Performing Ago HITS-CLIP on BCBL-1 and BC-3 cells produced a catalogue of putative cellular and viral miRNA targets (Table S3
). We carefully modified the original Ago HITS-CLIP protocol 
by adding more stringent wash steps(see Text S1
) to the Ago immunoprecipitation. In addition, we constructed both the miRNA and mRNA libraries from the 130 kDa complex, while previous studies isolated miRNAs from the 110 kDa complex 
. Three biological replicates and two technical replicates were sequenced to monitor biological variation and assay reproducibility. Finally, we excluded from the targetome analysis all KSHV miRNAs that were recovered at very low levels (). For the miRNA libraries, which are less complex, cross-validation was very high across biological replicates(R2
>0.9; Figure S1A, S1B
). For the mRNA libraries reproducibility was lower (R2
≥0.53; Figure S1E
, F) but well within the observed variation of previously published studies 
. Variability from the complex experimental procedure (see also below) and the small amount of Ago-associated RNA extracted and sequenced yields libraries that are not 100% representative; hence biological replicates add to both stringency and depths of the targetome analysis.
We retrieved 16 of 31 previously published KSHV miRNA targets (Table S5A
) and recovered 33 of 114 genes identified in PEL cells by RIP-CHIP without prior UV cross-linking(Table S6A
. Very recently, Gottwein et al. reported the identification of more than 2000 putative KSHV miRNA targets by PAR-CLIP 
. While our study as well as the work from Gottwein et al. analyzed BC-3 cells, the remaining targets were defined in two very different cell lines, BCBL-1 (our study) and BC-1 (Gottwein et al.). BCBL-1 cells are infected only by KSHV, BC-1 cells also express a large number of EBV-encoded miRNAs. Moreover, Gottwein et al. allowed for 7mer1A seed matches, which were not included in our analysis. Comparison revealed42% overlapping targets between the PAR-CLIP BC-3 target list and our data set (BCBL-1 and BC-3; Table S6B
). Moreover, enriched GO terms are similar between both studies. It is not well understood how both experimental platforms compare. Certainly, the cross-linking method (requirement for the presence of a uridine at the cross-linking position in PAR-CLIP, but not in HITS-CLIP), the nucleotide-specificity of the RNase used for clipping (RNase T1 vs. RNase A) and the extent of RNAse digest 
, as well as the choice of linkers (ligation bias; 
), and finally the number of PCR cycles all contribute to differences in the composition of HITS-CLIP and PAR-CLIP libraries. Moreover, the experimental procedure contains two steps with a strong inherent variability: the excision of the Ago-miRNA-mRNA complexes, and the excision of the PCR products, which add substantial variation. On the bioinformatics side, the algorithms and parameters chosen for alignment to the reference genome, algorithms and cut-off criteria used for cluster calling, and definition of seed matches influence which targets will be present in the final target lists. To date, only one study directly compared HITS-CLIP and PAR-CLIP data sets and determined a cross-validation index of R2
>0.4 to 0.65 between the two platforms 
, which is in good agreement with the overlap between our HITS-CLIP and the recent PAR-CLIP data 
.Notably, both data sets in the study by Kishore et al. 
were analyzed by the same analysis pipeline. This suggests that PAR-CLIP and HITS-CLIP are both specific and that variations in the recovered miRNA targetome are mostly due to experimental rather than bioinformatics differences. To uncover the complete miRNA targetome may therefore require the combination of multiple approaches. Hence, the 1170 and 950 targets identified in 2 of 3 repetitions for BCBL-1 and BC-3, respectively, partially validate and moreover complement the PAR-CLIP data set, which lacked biological replicates 
KSHV miRNAs may contribute to and reinforce the regulation of key pathways important for viral biology
The best characterized KSHV miRNA targets so far are mostly involved in regulating immune evasion (MICB
), pro-apoptotic pathways (BCLAF1
), and cell cycle control (BACH1
, and C/EBPβ
); for review see 
. The Ago HITS-CLIP-derived targetome shows strong enrichment for genes involved in these pathways, thus significantly expanding what to this point was solely based on single target gene studies. In addition, GO analysis suggests new host cell pathways to be targeted, such as glycolysis, lymphocyte activation and the ubiquitin/proteasome pathway, opening up additional interesting themes for functional studies. Finally, one clearly emerging concept from this HITS-CLIP data set is that multiple key pathways and processes such as the NFκB pathway, MHC class I-mediated immune surveillance, and cell cycle control can be co-regulated by both virally encoded proteins and miRNAs.
BCBL-1 and BC-3 cells differ with respect to miRNA expression and targeting
MiRNA library analysis revealed strong differences in Ago-associated miRNAs in BCBL-1 and BC-3 cells, with KSHV miRNAs comprising 18% of all miRNA reads in BCBL-1, and an astonishing 73% in BC-3, and numbers of single KSHV miRNAs being up to 10-fold higher in BC-3. Similar results for the overall KSHV versus human miRNA count in both cell lines were obtained by the recent PAR-CLIP study 
). Interestingly, several studies have analyzed KSHV miRNA expression in additional PEL cell lines and found differences not only with respect to overall expression levels but moreover also differences in the relative abundance of specific viral miRNAs 
. The fact that such expression differences likely affect targeting further supports the notion that miRNA targetomes are strictly context dependent.
Surprisingly, despite the much higher levels of KSHV miRNAs in BC-3 cells compared to BCBL-1, we identified similar KSHV miRNA target numbers in both cell lines, which were even 15–20% lower in BC-3. Only the number of transcripts exclusively targeted by KSHV miRNAs was slightly higher in BC-3 (). In contrast, we found that the number of genes targeted by human miRNAs (either exclusively or with additional KSHV sites), was almost 2-foldhigher in BCBL-1 than in BC-3. Thus, while the presence of more human miRNAs is correlated with more putative targets, the same appears to not be true for KSHV miRNAs. In this context it is interesting to note that we observed some differences between reported relative miRNA frequency observed by small RNA cloning 
and the relative frequency by which they were associated with Ago in BCBL-1 and BC-3 cells. Specifically, KSHV passenger strand miRNAs (miR-K12-3*, -5*, -8*, as well as -9* in BCBL-1), but also guide strand miRNAs (miR-K12-3, -10a, and 10b)are very modestly expressed, but have a relatively higher Ago-association rate (Figure S7
). Moreover, for two of the three miRNAs with the highest incorporation-to-expression ratio and also an overall high incorporation level, miR-K12-3* and -8*, we identified only few targets. This raises the possibility of an additional function of some viral miRNAs besides seed sequence-specific target silencing: by being present in very high numbers in KSHV-infected PEL cells (especially in BC-3, but to a lesser extent also in BCBL-1as well as in BC-1 
), they might prevent human miRNAs from accessing RISCs, which would lead to a global de-repression of host genes. Indeed, we observed a strong impact on the target numbers of human miRNAs. Read counts of miR-142-3p and the miR-30 family, which are the most frequent Ago-associated miRNAs in BCBL-1, were reduced 4–5-fold in BC-3(). Accordingly, we also identified about 4-fold less targets in BC-3. Gene Ontology analysis showed that a significant fraction of the BCBL-1-specific miR-142-3p and miR-30 targets(many of them targeted by both miRNAs/families) are involved in protein transport and localization, chromatin organization, macromolecule catabolic processes, and protein degradation. Hence, these processes might be de-repressed in BC-3.Recent very elegant studies interrogating the quantitative aspects of miRNA targeting documented how shifting the ratio between miRNA and target mRNA copy numbers profoundly affects silencing efficiency 
. Hence, flooding host cells with viral miRNAs, a phenomenon first described by Dolken et al. in the context of denovo
HCMV infection 
, maybe an additional mechanism by which herpesviruses induce cells into an activated state. Together with the fact that miRNAs from different viruses have evolved to target common pathways (i.e. apoptosis and cell cycle control) by direct silencing, this suggests that specific gene targeting and global inhibition of host miRNA function both contribute to gene expression differences in KSHV-infected cells.
In summary, our stringent and well-controlled approach provides a working list for functional follow-up studies to decipher viral (and host) miRNA function in KSHV-infected cells. In addition, the data strongly demonstrate that the KSHV miRNA targetome can significantly vary based on the miRNAs' overall abundance and RISC-incorporation, and by transcriptome differences between different PEL cell lines. As a consequence the putative PEL miRNA target catalogues presented by our HITS-CLIP data and the recently reported PAR-CLIP data 
represent an important starting point for many mechanistic studies. However, a full understanding of the role that KSHV miRNAs play in viral biology will require the combination of viral genetics with ribonomics approaches performed in all cell types associated with KSHV pathogenesis as well as in primary tumor biopsies.