ADAR1 Biology Can Hinder Effective Antiviral RNA Interference

ABSTRACT

Viruses constantly evolve and adapt to the antiviral defenses of their hosts. The biology of viral circumvention of these selective pressures can often be attributed to the acquisition of novel antagonistic gene products or by rapid genome change that prevents host recognition. To study viral evasion of RNA interference (RNAi)-based defenses, we established a robust antiviral system in mammalian cells using recombinant Sendai virus designed to be targeted by endogenous host microRNAs (miRNAs) with perfect complementarity. Using this system, we previously demonstrated the intrinsic ability of positive-strand RNA viruses to escape this selective pressure via homologous recombination, which was not observed in negative-strand RNA viruses. Here, we show that given extensive time, escape of miRNA-targeted Sendai virus was enabled by host adenosine deaminase acting on RNA 1 (ADAR1). Independent of the viral transcript targeted, ADAR1 editing resulted in disruption of the miRNA-silencing motif, suggesting an intolerance for extensive RNA-RNA interactions necessary for antiviral RNAi. This was further supported in Nicotiana benthamiana, where exogenous expression of ADAR1 interfered with endogenous RNAi. Together, these results suggest that ADAR1 diminishes the effectiveness of RNAi and may explain why it is absent in species that utilize this antiviral defense system.

IMPORTANCE All life at the cellular level has the capacity to induce an antiviral response. Here, we examine the result of imposing the antiviral response of one branch of life onto another and find evidence for conflict. To determine the consequences of eliciting an RNAi-like defense in mammals, we applied this pressure to a recombinant Sendai virus in cell culture. We find that ADAR1, a host gene involved in regulation of the mammalian response to virus, prevented RNAi-mediated silencing and subsequently allowed for viral replication. In addition, the expression of ADAR1 in Nicotiana benthamiana, which lacks ADARs and has an endogenous RNAi system, suppresses gene silencing. These data indicate that ADAR1 is disruptive to RNAi biology and provide insight into the evolutionary relationship between ADARs and antiviral defenses in eukaryotic life.

INTRODUCTION

The evolutionary arms race between viruses and the hosts they infect has established a multitude of different antiviral defense systems in both prokaryotic and eukaryotic life (1, 2). In eukaryotes, antiviral defenses that recognize and respond to viral RNA generally comprise one of two systems: an RNA- and a protein-based system. RNA-based defenses rely on host processing of pathogen genetic material, which is then coupled to a host nuclease to mediate its antiviral activity. This biology, generally referred to as RNA interference (RNAi), is widely found in plants, insects, and nematodes (3–6). In contrast, protein-based systems rely on signal transduction following viral recognition. This defense hinges on host capacity to recognize pathogen-associated molecular patterns (PAMPs) mediated by pattern recognition receptors (PRRs) (7). PRR activation culminates in the transcriptional induction of a subset of genes that together coordinate a defensive response (7, 8). In vertebrates, this system induces the expression of a family of cytokines known as interferons (IFN), which induce the expression of hundreds of antiviral IFN-stimulated genes (ISGs) (9–11). These genes are diverse in nature but include transcription factors, proinflammatory cytokines, and a large group of antiviral effector proteins that directly target the pathogen (12).

Despite their differences, both RNAi and the IFN systems can be activated by host detection of double-stranded RNA (dsRNA), a potent PAMP throughout all eukaryotic life (13). In plants, insects, and nematodes, dsRNA recognition leads to the production of diverse small interfering RNAs (siRNAs) (14). In vertebrates this same PAMP serves as a substrate for PRR aggregation and subsequent induction of antiviral gene products (15). While in general these two systems each embody a self-contained antiviral system utilized by a distinct phylum, some groups have reported that antiviral RNAi is functional alongside the IFN response in mammals (16–20), although this remains controversial (21).

In response to these defenses, eukaryotic viruses have evolved countermeasures to navigate host biology successfully. In plants and insects, these antagonistic strategies are often mediated by virus-encoded proteins generally referred to as viral suppressors of RNAi (VSRs). Strategies encompassing the VSR family include small RNA binding proteins, proteins that inactivate or degrade the nucleases, and other host components essential to mediating this defense (22–24). Similarly, virus evolution has generated a number of diverse strategies that antagonize the IFN system. Like RNAi, examples of viral IFN antagonism include masking RNA substrates, interfering with signaling pathways, and/or disrupting other essential aspects of host biology (25).

The evolutionary emergence of RNAi coincides with eukaryogenesis, a probable by-product of the rapidly expanding RNA virosphere in contrast to the DNA phages that dominate the prokaryotic world (26). Based on evolutionary conservation, the biogenesis of antiviral RNAi was the by-product of repurposing prokaryotic proteins involved in nucleic acid metabolism (27). The antiviral RNAi system that emerged was later repurposed, providing an additional level of posttranscriptional control of host genes in both plants and animals (27). Referred to as microRNAs (miRNAs), this biology is thought to have enabled multicellularity and has been shown to play critical roles in development and maintenance of cell state (28). In both small RNA pathways, Dicer, an RNase III ortholog found in all multicellular eukaryotes, produces the resulting RNA duplex of 19 to 21 nucleotides (nt) which is loaded into an RNA-induced silencing complex (RISC) and mediates posttranscriptional control (29–31). Despite these similarities, antiviral RNAi elicits complete silencing of its target through enzymatic cleavage, whereas miRNAs are host-derived fine-tuners of translation mediated by deadenylation or steric hindrance of the ribosome (32). The underlying biology defining the potency of repression between these systems is a product of small RNA complementarity in addition to the capacity of the RISC to cleave its target. In antiviral RNAi, where the duplex RNA was generated directly from the pathogen, complementarity will always be perfect, enabling an A-form helical structure that is amenable to cleavage by a member of the argonaut (AGO) nuclease family (33, 34). The capacity to induce cleavage further enables enzymatic activity to potently silence all of its targets, in contrast to being stoichiometric, where the RISC is dedicated to the translational repression of a single mRNA (35). In addition to complementarity, RISC-mediated cleavage is also defined by the Ago associated with it. Mammals encode four Ago homologs, Ago1 to -4, of which only a single member, Ago2, possesses enzymatic activity, reflecting the fact that miRNAs are never perfectly complementary to their targets (36).

Despite the fact that mammalian host small RNA silencing is limited to nonenzymatic targeting, Ago2-mediated cleavage can occur should a perfectly complementary target be provided (15). In fact, previous work has demonstrated that incorporating miRNA target sites into viruses can essentially recreate antiviral RNAi in mammals and can successfully negate the need for the IFN system (37). The capacity to impose this same RNAi selective pressure on vertebrate viruses that typically navigate the IFN defense was recently used to compare diverse RNA viruses to define their adaptation capacity (38). This study demonstrated that positive-sense single-strand RNA (+ssRNA) viruses in which 5 miRNA perfect targets were embedded into essential viral genes escaped the pressure of RNAi via homologous recombination. In this process the viral RNA-dependent RNA polymerase will stochastically template switch from one viral genome to another or skip over the inserted miRNA complementary sites in the viral genome, allowing for viral escape from silencing (39, 40). In contrast, when this same exact pressure was imposed onto negative-sense single-strand RNA (−ssRNA) viruses, no escape was observed, denoting a general inability to undergo homologous recombination within this system (38).

Given the extraordinarily high capacity to adapt to a selective pressure, we reasoned that even in the absence of homologous recombination, −ssRNA viruses may find a way to escape. This idea is supported by the existence of this group of viruses in hosts with functional antiviral RNAi (41, 42). Given this, we again tested the capacity of a recombinant negative-strand paramyxovirus to escape miRNA-mediated silencing, by imposing different degrees of selection pressure in addition to the time to adapt. To this end, we used a recombinant enhanced green fluorescent protein (eGFP)-expressing Sendai virus (rSeV) comprised of nucleocapsid protein (N), eGFP, phosphoprotein (P/C/V), matrix protein (M), fusion protein (F), hemagglutinin-neuraminidase (HN), and large protein (L), of which miRNA-mediated selective pressure was imposed on N or P. Here, we demonstrate that while weak targeting can be circumvented by viral polymerase-induced point mutations, even prolonged periods of time failed to produce evidence of homologous recombination and removal of the miRNA cassette as was commonly observed for +(RNA) viruses. However, despite the inability to remove this targeting cassette from the viral genome, regardless of whether N or P was targeted, here we describe host-mediated RNA editing events that inadvertently relieve this selective pressure and enable viral replication.

RESULTS

miRNA-targeted SeV escapes silencing pressure given sufficient time.

Given the existence of diverse RNA viruses of negative polarity that successfully navigate the RNAi systems of both plants and arthropods, we reasoned that even in the absence of homologous recombination or a VSR, viral escape from RNAi must be possible. To address this question, we utilized a recombinant Sendai virus containing enhanced green fluorescent protein (rSeV) in which we could further manipulate the genome to impose an RNAi-based selective pressure (43). To this end, we grafted five unique miRNA target sites into the 3′ untranslated region (UTR) of the nucleoprotein (rSeV-N5T), which was previously shown to be potently silenced in epithelial cells and cleared in vitro (37). In this construct, miRNAs can bind to positive-sense viral RNA, including the antigenome and N mRNA. To determine if reduced selective pressure would allow for escape, we additionally designed a Sendai virus variant removing all but a single miRNA target site (rSeV-N1T) and compared both viruses to one encoding the reverse complement of the same 3′ UTR as a negative control (rSeV-N5R) (Fig. 1A). In the rSeV-N5R construct, miRNAs can theoretically engage negative-sense viral genome target sites but are unable to do so as ribonucleoprotein (RNP) structures render the RNA inaccessible and resistant to cleavage as reported elsewhere (44). To initially assess the response to selective pressure, we repeated the same passaging experiments in which N5T was not detected following sequential passaging (38). To further increase escape probability, we performed this experiment using a 96-well plate format for each virus construct including rSeV-N5T, -N5R, and -N1T, transferring the supernatant from an initial infection to naive cells every 48 h for a total of 144 h. To assess the presence of virus (here designated N5R, N1T, and N5T), we utilized high-content microscopy to detect the presence of GFP. In agreement with published work, these efforts resulted in uniform fluorescence following the addition of N5R whereas N5T could not be detected (see Fig. S1A in the supplemental material). As anticipated, this same experiment demonstrated that reducing RNAi-mediated silencing to a single target resulted in the appearance of individual GFP-positive cells, suggesting that N1T had successfully navigated small RNA-mediated silencing (Fig. S1A).

FIG 1.

miRNA-targeted rSeV escapes selective pressure of RNAi with sufficient time. (A) Diagram of recombinant SeV-N5R, -N1T, and -N5T with miRNA cassettes and “reverse” control inserted after the N ORF to encode miRNA binding sites into the 3′ UTR of rSeV-N mRNA. (B) Representative single-well images from 96-well-plate sustained infections of A549 cells with rSeV at an MOI of 5, imaged at 2, 4, and 6 days postinfection for DAPI and eGFP expression. (C) Flow cytometry analysis of sustained infections with rSeV-N5R and rSeV-N1T in A549 cells for 1, 2, 4, and 6 dpi and with rSeV-N5T for 1, 2, 4, 6, and 8 dpi. Stacked modal histograms show eGFP fluorescence intensity on the x axis and percentage of maximum cell count within each condition on the y axis. (D) Western blotting of whole-cell lysates from A549 cells infected with rSeV-N5R from blotting for beta-actin and SeV-N. Samples for rSeV-N5R were collected at 1, 2, and 4 dpi. Samples for rSeV-N1T were collected at 1, 2, 4, and 6 dpi. Samples for rSeV-N5T were collected at 1, 2, 4, 6, and 8 dpi.

Given the sparse GFP positivity of N1T-infected cells, we next repeated the experiment in the absence of passaging to discern between bona fide escape of a replication-competent virus and a by-product of overwhelming miRNA biology as a result of high multiplicity of infection (MOI). Again, we inoculated epithelial cells with N5R, N1T, or N5T but did not transfer the supernatants of each condition. Instead, cell cultures were maintained for 144 h, the same time allotted for the passaging experiment. Epithelial infection with N5R yielded robust GFP expression at 2 days postinfection (dpi) followed by cell death thereafter (Fig. 1B and Fig. S1B). These same conditions demonstrated that N1T again enabled GFP positivity over time, with viral replication reaching levels comparable to N5R by 6 days postinfection (6 dpi). Unexpectedly, when this experimental setup was performed on N5T, GFP positivity became apparent at 6 dpi, suggesting that escape was possible but required more time than sequential passaging provided (Fig. 1B). As this experiment was performed in triplicate 96-well dishes per virus construct, we calculated escape frequencies, which demonstrated N5T escaped in ~8% of individual wells in contrast to N5R or N1T, where GFP positivity was uniform under all conditions by 6 dpi (Fig. S1C). To assess this biology in a more quantitative manner, we repeated the same sustained infection using N5R, N1T, and N5T but assessed GFP by flow cytometry. These data recapitulate what was observed by microscopy, showing that N1T resulted in robust GFP expression by 6 dpi, which could also be observed for N5T if the infection was extended to 8 dpi (Fig. 1C). These data could be further corroborated when examining the expression of the miRNA-targeted viral transcript as the appearance of GFP by microscopy correlated with the detection of SeV-N by Western blotting (Fig. 1D).

Escape of SeV-N5T from RNAi is mediated by multiple A-to-G changes in miRNA target sites.

To determine the mechanism of escape from RNAi-mediated silencing, we next performed RNA sequencing (RNA-Seq) on the transcriptome of infected cells which expressed GFP indicative of viral escape (Fig. 2). Aligning total reads against the positive sense of the N5R genome revealed the characteristic viral gene expression pattern for paramyxoviruses with diminished viral mRNA levels observed for L (Fig. 2A) (45). Extrapolation of minority variants captured in these RNA-Seq data suggested the fidelity of rSeV to be ~1:10,000, in agreement with past efforts (38). These data could be observed within the RNA-Seq data following N1T infection, which additionally demonstrated an amalgamation of point mutants located specifically on the miR-21 target site (Fig. 2B). Approximately 75% of reads at the miR-21 binding location implicated an A-to-U point mutation in the miRNA binding seed region in the virus genome (Fig. 2B). In parallel, we also observed other individual point mutants, including three shared U-to-C transitions found in ~5 to 10% of captured reads, two of which aligned to the seed region of the miRNA target (Fig. 2B). Of note, we did not observe a higher frequency of mutations in the vicinity of the miR-21 target site, including the N 3′ UTR (Fig. S2A).

FIG 2.

Mutations leading to rSeV-N5T escape from miRNA targeting. (A) Relative read depth (black line) and percentage of nonreference base reads (color bar graphs) plotted along the rSeV-N5R genome from infected A549 cells at 4 dpi. (B) Percentage of nonreference base reads plotted along the 22-bp-long miR-21-targeted site in the N 3′ UTR of the rSeV-N1T genome from infected A549 cells with confirmed viral escape. (C) Relative read depth (black line) and percentage of nonreference base reads (color bar graphs) plotted along the rSeV-N5T genome from infected A549 cells with confirmed viral escape. (D) Percentage of nonreference base reads plotted along the 5T miRNA cassette from amplicon sequencing of rSeV-N5T-infected A549 cells at 2, 4, and 8 dpi.

Given the low probability of simultaneously generating point mutations in all five miRNA targets, we next sequenced N5T with the expectation that aligning the resulting escape mutant would demonstrate evidence for a deletion event akin to what has been documented for +ssRNA viruses (38). However, we observed neither a deletion nor a cluster of random point mutants within the miRNA targeting cassette. We instead documented a prominent number of U-to-C transitions (corresponding to A-to-G changes in the negative sense/viral genome) within each target site (Fig. 2C and Fig. S2B). To determine the kinetics of these genome edits, we repeated the assay but monitored the process at 2, 4, and 8 dpi by performing amplicon sequencing on a PCR product of the miRNA-targeted region of the N5T transcript. These data demonstrated the presence of low-frequency U-to-C edits in the positive sense as early as 2 dpi (Fig. S2B). These edited viral genomes, which likely represent a few different hypermutation events, rapidly outcompeted the nonedited viral genomes and became dominant in the population by 4 and 8 dpi (Fig. 2D).

Host ADAR1 is responsible for hypermutations in miRNA-targeted sites.

As A-to-I transitions in RNA produce U-to-C mutations that become A-to-G after viral replication, these observed edits appeared to be the hallmark of adenosine deaminases acting on RNA (ADARs). Therefore, we next examined whether disruption of this host gene would result in loss of N5T escape. Mammals encode three paralogs of the ADAR family, notably ADAR1, ADAR2, and ADAR3 (46). ADAR1 is ubiquitous and is expressed constitutively as a 110-kDa product localized to the nucleus (p110) or as a 150-kDa IFN-induced product which is found in the cytoplasm (p150) (47, 48). We ruled out ADAR2 as a candidate due to its nuclear localization and rapid degradation in the cytoplasm (49) and ADAR3 because it is expressed only in the brain and has been reported to lack RNA editing activity (50, 51). To ascertain if ADAR1 enabled N5T escape, we generated ADAR1 knockout (KO) cells using a previously established STAT1 KO human alveolar lung epithelial cell line (A549) (Fig. S3A and B) (52). The requirement for STAT1 knockout cells was to circumvent engagement of IFN and subsequent cell death, which has been reported following the disruption of ADAR1 expression (53). We next sought to characterize how loss of STAT1 impacted the expression of ADAR1 isoforms, finding that both p110 and p150 were constitutively produced in both wild-type (WT) and STAT1-deficient A549 cells, consistent with what has been reported elsewhere in murine fibroblasts (Fig. S3A) (54). To further ensure that loss of STAT1 expression would not prevent ADAR1-mediated viral escape, we performed sustained infection experiments in WT and STAT1 KO lines. These data demonstrated that in the absence of STAT1, N5T remained capable of escape from RNAi pressure as noted by both GFP fluorescence and NP protein expression (Fig. 3A and B and Fig. S3C). Additionally, sequencing of escape mutants of N5T from STAT1 KO cells confirmed that U-to-C edits indicative of ADAR1 editing were still responsible for loss of silencing (Fig. 3C). In contrast, when STAT1 KO cells were further disrupted for ADAR1 expression, N5T escape variants were not observed in any of the 96 wells that were measured, implicating its role for editing of the viral genome (Fig. 3A and B and Fig. S3C). Lastly, to ensure the observed phenotype was not the by-product of an off-target or cell clonal event, we also reconstituted ADAR1 using an adenovector in the double knockout cells, confirming reconstitution by RNA-Seq (Fig. S2B). In agreement with our hypothesis, these data found that ADAR1 reconstitution of KO cells, which included expression of both the p110 and p150 isoforms, resulted in restoration of the escape phenotype (Fig. 3A, B, and D and Fig. S3C).

FIG 3.

Knockout of ADAR1 prevents SeV-N5T escape. (A) Representative single-well images from 96-well plates of WT, STAT1 KO, STAT1/ADAR1 double-KO (DKO), and STAT1/ADAR1 transduced with AdV-mCherry-hADAR1. Cells were infected with rSeV-N5R and fixed at 2 dpi or infected with rSeV-N5T and fixed at 8 dpi. Images were taken with channels to detect DAPI, eGFP, and mCherry. (B) Western blotting of whole-cell lysate from respective A549 lines with blotting for beta-actin and SeV-N in mock-infected cells and cells infected with rSeV-N5R at 2 dpi, SeV-N5T at 2 dpi, or SeV-N5T at 8 dpi. (C) Relative read depth (black line) and percentage of nonreference base reads (color bar graph) plotted along the SeV-N5T genome from infected STAT1 KO A549 cells with confirmed viral escape. (D) Relative read depth (black line) and percentage of nonreference base reads (color bar graph) plotted along the SeV-N5T genome from infected, AdV-mCherry-hADAR1-transduced, STAT1/ADAR1 DKO A549 cells with confirmed viral escape.

ADAR1 editing enables escape of Sendai virus harboring a miRNA-targeted P open reading frame (ORF).

As host antiviral defenses both sense and target the genomes of negative-sense RNA viruses, this family uniformly masks their genomes in an RNA binding protein such as N, generating a ribonucleoprotein (RNP) complex (55). For this reason, disruption of N would expose the genome, resulting in both antiviral host engagement and putative host targeting by ADAR1 (56). This dynamic would suggest that ADAR1-mediated escape of Sendai virus harboring an miRNA-targeted N transcript may be unique as a result of this dynamic. To address this possibility, we created a second rSeV where the 5T targeting cassette was removed from the 3′ UTR of N and was transposed into the same position of the phosphoprotein (P) transcript (rSeV-P5T). This gene was selected as it is also essential for viral transcription but, unlike the loss of N, would not be predicted to expose viral RNA (Fig. 4A). To initially confirm that rSeV-P5T (here P5T) both resulted in attenuation and would not engage the host cell antiviral defenses, we performed RNA-Seq on epithelial cells infected with N5R, N5T, or P5T at 1 and 2 days postinfection (Fig. 4B and C). Initial alignment of the sequencing results revealed that relative levels of viral RNA from P5T were comparable to those from N5T, with both demonstrating a more-than-100-fold decrease compared to the control N5R virus (Fig. 4B). Consistent with previous characterization of N5T, alignment of RNA-sequencing reads to the host genome revealed a robust engagement of host antiviral defenses in N5T which paralleled that of N5R despite producing significantly less PAMP (Fig. 4C) (56). In contrast, while P5T showed the same attenuation as N5T, this occurred in the absence of a strong IFN response, suggesting that genomic viral RNA remained protected as predicted (Fig. 4B and C).

FIG 4.

Targeting the SeV P gene leads to ADAR1-mediated viral escape by editing of viral antigenome. (A) Diagram of rSeV-P5T, with 5T miRNA cassette inserted at the end of the P ORF. (B) Reads per million of rSeV RNA from RNA-Seq of rSeV-infected (MOI of 5) A549 cells at 1 and 2 dpi. Statistical significance was determined using a one-way analysis of variance with the P value corrected for multiple comparisons using Tukey’s test (****, P < 0.0001; ns, not significant). Error bars indicate standard deviation. (C) Heatmap of log₂ fold change (FC) in differential expression of top 20 ISGs from same RNA-Seq data as in panel B. (D) Representative single-well images of rSeV-P5T escape from sustained infection in A549 cells at 2, 4, and 6 dpi, with imaging for DAPI and eGFP expression. (E) Relative read depth (black line) and percentage of nonreference base reads (color bar graph) plotted along the SeV-P5T genome from infected A549 cells with confirmed viral escape.

The lack of innate immune engagement of P5T compared to N5T is indicative of the latter targeting event resulting in the exposure of viral genomic PAMP. Given this, we reasoned that if ADAR1-mediated escape was the by-product of this unique dynamic between N and the viral genome, these criteria would not materialize in P5T and therefore escape would not be expected. To discern whether genome exposure in N5T contributed to viral escape, we performed sustained infections with P5T in ADAR1-expressing epithelial cells and found evidence for GFP-positive cells as early as 4 dpi (Fig. 4D). Of note, the frequency of P5T escape mutants was lower than that observed in N5T (Fig. S1B and S4A). Given this decreased frequency, we next sought to verify that the underlying mechanism of escape was shared with that of N5T. To this end, we sequenced RNA from P5T-infected cells and found similar mutations all represented by A-to-G transitions across the miRNA targeting cassette (Fig. 4E). Of note, while these findings are consistent with ADAR biology, sequenced P5T escape mutants showed editing of the viral antigenome rather than the genome, which was consistently edited in N5T (Fig. 4E and Fig. S4B). Sequenced P5T escape mutants also demonstrated a more heterogenous profile of G-to-A transitions, which did not always disrupt all miRNA targets (Fig. S4C). These results suggest that unlike N protein, low levels of P may be more tolerable for limited viral replication. Taken together, these data would suggest that ADAR-mediated escape may be not the by-product of miRNA-mediated cleavage and generation of dsRNA but rather a potential active interaction between host ADAR1 and the miRNA machinery, which has been suggested by others (57).

The evolutionary relationship between ADARs and cellular defense systems.

Given that ADAR1 impaired the capacity to harness RNAi-based antiviral defenses in mammals, we investigated whether the evolutionary expansion of this gene family correlated with the utilization of IFN. To this end, we selected some well-characterized species representing a wide range of eukaryotic life that utilize interferon, RNAi, and/or other defenses. As previously reported, phylogenetic analysis of ADARs, adenosine deaminase domain-containing proteins (ADADs), and adenosine deaminases acting on tRNAs (ADATs) demonstrated that this gene family is well represented throughout the phylogenetic tree of life (Fig. 5A). These data found that while ADATs are uniformly utilized by all species examined, the ADAR1 ortholog is unique in organisms with a known IFN- or IFN-like defense system and absent in species with well-characterized antiviral RNAi. While this analysis is limited in scope, it is also in agreement with previous and more extensive evolutionary analysis of the ADAR gene family (58, 59).

FIG 5.

(A) Phylogenetic tree of ADARs, ADADs, and ADATs including select species representing a wide range of multicellular eukaryotic evolution. Species with a prominent and well-characterized antiviral RNAi system are labeled with red dots. ADAR1 proteins are marked by an extruded color indicator. (B) Representative single-well images of rSeV-N5R infection at 2 dpi and rSeV-N5T infection and viral escape at 8 dpi in MDCK cells, with imaging for DAPI and eGFP expression. (C and D) Relative read depth (black line) and percentage of nonreference base reads (color bar graph) plotted along the rSeV-N5T genome (C) and along the 5T cassette (D) from infected MDCK cells with confirmed viral escape. (E) GFP expression in patches of N. benthamiana agroinfiltrated with plasmid(s) to express GFP, GFP and hADAR1p110, GFP and hADAR1p150, and GFP and P19. Images taken under UV lamp at 7 days postagroinfiltration.

Given the presence of ADAR1 across vertebrates, we first wanted to verify if nonhuman ADAR1 would produce the same rSeV-5T escape phenotype. To accomplish this, we infected Madin-Darby canine kidney (MDCK) cells and mouse embryonic fibroblasts (MEFs) with SeV-N5R and SeV-N5T and found that, consistent with human cells, the 5T cassette suppressed initial viral replication but also produced escape mutants (Fig. 5B and Fig. S5A and B). Subsequent sequencing of escape mutants from both MDCK cells and MEFs revealed that the ADAR1 of both these nonhuman species was responsible for the escape phenotype (Fig. 5C and Fig. S5C). Of note, ADAR1 editing of N5T from these species produced escape mutants with fewer edits throughout the 5T cassette (Fig. 5D and Fig. S5D). This decrease in edits likely reflects the fact that a subset of miRNAs used in the targeting cassette are absent in these cell lines as previously reported (43).

Finally, given the apparent lack of ADAR1 in species with well-characterized antiviral RNAi, we wanted to assess if its introduction into such a species would disrupt the host RNAi biology. To this end, we created codon-optimized hADAR1p110 and hADAR1p150 to express in wild-type Nicotiana benthamiana via agroinfiltration. We assessed the effects of these ADAR1 isoforms on gene silencing by coinfiltrating N. benthamiana leaves with a GFP expression plasmid. The expression of GFP, as a nonhost gene, is silenced by host RNAi biology unless a VSR, such as the tomato bushy stunt virus P19, is coinfiltrated (60–62). Using P19 as a positive control for gene silencing suppression, we observed that hADAR1p110, and hADAR1p150 to a lesser extent, was capable of suppressing RNAi, allowing for an increase in GFP expression (Fig. 5E and Fig. S5E). Moreover, the fact that we observe differential interference between P19, hADAR1p110, and hADAR1p150 also allows us to conclude that loss of GFP expression is not the result of direct competition for GFP for the plant’s translational machinery. These results suggest that both isoforms of ADAR1 are disruptive to endogenous RNAi biology.

DISCUSSION

To better understand how viruses can adapt to the selective pressures of cellular defenses, we characterized the evasion of an −ssRNA virus from RNAi in a mammalian system. Unexpectedly, our results provided evidence of a host factor that disrupted this suppressive biology rather than a direct adaptation of rSeV itself. These observations inform us about both the biology of −ssRNA viruses and the relationship between ADAR1 and antiviral RNAi systems. In a previous study we observed that −ssRNA viruses with a strong silencing pressure on essential gene expression from miRNA-based RNAi appeared incapable of escape (38). Although we demonstrate here that lowering the silencing pressure to a single miRNA target allows −ssRNA viruses to escape via point mutation, our results demonstrating lack of escape of rSeV-N5T in ADAR1 KO human epithelial cells reinforce our previous conclusions about the limitations of −ssRNA viruses in responding to RNAi compared to their +ssRNA counterparts. Of note, as in the previous study, we also attempted to generate a 5T recombinant influenza A virus escape mutant but observed neither ADAR1-mediated editing nor viral escape from miRNA-mediated silencing (56). This may be due to the difference in localization as it relates to the site of virus replication and/or a by-product of genomic packaging (63).

Going beyond the biology of −ssRNA viruses and miRNA-mediated RNAi, the results of this study provide insight into the effects of ADAR activity on RNAi biology. ADARs have previously been predicted to antagonize RNAi (64), which was experimentally supported thereafter using both in vitro and in vivo assays (65–67). Past reports focused on the impact of hADAR1 in Drosophila RNAi biology showed findings similar to our results in N. benthamiana, although they found interference to be mediated by ADAR1p150 and not the p110 isoform (66). These studies, combined with our experiments in N. benthamiana, provide evidence of ADAR1 disruption of endogenous RNAi systems.

It should be noted that our miRNA-mediated system establishes a form of RNAi that is functionally identical to endogenous RNAi systems with two exceptions. First, siRNAs generated in the context of endogenous RNAi require early replication to provide a substrate for the RNase III machinery. This dynamic, which would provide time for the virus to antagonize this biology, does not exist within the miRNA-mediated silencing system leveraged here. Second, the generation of endogenous siRNAs tends to be biased to the ends of the viral genome as a by-product of accessibility. In contrast, our miRNA-targeted designs engage viral products not found on the terminal ends of the genome. While the timing of small RNA-mediated silencing does not impact the findings of this paper as they might relate to endogenous RNAi, it remains possible that the location of targeting by ADAR1 would be affected. However, in this regard, the capacity of ADAR1 to interfere with endogenous RNAi in our N. benthamiana experiment would again suggest this is not a significant variability.

Although ADAR1 editing is responsible for the escape of rSeV-5T from RNAi in our system, the dynamics of how this editing occurs are still somewhat uncertain. It is tempting to speculate that miRNA binding is involved in the editing of SeV-N5T genomic RNA, as it has previously been reported that ADAR1 associates with the mammalian RISC through a protein-protein interaction with Dicer (57). However, the mutations in the SeV-N5T escape mutants are consistent with A-to-G edits in the viral genome, which is not targeted or bound by miRNAs, in contrast to what is observed for P5T, which would conform to this model. Based on the N5T data, one could postulate that the dsRNA substrate targeted by ADAR1 consists of negative-sense genome associated with antigenome, cleaved viral mRNA, or a highly complementary cRNA such as a defective viral genome (DVG), which has been previously described (68, 69). Alternatively, it also remains possible that the RISC engages the nascent mRNA generated by the RNA-dependent RNA polymerase, enabling ADAR1 to be recruited alongside the genomic template which spatially would align perfectly to the target sites where editing is observed.

ADAR1 editing of viral RNA is consistent with previous observations of paramyxovirus biology (70–72) but could be more active in N5T as a result of exposed genomes, an indirect consequence of miRNA targeting (56). In contrast, the mutations in SeV-P5T indicate editing in positive-sense RNA and therefore could implicate a direct role for miRNA-mediated recruitment of ADAR1 in this process. Differentiating between an indirect and a direct role for miRNAs in ADAR1 editing will be difficult as the frequency of escape is low and therefore not amenable to biochemical approaches and there are few alternatives to induce this level of attenuation in a manner that will also generate dsRNA. Despite these difficulties, one means of delineating the role of miRNAs in this biology is to test ADAR1-mediated editing in cells lacking a subset of the cognate miRNAs used in our targeting cassette. While the miRNA profile of MEFs is unknown, MDCK cells lack miR-192 and miR-31, which represent two of the five miRNAs in the 5T cassette (43). Given the absence of editing in these sites as it relates to N5T escape in MDCK cells (Fig. 5D), it is tempting to speculate that ADAR1 editing is a direct result of miRNA biology. This phenotype would not be anticipated if ADAR1 editing was the product of double-stranded viral RNA formed from genome and antigenome, in which case mutations would be found throughout the targeting cassette even in the absence of a subset of miRNAs. While the context of genomic editing remains unclear, one possible mechanism might include RISC engagement on the nascent mRNA which could recruit ADAR1 to the corresponding genomic target sites for editing. Regardless of the exact mechanism/target of ADAR1 editing of rSeV-5T viruses, our results indicate that ADAR1 activity creates a barrier to effective RNAi of paramyxoviruses in mammals.

Studies of ADAR1 in mice and humans have revealed it to be a suppressor of the IFN response by regulating erroneous production of endogenous dsRNA (73). In mice, disruption of ADAR1 expression or its editing activity results in elevated IFN levels and is embryonic lethal, but the additional knockout of MDA5 or MAVs can reverse this phenotype (74, 75). In humans, ADAR1 serves to prevent endogenous dsRNAs from shutting down translation through the activation of PKR in somatic cells and the expression of IFN-β in neuronal progenitor cells through MDA5 (76). And in the context of viral infection, the p150 isoform is necessary to prevent sustained IFN expression and apoptosis of cells through RIG-I induction (77). Intriguingly, Caenorhabditis elegans ADARs (Adr-1/Adr-2) seem to serve a function similar to that of mammalian ADAR1 in regulating their host antiviral response, which is RNAi based instead of IFN based. As with ADAR1 in mammals, the ADARs of C. elegans bind to and edit host dsRNA; however, this serves to prevent Dicer from processing these regions of host transcripts into siRNAs that would induce silencing (67). Taken together, these studies suggest an evolutionarily conserved relationship between ADARs and regulation of the host antiviral response.

It is possible that the ADAR1 activity we observe is a barrier to effective antiviral RNAi, which is why we do not see ADAR1 in insects or any ADARs in plants, where this defense is essential. However, with multiple studies in recent years suggesting that RNAi may serve an antiviral function in a subset of mammalian cells, how ADAR1 may function in this context remains unclear. One possibility, should this biology exist, is that it serves as a negative regulator for both IFN and RNAi systems which would need to be controlled in specific cellular contexts, similar to what has been described in C. elegans (67). While this remains possible, it is also important to note that ADAR1 gene evolution has shown evidence for both gains and losses in a variety of species which track with their type of antiviral defense. For example, the animal kingdom shows a loss of ADAR1 in RNAi-enabled animals such as Drosophila melanogaster whereas coral and vertebrate genomes maintain expression and are known to utilize IFN-based or IFN-like-basis defenses (58). Though there are many potential explanations for why some genes are conserved and lost through the course of eukaryotic evolution, this study highlights the delicate balance between minimizing dsRNA production and enabling antiviral defense systems that pivot on its detection or rely on its direct usage.

MATERIALS AND METHODS

Virus design, rescue, and quantification.

rSeV genomes and viral stocks for both SeV-N5T and SeV-N5R were used from a previous study (78). The SeV-N1T construct was generated in the same way as SeV-N5T and SeV-N5R as previously described with a single miR-21 site being inserted into the 3′ UTR of the N gene using a previously inserted SnaBI site using the In-Fusion cloning system (TaKaRa Bio) (78). To construct SeV-P5T, a gBlock containing a fragment of the SeV-P gene and the SeV-P 3′ UTR containing the 5T cassette was synthesized (Integrated DNA Technologies). This was inserted into the rSeV genome using SmaI sites found in the SeV-P gene coding region and SeV-P gene 3′ UTR using the In-Fusing cloning system (TaKaRa Bio). SeV-N1T and SeV-P5T were rescued by cotransfecting their respective genome constructs into BSRT7 cells along with a codon-optimized T7-expressing plasmid and the following plasmids for specific SeV genes under T7 promoters: SeV-N, SeV-P, and SeV-L. Transfections of rescue plasmids were performed with Lipofectamine LTX and Plus Reagent (Thermo Fisher) according to the manufacturer’s instructions. Two days posttransfection, supernatant from transfected BSRT7 cells was transferred to NoDice HEK293T cells to allow for viral replication in the absence of functional miRNA biology. Supernatant from NoDice HEK293T cells was harvested 4 to 6 days postinfection. The titers of all rescued rSeVs were then determined on NoDice HEK293T cells by 50% tissue culture infective dose (TCID₅₀) with infectious dose to infect 50% of cells being determined by eGFP expression under fluorescence microscopy.

Cell culture and reagents.

All cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco 11995-065) with 10% fetal bovine serum (FBS) and penicillin/streptomycin at 37°C in 5% CO₂. All human epithelial cell lines described refer to A549 cells (ATCC CCL-185), including wild-type, Cas9-expressing, and knockout lines. STAT1 KO A549 cells were a kind gift of M. Dittmann from New York University, Grossman School of Medicine, New York, NY (52). NoDice HEK293T cells were a kind gift from B. R. Cullen (79). Nonhuman cell lines used include baby hamster kidney cells (BHK-21) (ATCC CCL-10), Madin-Darby canine kidney (MDCK) cells (ATCC CCL-34), mouse embryonic fibroblasts (ATCC SCRC 10-40), and BHK-21 cells expressing a T7 RNA polymerase gene (BSRT7 cells), which were a kind gift of the Lee lab (Icahn School of Medicine at Mount Sinai).

Generation of Cas9 and ADAR1 KO lines.

Cas9-expressing A549 and STAT1 KO A549 cells were generated using the following lentivirus from Addgene: lentiCas9-BLAST was a gift from Feng Zhang (Addgene viral preparation no. 52962 LV; https://www.addgene.org/52962/; RRID, Addgene_52962) (80). Stable Cas9-expressing lines were selected for with blasticidin. ADAR1 KO cell lines were generated in Cas9-expressing STAT1 KO A549 cells by transfecting a set of three guide RNAs (gRNAs) to human ADAR1 (Horizon; Dharmacon reagent SQ-008630-01-0010) using Lipofectamine RNAiMAX transfection reagent according to the manufacturer’s instructions. Following gRNA transfection, cells were diluted to a concentration of 0.3 cells per mL. One hundred microliters of diluted cells was plated in each well of multiple 96-well plates to isolate single cell clones. Cells were cultured for 2 to 3 weeks with regular medium replacement and then verified for ADAR1 KO via Western blotting.

Flow cytometry.

Cells were trypsinized 1, 2, 4, 6, or 8 days postinfection using 0.25% trypsin for 5 min at 37°C and washed 2 times with 1× Dulbecco phosphate-buffered saline (DPBS), centrifuging between washes at 300 × g. Cells were then fixed by resuspension in 250 μL of 1× BD Cytofix fixation buffer (catalog no. 554655) and incubation on ice for 30 min. After incubation, cells were washed and suspended in BD stain buffer (FBS) (catalog no. 554656). A BD FACSCalibur cytometer was used to acquire GFP signal for samples, and resulting fluorescence-activated cell sorter (FACS) data were analyzed using FlowJo.

RNA-Seq of viral infections.

Preparation of RNA samples for RNA-Seq was performed using the TruSeq stranded mRNA library preparation kit (Illumina) and TruSeq stranded total RNA, with the RiboGold Zero library preparation kit (Illumina) according to the manufacturer’s instructions. Poly(A) RNA-Seq libraries were sequenced on an Illumina NextSeq 500 sequencer, while total RNA-Seq libraries were sequenced on an Illumina NextSeq 500 and a NovaSeq 6000 sequencer.

Amplicon sequencing.

A gene-specific primer to the 5T cassette of SeV N (5′-GTTATCTAGATCCGGTGGATCCCTTG-3′) was used for reverse transcription with SuperScript II reverse transcriptase (Thermo Fisher) on RNA isolated from infected cell lines at various days postinfection according to the manufacturer’s instructions. Resulting cDNA was PCR amplified using CloneAmp HiFi PCR premix (TaKaRa Bio) according to the manufacturer’s instructions. PCR was performed with the following primers to generate 204-bp-long amplicons: forward primer 5′-CGGAGACAATGCAGAGGAGTACG-3′ and reverse primer 5′-GGACCGAAATCATAAGGATGATGGAAAG-3′. The resulting PCR amplicon was sent to Genewiz (Azenta Life Sciences) for “Amplicon-EZ” sequencing.

Bioinformatic and statistical analysis.

For differential gene expression analysis, the RNA-Seq alignment application on Basespace (Illumina) was used to align reads from RNA-Seq to a reference human genome (h19), and the resulting alignments were used with DESeq2 (81). A heatmap for differentially expressed genes (DEGs) was generated using ggplot2 and a custom R script (82). Alignments of viral reads to rSeV genomes and 5T cassette amplicon reference were performed using Bowtie 2 (83). All other statistical analysis was performed using Prism 9 (GraphPad Software), with details of statistical methods used given in the corresponding figure legends.

Sequencing readouts of viral 5T insertion sites were generated via amplicon sequencing as described above. Due to the large accumulation of point mutations across samples, amplicon reads were aligned to a reference unmodified 5T sequence using Magic-BLAST (NCBI). SAM files of Magic-BLAST alignments and BAM files of full viral genome Bowtie alignments were visualized using a custom R script.

Viral passaging and sustained infection.

For both types of infections, cells were initially infected with rSeV at an MOI of 5. For passaging experiments in 96-well plates, an entire 100 μL per well of supernatant was transferred from the infected-cell plate to naive cells every 48 h for a total of 2 passages. For sustained infections, cells were split 1:2 48 h postinfection with additional splitting at later time points as needed to prevent overconfluence.

Western blots.

Cells in either 12- or 6-well plates were washed two times with 1× PBS and lysed in 150 μL of radioimmunoprecipitation assay (RIPA) buffer with 1× cOmplete protease inhibitor cocktail (Roche) and 1× phenylmethylsulfonyl fluoride (Sigma-Aldrich). The lysate was incubated on ice for 10 min and then spun at 17,000 × g at 4°C for 15 min to pellet cell debris. The supernatant was collected, and Laemmli sample buffer (Bio-Rad) containing 2-mercaptoethanol was added at a ratio of 3:1 (buffer to sample). Samples were then heated at 100°C for 5 min before loading into an SDS-PAGE gel. Samples were run on an SDS-PAGE gel and transferred onto a nitrocellulose membrane using the standard transfer settings on the Trans Blot Turbo transfer system (Bio-Rad). Proteins were detected using the following primary antibodies: beta-actin (Cell Signaling, 4967L), beta-actin (Thermo Fisher, AM4302), ADAR1 (Cell Signaling, 81284S), STAT1 (N terminus) (BD Transduction, 610116), and SeV N (6H4), which was provided by the Center for Therapeutic Antibody Discovery at the Icahn School of Medicine. All primary antibodies were diluted 1:1,000 in 5% milk in 1× Tris-buffered saline with 1% (vol/vol) Tween 20 (TBST). Incubations of membranes in primary antibodies were done at 4°C overnight. Anti-mouse (Cytiva, NXA931) and anti-rabbit (Cell Signaling, 7074P2) horseradish peroxidase (HRP)-conjugated secondary antibodies were diluted 1:5,000 in 5% milk in TBST. HRP-conjugated antibodies were detected using the Pierce ECL Western blotting substrate (Thermo Fisher) or the SuperSignal West Femto maximum-sensitivity substrate (Thermo Fisher) according to the manufacturers’ instructions. Fluorescent anti-mouse Alexa 680 secondary antibody (Li-Cor, 926-68070) was also used at a 1:5,000 dilution in 5% milk in TBST. Images of chemiluminescence were taken via film exposure or the ChemiDoC MP imaging system (Bio-Rad). Images of blots using a fluorescent secondary antibody were taken using a ChemiDoC MP imaging system (Bio-Rad). Western blot samples treated with human IFN-β (National Institute for Biological Standards and Control, Gb23-902-531) were treated with 1,000 U/mL 24 h prior to harvesting of cell lysate.

RNA isolation.

Viral and cellular RNA from mammalian cell lines was harvested using either TRIzol reagent (Thermo Fisher) or RNeasy minikits (Qiagen) according to the manufacturer’s instructions. For plant RNA extraction, 100 mg of plant tissue was collected from each infiltrated region at 8 days postinfiltration and frozen immediately in liquid nitrogen. The frozen tissue was ground in a 1.5-mL centrifuge tube using a pestle and resuspended in 1 mL TRIzol (Invitrogen, Inc.) immediately. The RNA extraction from the tissue using TRIzol followed the manufacturer’s protocol.

Viral eGFP assay by microscopy.

Ninety-six-well plates were imaged on a CellInsight CX7 high-content screening platform (Thermo Fisher) with high-content software (HCS) to obtain whole-well images. Cells were plated onto Costar 96-well assay plates (3904) and fixed at various time points postinfection by removing supernatant and adding 50 μL of 4% paraformaldehyde (PFA) for 15 min. PFA was then removed, and cells were washed 2 times with 1× PBS. Cells were then stained with 4′,6-diamidino-2-phenylindole (DAPI) (Sigma-Aldrich) at a concentration of 10 μg/mL in 50 μL of 1× PBS per well for 30 min. DAPI stain was then removed, and wells were washed 2 times with 1× PBS before imaging. Images were taking using HCS analysis protocol Target Activation V4, imaging with channels for DAPI and 488 (eGFP) in all experiments, and additionally with channel 581 (mCherry) in experiments where adenovector expression and mCherry reporter gene were used.

Adenovector transductions.

A549 cell lines were plated prior to transduction in DMEM containing 10% FBS until settled. Cells were then washed twice in 1× PBS to remove all FBS. Adenovector for human ADAR1 (Ad-mCherry-hADAR) from Vector Biolabs was diluted in Opti-MEM reduced-serum medium (Gibco) and added to cells for an MOI of 100 per well. Expression of adenovector-delivered genes was verified 48 h after transduction on an EVOS M5000 imaging system by imaging of the fluorescent mCherry reporter gene, and ADAR1 expression was confirmed by RNA-Seq.

Plant codon-optimized ADAR1 plasmid construction.

The two inserts, 2,796 nt of ADARp110 and 3,681 nt of ADARp150, were amplified from the provided cDNA using the primer sets p110F/p110R and p150F/p150R, respectively (Table 1). The forward primers introduced 20 nt of the recombination region (overlapping) at the 5′ end of the amplicon that is compatible with the 3′ end of the tobacco etch virus (TEV) leader. The reverse primers introduced 20 nt of the recombination sequence at the 3′ end of the amplicon that is compatible with the 5′ end of the S35 terminator. The plasmid backbone, pCB301TEVL-t35S, was linearized through inverse PCR using the primer set ADAR_B.B_F/R (Table 1). Each insert amplicon was cloned into the linearized backbone using the NEBuilder HiFi DNA assembly reaction (New England Biolabs [NEB] Inc.).

TABLE 1.

Primers for plasmid construction

Primer	Sequence, 5′→3′
P110F	accatttacgaacgatagcaATGGCTGAGATAAAAGAAAAGATTT a
P110R	ctacacgggactacacgggaCTAAACTGGGCACAGATAGAAGTTT
P150F	accatttacgaacgatagcaATGAACCCACGACAAGGGTATTCTT
P150R	ctacacgggactacacgggaCTACACGGGACAGAGATAGAAATTC
ADAR_B.B_F	TCCCGTGTAGTCCCGTGTAGTCTAGAGTCC
ADAR_B.B_R	TGCTATCGTTCGTAAATGGTGAAAATTTTC

^aLower case in sequence indicates overlapping region with vector for recombination reaction.

Agroinfiltration.

pCB301 ADARp110 and pCB301 ADARp150, pZP-GFP, and pCB301-P19 (tomato bushy stunt virus gene silencing suppressor, positive control) were introduced into electrocompetent Agrobacterium strain GV3101 using a 2-kV pulse with ECM395 (BTX Inc.). The transformed agrobacteria were spread on the selective laked blood agar (LBA) medium containing appropriate antibiotics and incubated at 29°C until visible colonies formed (about 48 h). A single colony was picked using a disposable micropipette tip and inoculated into 2 mL of fresh LB medium containing appropriate antibiotics and cultured overnight. Ten microliters of overnight culture was inoculated into fresh LB (containing 10 mM morpholineethanesulfonic acid [MES], 20 μM acetosyringone) and cultured until the value for optical density at 600 nm (OD₆₀₀) reached 1.0. The culture was centrifuged at 5,000 × g for 5 min, and the pellet was resuspended in infiltration buffer (10 mM MgCl₂, 10 mM MES, 100 μM acetosyringone). The final OD₆₀₀ of infiltration culture was adjusted to 0.8. Two milliliters of infiltration culture of pZP-GFP was mixed with the same volume of pCB301ADARp110, pCB301ADARp150, pCB301-P19 infiltration culture, and blank infiltration buffer (negative control), respectively. In the mixture, the final concentration of each transformant was 0.4 (OD₆₀₀). The mixture was infiltrated on the abaxial surface of a Nicotiana benthamiana leaf with a 1-mL needleless syringe.

Imaging of plant tissue.

GFP was coexpressed with ADARp110, ADARp150, and P19 to visualize the gene silencing effect of ADARp110 and ADARp150 compared to P19. The single expression of GFP was considered a negative control. The GFP signal strength from each combination with a gene silencing suppressor was imaged 7 days postinfiltration from the abaxial surface of the infiltrated leaf using a UVP Black-Ray B-100A UV lamp and imaging device. The dark background in the image of agroinfiltrated leaves was removed using Adobe Photoshop with no edits made to the leaves themselves or the intensity of GFP in agroinfiltrated patches.

ADAR phylogenetic tree and immune protein table.

FASTA files of proteomes were downloaded from UniProt for the following reference organisms: Danio rerio (UP000000437) (zebrafish), Gallus gallus (UP000000539) (chicken), Mus musculus (UP000000589) (mouse), Drosophila melanogaster (UP000000803) (fruit fly), Caenorhabditis elegans (UP000001940) (nematode), Canis lupus familiaris (UP000002254) (dog), Homo sapiens (UP000005640) (human), Nicotiana tabacum (UP000084051) (tobacco), Xenopus laevis (UP000186698) (frog), Mesocricetus auratus (UP000189706) (hamster), Octopus vulgaris (UP000515154) (octopus), and Sepia pharaonis (UP000597762) (cuttlefish). Reference proteome files were queried for sequences which contained adenosine deaminase editase domains using the HMMSearch function of HMMER3 to flag matches to the Pfam “adenosine deaminase editase” family (PF02137) hidden Markov model (HMM) (84). Sequences which met default statistical cutoff thresholds of HMMSearch were included in downstream analyses. Sequences were manually reviewed, and any exact duplicate sequences, or sequences with significant overlap, for a given organism were collapsed down to a single sequence identity. All remaining sequences were then aligned using the MAFFT multiple sequence aligner with default settings. Multiple sequence alignment was then processed using FastTreeMP to generate a phylogenetic tree. Phylogenetic tree data were imported into the Interactive Tree of Life web application and rerooted at midpoint root (85). The rerooted tree was then exported in Newick tree format and visualized using ggtree (86). Adenosine deaminase proteins were manually labeled into families using prior annotation metadata made available on the UniProtKB website.

Data availability.

The raw sequencing data generated for this study can be found on the NCBI Gene Expression Omnibus (GEO) server under the accession number GSE227592.