Table 1.
Sequencing panel characteristics.
Figure 1.
Virana's approach to identifying viral transcripts in human tumors.
a) Transcriptome sequence samples are first mapped to a combined set of human and viral reference sequences in a splicing-aware fashion. b) Unmapped or discordantly mapped read pairs are further processed by assembly methods to detect novel viruses or transcript chimeras that may indicate proviral integration events. c) Reads mapping to one or more viral genomes (HITs) are analyzed in an integrated fashion by considering human homologous mapping locations and viral taxonomies. This process results in a number of homologous regions (HOR) for each viral family. HORs are represented as multiple sequence alignments incorporating a wealth of sequence information. Alignments are further enriched by taxonomic annotations and phylogenetic analyses.
Figure 2.
Detection of divergent viruses.
Performance comparison of Virana, CaPSID, and RINS at detecting viral reads at different rates of simulated sequence divergence among a background set comprising human genomic reads. The background set without any spike-ins of viral reads serves as negative control. Left panel: stacked bars represent absolute numbers of detected reads grouped by sequence divergence, correctness of classification (TP: true positive, FP: false positive), and detection method. Falsely classified reads not assigned to any of the viral families present in the validation are labeled as false positives (FP). Colored segments indicate to which viral families the reads were assigned. Each condition allowed for the correct detection of up to reads. Right panel: color coded markers for each condition and detection method indicating which viral families were identified. A maximum number of
viral families could be correctly identified in each condition.
Figure 3.
Time required for data analysis.
Cumulative time in minutes required for analysis of the divergence validation set. Times are reported for the negative control without viral spike-ins as well as for four mixed data sets consisting of negative control background set with viral spike-ins at different divergence rates. Segments within bar plots represent different analysis processes employed by the three viral detection methods Virana, CaPSID, and RINS. All measurements are based on a single CPU Intel(R) Xeon(R) E5-4640 clocked at 2.40 GHz.
Figure 4.
Detection of low-coverage, homologous, and chimeric viral transcripts.
Displayed are performances of Virana, CaPSID, and RINS at detecting the three human-viral homologous gene regions Bo17, gag, and vABL. Performance is quantified in terms of sensitivity (right panel) and absolute number of reads correctly identified (left panel) at differing sequencing coverages ( fold). Methods are validated at detecting both isolated gene regions (upper part) as well as at detecting human-viral fusion transcripts involving each of the three gene regions fused to the human TP53 proto-oncogene (lower part). Specificity of detection is 1.0 (100%) for all detection methods (not displayed).
Figure 5.
Estimation of required sequencing coverage for detection of a homologous region.
Probability of successful region construction by Virana depending on the lengths of the transcripts being sought, the region linkage parameter , as well as characteristics of the sequencing platform employed. Colored areas represent overlapping standard error bands of the mean, denoting the uncertainties of the estimations. The probability of Virana to detect a homologous region depends on the length of the viral transcript being sought, the linkage parameter
of the homologous region, as well as the transcript coverage and read length of the sequencing platform employed. Given characteristics of the sequencing process applied for NB1 sample panel, an average viral cDNA of length
bp requires a minimal transcript coverage of
in order to be reliably detected using a linkage parameter of
as employed in this study (upper left quadrant, dashed blue vertical line). Technologies affording longer read length as used for the NB2 panel typically also afford higher sequencing depths. However, at a fixed coverage these technologies generate a more highly fragmented region linkage due to a smaller number of longer reads, resulting in lower probability of generating contiguous homologous regions (lower left quadrant). Lower transcript coverage is sufficient for longer transcripts transcribed from a complete A-MuLV genome (upper right panel, dotted black vertical line) or smaller values of the region linkage parameter
.
Figure 6.
Estimation of required cellular transcript abundances for achieving a given transcript coverage.
Sequencing coverage of viral transcripts is depending on the average number of transcript copies per cell in the sequenced sample, on the length of the viral transcript being sought, and on characteristics of the sequencing process. In order to better visualize the optimal sequencing depth required for detection of viral factors, we estimated the required number of transcript copies per cell for different sequencing depths. These sequencing depths are expressed as factors relative to the depths employed for the NB1/NB2 panel generated in this study (which are here reported as a relative sequencing depth of 1).
Figure 7.
Overview of identified homologous regions in positive and negative experimental controls.
Left panel: cumulative numbers of reads assigned to viral taxonomic families (log-scale). Each bar represents a homologous group (HOG) colored according to viral taxonomic family. Bars comprise several segments, each representing a homologous region (HOR). Heights of segments indicate the putative origin of reads assigned to this region (human, viral, or ambiguous). Viral families of bacteriophages are marked accordingly. Right panel: Analogous to left panel, but the lengths of bars represent relative rather than absolute abundances quantified in cumulative reads per million reads mapped (RPMM).
Figure 8.
Overview of identified homologous regions in neuroblastoma samples.
Left panel: cumulative numbers of reads assigned to viral taxonomic families (log-scale). Each bar represents a homologous group (HOG) colored according to viral taxonomic family. Bars comprise several segments, each representing a homologous region (HOR). Heights of segments indicate the putative origin of reads assigned to this region (human, viral, or ambiguous). Viral families of bacteriophages are marked accordingly. Right panel: Analogous to left panel, but the lengths of bars represent relative rather than absolute abundances quantified in cumulative reads per million mapped (RPMM).
Table 2.
Mapping rates.
Figure 9.
Human-viral phylogeny based on a HOR.
Phylogenetic tree of HOR #16 of the NB1 stage 4 panel. Viral reference sequences are indicated with red branches and associated tip labels (‘Virus’) while human factors are labeled with green branches. Blue branches represent consensus sequences of neuroblastoma reads (‘Sample’). The tree was generated by the maximum likelihood approach PhyML using the multiple sequence alignment of the HOR as input (see Materials and Methods). Distances between nodes are quantified as substitutions per site. As can be derived from the tree, neuroblastoma consensus sequences are tightly clustered in close proximity to the endogenous retrovirus HERVK9I and two human factors, thereby unambiguously indicating the human origin of these neuroblastoma reads. Clusters of other sequences represent well known sequence homologies, as for example between human ABL1/SRC genes and acutely transforming retroviruses.
Figure 10.
Reconstruction of novel transcripts by de-novo assembly.
Histograms display lengths of reconstructed sequence contigs assembled from unmapped reads of NB2 stage 4 and stage 4S samples (y-axis in log-space). Two independent assembly methods, Trinity and Oases, were used in the reconstruction. The grand total number of contigs reconstructed within each assembly is displayed in the rightmost column. Reconstructed contigs are annotated with their putative taxonomic origin as inferred by comparison with NCBI nucleotide (nt) and protein (nr) archives using TBLASTX database searches.