HSV-1 miRNAs are post-transcriptionally edited in latently infected human ganglia

To analyze the expression of HSV-1 miRNAs in latently infected human neurons, we sequenced small RNAs extracted from sections of 10 latently infected human trigeminal ganglia (TGs) derived from 10 individuals (Table S1). Briefly, we extracted RNA and DNA from all samples and confirmed the presence of the HSV-1 genome prior to sequencing (Fig. S1). Small RNA libraries were generated using the NEBNext Multiplex Small RNA Library PreSet for Illumina and sequenced on the NextSeq 500 (Illumina). The obtained sequencing data sets were processed with sRNAtoobox (49), an integrated collection of small RNA research tools, applying stringent parameters with exclusion of all bases with Phred score of <30.

We obtained between 2.6 × 10⁷ and 1.5 × 10⁷ high-quality sequencing reads per sample, of which between 2.6 × 10⁶ and 9.5 × 10⁶ represented miRNAs (Table 1). As expected, the vast majority of reads were human miRNAs (>99.8%), and only a small number of HSV-1 miRNA were obtained (2–6,884 reads/sample) (Table 1). HSV-1 miRNAs detected in each individual sample did not correlate with the number of human miRNAs detected and ranged from 0.00003% to 0.17% (Table 1). These results could indicate differences in viral load, differences in processed sections of individual TG, and other factors.

The expression of HSV-1 miRNAs in latently infected human ganglia has been comprehensively studied in only a few cases (10, 50), and in contrast to studies in cultured cells and animal models, only a small subset of miRNAs has been detected (miR-H2 to miR-H8). Our study shows the expression of these miRNAs in all samples (Fig. 1) with a remarkably similar pattern. In two samples, we detected a low number of HSV-1 miRNAs, which could indicate a lower viral load or a variation within each tissue section. Based on the number of reads, the most abundant miRNA was miR-H4 followed by miR-H3, miR-H2, miR-H7, miR-H6, miR-H8, and miR-H5, with a range of 348 reads per million (RPM) for the most abundant and <1 RPM for the least abundant miRNA. The number of reads does not necessarily correspond to absolute quantification, as these numbers may depend heavily on the sequence of the miRNA and the sequencing protocol (51, 52). Nonetheless, these LAT-associated miRNAs define the pattern of HSV-1 miRNA in latently infected human trigeminal ganglia. In addition to these miRNAs, we detected miR-H1 and miR-H17 in TG4 and TG7, respectively (Fig. 1). miR-H1 was detected with only three reads (0.23 RPM) in sample TG4 but the reads fully matched the miR-H1 sequence deposited in miRbase, indicating a confident result. Since miR-H1 is abundantly expressed during productive infection (9), this could suggest a possible reactivation from latency in one of the samples. On the other hand, miR-H17 was detected with only a single read and the sequence was shorter (19 nt) and had a mutation compared to miRNA in miRBase, which may indicate false detection. It is important to note that the abundantly expressed HSV-1 miRNAs (miR-H2, miR-H3, miR-H4, miR-H7, except miR-H6) were readily detected with both strands of the duplex (5p and 3p) in ratios of 1:4 to 1:100, which is well above the average host miRNA strand bias (>1:1,000). These results suggest, at least for some HSV-1 miRNA, that both arms of the duplex may regulate specific targets. Indeed, this hypothesis is supported by the analysis of RNA-inducing silencing complex (RISC)-associated HSV-1 miRNAs in which both strands of the duplex were associated with the RISC [reference (30), and not shown].

In our previous study, we identified particular sequence variants of HSV-1 miRNAs that could result from active post-transcriptional modifications by ADAR enzymes (50). To investigate this further, we analyzed the sequence variants of HSV-1 miRNAs in our data set. In total, we obtained approximately 29,000 reads representing HSV-1 miRNAs (ranging from 2 to 6,884/sample) under our threshold settings (i.e., maximum of two mismatches, allowing 5′ and 3′ fluctuations). Detailed analysis of the reads showed that 13.1% of all miRNAs had one or two nucleotide substitutions. This is comparable to the substitutions detected in the host miRNA data set (i.e., 11.5%). Remarkably, among all substitutions within viral miRNAs, we observed a very high frequency of A-to-G substitutions, which accounted for more than 50% of all nucleotide substitutions (Fig. 2A). Importantly, the frequency of nucleotide substitutions was comparable between host and viral miRNAs, and most types of substitutions were relatively rare. However, A-to-G substitutions were the most common substitution in both host and virus, and surprisingly, they occurred twice as frequently in virus as in host (Fig. 2B). It is possible that some of the detected substitutions are sequencing errors, but the observed high frequency of A-to-G substitutions indicated an active enzymatic process. Next, we carefully analyzed the A-to-G substitutions for each miRNA and found that the frequency of A-to-G substitutions for most viral miRNAs was very low, similar to the baseline level (<1%) (Fig. 2C). On the other hand, miR-H2-3p was in each sample represented with 35–48% reads having A-to-G substitutions and accounted for nearly 80% of all A-to-G substitutions in HSV-1 miRNAs (Fig. 2C and D). We observed the vast majority of A-to-G substitutions at the nucleotide positions N5 (miR-H2-3p-e5) (Fig. 2D). In addition, we detected a lower level of editing at the N9 nucleotide position (<10%). The pattern of N5 and N9 nucleotide substitutions was conserved in all latently infected TG with detectable miR-H2 (Fig. 2E). Other types of substitutions were at negligible levels (Fig. 2D).

This result was quite surprising, as HSV-1 miRNAs are considered highly conserved. Indeed, comparison of the pre-mR-H2 sequence of laboratory and clinical strains showed strict conservation (not shown). Moreover, the reproducible pattern of the miR-H2 editing in all analyzed samples strongly indicates a specific modification. However, to exclude the possibility that the sequence variants were due to heterogeneous infection, i.e., different virus strains carrying nucleotide variations in the miR-H2 gene, we sequenced PCR amplicons generated on the DNA template of three TGs analyzed (TG2, TG6, and TG10). We also sequenced amplicons obtained from the cloned genome of the HSV-1 strain KOS and DNA extracted from cells infected with HSV-1 in culture, which served as a monoclonal control. In the case of heterogeneous infection, one would expect overlapping sequencing signals for nucleotides A and G. In contrast, we found no evidence of variation at the indicated editing sites in any of the samples (Fig. 2F and G). Overall, our results suggest that miR-H2 variants arise by post-transcriptional sequence-specific editing in latently infected neurons, which is similar to the signature of A-to-I deamination by ADAR proteins (53). Curiously, we also observed other nucleotide substitutions in viral miRNAs, such as T-to-C (17% overall) or C-to-T (5.7%) (Fig. 2A), which were distributed across all detected viral miRNAs with low frequency for individual miRNAs (not shown) and were therefore not analyzed further. Figure 2D shows the substitutions per nucleotide in miR-H4, an example of a highly expressed viral miRNA that, unlike miR-H2, shows no signs of hyperediting.

ADAR1 and ADAR2, the two enzymatically active members of the ADAR protein family, are known to be ubiquitously expressed in various tissues, including the brain (54). To confirm that these genes are expressed in the tissue where we detected potential post-transcriptional editing, we analyzed RNA and proteins extracted from human TG tissue. Using reverse transcription quantitative real-time PCR (RT-qPCR), we confirmed the expression of ADAR1 and ADAR2 in human TGs (Fig. 3). On the other hand, we detected higher levels of ADAR1 p110 protein compared with the interferon-inducible p150 form, which could be detected only with an isoform-specific antibody (Fig. 3). The expression of ADAR2 was below our detection limit, consistent with relatively low levels of the ADAR transcript. In addition, we detected nuclear and moderate cytoplasmic staining of neuronal cells with immunohistochemistry and two different ADAR1 antibodies. ADAR1 expression was also observed in small surrounding cells that may correspond morphologically to satellite glial cells (two middle panels). ADAR2, on the other hand, was observed almost exclusively in nuclei and was also present in surrounding satellite cells (lower panel). Immunohistochemical signals were not detected in stromal connective tissue. Taken together, these results suggest that ADAR proteins might process miR-H2 in latently infected human TG.

We demonstrated that sequence-specific hyperediting of miR-H2 occurs during long-lasting latency in humans, which might have important implications for understanding the molecular mechanisms regulating latency. We next asked whether editing of miR-H2 could also be observed during productive infection in cultured cells, which would strongly suggest sequence-specific deamination. Importantly, under such experimental conditions, the infection can be considered monoclonal with respect to sequence variations. To address these questions, we reanalyzed our previous sequencing of smRNAs from productively infected human foreskin fibroblast (HFF) with the HSV-1 strain KOS at a multiplicity of infection (MOI) of 5 (ref) at 8 and 18 h post-infection (h.p.i.). As expected, we detected almost all previously reported HSV-1 miRNAs (except miR-H3, miR-H18, and miR-H27), which differed significantly in abundance (Fig. 4A). HSV-1 miRNAs were represented with 243 714 reads and 8.71% had one or two nucleotide substitutions, consistent with the basal frequency of substitutions observed in human latency samples. T-to-C (25.0%) and C-to-T (29.3%) nucleotide substitutions were the most common (Fig. 4B), but these were all randomly scattered across all detected miRNAs (not shown). Although it is possible that some of these represent sequencing errors, it cannot be excluded that C-to-T nucleotide substitutions are the result of random APOBEC3 activity.

In addition, we detected high levels of A-to-G substitutions (12.4%), which were significantly higher than most of the other substitutions, but still the phenomenon of hyperediting was less pronounced than in latently infected human ganglia (Fig. 4B). In contrast to latency in humans, miR-H2 editing was not highly overrepresented (14% of all A-to-G editing), and it was slightly above the frequency of A-to-G substitutions in most detected miRNAs. However, we observed two important correlations with the latent samples. First, miR-H2 editing showed a slight but reproducible increase from early (8 h) to late (18 h) times post-infection, i.e., from 2.7% to 5.9% (Fig. 4D). Second, miR-H2 editing occurred at the same nucleotides (N5 and N9) and in similar ratios between these two substitutions as observed for latency (Fig. 4D). In addition, we observed higher frequency of A-to-G substitutions for miR-H11, miR-H12, and miR-H13 (Fig. 4C), miRNAs that arise from transcripts spanning oriS and oriL regions (9). Although we cannot determine whether these substitutions represent true editing events, due to the very small number of reads, it is possible that palindromic sequences of oriS and oriL could be genuine substrates for ADAR binding and editing.

To investigate whether the representative animal model recapitulates the miRNA editing profile in humans, we analyzed miRNAs from acutely and latently infected mouse TGs [3, 14, and 30 days post-infection (d.p.i.)]. As expected, we found an increasing number of miRNAs and reads for each miRNA detected with increasing time post-infection, consistent with the previously determined kinetics of virus replication and establishment of latency in the mouse model after corneal scarification (Fig. 5A). The pattern of miRNAs and nucleotide substitutions within miRNAs resembled the pattern found in human tissue. A-to-G substitutions accounted for the majority of nucleotide substitutions (32%) (Fig. 5B), but were not as pronounced as in human tissues. Nevertheless, miR-H2 changes represented the vast majority of these substitutions (72% and 90% at D14 and D30, respectively), and the pattern was similar to that found in humans (i.e., miR-H2-e5 10.7% and miR-H2-e9 3.5% at D30) (Fig. 5C and D). These results indicate that HSV-1 latency in mice, at least in terms of miRNA expression and the ability of mouse proteins ADAR to edit miR-H2, largely recapitulates human latency.

We were also interested in whether hsv2-miR-H2-3p encoded by HSV-2, the sequence and position homolog of hsv1-miR-H2-3p, also shows evidence of editing in productive and latent infections. We use our previously described data sets (9) to reanalyze them and find evidence of hyperediting. Briefly, for analysis of productive infection, HEK-293 cells were infected with the HSV-2 strain 186ΔKpn, a thymidine kinase-negative viral mutant that can establish latency in TGs without lethality to mice, with an MOI of 1, and samples for sequencing were collected 18 h.p.i. For analysis of latency, mice were infected internasally and TGs harvested 30 d.p.i. As expected, we detected a large number of miRNAs during productive infection (Fig. 6A) and only a limited number of miRNAs during latency (hsv2-miR-H2, hsv2-miR-H7, and hsv2-miR-H24). Similar to productive infection of HSV-1, A-to-G substitutions accounted for approximately 12% and 23% of all substitutions in productive and latent infection, respectively (Fig. 6B and C). hsv2-miR-H2-3p was detected with 40 reads, but we were unable to detect any read with the A-to-G substitution. We are aware that this is a relatively small number of reads, but we were still surprised that we did not find any read with A-to-G substitutions (Fig. 6D), as pre-miRNAs are relatively conserved between these two viruses (Fig. 6E). Interestingly and similar to HSV-1, miRNA arising from a transcript spanning the oriL palindrome, hsv2-miR-H11, shows a relatively high frequency of A-to-G substitutions (16.2%; blue color, Fig. 6D) in productive infection. Overall, the most abundant substitutions in the HSV-2 experiment were G-to-A (Fig. 6B) for the productive infection and G-to-T for the latent sample (Fig. 6C). However, we did not observe confident higher representation of specific substitutions for any of miRNAs (i.e., all below 1%) but rather scattered across all detected reads.

We show evidence of miRNA editing in latently infected human ganglia and in latently infected mice. Moreover, editing can also be observed during productive infection in culture, the system in which miRNA plays a minor role in efficient viral replication. Overall, it is expected that it will be very difficult to determine the biological significance of these events using current experimental models. miR-H2 deletion mutants, e.g., show little phenotype in mice (28). Nonetheless, we wanted to investigate the potential of edited miRNAs to regulate additional viral or host targets.

First, we asked whether the edited form is efficiently associated with the RISC, which should indicate their inhibitory potential. To this end, we infected HFF cells in triplicate at an MOI of 10 and extracted 12 h.p.i. RNAs from whole cell lysates or immunoprecipitated Ago2, the major component of the RNA-inducing silencing complex. We postulated that a consistent ratio of canonical to edited miRNA in Ago2 pull-down and total miRNA would indicate efficient loading of edited miRNA into RISC (55). Indeed, the percentage of edited miRNA (miR-H2-3p-e5) in the total miRNA samples was 8.09%, while it was 10.23% in the Ago2 pull-down, indicating an efficient association with RISC (Table 2).

Next, we predicted the host and viral targets of these miRNAs. Not surprisingly, the targets of miR-H2 and miR-H2-e5 overlapped only slightly using two common prediction algorithms, TargetScan and miRDB (Fig. 7A; Table S4). These results suggest that editing significantly increases the potential of miR-H2 to regulate host genes. Of note, miR-H2-e9 was not included in the analysis because the host target prediction algorithms are largely based on the seed sequence, and miR-H2-e9 has the same sequence as miR-H2. However, this does not exclude a possibility that miR-H2-e9 regulated different transcripts.

Regarding potential viral targets, miR-H2 is fully complementary to the ICP0 transcript and likely represents the main target of miR-H2-3p (8). We used the Bowtie alignment tool (58) to predict potential viral targets of miR-H2-3p and its edited forms. We identified 17 potential binding sites for miR-H2-3p, including the previously identified ICP0. Surprisingly, one nucleotide substitution within the seed region of miR-H2 (i.e., miR-H2-e5) results in a dramatic increase in predicted targets (i.e., 79 loci), which include ICP0, ICP4, and a large number of other productive infection gene transcripts (Table S2). Although the functional characterization of miR-H2-e5 is experimentally challenging and beyond the scope of this manuscript, we carefully analyzed the potential binding sites for miR-H2 and miR-H2-e5 within the ICP0 and ICP4 transcripts using RNAhybrid and allowing G-U base pairing. Similar to Bowtie prediction, editing miR-H2 increased the number of potential target sites in both transcripts (Table S3). Indeed, the canonical miR-H2 has only one fully matched seed-target site in the ICP0 and ICP4 transcripts, whereas editing at position N5 results in four and seven potential seed matches in the ICP0 and ICP4 transcripts, respectively.

To preliminary test whether miR-H2-3p edited at positions 5 and/or 9 has the potential to regulate the expression of ICP0 or ICP4, we performed co-transfection experiments. Briefly, we co-transfected HEK293 cells with a wild-type ICP0 or ICP4 expression plasmid together with a pEGFP plasmid (an indicator of non-selective targeting) and microRNA mimics including canonical miR-H2-3p, miR-H2-3p with scrambled seed sequence (miR-H2-3p-mut), negative control mimic (scrambled sequence), edited miR-H2-3p at position 5 and/or 9, or both (miR-H2-3p-e5, miR-H2-3p-e9, miR-H2-3p-e5-e9; Fig. 8). As shown in Fig. 8, canonical miR-H2-3p significantly decreased ICP0 levels and to laser extent ICP4 levels. Downregulation of ICP4 was expected based on one seed match within the ICP4 transcript (Table S3). On the other hand, miR-H2-e5 strongly decreased the levels of both ICP0 and ICP4 (Fig. 8B through E), whereas miR-H2-e9 had a slight effect on ICP0 only. Interestingly, miR-H2 edited at both sites and miR-H2-e5-e9 showed no effect on either transcript. The levels of these proteins were not affected by miR-H2-3p-mut and negative control. These initial results, albeit in an overexpression system, suggest that miR-H2 may play a broader role in repressing early genes during latency.

Our analysis of 10 latently infected human ganglia reveals remarkably similar patterns of HSV-1-encoded miRNAs in all samples, largely confirming the results of previous studies (10, 50). These include miR-H2 to miR-H8, which likely represent the HSV-1 miRNA latency pattern. We cannot rule out the possibility that with increasing depth (higher number of reads) other miRNAs are not found but would certainly be underrepresented. Similar to humans, latently infected ganglia from mice have the same miRNA expression pattern. It is important to note that standard latent infection models in mice are based on 30-day infection, which is far less that latency studied in humans. Also, latency in mice is established after massive productive infection within TG, which is not the case in humans. Therefore, there is increased potential for residues of productive infection transcripts to exist at 30-d.p.i. mice compared to the human samples, including appearance of some miRNA. The exact function of these miRNAs is still quite enigmatic, mainly because studies on HSV-1 miRNAs during the latent phase of infection have multiple challenges and drawbacks. An important advance in understanding the role of HSV-1 miRNAs has been made using viral mutants lacking single miRNAs, but the observed phenotypes are generally mild (24, 28 – 31, 59). This is to be expected because miRNAs likely regulate the robustness of the system (i.e., sensitive on/off switches) rather than enabling latency or reactivation. Interestingly, in HSV-1, miRNAs present in latency have both arms of the miRNA duplex, suggesting that viruses use the miRNA machinery particularly efficiently.

Our study provides several lines of evidence that miR-H2 is hyperedited during latency, further increasing its targeting potential. First, the sequencing data were analyzed under stringent cutoff values, resulting in highly reliable sequence information. Second, sequencing of DNA in all samples analyzed revealed no evidence of heterogeneity within the precursor of the HSV-1 miR-H2 (pre-miR-H2) locus, suggesting that miR-H2 heterogeneity is due to post-transcriptional modifications of its transcripts. Third, the pattern of miR-H2 expression (ratio of edited to unedited forms) was remarkably consistent in all samples. It is highly unlikely that all patients carry similar genomic variants. Fourth, evidence from sequencing of numerous viral genomes suggests that the miR-H2 locus is highly conserved. Fifth, we found the same editing pattern (i.e., miR-H2 positions N5 and N9) in productively infected cells in culture, albeit at lower frequency, and in the mouse model, suggesting a specific editing process. We believe that the lower frequency of miR-H2 editing in productive infection is due to the ADAR proteins being saturated with the overwhelming viral transcript. This is in contrast to the latency phase, where fewer transcripts are present and can be efficiently edited.

Interestingly, we found no evidence of site-specific hyperediting for the encoded HSV-2 homolog; however, in the context of limitations this study, we cannot rule out the possibility that hsv2-miR-H2 is edited, especially as mice were infected with a TK-null virus, which would limit viral replication in the ganglia prior to latent establishment. In the HSV-2 data set, we found a relatively small number of reads representing miR-H2, and unfortunately, we were unable to obtain human tissues latently infected with HSV-2.

How the ADAR1 and ADAR2 editors select their targets is not clear, but the length and structure of the dsRNA (bulges, loops, and mismatches) play an important role. The precursors of the pre-miRNAs of HSV-1 and HSV-2 are very similar (9), and the adenosine targeted by editing at position N5 is conserved, and we found no clear evidence or motifs for differential specificity of editing. For the pre-miR-H2, the optimal secondary structure was predicted to have a slightly higher minimum free energy (−37.50 kcal/mol), higher than all other commonly expressed miRNAs and higher than HSV-2 miR-H2 (−40.40 kcal/mol). Editing at the N5 position slightly lowers the folding energy. Nevertheless, detailed studies on host miRNAs and viral miRNAs are needed to establish such a correlation. In addition to the A-to-G substitutions, we observed relatively many other substitutions in our analysis. Some of these could be due to technical sequencing errors (51), while others could be active processes. The APOBEC (apolipoprotein B mRNA editing catalytic polypeptide-like) family of proteins binds to both RNA and single-stranded DNA and may be responsible for some of the C-to-T substitutions (60), but investigating this is beyond the scope of this study.

The main question is whether editing of miR-H2 has biologically relevant functions. Our results indicate that the edited miRNA is loaded into the RISC as efficiently as the canonical sequence, suggesting functional properties. Furthermore, we show that editing miR-H2, similar to other edited miRNAs, could increase its targeting potential. Indeed, using co-transfection assays, we demonstrated that canonical miR-H2-3p and miR-H2-3p-e5 can target ICP0 and ICP4 transcripts and limit their protein levels. Moreover, miR-H2-3p-e9 was effective against ICP0. These results could be explained by the proportional number of predicted miRNA binding sites within these transcripts (Table S3). However, an miRNA in which both sites were edited was not effective against either transcript, and the prediction would indicate binding. At this stage, we cannot explain the discrepancy between the computational predictions and the experimental results. Clearly, more sophisticated techniques are needed to determine the exact binding sites and target transcripts and to investigate the functions of these miRNAs. Nonetheless, it is important to note that viruses lacking miR-H2 expression, which includes both canonical and edited miRNA, exhibit a mild phenotype in latency models (28, 31, 34), suggesting that editing may at best contribute to the observed phenotype. Current latency models rely on relatively harsh experimental conditions, including viral reactivation in explanted tissues, which may not provide the resolution necessary to study miRNA-regulated processes. However, recently developed models in primary neurons of latency establishment and reactivation, where reactivation is induced on intact neurons, may in the future prove useful for investigating the function of ADAR in HSV neuronal infection (61). One potential caveat, however, is that studying the biological role of editing is challenging because it is challenging to generate ADAR1 KO in most cells. In addition to the evidence that editing affects targeting properties, it is also possible that editing of the miR-H2 precursor affects its biogenesis, as shown for EBV- and KSHV-encoded and edited miRNAs (43, 45, 46). ADAR1 was shown to directly interact with DICER and affect miRNA processing, maturation, RISC loading, and silencing of target RNAs, all of which remain to be investigated. To our surprise, we did not find evidence of HSV-2 miR-H2 editing, nor in infected cells in culture, nor in latently infected trigeminal ganglia. At this stage, because of the relatively small number of reads, we cannot rule out the possibility that hsv2-miR-H2 is edited. We can speculate that editing may be the molecular mechanism that determines some biological differences between HSV-1 and HSV-2. Therefore, it is critical to sequence relevant human samples, such as the dorsal root ganglia, to determine the status of HSV-2 miRNAs during latency. In addition, our study suggests that other non-coding RNAs, i.e., LAT intron, can be edited during latency, which may explain some of the peculiar properties of this molecule. For example, editing the LAT intron could prevent this abundantly expressed molecule from being recognized as non-self, which would lead to apoptosis (62). Editing could also affect its expression and stability or be essential for its interactome (63), and yet there is still much to learn about the importance of post-transcriptional editing in herpesvirus infection.

Human Embryonic Kidney (HEK293, CRL-1573) and human foreskin fibroblast (generous gift of S. Jonjić, Faculty of Medicine, University of Rijeka) were grown in Dulbecco’s Modified Eagle Medium (PAN-Biotech) supplemented with 10% fetal bovine serum (PAN-Biotech), penicillin/streptomycin 100 µg/µL, 2-mM L-glutamine (Capricorn), and 1-mM sodium pyruvate (Capricorn), under standard conditions in humidified incubator at 37°C and in the presence of 5% CO₂. Wild-type HSV-1 strain KOS (kindly provided by Professor Donald M. Coen and David Knipe, Harvard Medical School, Boston, USA) was prepared in Vero cells and stored at −80°C as previously described (64). For productive infection, cells were seeded on the dish or plate the day before the experiment. After 24 h, cells were infected with specified viruses from a viral stock, with indicated multiplicity of infection or mock infected (uninfected). The medium in all samples was replaced with fresh growth medium 1 h after infection. Samples were collected at specified hours post infection. For latent infection, 6-week-old female CD-1 mice (Charles River) were anesthetized by intraperitoneal injection of ketamine hydrochloride (80 mg/kg of body weight) and xylazine hydrochloride (10 mg/kg) and inoculated with 1.5 × 10⁶ PFU/eye of KOS strain HSV-1 (in a 5-µL volume) onto scarified corneas, as described previously (65). Mice were housed in accordance with institutional and National Institutes of Health guidelines on the care and use of animals in research, and all procedures were approved by the Institutional Animal Care and Use Committee of the University of Virginia. Trigeminal ganglia were harvested 3, 14, and 30 d.p.i. and immediately snap frozen in liquid nitrogen. For RNA extraction, lysis was carried out by addition of RNA lysis buffer (Zymo) and homogenization for 60 s using the BeadBug microtube homogenizer. RNA was isolated from the homogenized mixture using the Quick-RNA miniprep kit (Zymo).

The ICP0-WT plasmid (pRS-1) was a generous gift of Rozanne Sandri-Goldin (University of California). This plasmid includes the entire ICP0 gene, with its promoter, cloned into pUC-18 (35, 66). pcDNA-1-ICP4 contains the entire ICP4 gene (67). pEGFP-N1 was obtained from Clontech-Takara.

Trigeminal ganglia were thawed and small sections homogenized and dissolved in TRIreagent (Ambion). DNA and RNA were extracted from the same sample according to the manufacturer’s instructions. Briefly, chloroform was added in TRIreagent to separate the RNA in the aqueous phase and precipitated with isopropanol. After washing with 70% ethanol, the precipitate was resuspended in nuclease free water and treated with DNase for 30 min. The remaining sample from which the aqueous phase was removed was used for DNA precipitation with absolute ethanol. HEK293T and HFF samples were homogenized in TRIReagent solution (Ambion) on ice, and total RNA was extracted according to the manufacturer’s protocol. DNA and RNA concentration and quality were measured using UV/VIS spectrophotometer (BioDrop μLITE, UK).

Small RNA sequencing libraries were prepared using NEBNext Multiplex Small RNA Library Prep Set for Illumina using standard protocol according to the manufacturer’s instructions. Libraries were size selected by using AMPure XP beads to selectively bind DNA fragments 100 bp and larger to paramagnetic beads. Samples were sequenced on NextSeq 550 NGS platform (Illumina) using single-end sequencing mode with total of 36 cycles. RNAs extracted from immunoprecipitated complexes using Ago2 and GFP antibody were used to generate small RNA sequencing libraries using TruSeq Small RNA Kit (Illumina), according to the manufacturer’s instructions. cDNA libraries were size selected using 10% PAGE (BioRad), and species from 15 to 30 nt were isolated. cDNA libraries were then sequenced by single-read sequencing (50 nt) in Illumina HiSeq 2,500 sequencer.

For quality control check on raw sequences, we used FastQC. The sequencing data were analyzed using sRNAbench (Computational Epigenomics Lab, Evolutionary Genomics and Bioinformatics Group, Department of Genetics, Institute of Biotechnology, University of Granada, Spain) (49, 68, 69). Briefly, 3′ flanking sequencing adapter sequence was removed from all the reads, and reads were aligned against the human reference genome (Genome Reference Consortium Human Build 38 patch release 13) and HSV-1 strain 17 (NC_001806.2) or HSV-1 KOS strain (JQ673480.1). Reads smaller than 18 were excluded, as were the reads with a sequencing quality score lower than 30, to obtain high-quality reads. To assign the reads to known miRNAs, reads were aligned against known miRNAs from the miRBase sequence database (release 22), human and HSV-1, while published HSV-1 miRNAs not represented within the miRBase, hsv1-miR-H28, and hsv1-miR-H29 were added to the database file based on the published sequence. We allowed for two nucleotide mismatches in the first 19 nucleotides mapped but also accepted all sequences that started at most three nucleotides upstream and ended at most five nucleotides downstream of the reference sequence.

For the prediction of the host targets, prediction tools such as TargetScanHuman v.5.2 Custom (70) and miRDB (57) were used to search for the presence of sites that match the seed region of miRNA, nucleotides from 2 to 8, and for the viral targets, we also searched for the presence of sites complementary to the seed region of miRNA using Bowtie alignment tool (58). In addition, we used RNAhybrid (71), applying standard settings to visualize 10 miRNA-target bindings in the ICP0 and ICP4 region.

Immunohistochemical staining of ADAR proteins was conducted using DAKO EnVision + System, Peroxidase [3,3′-diaminobenzidine (DAB) kit (DAKO Cytomation, Santa Clara, CA, USA) on 4-µM-thick serial sections of paraffin-embedded TG tissues. Briefly, after deparaffinization and rehydration, tissues underwent heat-mediated epitope retrieval by microwave heating in a 10-mM citrate buffer, pH 6.0. Having subsequently been treated with blocking solution, slides were incubated with rabbit monoclonal α-ADAR1 IgG (Cell Signaling Technology; diluted 1:100 in 1% bovine serum albumin (BSA) in phosphate-buffered saline (PBS), mouse monoclonal α-ADAR1 IgG (Santa Cruz Biotech; diluted 1:100 in 1% BSA in PBS) and mouse monoclonal α-ADAR2 IgG (Santa Cruz Biotech; diluted 1:100 in 1% BSA in PBS) over 12 h in a humid chamber at 4°C. As a secondary antibody, a peroxidase-labeled polymer linked to goat α-rabbit and α-mouse immunoglobulins was applied for 30 min at room temperature. Immunoreactions were visualized by DAB, and slides were counterstained with hematoxylin. After being dehydrated, slides were mounted with Entellan (Sigma-Aldrich, Hamburg, Germany) and analyzed by an Olympus BX51 microscope equipped with a DP50 camera and Cell^F software (Olympus, Japan). The specificity of antibody binding was verified performing negative controls by substitution of ADARs antibodies with isotype-matched control antibodies applied in the same conditions. Negative control slides showed no immunohistochemical signals.

Single-stranded RNAs with modified 2′Ome designed to mimic canonical miR-H2-3p (miR-H2-3p-wt—5′-CCUGAGCCAGGGACGAGUGCGACU-3′), mutated seed sequence of the miR-H2-3p (miR-H2-3p-mut—5′-CGCCAUCGAGGGACGAGUGCGACU-3′), edited miR-H2-3p on positions 5 and/or 9 or both (miR-H2-3p-e5—5′-CCUGIGCCAGGGACGAGUGCGACU-3′, miR-H2-3p-e9—5′-CCUGAGCCIGGGACGAGUGCGACU-3′, miR-H2-3p-e5-e9—5′-CCUGIGCCIGGGACGAGUGCGACU-3′), and negative control mimic were ordered from GenePharma. Plasmid (pRS-1-ICP0/pcDNA-1-ICP4/pEGFP-N1) and miRNA mimic co-transfections were performed in HEK293T cells using Lipofectamine 3000 according to the manufacturer’s instructions without the addition of the P3000 reagent. Briefly, cells were seeded in a 24-well plate to be 70% confluent for the transfection. Cells were co-transfected with 70 ng per well of ICP0 expression plasmid or 50 ng per well of the ICP4 expression plasmid, 30 ng per well of the pEGFP-N1 expression plasmid, and with 30 nM of the mimic or negative control, using 1 μL of Lipofectamine 3,000 per well. After 24 h, samples were extracted for Western blot analysis.

To extract the proteins, cells were collected at indicated timepoints and lysed in RIPA buffer [150-mM NaCl, 1% NP-40, 0.5% Na deoxycholate, 0.1% SDS, 50 mM Tris (pH 8.0), with protease inhibitors (cOmplete, Roche, Basel, Switzerland)], mixed with 2× Laemmli buffer with β-mercaptoethanol (Santa Cruz Biotech, Dallas, TX, USA) and denatured for 6 min at 95°C. Proteins were separated in 10% SDS–PAGE gels and transferred onto a nitrocellulose membrane (Santa Cruz Biotech). Membranes were blocked in 5% (wt/vol) non-fat dry milk in 1× Tris-buffered saline (TBS) for 30 min at room temperature followed by the incubation at 4°C with gentle rotation overnight with the primary antibodies with specified dilutions: α-actin (MilliporeSigma, Burlington, MA, USA)—1:10,000; α-ICP0 (Abcam, Cambridge, UK)—1:2,000; α-ICP4 (Abcam, Cambridge, UK)—1:2,000; α-ADAR1 (Santa Cruz Biotech)—1:1,000; and α-ADAR1p150 (Cell Signaling Technology, Inc., Danvers, MA, USA)—1:1,000. Blots were washed for 30 min with TBS-0.05% Tween 20, and primary antibodies were detected using horseradish peroxidase-conjugated goat anti-mouse secondary antibody diluted 1:2,000 (Cell Signaling Technology, Inc.) and incubated at room temperature for 1 h. Blots were again washed for 30 min and visualized using the Amersham ECL reagent or SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific, Waltham, MA, USA) and ChemiDoc MP (Bio-Rad Laboratories, Hercules, CA, USA).

HFF cells were seeded in 10-cm dishes in triplicates for the infection with HSV-1 KOS strain and extracted at 12 h.p.i. From each sample, a fraction (1/10) was taken for the total population of miRNAs before the immunoprecipitation protocol, and the rest of the sample was used for Ago and GFP immunoprecipitation (control). Immunoprecipitation of RNA bound to Argonaute (Ago) proteins was performed according to the Dynabeads Protein G protocol (Thermo Fisher Scientific). Briefly, Dynabeads were incubated with the 10 µg of α-Ago antibody (Millipore) or α-GFP (Millipore) that acted as a negative control, diluted in PBS with 0.1% Tween-20 overnight at 4°C. After incubation, the supernatant was removed using a magnet, and beads were washed with PBS with 0.1% Tween-20. HFF samples were collected 12 h.p.i. in triplicates using NP-40 lysis buffer, and the supernatant from the cell lysis was added to the beads-antibody complex and incubated overnight at 4°C. After incubation, beads were washed and resuspended in TriReagent solution, and total RNA was extracted according to the manufacturer’s instructions.