Fig 1.
High expression of viral miRNAs in EBV-associated tumors.
Raw long and short fraction RNA sequencing reads were obtained from the Burkitt Lymphoma Genome Sequencing Project[47] (BL) and The Cancer Genome Atlas[59] (GC). Long fraction sequences were aligned to the combined human (GENCODE GRCh38.p13)[44] and EBV[45] transcriptomes. Short fraction sequences were aligned to human and viral miRNA sequences (miRbase v22[34]). (A) Principal component analysis (PCA) of BL and GC gene expression. (B) The viral percentage of total gene expression in each EBV+ tumor sample (BL, N = 70; GC, N = 24), . (C) The structure of the EBV RPMS1 locus, including the BART miRNA clusters, plotted with SpliceV[106]. The coverage track and splice junction counts were derived from the aligned RNA-Seq reads of an EBV+ BL patient (BLGSP-71-06-00281). (D) Relative abundance of viral miRNAs, plotted as a percentage of all miRNAs expressed
. Each bar represents one EBV+ tumor (BL, N = 69; GC, N = 38). Akata and SNU719 labels indicate the viral percentage of small fraction sequencing reads in Akata and SNU719 cell lines. Note: for TCGA GC dataset, more samples were sequenced by short fraction sequencing than for long fraction sequencing facilitating the analysis of a greater number of EBV positive miRNA datasets than EBV positive long fraction datasets analyzed for panel B.
Fig 2.
EBV miRNAs are over-represented in RNA induced silencing complexes (RISCs).
CLASH and miRNA-sequencing were performed in triplicate for SNU719 and Akata cells. (A-B, left) The viral percentage of total miRNA expression in each sample, (yellow triangles), and the percent of all miRNA-mRNA hybrids containing a viral miRNA,
(green triangles), were calculated. The indicated P-values were calculated using unpaired Student’s t-tests. (A-B, right) The average number of miRNA-mRNA hybrids formed for each miRNA was normalized to its baseline expression level,
. These mRNA-bound proportions were plotted in order of rank on the corresponding x-axis. Each circle represents an individual miRNA; circle size represents expression level; red circles indicate viral miRNAs. The indicated P-values were calculated using the Kolmogorov–Smirnov test (KS), comparing viral and human miRNAs.
Fig 3.
EBV miRNAs have high targeting efficacy.
(A) Venn diagrams comparing genes found to be targeted in Akata and SNU719 cells by cellular (top) and viral (bottom) miRNAs. Only high confidence (>100 h.c.p.m. in Akata cells; >30 h.c.p.m. in SNU719 cells) 3’-UTR targets were considered. (B) h.c.p.m values for each interaction pair were compared between Akata and SNU719 cells, considering the high confidence interactions found in both cell lines. The log2-transformed values from each cell line were correlated, resulting in a correlation coefficient of ρ = 0.28 (Spearman). (C) Schematic of the targeting efficacy calculation. (D) The log2-transformed targeting efficacies were correlated between SNU719 and Akata cells, resulting in a spearman correlation coefficient of ρ = 0.82. (E; left) The distribution of targeting efficacies for each interaction in each CLASH replicate, comparing viral and cellular interactions. P-values were calculated using the KS test. (E; right) The median targeting efficacy of all three replicates. P-values were calculated using paired Student’s t-tests.
Fig 4.
EBV miRNA hybrids are thermodynamically stable.
Interactions in SNU719 cells were categorized by seed match type (NS, no seed; mm8m, mismatch 8mer; 6m, 6mer; 7mM8, 7mer base 8 match; 7mA1, 7mer A1 (A opposite base 1); 8m, 8mer), with “supplemental” interactions requiring >3 bases of complementarity between bases 13–17 of the miRNA and its target (similar results were obtained in Akata cells). (A) Median targeting efficacy for each type of seed match comparing viral and cellular miRNA interactions. (B) The targeting efficacy of each interaction binned by flanking A/U content of each miRNA target site. The two most proximal bases on each side of the seed binding region were considered. P-values were calculated using the KS test. (C) The mean number of flanking A/Us (max = 4) for cell and EBV miRNA target sites. P value was calculated via KS test. (D) The predicted local site accessibility score of each miRNA target site, using RNAplfold (using the following parameters: -W 80 -L 40 -u). The scores indicate the probability that all 14 bases, centered on position 7 of the miRNA target site, will be unpaired. P-values were calculated using the KS test. (E) Predicted minimum free binding energies (ΔG) were calculated for each hybrid using the RNAcofold function of the Vienna RNA Suite[107]. As a control, ΔG calculations were performed on shuffled sequences, with 100 permutations performed for each hybrid pair. The ΔG values of cellular and viral hybrids were compared by KS test.
Fig 5.
EBV miRNAs and their targets have strong inverse correlations in primary BLs and GCs.
For each CLASH miRNA-mRNA hybrid, mRNA and miRNA expression levels (c.p.m.) were correlated across all EBV-positive BL and GC tumors. Correlations were performed on hybrids that met the following criteria: 1. Average miRNA expression in tumors > = 10 c.p.m., 2. Average mRNA expression in tumors > 5 t.p.m., 3. > 3, and 4. miRNA-mRNA interactions occurred within the 3’-UTR, where miRNA mediated repression is more often effective[108]. Cumulative distribution plots were generated using ranked Spearman correlation coefficients for each species. As a control, expression levels of the miRNA of each hybrid pair analyzed was correlated with expression levels of a random mRNA across the same tumors. For each randomized interaction, 100 permutations were run. Correlation values of EBV miRNA-mRNA pairs were compared to randomized controls; P-values were generated using the KS test.
Fig 6.
γHV miRNAs target components of immune signaling pathways.
(A) SNU719 hybrids containing EBV miRNAs. Each interaction was represented by a circle; circle size corresponds to the total number of hybrids formed (); The y-axis values represent the percent of all hybrids that contain the indicated miRNA,
. (B) The distribution of y-axis values from (A), extended to all hybrids. Cellular and viral hybrids were compared; P-value was generated using the KS test. (C) The fraction of individual miRNAs hybridizing with each mRNA,
, comparing viral and human distributions; P-value was generated using the KS test. (D) Pathways targeted by EBV and cellular miRNAs. Protein-protein interaction networks of each of the top 20 EBV or cellular target genes were obtained from StringDB[53], and resulting protein names were submitted to Enrichr[54] for pathway enrichment analysis (interrogating pathways included in the Reactome database). All statistically significant pathways (FDR < 0.05) were assigned to the target gene.
Fig 7.
Viral miRNAs interfere with anti-viral immunity.
The sum of all viral miRNA c.p.m. was correlated to expression levels of each mRNA (c.p.m.) in EBV-positive BL and GC tumors. The correlations were ranked, then gene lists were interrogated using GSEA. (A) Stacked GSEA curves of the Hallmarks IFNγ signaling pathway (left) and the TNFα-NFκB signaling pathway (right). NES, nominal enrichment score, FDR, false discovery rate. (B) Immune cell infiltrates were inferred using CIBERSORTx. Each cell type was correlated (spearman) to the sum of EBV miRNAs across EBV-positive BLs and GCs. Circle size represents the average CIBERSORTx absolute score across all tumors, filled circles represent statistically significant (P < 0.05) spearman correlations. (C) T-cell clonotypes were obtained for each EBV-positive tumor sample using MIXCR; The number of unique clonotypes was correlated with the sum of viral miRNAs in EBV-positive BLs and GCs.