outputs generated by blastn and blastx alignments are processed by the “Parse blast output and compile hits” tool (S1 Table), which returns 4 files, namely “blast analysis, by subjects”, “hits”, “Blast aligned sequences” and “Blast unaligned sequences”.

In the “blast analysis, by subjects” file (S2 Fig), the subject sequences in the viral nucleotide or protein blast databases that produced significant blast alignments (hits) with de novo assembled contigs are listed, together with those contigs and hit information (% Identity, Alignment Length, start and end coordinates of hits relatively to the subject sequence, percentage of the contig length covered by the hit, E-value and Bit Score of the hit). In addition, for each subject sequence in the list, the length in nucleotide or amino-acid of the subject sequence (Subject Length), the summed coverage of the subject by all contig hits (Total Subject Coverage) as well as the fraction of the subject length that this coverage represents (Relative Subject Coverage), and the best (Best Bit Score) and mean (Mean Bit Score) bit scores produced by contig hits are computed and indicated. A simplified output can be generated without contigs and blast information by using the “compact” option for the reporting mode of the “Parse blast output and compile hits” tool. Note that the total and relative subject coverages indicate how much of the virus sequence is covered by the reconstructed contigs, whereas the Bit scores allow to estimate the distances between the reconstructed contigs and the subject sequence.

The “hits” file contains the sequences of contig portions that produced significant alignment in the BLAST step (i.e. query hit sequences), flanked by additional contig nucleotides 5’ and 3’ to the hit (the size of these margins is set to 5 by default and can be modified by the user). These margins allow to include sequences that might not have significant homology but could still be of viral origin.

Finally, the “Blast aligned sequences” file contains contigs that produced significant blast hits, whereas the “Blast unaligned sequences” file contains those that did not.

Use Cases 1–1, 1–2 and 1–3: detection and reconstruction of the Nora virus genome in small RNA sequencing datasets.

Using small RNA sequencing libraries SRP013822 (EBI ENA) and the Paparazzi software [4] we were previously able to propose a novel reference genome (NCBI JX220408) for the Nora virus strain infecting Drosophila melanogaster stocks in laboratories [3]. This so-called rNora genome differed by 3.2% nucleotides from the Nora virus reference NC_007919.3 and improved the alignment rate of viral siRNAs by ~121%. Thus, we first tested Metavisitor on the small RNA sequencing datasets SRP013822 using the Oases de novo assembly tool which is well suited to assembly of short read [9].

Three Metavisitor workflows were run on the merged SRP013822 small RNA sequence reads and the NC_007919.3 genome as a guide for final scaffolding. The workflow for Use Case 1–1 (S3 Fig) used raw reads collapsed to unique sequences (experimental procedures) to reconstruct a Nora virus genome referred to as Nora_MV (S1 File). In a second workflow for Use Case 1–2 (S4 Fig), we did not collapse the SRP013822 reads to unique sequences, which allowed the reconstruction of a Nora_raw_reads genome (S2 File). Finally, the workflow for Use Case 1–3 (S5 Fig) normalized the abundances of SRP013822 sequence reads using the Galaxy tool “Normalize by median” [27] and reconstructed a Nora_Median-Norm-reads genome (S3 File).

All three reconstructed genomes as well as the Paparazzi-reconstructed JX220408 genome had a high sequence similarity (>96.6% nucleotide identity) with the NC_007919.3 guide genome (S4 File). The final de novo (capital letters) assemblies of both the Nora_raw_reads and Nora_Median-Norm-reads genomes entirely covered the JX220408 and NC_007919.3 genomes (both 12333 nt), whereas the de novo assembled part of the Nora_MV genome was marginally shorter (12298 nt, the 31 first 5’ nucleotides are in lowercase to indicate that they were not de novo assembled but instead recovered from the guide genome). To evaluate the quality of assemblies, we remapped the SRP013822 reads to the 3 reconstituted Nora virus genomes as well as to the JX220408 guide genome using the “workflow for remapping in Use Cases 1–1,2,3” (S6 Fig). As can be seen in Fig 2, SRP013822 reads matched the genomes with almost identical profiles and had characteristic size distributions of viral siRNAs with a major peak at 21 nucleotides. Importantly, the numbers of reads re-matched to the Nora virus genomes were 1,578,704 (Nora_MV) > 1,578,135 (Paparazzi—JX220408) > 1,566,909 (Nora_raw_reads) > 1,558,000 (Nora_Median-Norm-reads) > 872,128 (NC_007919.3 reference genome guide).

Download:

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

Fig 2. Realignments of small RNA sequence reads to reconstructed (Nora_MV, Nora_raw_reads and Nora_Median−Norm−reads) or published (JX220408.1 and NC_007919.3) Nora virus genomes.

Plots (left) show the abundance of 18–30-nucleotide (nt) small RNA sequence reads matching the genome sequences and histograms (middle) show length distributions of these reads. Positive and negative values correspond to sense and antisense reads, respectively. Total read counts are indicated to the right hand side.

https://doi.org/10.1371/journal.pone.0168397.g002

Thus, Metavisitor reconstructed a Nora virus genome Nora_MV whose sequence maximizes the number of vsiRNA read alignments which suggests it is the most accurate genome for the Nora virus present in the datasets. Of note, the Nora_MV genome differs from the JX220408 rNora genome generated by Paparazzi by only two mismatches at positions 367 and 10707, and four 2nt-deletions at positions 223, 365, 9059 and 12217 (see S4 File). These variations did not change the amino acid sequence of the 4 ORFs of the Nora virus. We conclude that Metavisitor performs slightly better than Paparazzi for a known virus, using de novo assembly of small RNA reads followed by blast-guided scaffolding. We did not observe any benefits of using raw reads or normalized-by-median reads for de novo assembly with Oases, but rather a decrease in the accuracy of the reconstructed genome as measured by the number of reads re-mapped to the final genomes (Fig 2).

Use Case 1–4: detection of multiple viruses in small RNA sequencing datasets.

In order to show the ability of Metavisitor in detecting multiple known viruses in small RNA sequencing datasets, we built another workflow Case that performs blastn alignments of Oases contigs on the vir1 reference and reports for all significant alignments without filtering (S7 Fig). Applying this workflow to the SRP013822 sequence datasets produced a list of alignments which contains, as expected, the Nora virus. In addition, contigs were found to align with high significance (Mean BitScore > 200) to the Drosophila A virus and to the Drosophila C virus (S5 File and Table 1), strongly suggesting that the fly stocks analyzed in our previous work were also subject to persistent infection by these viruses [3].

Download:

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

Table 1. Report table generated by the “Parse blast output and compile hits” tool in Use Case 1–4 showing the presence of Drosophila A virus and Drosophila C virus in addition to the Nora virus in the small RNA sequencing of laboratory Drosophila.

See Method section for a description of the columns.

https://doi.org/10.1371/journal.pone.0168397.t001

Use Case 2–1: identification of new viruses in small RNA sequencing datasets.

Using Metavisitor, we recently discovered two novel viruses infecting a laboratory colony of Anopheles coluzzii mosquitoes [28]. In this case, a workflow (S8 Fig) was used to process small RNA datasets from these mosquitoes (EBI SRA ERP012577) and to assemble a number a Oases contigs that show significant blastx hits with Dicistroviridae proteins, including Drosophila C virus (DCV) and Cricket paralysis virus (CrPV) proteins (S6 File).

The viral family of Dicistroviridae was named from the dicistronic organisation of their genome with a 5’ open reading frame encoding a non-structural polyprotein and a second non-overlapping 3’ open reading frame encoding the structural polyprotein. In order to construct a potential new A. coluzzii dicistrovirus genome, the “Pick Fasta Sequences” tool (S8 Fig) collected blastx hits showing significant alignment with both Drosophila C virus and Cricket paralysis viral polyproteins (S7 File) that were further clustered with the “cap3 Sequence Assembly” tool in 4 contigs of 1952, 341, 4688 and 320 nt, respectively (S8 File). These 4 contigs were further aligned to the DCV genome NC_001834.1 sequence with tblastx and integrated in this scaffold sequence with the “blast_to_scaffold” tool to produce a final assembly (S9 File). Re-mapping of the ERP012577 small RNA reads using the “Workflow for remapping in Use Cases 1–1,2,3” (S6 Fig) showed that they mostly align to de novo assembled regions (uppercase nucleotides) of this chimeric genome and have a typical size distribution of viral derived siRNA (S9 Fig), suggesting that the NC_001834.1 DCV sequences of the scaffold (lowercase nucleotides) are loosely related to the actual sequence of the novel A. coluzzii dicistrovirus. Nevertheless, the composite assembly allowed designing primers in the de novo assembled regions to PCR amplify and sequence the regions of the viral genome that could not be de novo assembled [28].

Several teams have used siRNA signature (a peak at 21 nt in the size distribution of re-aligned small RNA sequences) as an alternate approach to sequence similarity to identify contigs of potential viral origin [12,13]. In order to further illustrate the flexibility of Metavisitor for implementing this strategy, we built a workflow (S10 Fig) to realign ERP012577 small RNA sequences to Oases contigs of length higher than 300 nt and to generate in batch read maps and read size distributions for these contigs using the “Generate readmap and histograms from alignment files” tool (S1 Table). We manually inspected these read maps and size distributions (S10 File) and collected all contigs with a clear siRNA signature (a pick at 21nt for both forward and reverse strands of contig sequences), 2 sets of contigs with a modest excess of 21nt reads from the forward strand only and 3 sets of contigs with no siRNA signature as negative controls (S3 Table). With the notable exceptions of loci 3 and 46 contigs, all contigs with a clear siRNA signature blastx aligned to vir1 viral sequences (S3 Table and see here the public Galaxy history for details). Loci 3 and 46 contigs did not align either to the non-redundant protein database of the NCBI and may therefore be of potential viral origin (S3 Table). All 5 negative control contigs with unclear or no RNA signature only aligned significantly to non-viral proteins (S3 Table). Together, these results illustrate the use of Metavisitor to implement a sequence-independent strategy based on siRNAs for virus identification [12,13].

Use Case 2–2: identification of new viruses in mRNA sequencing datasets.

In our study [28], we also used RNAseq libraries from the same A. coluzzii colony (EBI-SRA, ERS977505), demonstrating the use of a Metavisitor workflow for long RNA sequencing read datasets (S11 Fig). Thus, 100nt reads were aligned without adapter clipping to the Anopheles gambiae genome using bowtie2, and unmatched read were subjected to Oases assembly (kmer range 25 to 69, to take into account longer reads). Oases contigs were then filtered for a size > 5000 nt and aligned to the protein viral reference using blastx. Parsing of blastx alignments with the “blast analysis, by subjects” tool repeatedly pointed to a 8919nt long Oases contig that matched to structural and non-structural polyproteins of DCV and CrPV (S11 File). This 8919nt contig (S12 File) completely includes the contigs generated with the small RNA datasets (S8 File) and shows a dicistronic organization which is typical of Dicistroviridae and is referred to as a novel Anopheles C Virus [28]. The sequence of this Anopheles C Virus is deposited to the NCBI nucleotide database under accession number KU169878. As expected, when realigned to this genome (S12 Fig), the ERP012577 small RNA reads now show a typical alignment profile all along the AnCV genome sequence with a size distribution peaking at the 21nt length of viral derived siRNAs and no gap (Fig 3). Of note, we tested in Use Case 2–2 two alternate workflows substituting the Oases assembly tool with Trinity (S13 Fig) and SPAdes (S14 Fig), respectively. Both these workflows were equally able to assemble the genome KU169878 of the Anopheles C Virus (S13 File).

Download:

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

Fig 3. Alignments of small RNA sequence reads to the Anopheles C virus genome reconstructed in Use Case 2–2.

Plot shows the abundance of 18–30-nucleotide (nt) small RNA sequence reads matching the genome sequence and histogram shows the length distribution of these reads. Positive and negative values correspond to sense and antisense reads, respectively.

https://doi.org/10.1371/journal.pone.0168397.g003

Taken together, the Metavisitor Use Cases 2–1 and 2–2 illustrate that when short read datasets do not provide enough sequencing information, an adapted Metavisitor workflow (S11 Fig) is able to exploit long reads of RNA sequencing datasets, if available, to assemble a complete viral genome [28].

Use Case 3–1.

Innate lymphoid cells (ILCs) play a central role in response to viral infection by secreting cytokines crucial for immune regulation, tissue homeostasis, and repair. Therefore, the pathogenic effect of HIV on these cells was recently analyzed in infected or uninfected patients using various approaches, including transcriptome profiling [30]. ILCs are unlikely to be infected in vivo by HIV as they lack expression of the CD4 co-receptor of HIV and they are refractory in vitro to HIV infection. However, we reasoned that ILCs samples could still be contaminated by infected cells. This might allow Metavisitor to detect and assemble HIV genomes from patient’s ILC sequencing data (EBI SRP068722).

We imported 40 ICL sequence datasets from the EBI SRP068722 archive and merged the datasets belonging to the same patients. As the data contained short 32 nt reads that in addition had to be 3’ trimmed to 27 nt to retain acceptable sequence quality, we designed a workflow for Use Case 3–1 (S15 Fig) that is similar to the workflows used in cases 1–1 and 2–1 for small RNA sequencing data. Thus, the sequencing datasets were depleted from reads aligning to the human genome (hg19) and viral reads were selected by alignment to the NCBI viral sequences using the sRbowtie tool (S1 Table). These reads were further submitted to Oases assembly (kmers 11 to 27, to take into account short reads) and the resulting contigs were aligned to the Nucleotide Viral Blast Database using blastn. Alignments were parsed using the “Parse blast output and compile hits” tool, removing alignments to NCBI sequences related to patents to simplify the report (“Patent” term in the filter option of the “Parse blast output and compile hits” tool). A final report was generated by concatenating the reports produced by this tool for each patient (S14 File and Table 2). In summary, we were able to detect HIV RNAs in samples from 3 out of 4 infected patients whereas all samples from control uninfected patients remained negative for HIV. This Metavisitor workflow was able to accurately detect HIV RNA, even in samples where the number of sequence reads was expected to be low.

Download:

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

Table 2. HIV detection in ILC patient samples of Use Case 3–1.

The table summarizes the report generated by Metavisitor from a batch of 40 sequence datasets (S14 File). Metadata associated with each indicated sequence dataset as well as the ability of Metavisitor to detect HIV in datasets and patients are indicated.

https://doi.org/10.1371/journal.pone.0168397.t002

Use Case 3–2.

Yozwiak et al. searched the presence of viruses in RNA Illumina sequencing data from serums of children suffering from fevers of unknown origins [29]. In this study, paired-end sequencing datasets were depleted from reads aligning to the human genome and the human transcriptome using BLAT and BLASTn, respectively, and the remaining reads were aligned to the NCBI nucleotide database using BLASTn. A virus was considered identified when 10 reads or more aligned to a viral genome which was not tagged as a known lab contaminant.

For a significant number of Patient IDs reported in Table 1 of the article [29], we were not able to find the corresponding sequencing files in the deposited EBI SRP011425 archive. In addition, we did not find the same read counts for these datasets as those indicated by the authors. With these limitations in mind, we downloaded 86 sequencing datasets that could be further concatenated and assigned to 36 patients in Yozwiak et al [29]. As sequence reads in SRP011425 datasets are 97 nt long, we adapted a workflow for this Use Case 3–2 (S16 Fig) from the one used in the Use Case 3–1 with the following modifications: (i) sequences reads were depleted from human sequences and viral reads were selected by alignment to the NCBI viral sequences using the Galaxy bowtie2 tool (S1 Table) instead of the sRbowtie tool; (ii) viral reads were submitted to Oases assembly using kmer values ranging from 13 to 69 to take into account long reads; (iii) the SAM file with reads alignments to the vir1 bowtie2 index was parsed using the “join” and “sort” Galaxy tools in order to detect putative false negative datasets with viral reads that fails to produce significant Oases viral contigs.

This workflow generated a report file (S15 File) summarized in Table 3. The results show that Metavisitor detected the same viruses as those reported by Yozwiak et al. in 17 patients. Although viral reads were detected in 16 other patients, they were not covering sufficient portions of viral genomes to produce significant viral assemblies. Finally, in the three remaining patients (patients 363, 330 and 345 in Table 3), we detected viruses (Dengue virus 2, Stealth virus 1 and Dengue virus 4, respectively) other than those identified by Yozwiak et al. These discrepancies are most likely due to misannotation of some of the deposited datasets, which precludes further detailed comparisons.

Download:

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

Table 3. Summary of virus detection in 36 traceable patients of the Use Case 3–2.

The Data of this table were extracted from the Metavisitor report file available as S15 File. Values of the column “Coverage of complete viral genome (%)” correspond to the fractions (in %) of the complete viral genomes that are covered by blast hits of viral contigs to these genomes and values of the column “Mean blast bit score” correspond to the mean values of the bit scores observed for these blast hits. Note that blast alignments to incomplete viral genomes were not taken into account. For detection of false positives, reads were aligned to the bowtie2 vir1 index before de novo assembly and counts of these reads were reported in the column “Read mapping to vir1 using bowtie2”).

https://doi.org/10.1371/journal.pone.0168397.t003

Use case 3–3.

Matranga et al. recently improved library preparation methods for deep sequencing of Lassa and Ebola viral RNAs in clinical and biological samples [31]. Accordingly, they were able to generate sequence datasets of 150 nt reads providing high coverage of the viral genomes. We used these datasets, relevant in the context of Lassa and Ebola outbreak and epidemic response, to demonstrate the versatility of Metavisitor as well as its ability to generate high throughput reconstruction of viral genomes.

In order to take into account longer reads and higher viral sequencing depths in the available datasets [31], we adapted a Metavisitor workflow for Use Case 3–3 (S17 Fig) as follows: (i) sequencing reads were directly aligned to vir1 sequences using bowtie2, without prior depletion by alignment to the human or rodent hosts; (ii) the Trinity de novo assembler [23] that performs well with longer reads was used instead of Oases (S1 Table); (iii) reconstruction of Lassa and Ebola genomes from the sequences of the blast hits with the nucleotide viral blast database was directly performed with the “blast to scaffold” tool without CAP3 assembly since the Trinity contigs were already covering a significant part of the viral genomes; (iv) the reports generated by our “Parse blast output and compile hits” tool as well as the reconstructed genome generated for each sample were merged in single datasets for easier browsing and subsequent phylogenetic or variant analyses; (v) for adaptability of this workflow to any type of virus, we allowed users to specify two input variables at runtime: the name of the virus to be searched for in the analysis and the identifier of the sequence to be used as guide in genome reconstruction steps.

We imported 63 sequence datasets available in the EBI SRA PRJNA254017 and PRJNA257197 archives [31] and grouped these datasets in Lassa virus (55 fastq files) and Ebola virus (8 fastq files) dataset collections (see Table 4 for description of the sequence datasets). On the one hand, we executed the workflow (S17 Fig) taking the Lassa virus dataset collection as input sequences, “Lassa” as a filter term for the “Parse blast output and compile hits” tool and the NCBI sequence NC_004297.1 as a guide for reconstruction of the Lassa virus segment L. On the other hand, we executed the workflow taking the Ebola virus dataset collection as input sequences, “Ebola” as a filter term for the “Parse blast output and compile hits” tool and the NCBI sequence NC_002549.1 as a guide for reconstruction of the Ebola virus genome.

Download:

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

Table 4. Summary of detection of Ebola and Lassa viruses in Use Case 3–3.

The table summarizes the Metavisitor report files available as S16 and S17 Files.

https://doi.org/10.1371/journal.pone.0168397.t004

The results of both analyses are summarized in Table 4. Metavisitor was able to detect Ebola virus in all corresponding sequence datasets (S16 File) as well as Lassa virus in 53 out the 55 sequence datasets generated from Lassa virus samples (S17 File). Consistently, Matranga et al did not report reconstructed Lassa genomic segments from the two remaining datasets, which likely reflects high read duplication levels in the corresponding libraries [31]. The reconstructed Lassa virus L segments and Ebola virus genomes are compiled in S18 and S19 Files, respectively. In these sequences, de novo assembled segments in uppercases are integrated in the reference guide sequence (lowercase) used for the reconstruction. To note, for viruses with segmented genomes such as Lassa virus, the workflow has to be used separately with appropriate guide sequences for the segment to be reconstructed. As an example, we used this workflow with the filter term “Lassa” for the “Parse blast output and compile hits” tool and the Lassa S segmentNC_004296.1 for guiding the reconstruction (S20 File).

At this stage, users can use the genomic fasta sequences for further analyses. For instance, multiple sequence alignments can be performed for phylogenetics or variant analyses, or reads in the original datasets can be realigned to the viral genomes to visualize their coverages, as has been done in Use Cases 1 and 2.