In-Depth Analysis of Kaposi's Sarcoma-Associated Herpesvirus MicroRNA Expression Provides Insights into the Mammalian MicroRNA-Processing Machinery^▿

Cells, TPA induction, and RNA.

BC-3 cells were maintained in RPMI medium supplemented with 20% fetal bovine serum (FBS). BC-1 and BJAB cells were maintained in RPMI medium supplemented with 10% FBS. Where appropriate, cells were induced with tetradecanoyl phorbol acetate (TPA) at a final concentration of 20 ng/ml for 16 h. Total RNA was isolated from both induced and uninduced cells by using Trizol (Invitrogen), according to the manufacturer's directions.

cDNA library preparation and data analysis.

cDNA libraries for Solexa/Illumina sequencing were prepared as previously described (34). Briefly, small RNAs were isolated by preparative gel electrophoresis and sequentially ligated to 3′ and then 5′ linkers. Primers complementary to the linker sequences were used for reverse transcription and PCR in order to generate cDNA libraries for deep sequencing. Raw sequencing data were filtered to remove reads lacking identifiable 3′ linker sequences and/or reads falling outside of the predicted miRNA size range. A list of final usable reads was then collapsed down into a list of unique sequences which was analyzed against the KSHV genome (NC_003409) and the mature miRNA database from miRBase (release 9.2) by MegaBLAST using the formatdb, megablast, blastoutparse, and filter_alignment scripts of the miRDeep software package (9).

qRT-PCR for Rta and Mta mRNA.

Rta and Mta mRNA levels were analyzed using the Power SYBR green quantitative reverse transcription-PCR (qRT-PCR) system (Applied Biosystems), according to the manufacturer's directions, using 275 ng of total RNA. Total RNA was DNase treated and reverse transcribed using a random 9-mer primer mix. All values are expressed as a multiple of the background signal detected using RNA derived from the KSHV-negative BJAB B-cell line. Values were normalized to levels of β-actin, and assays were performed in triplicate. The following primers were used for PCR: Rta, forward primer 5′-AGGCATCCCAAGGCATTA-3′ and reverse primer 5′-ATGCGGCTGTCAGAAATG-3′; Mta, forward primer 5′-ACACTGGAAGACGAGCAA-3′ and reverse primer 5′-GCAGTTGAGAACGACCTT-3′; β-actin, forward primer 5′-CACACCTTCTACAATGAGCTGCGTG-3′ and reverse primer 5′-ATGATCTGGGTCATCTTCTCGCGGT-3′.

Stem-loop qRT-PCR for KSHV miRNAs.

Viral miRNA levels were quantified using TaqMan MicroRNA assays (Applied Biosystems) according to the manufacturer's directions using 10 μg of total RNA per RT reaction. Values are expressed as fold expression over an RNA sample derived from uninfected BJAB cells. All values were normalized to levels of β-actin, and assays were performed in triplicate. Custom primers for each miRNA were designed against the most common isoform identified by deep sequencing.

Sequencing of the miR-K9 locus.

The following primers were used to amplify a portion of the KSHV DNA genome encompassing the pri-miR-K9 hairpin from both BC-3 and BC-1 cells: 5′-GGCCAACATAAAGTGTGGATGGCC-3′ and 5′-ATCGAACAACCACGCTTGGCTAAC-3′. The resultant DNA fragments were then sequenced using standard techniques.

Deep sequencing of KSHV miRNA species.

Previous efforts to identify KSHV-encoded miRNAs by conventional sequencing used the primary effusion lymphoma (PEL) B-cell lines BC-1 and BCBL1, which are latently infected with KSHV and Epstein-Barr virus (EBV) or with KSHV alone, respectively (7, 25, 28). For deep sequencing of small RNAs of both viral and cellular origin in a latently KSHV-infected cell context, we selected a third latently KSHV-infected PEL cell line, BC-3, which is not coinfected with EBV and which has been shown to spontaneously produce low levels of KSHV virions in culture (1). We initially harvested total RNA from BC-3 cells and then isolated ∼18- to 24-nt-long RNAs by gel electrophoresis. These RNAs were sequentially ligated to 3′ and 5′ linkers and subjected to reverse transcription followed by PCR amplification. The resultant cDNAs were then directly sequenced using a Solexa/Illumina apparatus.

Deep sequencing resulted in 26,663,462 total sequence reads, of which 19,398,417 were used for subsequent analysis, based on the detection of the intact 3′ linker sequence and an appropriately sized cDNA insert. Of these, 13,469,484 were found to derive from one of the 12 known KSHV pre-miRNAs, while 1,151,608 were found to represent cellular miRNA sequences. The KSHV miRNA therefore represented 92% of the total miRNA pool present in BC-3 cells. The remaining 4,775,545 reads either represented fragments of cellular mRNAs or noncoding RNAs (rRNAs, tRNAs, snRNAs, etc.) or were not present in the databases used. A significant level (1 to 2%) of unassignable reads is expected, as the deep sequencing technology is somewhat error prone and the sequenced cDNA samples had undergone both reverse transcription and PCR amplification, which also generates sequencing errors. Finally, we also detected 1,780 KSHV sequencing reads which did not map to any of the 12 known KSHV pre-miRNAs. Almost all of these represented fragments of the ∼10.5-kb primary miRNA that gives rise to all 12 KSHV miRNAs in latently infected cells (5). Importantly, we did not repeatedly recover any additional unique KSHV sequence distinct from the known miRNAs, as would be expected if KSHV encoded additional miRNAs.

An overview of the major miRNA and miRNA* sequence variants and their origins is provided in Table 1, which shows that we recovered not only 11 of the known KSHV miRNAs but also all 11 matching miRNA* sequences, most of which have not been previously detected. We also recovered the previously reported edited version of miR-K10a-3p, called miR-K10b-3p, which contains a G residue in place of the genomically encoded A in the miRNA seed (25). We did not, however, recover any examples of miR-K9, which has previously been identified in the KSHV-infected cell lines BC-1 and BCBL1, as well as in vivo (7, 22, 25, 28). As described below, this is due to the mutational inactivation of miR-K9 in the KSHV genome present in BC-3 cells.

TABLE 1.

Major KSHV miRNA and miRNA* sequence variants

Open in a new tab

^a

Major species of each KSHV miRNA comprising at least 10% of the total mature miRNA population are included.

^b

Genomic coordinates provided based on NC_003409 genome.

Although all 12 KSHV miRNAs derive from a single pri-miRNA in latently infected cells (5), we observed dramatic differences in their levels of expression, as determined by the number of sequence reads recovered (Fig. 1). Specifically, miR-K4 was found to represent ∼78% of the 13,469,484 KSHV miRNA sequence reads obtained, with miR-K3 representing a further ∼12%. The remaining nine KSHV miRNAs represented from 2.4% to 0.19% of the total number of sequence reads. This remarkable discrepancy could reflect differential stability and/or differential processing of the single KSHV pri-miRNA. However, as the prevalence of the miRNA* strands tended to correlate with the prevalence of the mature miRNA strand (Fig. 1), we feel that the latter mechanism is likely to be more important.

A comparison of the recovery of miRNA versus miRNA* strands for each KSHV miRNA shows that in most cases the former is present at 10²- to 10³-fold-higher levels than the latter. The major exception to this generalization is miR-K7, where the dominant miR-K7-3p strand was recovered only ∼2-fold more frequently than the less common miR-K7-5p strand (Fig. 1).

RT-PCR analysis of viral miRNA expression.

To confirm and extend these deep sequencing data, we performed qRT-PCR to detect the various known KSHV miRNAs in BC-3 cells and also in BC-1 cells. In both cases, we analyzed both uninduced cells and cells treated with TPA, which induces entry of a substantial percentage of cells into the productive KSHV replication cycle (36). As shown in Fig. 2A, TPA treatment indeed strongly induced expression of the KSHV mRNAs encoding the immediate-early Rta and Mta proteins, both of which play a key role in initiating the viral productive replication cycle (21, 24, 32). We note, however, that both mRNAs are also detectable in uninduced cells, consistent with the known low level of spontaneous induction of productive KSHV replication seen with both BC-3 and BC-1 (1, 7).

As shown in Fig. 2B, we were able to detect all mature KSHV miRNA species except miR-K9 in BC-3 cells, consistent with the sequencing data presented in Table 1. TPA treatment had little or no effect on the level of expression of miR-K1 through miR-K8 plus miR-K11, which are expressed under the control of a latent KSHV promoter, but did modestly induce expression of miR-K10 and miR-K12, which both lie 3′ to a promoter induced during productive replication (Fig. 2B) (5, 10). Similar data were also obtained for BC-1 cells (Fig. 2C), with the notable distinction that miR-K9 was, as expected, readily detectable in BC-1 cells.

Perhaps surprisingly, there was only a moderate correlation between the levels of KSHV miRNA expression detected by deep sequencing (Fig. 1) and those detected by qRT-PCR (Fig. 2B). Possible reasons for this include differences in the ligation efficiency of linkers to the various viral small RNAs and differences in RT and/or PCR primer annealing efficiency. Nevertheless, the overall correlation between these two very different techniques was reasonable, given these technical constraints.

KSHV miR-K9 is mutationally inactivated in BC-3 cells.

Based on both deep sequencing data (Table 1) and qRT-PCR (Fig. 2B), it was clear that the KSHV miRNA miR-K9 was absent from BC-3 cells. To illuminate why this might be the case, we used PCR to clone and then sequence the relevant segment of the KSHV genome, using DNA isolated from BC-3 cells. As shown in Fig. 3A, we observed a substantial number of missense mutations, and two single nucleotide deletions, in the predicted pri-miR-K9 sequence in BC-3 compared to the canonical KSHV genomic sequence present in BC-1 cells, which express wild-type miR-K9. In particular, these data predict seven mutations in the mature miR-K9-5p miRNA. Analysis of the “pri-miR-K9” sequence present in BC-3 cells nevertheless predicted that this should fold into a stable RNA stem-loop structure (Fig. 3B). As the predicted miR-K9 sequence in BC-3 cells is substantially different from the sequence present in BC-1 cells and most other KSHV isolates, we were concerned that we might have missed a novel variant of miR-K9 during analysis of our deep sequencing data. However, reexamination of these data using the sequence information shown in Fig. 3 again failed to identify any mature “miR-K9” variants in BC-3 cells. It is unclear why the predicted pri-miR-K9 stem-loop shown in Fig. 3 is not recognized by the microprocessor, but it has been reported that the microprocessor requires a large (≥10-nt) terminal loop for efficient pri-miRNA recognition and cleavage (38).

Sequence variation at the ends of KSHV miRNAs.

As noted above, target mRNA recognition by miRNAs is usually highly dependent on full sequence complementarity between the mRNA and the miRNA seed sequence, located between nucleotides 2 and 8 from the 5′ end of the miRNA (4). As a result, sequence differences at the 5′ end can greatly affect the ability of an miRNA to target a given mRNA. Conversely, while there have been reports documenting a role for the 3′ end of miRNAs in target selection, these sequences are generally less important for target discrimination. Accordingly, deep sequencing of miRNAs in Drosophila and in Caenorhabditis elegans has documented the presence of very little sequence variation at the 5′ ends of most cellular miRNAs, while variation at the 3′ ends was fairly common (26, 27).

As shown in Fig. 4A, a combined analysis of the mature KSHV miRNAs listed in Table 1 also revealed almost no sequence variation at their 5′ ends and yet significant variation at the 3′ end. Analysis of the sequence variation of the miRNA* strands reported in Table 1 revealed a somewhat higher level of sequence variation (Fig. 4B), which is curious, given that the mature miRNA and miRNA* strands are generated by the same processing events. We did not see any difference in the level of 5′ sequence heterogeneity dependent on whether the mature miRNA derived from the 5p arm of the pre-miRNA, in which case the 5′ end would have been generated by Drosha cleavage (Fig. 4C), or from the 3p arm, in which case the 5′ end would have been generated by Dicer cleavage (Fig. 4D). However, when the 3′ end was generated by Dicer (Fig. 4C), somewhat more sequence variability was observed than when the miRNA 3′ end was generated by Drosha (Fig. 4D).

Although 10 of the 11 KSHV miRNAs showed a tightly constrained 5′-terminal sequence, one exception was noted. Specifically, we observed a high level of sequence variation at the 5′ end of the mature miR-K10-3p miRNA, with ∼40% of all the cDNAs sequenced being 1 nt shorter at the 5′ end. This difference was the same for both the edited and unedited forms of miR-K10-3p (Table 1) and has the potential to strongly affect miR-K10-3p function.

Comparison of the dominant sequence variants of the 11 miRNAs recovered from BC-3 cells with the sequences recorded on the miRBase website (12) revealed two important discrepancies. The first is that miRBase gives the sequence of miR-K8-3p—which is the dominant form of miR-K8—as 5′-UAGGCGCGACUGAGAGAGCACG-3′, while our data recovered 5′-CUAGGCGCGACUGAGAGAGCA-3′ as the most prevalent form. This form is 1 nt longer at the 5′ end and 2 nt shorter at the 3′ end than the version in miRBase. Of note, an additional 1 nt at the 5′ end of miR-K8-3p was also reported by two of the three groups that used conventional sequencing to analyze the KSHV miRNAs present in BC-1 and BCBL1 cells (7, 28), and so this is unlikely to represent a BC-3-specific phenomenon.

A more dramatic sequence difference with miRBase is seen with miR-K12. In miRBase, the dominant form of miR-K12 is predicted to be miR-K12-5p. However, our cloning data reveal that the dominant form of miR-K12 is actually miR-K12-3p, which was recovered ∼100-fold more often than miR-K12-5p (Table 1 and Fig. 1, 2, and 5). We note that miR-K12 has not been sequenced previously and was instead detected only by Northern analysis (13), so that the previously reported mature miRNA sequence was actually hypothetical.

MicroRNA offset RNAs, loops, and reverse complements.

Additional analysis of the small RNA sequence reads revealed several short RNAs that are likely to represent by-products of KSHV pri- or pre-miRNA processing or that lie antisense to the known KSHV miRNAs (Table 2). The most common KSHV RNA sequences obtained, other than the miRNA and miRNA* strands themselves, were ∼22-nt RNAs that lie immediately 5′ or immediately 3′ to one of the KSHV miRNAs. These so-called moRs (30) were initially observed upon deep sequencing of small RNAs expressed in the simple chordate Ciona intestinalis, where a substantial number of small RNAs that were found to immediately flank mature miRNA sequences were recovered. moRs have also been detected at very low levels in Drosophila (27) but have not previously been reported in mammalian somatic cells. Because moRs derive from RNA sequences that lie directly adjacent to either the mature miRNA or the miRNA* strand, it is apparent that one end of each moR must arise from Drosha cleavage (Fig. 5). However, how the other end arises is currently unclear. In our hands, we obtained moRs derived from nine of the KSHV pre-miRNAs, with both 5p and 3p moRs obtained in the case of miR-K4 and miR-K10 (Table 2). As shown in Fig. 5, these are far less prevalent than the adjacent miRNA or miRNA* strands and their level of RISC association is currently unclear.

In addition to a fairly large number of KSHV moRs, we also obtained a number of short RNAs that appear to represent the terminal loops of KSHV pre-miRNAs and that therefore presumably represent unstable by-products of Dicer cleavage of viral pre-miRNAs (Table 2). Finally, we also detected a small number of “reverse complement” (RC) RNAs, in particular in the case of miR-K8 (Fig. 5). RC miRNAs have previously been reported in both Drosophila and herpes simplex virus type 1 (HSV-1) (31, 33, 35) and arise when both strands of a genomic region are transcribed. Bidirectional transcription of a single miRNA locus can result in the generation of “antisense” pri-miRNA hairpins that are then recognized by the microprocessor, although not all antisense pri-miRNA sequences are able to fold appropriately. As shown in Fig. 5, the miR-K8 RC stem-loop looks like a normal pri-miRNA stem-loop and the miR-K8-5p RC and miR-K8-3p RC RNAs that we identified are also located where one would expect. Although the sequences that encode the KSHV miRNAs do not lie antisense to any known viral transcript, these must presumably exist at a low level in latently and/or productively KSHV-infected cells.