Selective inhibition of microRNA processing by a herpesvirus-encoded microRNA triggers virus reactivation from latency

Herpesviruses have mastered host cell modulation and immune evasion to augment productive infection, life-long latency and reactivation thereof 1,2. A long appreciated, yet elusively dened relationship exists between the lytic-latent switch and viral non-coding RNAs 3,4. Here, we identify miRNA-mediated inhibition of miRNA processing as a novel cellular mechanism that human herpesvirus 6A (HHV-6A) exploits to disrupt mitochondrial architecture, evade intrinsic host defense and drive the latent-lytic switch. We demonstrate that virus-encoded miR-aU14 selectively inhibits the processing of multiple miR-30 family members by direct interaction with the respective pri-miRNA hairpin loops. Subsequent loss of miR-30 and activation of miR-30/p53/Drp1 axis triggers a profound disruption of mitochondrial architecture, which impairs induction of type I interferons and is necessary for both productive infection and virus reactivation. Ectopic expression of miR-aU14 was sucient to trigger virus reactivation from latency thereby identifying it as a readily drugable master regulator of the herpesvirus latent-lytic switch. Our results show that miRNA-mediated inhibition of miRNA processing represents a generalized cellular mechanism that can be exploited to selectively target individual members of miRNA families. We anticipate that targeting miR-aU14 provides exciting therapeutic options for preventing herpesvirus reactivations in HHV-6-associated disorders like myalgic encephalitis/chronic fatigue syndrome (ME/CFS) and Long-COVID. addition viral the master of mitochondrial fusion ssion, we that the miR-155 35 inhibits pri-miR-148b processing. previous of these two important human miRNAs . The putative binding site miR-155 within the pri-miR-148b closely anks the pre-miR-148b stem. This implies that miR-155 binding disrupts the hairpin structure and thereby sterically inhibits pri-miR-148b cleavage Drosha in the We therefore propose a model that miRNAs can either sterically inhibit the binding of DGCR8 to the pri-miRNA hairpin loop or pri-miRNA cleavage by Drosha via base pairing with energetically favorable unpaired sequences in the respective pri-miRNAs. As we exemplify for two let-7 family members, miRNA-mediated inhibition of miRNA processing can be readily exploited to specically target individual miRNAs of large miRNA families that so far could not be individually targeted.


Main Text
MicroRNAs (miRNAs) are important regulators of gene expression that have been implicated in all major cellular processes of life ranging from embryonic development to tissue homeostasis and cancer 5,6 .
Their biogenesis is tightly regulated at all levels 7 . Shortly after the discovery of cellular miRNAs, a number of viruses, predominantly of the herpesvirus family, were identi ed to encode and express their own set of viral miRNAs 4,8 . One of these is human herpesvirus 6A (HHV-6A), which has a seroprevalence of >90% in the human population. HHV-6A establishes latency by integrating into the subtelomeric regions of host chromosomes 9 . Virus reactivation has been associated with cardiac dysfunction and graft rejection as well as neuronal disorders including myalgic encephalitis and chronic fatigue syndrome (ME/CFS) 10 . The later has recently sparked interest because of HHV-6A reactivations as a putative cause of ME/CFS-like symptoms observed in Long-COVID. Here, we reveal miRNA-mediated inhibition of miRNA processing as a novel cellular mechanism that HHV-6A exploits to disrupt mitochondrial architecture, evade the induction of type I interferons and facilitate virus reactivation from latency.

HHV-6A induces mitochondrial fragmentation
Mitochondria play a key role in the cell intrinsic defense against viruses. They constantly undergo ssion and fusion events that help maintain functional mitochondria in cells under metabolic and environmental stress 11 . To examine whether HHV-6A affects mitochondrial architecture, we infected human umbilical vein endothelial cells (HUVEC) with wild-type HHV-6A and imaged mitochondria using a constitutively expressed, mitochondrially targeted GFP (mitoGFP) 12 . Lytic HHV-6A infection resulted in extensive mitochondrial fragmentation by 24 hours post infection (hpi) (Fig. 1a, Extended data Fig. 1a). The same was observed upon reactivation of latent HHV-6A in U2-OS bone osteosarcoma cells induced by Trichostatin-A (TSA) treatment (Extended data Fig. 1b). Mitochondrial fusion-ssion dynamics are governed by the activity of Dynamin-related protein 1 (Drp1) 13 . Helical oligomers of Drp1 form a ring around the outer mitochondrial membrane and fragment it 14 . Mitochondrial fragmentation was re ected by increased Drp1 expression during both lytic HHV-6A infection (Fig. 1b) and virus reactivation (Extended data Fig. 1c) as well as colocalization of Drp1 on mitochondrial surfaces in the virusreactivated cells (Extended data Fig. 1d). Drp1 levels are directly controlled at transcriptional level by the p53 tumor suppressor protein 15 . Accordingly, both lytic HHV-6A infection and virus reactivation exhibited increased p53 expression (Fig. 1b, Extended data Fig. 1c) indicating that HHV-6A induces mitochondrial fragmentation via the canonical p53/Drp1 axis 15 .
Viral miR-aU14 shares sequence homology to the miR-30c hairpin loop Unlike lytic infection, HHV-6A reactivation in the U2-OS cell model does not progress to fully productive virus replication but is restricted to the expression of some viral miRNAs and a few viral mRNAs 16 . We thus asked whether any of the HHV-6A small non-coding RNAs might be involved in the observed miR-30 processing defect. Manual sequence inspection revealed an interesting complementarity between HHV-6A miR-U14 and the hairpin loops of pre-miR-30c, pre-miR-30a and pre-miR-30d (Fig. 1e, Extended data Fig. 3a). This viral miRNA is expressed at very high levels during both productive infection 17 and virus reactivation 16 . However, it is encoded antisense to the U14 ORF. Hence, we decided to rename it as miR-aU14. The miR-30c hairpin loop showed the strongest sequence complementarity to miR-aU14 (Extended data Fig. 3b-3c) 18 . Small RNA sequencing of Argonaute (Ago)-bound RNA from HHV-6A-infected HSB-2 cells con rmed this so far incompletely characterized viral miRNA to represent one of the two most abundant viral miRNAs in HHV-6A-infected cells (Extended data Fig. 4a). In addition, transcription start site pro ling (TSS) using differential RNA-seq (dRNA-seq; Extended data Fig. 4b) 19,20 identi ed the pri-miR-aU14 transcript, which initiates 158 nt upstream of the miRNA (Extended data Fig. 4b) consistent with the unusual 130 nucleotide pre-miR-aU14 hairpin predicted by mFold 21 (Extended data Fig. 4c). Of note, pre-miR-aU14 is highly conserved between HHV-6A and HHV-6B with the region complementary to the pre-miR-30c hairpin loop showing 100% conservation (Extended data Fig. 4d).
Viral miR-aU14 is su cient to induce the miR-30c processing defect and mitochondrial fragmentation To assess the role of miR-aU14 in miR-30c processing, we employed a wild-type miR-aU14 miRNA mimic (Wt mimic) and a mutant thereof (Mut mimic) (Fig. 1f). Strikingly, transfection of the wild-type but not the mutant mimic into U2-OS cells reproduced both the loss of mature miR-30c and the concomitant increase in pri-miR-30c (Fig. 1g). To further validate that miR-aU14 was responsible for the miR-30c processing defect, we generated HeLa cells with a doxycycline (dox)-inducible miR-aU14 expressed from a Pol III promoter-driven shRNA and a mutant version (HeLa-Mut) thereof (Extended data Fig. 5a). Dox-induced expression of miR-aU14 (HeLa-Wt, Extended data Fig. 5b) but not of its mutant (HeLa-Mut, Extended data Fig. 5c) fully reproduced the miR-30c processing defect.
We next asked whether expression of miR-aU14 was su cient to disrupt mitochondrial architecture via the p53/Drp1 axis. Interestingly, both transfection and dox-induced expression of miR-aU14 but not of the mutant thereof induced p53 and Drp1 expression (Extended data Fig. 5d, 5e). Furthermore, transfection of the miR-aU14 mimic fully recapitulated the mitochondrial ssion phenotype (Extended data Fig. 5f).
To validate this effect in the virus context, we generated a mutant virus genome with discrete nucleotide substitutions within miR-aU14 (Extended data Fig. 6a). Mutations were designed not to alter the amino acid sequence of the U14 ORF. Deep sequencing of both the wild-type (HHV-6A-wt) and mutant (HHV-6Amut) bacterial arti cial chromosomes (BAC) con rmed the introduced mutations to represent the only differences between the two viral genomes. In contrast to wild-type HHV-6A, we were unable to reconstitute the miR-aU14 mutant virus despite multiple attempts indicating that the loss of miR-aU14 severely reduced viral tness. Hence, we generated polyclonal U2-OS cells that either carried chromosomally integrated latent wild-type HHV-6A or its miR-aU14 mutant by selection of cells stably transfected with the respective BACs. Upon virus reactivation with TSA, wild-type HHV-6A but not its miR-aU14 mutant impaired pri-miR-30c processing (Fig. 1h), induced Drp1 expression (Extended data Fig. 6b) and triggered mitochondrial ssion (Fig. 1i, Extended data Fig. 6c).

Viral miR-aU14 inhibits miR-30 processing by direct RNA:RNA interaction
To test whether miR-aU14 directly interacts with the pre-miR-30c hairpin loop, we a nity puri ed radioloabeled synthetic miR-aU14 with biotinylated pre-miR-30c. HeLa cells were transfected with equimolar amounts of radiolabeled synthetic miR-aU14, or two mutants thereof, and synthetic biotinylated pre-miR-30c (Fig. 2a, 2b). Radiolabeled RNA a nity puri ed 16 h later showed stronger enrichment of wild-type miR-aU14 than of its two mutants (Fig. 2c). In contrast, a radiolabeled small RNA control did not co-elute with pre-miR-30c.
We next asked whether the presence of the pre-miR-30c hairpin loop was su cient to mediate its inhibitory effects on miRNA processing. We designed two arti cial target pre-miRNAs that carried the original hairpin loop sequence of pre-miR-30c but contained arti cial miRNA stem duplex sequences (termed: miR-A and miR-B) (Fig. 2d). We then generated polyclonal HeLa cells with dox-inducible Pol-IIIdriven expression of miR-A or miR-B as well as stable transduced with lentiviruses that express either wild-type or mutant miR-aU14 (as shown in Extended data Fig. 5a). Consistent with the predicted interaction of miR-aU14 with the pre-miR-30c hairpin loop, induction of miR-aU14, but not of the mutant thereof, strongly repressed both miR-A and miR-B processing (Fig. 2e, 2f). Viral miR-aU14 inhibits the induction of type I interferons Healthy mitochondria play an important role in intrinsic immunity 22,23 . Upon activation of toll-like or RIG-I-like receptors, mitochondria serve as antiviral signaling hubs that govern the production of type I interferons (IFNs) 24 . RNA polymerase III can use cytosolic herpesvirus DNA as a template to produce 5'triphosphate RNAs, which induce type I IFN through the RIG-I pathway [24][25][26] . Enforced mitochondrial ssion dampens RIG-I/MAVS signaling and reduces the induction of type I IFNs 27 . We thus asked whether miR-aU14-mediated mitochondrial fragmentation impacts on the induction of type I interferons. Exposure of HEK293T cells, transfected with the miR-aU14 mimic, to the RIG-I pathway activator 3p-hpRNA (5' triphosphate hairpin RNA) 28 resulted in reduced mRNA levels of IFNbeta (Fig. 3a) as well as the IFN-responsive IFIT1 gene (Fig. 3b) in comparison to cells transfected with the miR-aU14 mutant.
We next asked whether miR-aU14 also plays a role in suppressing the production of IFNbeta upon HHV-6A reactivation. In addition to inducing virus reactivation by TSA, we treated cells with the JAK/STAT inhibitor Ruxolitinib to prevent secondary IFNbeta-mediated effects on virus reactivation. The latter was assessed by Northern blot for viral miR-aU14 and sncRNA-U77 (Fig. 3c). Ruxolitinib treatment enhanced TSA-induced virus reactivation resulting in a concordantly greater loss of miR-30c. Expression of viral sncRNA-U77 was signi cantly reduced for HHV-6A-mut indicative of impaired virus reactivation.
Accordingly, miR-30c levels remained unchanged. Nevertheless, the mutant virus induced signi cantly greater levels of IFNbeta than the wild-type virus (Fig. 3d). This was further increased upon Ruxolitinib treatment, presumably due to the inhibition of secondary negative feedback loops of IFN signaling. While reactivation of wild-type virus reduced IFIT1 mRNA levels by ~5-fold, IFIT1 levels were increased by 1.2fold for HHV-6A-mut relative to non-reactivated cells. This was partially inhibited by Ruxolitinib treatment (Fig. 3e).

Viral miR-aU14 governs the latent-lytic switch and augments productive infection
Considering the observed effects of miR-aU14 on the induction of type I IFNs, we asked whether ectopic expression of miR-aU14 would augment productive wild-type virus infection and rescue reactivation of the mutant virus. Transfection of miR-aU14 but not of a control miRNA mimic signi cantly increased the number of cells productively infected with wild-type virus by ≈4-fold (Fig. 3f). Furthermore, transfection of miR-aU14 e ciently rescued reactivation of the mutant virus even in the absence of TSA (Fig. 3g). The combination of both TSA and miR-aU14 showed enhanced virus reactivation indicating synergistic effects between the two. Similarly, transfection of miR-aU14 was substantially more effective at inducing reactivation of wild-type virus than TSA (Extended data Fig. 7a). Neither of the two mutant mimics had any effect on virus reactivation (Extended data Fig. 7b).

Targeting hairpin loops and other unpaired sequences of human pre-miRNAs by small synthetic RNAs interferes with miRNA maturation
In principle, miRNA-mediated inhibition of miRNA processing should be applicable to other cellular miRNAs. This is of particular interest as many important cellular miRNAs exist as miRNA families.
Targeting hairpin loops rather than the mature miRNA sequences would offer a unique opportunity for the development of more selective miRNA inhibitors. Many of the let-7 family members carry relatively large hairpin loops, which may comprise up to 30 nt. Hence, we designed synthetic miRNA mimics targeting two different regions of the hairpin loop of pre-let-7d (Fig. 4a). Upon transfection into cells, both miRNA mimics e ciently reduced mature let-7d levels consistent with impaired miRNA processing (Fig. 4b).
Similar data were obtained for two other miRNA mimics targeting the hairpin loop of let-7f1 (Extended data Fig. 8a, 8b).
Finally, we speculated that miRNA-mediated inhibition of miRNA processing should also be observable for cellular miRNAs. To identify such regulation, we carried out a systematic blast search of mature human miRNAs against known pre-miRNAs from miRBase. Several search results indicated interesting potential binding sites within pre-miRNAs (Extended data Table 1). However, majority of the respective miRNA pairs were not abundantly expressed in most of the standard human cell lines. We thus focused on one particular candidate pair of miRNAs, namely miR-155 and miR-148b. Our analysis indicated potential binding of miR-155 to pri-miR-148b just 5' of the pre-miR-148b hairpin (Fig. 4c). Transfection of a miR-155 mimic resulted in reduced levels of mature miR-148b. MicroRNA-mediated inhibition of miR-148b processing by cellular miR-155 thus at least partially explains the dichotomous expression of these two human miRNAs 29 .

Discussion
Here, we identify miRNA-mediated inhibition of miRNA processing through sequence-speci c RNA-RNA interactions as a novel cellular mechanism that governs miRNA processing (Extended data Fig. 9).
Regulation of miRNA processing by cellular proteins is well described 7 . The rst and best characterized example is the stem cell factor Lin-28, which interacts with the hairpin loop of let-7 family members and blocks their biogenesis [30][31][32] . Recently, a large unbiased screening approach identi ed ~180 RNA binding proteins that speci cally interact with distinct human pre-miRNAs 33 . RBP-mediated regulation of miRNA processing thus constitutes an important regulatory network that shapes miRNA activity and function.
Here, we show that miRNAs can take over similar functions and selectively inhibit miRNA processing in a sequence-speci c manner. Interestingly, the miR-aU14-mediated loss of miR-30c was accompanied by a marked increase of only pri-miR-30c levels. This implies that the inhibition occurs at the level of pri-miRNA processing within the nucleus. miR-30c is encoded from an intron of the NF-YC gene 34 .
Recognition and cleavage of the intronic pre-miRNA hairpin loop by the RNase II Drosha thus competes with the cellular splicing and RNA degradation machinery. Sterical interference of DGCR8 binding to the miR-30c hairpin by miR-aU14 in the nucleus and subsequent degradation of the parental intron upon splicing therefore is a likely explanation for the observed loss of mature miR-30c. In contrast, let-7d is expressed from its own primary transcript. Inhibition of let-7d processing by two arti cial miRNAs targeting the hairpin loop indicates impaired pre-miRNA processing in the cytoplasm. We thus cannot exclude that miRNA-mediated inhibition of miRNA processing may occur both in the nucleus and cytoplasm. In addition to viral inhibition of miR-30, the master regulator of mitochondrial fusion and ssion, we show that the in ammatory miR-155 35 inhibits pri-miR-148b processing. This explains previous reports of dichotomal expression of these two important human miRNAs 29 . The putative binding site of miR-155 within the pri-miR-148b closely anks the pre-miR-148b stem. This implies that miR-155 binding disrupts the hairpin structure and thereby sterically inhibits pri-miR-148b cleavage by Drosha in the nucleus. We therefore propose a model that miRNAs can either sterically inhibit the binding of DGCR8 to the pri-miRNA hairpin loop or pri-miRNA cleavage by Drosha via base pairing with energetically favorable unpaired sequences in the respective pri-miRNAs. As we exemplify for two let-7 family members, miRNA-mediated inhibition of miRNA processing can be readily exploited to speci cally target individual miRNAs of large miRNA families that so far could not be individually targeted.
Viral miR-aU14-mediated inhibition of miR-30 processing explained mitochondrial fragmentation during both lytic HHV-6A infection and virus reactivation via the miR-30/p53/Drp1 axis. This in turn impairs the induction of type I IFN and augments productive virus infection. Interestingly, multiple attempts to reconstitute a miR-aU14 mutant virus from BAC DNA failed, indicating that viral miR-aU14 is crucial for productive virus replication in vitro. Similarly, the miR-aU14 mutant virus was severely impaired in its ability to reactivate from latency. It is, however, important to note that reactivation of the miR-aU14 mutant virus by TSA resulted in a signi cantly stronger type I IFN response than observed for wild-type virus. This indicates that miR-aU14 may not be essential for the desilencing of the latent virus genomes but rather the inhibition of intrinsic cellular defense mechanisms that otherwise e ciently prevent successful virus reactivation. This is in line with the gross disruption of mitochondrial architecture by miR-aU14 via the miR-30/p53/Drp1 axis.
A peculiar feature of the miR-aU14 locus is that miR-aU14 is expressed antisense to the U14 open-reading frame from a novel pri-miRNA transcript that initiates within the front part of the U14 ORF. U14 encodes for a G2/M cell cycle checkpoint regulator of HHV-6 36 , which also interacts with p53 37 . While miR-aU14 thus has the potential to repress expression of the important U14 protein, mutational analysis indicates that both U14 and miR-aU14 are important for productive HHV-6A infection. Accordingly, transient transfection of miR-aU14 enhanced productive HHV-6A infection and fully rescued virus reactivation of a miR-aU14 mutant virus. The most striking nding, however, was that transfection of miR-aU14 triggered virus reactivation from latency at least as e ciently as the commonly employed histone deacetylase inhibitor TSA. While enhanced mitochondrial ssion and impaired intrinsic immunity via the miR-30/p53/Drp1 axis will augment successful virus reactivation, miR-aU14 may also target other cellular or viral genes that help trigger virus reactivation from latency.
In summary, our ndings reveal a surprising, drugable miRNA-mediated mechanism that a prevalent human herpesvirus usurped to interfere with intrinsic immunity, govern the latent-lytic switch and augment productive infection. Importantly, viral miR-aU14 should be readily drugable using antisense approaches (antagomiRs) 38 , thereby providing an interesting therapeutic option for preventing herpesvirus reactivation in ME/CFS patients and Long-COVID. with pre-miR-30c hairpin loop. Predicted sites of interaction are in grey boxes. f. Nucleotide sequences of wild-type (Wt mimic) and mutant (Mut mimic) miR-aU14 mimics. Point mutations are highlighted in blue within red boxes. g. Pri-miR-30c processing defect by wild-type or mutant miR-aU14 mimic. n=2. h. Pri-miR-30c processing defect during reactivation of wild-type HHV-6A (HHV-6A-wt) or mutant virus (HHV-6Amut). U2-OS cells without HHV-6A served as mock control. n=3. i. Average mitochondrial area in wild-type or mutant HHV-6A reactivating cells as a box plot. RNA and protein quanti cation was done by densitometric analysis (b, d, g, h). Box and whiskers (a, i) show minimum to maximum values with all independent replicates, centre denotes median, and the bounds denote the 25th to 75th percentiles. Data are mean ± s.e.m. *P ≤ 0.05, **P ≤ 0.005, ***+P ≤ 0.00005, two-way ANOVA with Tukey's multiple comparisons test (a, f), unpaired two-tailed Student's t-test (b, d, g, h).

Figure 2
HHV-6A miR-aU14 inhibits miR-30 processing by direct RNA:RNA interaction a. Schematic of the experimental set up. b. Nucleotide sequences of the wild-type miR-aU14 mimic (Wt mimic) and the mutant mimics (Mut and Mut2 mimic). Mutated sequences are in blue within the rectangular boxes. c. A nity puri cation of biotinylated pre-miR-30c co-precipitates miR-aU14 as in Fig. 2a. A random small RNA was used as a bait for control. The blot was probed for miR-30c for equal amounts of a nity puri ed biotinylated pre-miR-30c. d. Putative interactions of the two arti cial chimeric pre-miRNAs (pre-miR-A and pre-miR-B) with miR-aU14. The mature miRNA sequences (miR-A and miR-B) are indicated in red. The putative interactions of the transplanted miR-30c hairpin loop with miR-aU14 (blue) are indicated. e, f. miR-aU14 impairs processing of arti cial miRNAs carrying miR-30c hairpin loop.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.