Asymmetric bulges within hairpin RNA transgenes influence small RNA size, secondary siRNA production and viral defence

Abstract

Small RNAs (sRNAs) are essential for normal plant development and range in size classes of 21–24 nucleotides. The 22nt small interfering RNAs (siRNAs) and miRNAs are processed by Dicer-like 2 (DCL2) and DCL1 respectively and can initiate secondary siRNA production from the target transcript. 22nt siRNAs are under-represented due to competition between DCL2 and DCL4, while only a small number of 22nt miRNAs exist. Here we produce abundant 22nt siRNAs and other siRNA size classes using long hairpin RNA (hpRNA) transgenes. By introducing asymmetric bulges into the antisense strand of hpRNA, we shifted the dominant siRNA size class from 21nt of the traditional hpRNA to 22, 23 and 24nt of the asymmetric hpRNAs. The asymmetric hpRNAs effectively silenced a β-glucuronidase (GUS) reporter transgene and the endogenous ethylene insensitive-2 (EIN2) and chalcone synthase (CHS) genes. Furthermore, plants containing the asymmetric hpRNA transgenes showed increased amounts of 21nt siRNAs downstream of the hpRNA target site compared to plants with the traditional hpRNA transgenes. This indicates that these asymmetric hpRNAs are more effective at inducing secondary siRNA production to amplify silencing signals. The 22nt asymmetric hpRNA constructs enhanced virus resistance in plants compared to the traditional hpRNA constructs.

Graphical Abstract

Graphical Abstract.

Introduction

Small RNAs (sRNAs), including microRNAs (miRNAs) and small interfering RNAs (siRNAs), play an important role in development, stress response and pathogen defence (1–3). The process of forming sRNAs involves the production of double stranded RNA (dsRNA) or fold-back hairpin RNA, which is cleaved into shorter fragments. sRNAs are loaded into argonaute proteins (AGO) forming an RNA-induced silencing complex (RISC), which then targets RNA for silencing through the complementation of the sRNA to the target sequence resulting in either transcript cleavage or translational repression (4). Critical to the biogenesis of sRNAs are class 3 endoribonuclease (RNase) III enzymes, known as Dicers, which cleave dsRNA into sRNAs (5,6). In plants, Dicer-Like 1 (DCL1) cleaves short hairpin RNAs into miRNAs which are predominantly 21 nucleotides (nt) in length and play a critical role in plant development. DCL2, DCL3 and DCL4 process long dsRNA into 22, 24 and 21nt siRNAs respectively (7). DCL2 and DCL4 play an important role in antiviral defence where they can process viral dsRNA into viral siRNAs (vsiRNAs) (8–11). DCL3 cleaves dsRNA into 24nt repeat-associated siRNAs (rasiRNAs). rasiRNAs play a critical role in transcriptionally silencing repeat-associated DNA by recruiting DNA methylation enzymes to the cognate DNA sequence resulting in cytosine methylation (RNA directed DNA methylation) (12). A monocot specific DCL, DCL5, is expressed specifically in reproductive tissue where it processes long dsRNA into 24nt siRNAs (13,14).

The 22nt size class of sRNAs in plants possess a unique functional property: it can initiate the production of secondary siRNAs which are required for transitive and systemic gene silencing (15,16). These 22nt sRNAs and the resulting secondary siRNA represent a key aspect of plant defence against viruses (10,11), where DCL2-processed 22nt siRNAs promote the production of 21nt secondary siRNAs from the viral transcript via DCL4 (17–19). Similarly, several 22nt miRNAs found in plants are shown to initiate secondary siRNA production to amplify gene silencing. For instance, the Arabidopsis 22nt miRNAs, miR173 and miR828, target the transcripts of Trans-Acting siRNA (TAS) loci (TAS1, TAS2 and TAS4) for initial cleavage and secondary siRNA production, with secondary siRNAs silencing PPR transcripts (TAS1, TAS2) and MYB transcription factors (TAS4) in trans (20–23). Such endogenous secondary siRNAs play an important role in a diverse range of situations such as reproduction of monocotyledonous plants (13,24), regulation of defence response genes (25), and cross-kingdom interactions between the parasitic vine, Cuscuta campestris, and its host (26).

The 22nt miRNAs are produced through the presence of an asymmetric bulge in the precursor miRNA stem that, when processed by DCL1, results in a 22nt/21nt duplex (15,27,28). RISC’s loaded with 22nt miRNAs initiate the production of phased secondary siRNAs (phasiRNAs) from the original cleaved target (29). Upon initial cleavage of the target gene transcript, the 22nt miRNA-AGO1 complex recruits Suppressor of Gene Silencing 3 (SGS3) and RNA dependent RNA polymerase 6 (RDR6) to the cleaved transcript fragment generating long dsRNA, which is subsequently processed by DCL4 producing 21nt secondary siRNAs (30).

Long hairpin RNA (hpRNA) transgenes are the most widely used gene silencing technology in plants (31). A traditional hpRNA transgene consist of a perfect sense-antisense inverted repeat (forming the dsRNA stem of hpRNA) separated by a spacer sequence (forming the loop). hpRNA transgenes typically generate 21nt siRNAs and, to a lesser extent 24nt siRNAs (8,32), but usually with very low levels of 22nt siRNAs presumably due to the hierarchical nature of DCL4 outcompeting DCL2 for dsRNA substrates (10,16,17,33,34). In the present study, we created hpRNA transgenes with asymmetric bulges to express desired siRNA size classes as the dominant siRNA populations in plants, including the 22nt size class. By periodic nucleotide deletions in the sense strand to create asymmetric bulges in the antisense strand of hpRNA, we shifted the dominant antisense siRNA populations from 21nt to 22, 23, or 24nt in size. Most of these asymmetric hpRNA constructs efficiently silenced both a reporter transgene and an endogenous gene. Importantly, these asymmetric hpRNA transgenes effectively induced secondary siRNA production from both the reporter and endogenous target genes downstream of the hpRNA target region irrespective of the primary siRNA size class. Furthermore, an asymmetric hpRNA construct producing 22nt siRNAs conferred enhanced viral resistance compared to a traditional hpRNA construct, representing an important downstream application for these designs.

Materials and methods

Plasmid construction

Preparation of hpRNA[Δ22nt], hpRNA[Δ23nt], hpRNA[Δ24–1nt], hpRNA[Δ24nt-2nt] and hpRNA[Δ24–3nt] constructs involved adding base pair deletions into the sense strand of the hpRNA constructs resulting in a bulge within the antisense strand with no sequence change (Figure 1A, Supplementary Figure S1). All modified sense sequences for GUS, EIN2 and CHS were assembled by annealing the overlapping forward and reverse oligonucleotide primers (Supplementary Table S1) containing XhoI and KpnI sites, respectively. PCR extension of 3′ ends was performed using the high identity LongAmp Taq polymerase (NEB), the product was ligated into the pGEM-T Easy plasmid, and the correct nucleotide sequences were confirmed by sequencing. The modified PCR fragments of GUS, EIN2 and CHS were excised from respective pGEM-T Easy plasmids by digestion with XhoI and KpnI to replace the wild-type sense sequences from hpGUS[WT], hpEIN2[WT] and hpCHS[WT] (32) which were excised with the same restriction enzymes. The resulting binary plasmids were electroporated into E.coli strain DH5α and subsequently introduced into Agrobacterium strains GV3101 and LBA4404 for Arabidopsis and tobacco transformation, respectively.

Figure 1.

hpGUS asymmetric bulge transgenes produce altered siRNA size classes. (A) hpGUS transgenes were designed to vary the frequency and size of nt deletions (X) in the sense strand (pink), resulting in asymmetric bulges (^) present in the antisense strand (blue). Dotted green line represents G:U wobble base pairs. Long hpRNAs are not drawn to scale. Free energy values predict the thermodynamic stability of the different asymmetric bulge structures. (B) Northern blot of small RNAs expressed from the hpGUS transgenes agroinfiltrated into Nicotiana benthamiana leaves. A target GUS-over-expression construct was co-infiltrated. Mir168 was used as a loading and migration control of the siRNAs. (C) Small RNA northern blot of hpGUS transgenic tobacco lines demonstrating altered siRNA size classes between asymmetric bulge constructs. Three biological replicates were analysed per construct. U6 sRNA was used to demonstrate similar loading and migration of sRNA between lines. (D) MUG assay looking at GUS activity in several T0 individuals for each construct. The control are tobacco plants which express the GUS reporter gene. Each bar represents an individual plant with S.E.M. bars representing technical replicates for that plant.

hpRNAs targeting ∼400 bp of the GFP16C reporter gene and a hpRNA construct targeting 300 bp of CMV 2b gene and 300bp of the PVY NIA protease gene (hpCMV:PVY) were designed and sent for synthesis by SynBio Technologies. Both hpCMV:PVY[WT] and hpCMV:PVY[Δ22nt] sequences were the same apart from deletions within the sense strand and a GGG deletion in the antisense strand of the hpCMVPVY[Δ22nt] construct. hpRNA constructs were moved into a binary vector containing a kanamycin-resistance selectable marker and introduced into GV3101 and LBA4404 Agrobacterium strains for plant transformation assays.

Plant growth

Plants used in the experiments included Arabidopsis thaliana (ecotype Col-0), Nicotiana tabacum Wisconsin 38, transgenic Nicotiana tabacum Wisconsin 38 PPGH24 and Nicotiana benthamiana wildtype. Plants were grown on MS with 1% sucrose for 2 weeks at 25°C. Seedlings were transferred into a soil mix containing ¾ Osmocote premium plus mix (Scotts), ¼ seed raising mix (Debco) supplemented with 3 g/l Osmocote mini fertilizer. Plants were grown at 25°C under long day conditions (16 h light/8 h dark). Upon flowering N. tabacum and N. benthamiana lines were moved into a glasshouse and grown at 28°C under natural light conditions for seed collection.

Plant transformation

N. tabacum lines PPGH24 expressing the GUS gene (35) were transformed with GUS hpRNA constructs via the agrobacterium-mediated leaf-disk method (36). N. tabacum W38 plants and N. benthamiana plants were transformed with hpRNA constructs targeting GFP16C and CMV/PVY using the same Agrobacterium transformation method. Arabidopsis thaliana (ecotype Col-0) was transformed with EIN2 and CHS hpRNA constructs using the floral dip method (37). The collected seeds were surface-sterilized (38) and plated on MS media supplemented with 50 μg/ml kanamycin and 150 μg/ml timentin to select transgenic plants.

Agro-infiltration assays were carried out as previously described using N. benthamiana leaves (39). For assaying the processing of hpGUS, N. benthamiana leaves were infiltrated with the hpRNA construct with or without the GUS reporter construct, a GFP reporter construct and the V2 RNA silencing suppressor construct. The GUS reporter construct was included to assay for GUS silencing. The GFP reporter was used to enable visualization of successful agro-infiltration. The viral suppressor V2 was used to prevent the sense-induced post-transcriptional gene silencing naturally associated with agro-infiltration assays, allowing us to observe primary siRNA processing from the hpRNA transgenes. Infiltrated leaves were collected 3 days after infiltration for RNA analysis.

Gene silencing analysis

Quantification of GUS activity was carried out using 4-methylumbelliferyl-β-D-glucuronide (MUG) assay (38), using proteins isolated from three leaves of each plant. EIN2 silencing phenotypes were measured with the triple response assay as previously described (40). In brief, sterilized wild-type and T1 transgenic seeds were grown on MS salt media (without organics) supplemented with 5mg/L 1-aminocyclopropane-1-carboxylic acid (ACC) and kanamycin (except for the WT lines). The plates were sealed tightly with parafilm and exposed under light for 10 h at 22°C to promote germination, and then incubated in the dark for 4 days. Ten representative seedlings were selected for each line and transferred to agar plates to visualize hypocotyl length. Photographed hypocotyl length was measured using ImageJ (http://rsb.info.nih.gov/ij).

DNA methylation analysis

Methylation analysis using bisulphite sequencing was performed as previously described (32). The EpiTect Bisulphite kit (QIAGEN Cat. No. 59 124) was used to bisulphite convert approximately 2 μg of Arabidopsis thaliana DNA. A nested PCR was used to amplify the converted DNA with the following PCR protocol: 12 min at 94°C followed by 10 cycles of 1 min at 94°C, 2:30 min at 50°C, 1:30 min at 72°C, and 30 cycles with 1 min at 94°C, 1:30 min at 55°C, 1:30 min at 72°C, with a final extension of 10 min at 72°C. Following the nested PCR reactions the PCR product was purified using QIAquick PCR purification kit (Qiagen Cat No. 28 104) with ∼50–200 ng of the PCR product sequenced with BigDye Terminator V3.1 premix (Applied Biosystems) using one of the nested primers. The sequencing trace file was opened using the BioEdit software (https://bioedit.software.informer.com), exported to Microsoft Excel using the ‘Export trace values (tab-delimited text)’ feature, and the relative peak heights of cytosines and thymines used to determine methylation levels of each cytosine. Primer sequences are provided in Supplementary Table S1.

RNA analysis

Total RNA was extracted from tobacco leaves using TRIzol® Reagent (Ambion® USA) following the manufacturer's instructions. Northern blot analysis was performed as previously described (41). For small RNA northern blots, 5–20 μg of total RNA were separated in 17% denaturing polyacrylamide gel, blotted to HyBond-N⁺ membrane (GE Healthcare), and hybridized with ³²P-labeled 200-nt RNA probes. For mRNA northern blots, 2–5 μg of total RNA was separated on a formaldehyde-agarose gel. Probe sequences can be found in Supplementary Table S1.

For RT-qPCR, 1μg of total RNA was first treated with RQ1 DNase (Promega, Cat. # M6101) for 30 minutes followed by cDNA synthesis using Superscript III and oligo(dT) (ThermoFisher, Cat. # 18 080 051). The cDNA reaction was diluted with 180μl of RNase free water with 3μl of diluted cDNA used per reaction. qPCR reaction was run on a Corbett 2000 Rotor-Gene real-time PCR machine (QIAGEN) using FAST™ SYBR™ Green Master Mix (ThermoFisher, Cat. # 4 385 612), with 0.8 μM forward primer and 0.8 μM reverse primer. qPCR was performed using two technical replicates for each sample, in a 20 μL reaction. EIN2 transcript level in Arabidopsis was normalized to the house keeping gene FDH using the 2-ΔΔCt method. Primer sequences can be found in Supplementary Table S1.

Sequencing and analysis

Total RNA was extracted as described above and sent to The Australian Genomics Research Facility for small RNA sequencing. GSE252890 and GSE243255 libraries were created using the NEBNext® Small RNA Library Prep kit with GESE252890 sequenced on a NextSeq 500 machine—high output (75 bp reads) while GSE243255 was sequenced on a NovaSeq SP Lane (100) cycles. GSE243296 libraries were created using the NEXTFLEX Small RNA v3 Library Preparation with Bead Size Selection kit and sequenced on a NovaSeq SP Lane 100 cycles. For analyzing small RNAs from the GUS gene, one control transgenic line (hpGFP) and 3–6 different T0 individuals per hpGUS construct were sent for sequencing. For analyzing EIN2 siRNAs, seedlings were grown as described above, with a pool of T2 progeny plants from individual T1 lines used for RNA isolation and small RNA sequencing.

sRNA-seq reads had adapters removed using cutadapt followed by mapping using BOWTIE version 1.3.1 with options -a, -v0, –best and –strata. siRNA reads of N. tabacum lines were mapped to the GCA_002210045.1_Nitab4.5_genomic sequence plus the sequences of the PDK intron, GUS and the modified sense strand for each asymmetric bulge construct. For detecting EIN2 siRNAs, total reads were mapped to the TAIR10 genome sequence plus the sequences of EIN2 cDNA, the hpEIN2WT construct (including the upstream 35S promoter) and the modified sense strand. Reads were subsequently remapped to a custom reference containing EIN2 cDNA, the hpEIN2WT sequence and sense sequence of each asymmetric bulge construct stem. A Custom R script was used to separate the reads into different target gene regions (Upstream, Downstream and dsRNA Stem), sense and antisense strands, and different size classes, which were then normalized to reads per million and visualized using ggplot2. Bedtools genomecov function was used to isolate reads to the target gene (GUS or EIN2 cDNA), normalize the data to reads per million and create a bedgraph coverage file. A custom R-script was used to visualize sRNA abundance over the target gene via a graphical representation (ggplot2). These graphical representations exclude reads mapped to the modified asymmetric bulge sense strands due to different base-pair co-ordinates (as a result of the deletions).

Viral infection assays

Cucumber Mosaic Virus (CMV) and Potato Virus Y (PVY) were propagated in a wildtype N. tabacum or N. benthamiana plants (stock plant). CMV assays were undertaken in N. tabacum while PVY assays were undertaken in N. benthamiana. Transgenic lines were grown for ∼one month in soil before being infected with either CMV or PVY virus.

For CMV infection, infected leaf from the stock plant was ground in 0.01 M NaPO₄ pH 7.2 at a ratio of 1g/5ml in the presence of carborundum. Two expanded leaves per plant were sprinkled with carborundum and either 0.01 M NaPO₄ pH 7.2 (negative control) or the CMV viral extracts were rubbed onto the leaf. For the experiment with T0 plants, 5 WT plants were treated with buffer alone (mock inoculation), 10 WT plants inoculated with the CMV virus, and 25 transgenic line per construct were CMV inoculated. For the experiment with T1 progeny plants, two independent lines for hpCMV:PVY[WT] and hpCMV:PVY[Δ22nt] were selected. Twenty T1 plants per line were used with 2–5 individuals mock-inoculated while the rest were inoculated with CMV (>14 plants). Viral symptoms were scored once a week for three weeks based on the score range from – (no symptoms) to +++++ (severe symptoms).

For PVY assays infected leaf from the stock plant was ground in 0.01M NaPO₄ pH 7.2 at a ratio of 1g/10ml in the presence of carborundum. Two expanded leaves per plant were sprinkled with carborundum and either 0.01M NaPO₄ pH 7.2 (mock inoculation) or the PVY extract rubbed onto the leaf. Ten WT plants were mock-inoculated, 10 WT plants and >13 plants per transgenic line were PVY-inoculated. Viral symptoms were scored once a week for three weeks based on the range from – (no symptoms) to +++ (severe symptoms).

Results

hpRNA constructs with asymmetric bulges generate desired siRNA size classes and induce effective silencing of the β-glucuronidase (GUS) gene

We tested whether hpRNA transgenes could produce different siRNA size classes (22, 23 and 24nt siRNAs) through additions of asymmetric bulges into the hpRNA structure. Asymmetric bulges were introduced through the deletion of one or more nucleotides from the sense strand within every 22–24nts, resulting in periodic bulges along the unmodified antisense strand that retained target sequence complementarity (Figure 1A, Supplementary Figure S1). A traditional hpRNA construct (hpRNA[WT]) and a G–U base-paired hpRNA construct (32), with the same target sequence, were included as controls. hpRNA[Δ22nt], designed to express 22nt siRNAs, contained one asymmetric bulge in every 22nt; hpRNA[Δ23nt] carried two asymmetric bulges in every 23nts. For hpRNA[Δ24nt], three versions were designed with different patterns of 3nt deletions hence asymmetric bulges every 24nts. The first version, hpRNA[Δ24nt-1], had three single nucleotide bulges in the 24nt window separated by ∼6/7nts; the second version, hpRNA[Δ24nt-2], had a 1nt bulge followed by a 2nt bulge separated by ∼10/11nt; the last version, hpRNA[Δ24nt-3], had a single 3nt asymmetric bulge every ∼24nts (Figure 1A, Supplementary Figure S1B).

The asymmetric hpRNA constructs were first tested against the GUS reporter gene using Agrobacterium infiltration transient expression assays in Nicotiana benthamiana leaves with or without a target GUS construct. This assay included a construct expressing the viral RNA silencing suppressor V2, which binds to SGS3 and inhibits secondary siRNA biogenesis (42) thereby enabling a clear understanding of the primary processing of the different hpRNA constructs. The hpGUS[WT] and hpGUS[G:U] constructs produced 2 dominant siRNA bands consistent with the production of 21 and 24nt siRNAs (Figure 1B, Supplementary Figure S2A) (32). All the asymmetric hpRNA constructs produced siRNAs with different sizes from hpGUS[WT], becoming progressively larger with increasing number and size of the asymmetric bulges (Figure 1B, Supplementary Figure S2A). The change in siRNA size classes was not dependent on the presence of the target transcript (Supplementary Figure S2A). The hpGUS[Δ23/Δ24nt] constructs tended to produce fewer siRNAs than either the hpGUS[WT] or hpGUS[Δ22nt], suggesting that too many asymmetric bulges inhibit hpRNA processing through reducing the stability of hpRNA structure (Figure 1B, Supplementary Figure S2A).

The hpGUS constructs were then tested in stably transformed N. tabacum plants which contain a target GUS transgene (PPGH24) (32). Like the Agrobacterium infiltration experiments, increasing the number and size of the asymmetric bulges altered the gel migration of the transgene-derived siRNA indicating a change in siRNA size class from the antisense strand (Figure 1C). siRNAs from the sense strand of hpGUS[Δ22nt] were 21nt in size, the same as hpGUS[WT], suggesting that all constructs are likely processed by DCL4 or DCL1 (Supplementary Figure S2B). The amounts of both the sense and antisense siRNAs were comparable between hpGUS[Δ22nt] and hpGUS[WT] (Figure 1C, Supplementary Figure S2B), suggesting that asymmetric bulges present in the hpGUS[Δ22nt] molecule did not impact hpRNA processing. The stably transformed hpGUS[Δ23nt/Δ24nt] plants had reduced levels of siRNAs compared to the hpGUS[WT] and hpGUS[Δ22nt] plants (Figure 1C), consistent with the Agrobacterium infiltration results suggesting reduced efficiency of hpRNA processing.

A fluorometric 4-methylumbelliferyl-β-d-glucuronide (MUG) assay was used to examine the efficiency of GUS silencing by the different hpRNA constructs. The hpGUS[WT] and hpGUS[Δ22nt] plants showed strong GUS silencing, with a 89% and 96% reduction in GUS activity respectively across all the independent lines (Figure 1D). hpGUS[Δ23nt] plants had an average GUS knockdown of 75%, while hpGUS[Δ24nt-1], hpGUS[Δ24nt-2] and hpGUS[Δ24nt-3] plants showed an average GUS knockdown of 58%, 65% and 43%, respectively (Figure 1D). The degree of GUS silencing appeared to correlate with the amount of siRNAs produced by the different hpGUS constructs (Figure 1C, D).

Asymmetric hpRNA transgenes induce siRNA production downstream of the target sequence

To further analyse the siRNA profiles from the hpGUS transgenes, RNA samples from 3–6 independent plants for each transgene construct were sent for sRNA deep sequencing. sRNAs were mapped to the full-length target GUS sequence (1865 bp) including the 200bp hpRNA target region (nt.804 – 1004 of GUS ORF), with reads sorted based on size class and strand polarity. Consistent with the northern blot analysis, the size profile of siRNAs matching the sense strand of hpRNA was similar for all the hpGUS constructs, with 21nt being the dominant population (Figure 2A, Supplementary Figure S3A). In contrast, the predominant size class of siRNAs matching the antisense strand of hpRNA was dependent on the number and size of asymmetric bulges present. hpGUS[Δ22nt] displayed a size change of the dominant antisense siRNA population from 21 to 22nt, while hpGUS[Δ23nt] showed a size shift from 21 to 23nt (Figure 2A, Supplementary Figure S3A). All three hpGUS[Δ24nt] transgenes gave a dominant antisense siRNA size population of 24nt, despite the relatively low abundance (Figure 2A). Thus, all asymmetric hpGUS constructs generated the expected siRNA size classes.

Figure 2.

Asymmetric bulge transgenes can produce secondary siRNAs downstream of hpGUS target region. Deep sequencing reads over the target region of the GUS gene (A) and downstream of the target region (B). Errors bars represent the S.E.M. from three biological replicates. C) Graph representing normalized reads (reads per million) over GUS gene. Darker colour represents reads mapping to the hpRNA target region while lighter colours representing reads mapping either upstream or downstream of target region. Sense reads mapping to asymmetric bulge stems are not presented on these graphs.

An important question was whether siRNAs from the asymmetric hpRNA transgenes, especially the hpGUS[Δ22nt] transgene, could initiate the production of secondary siRNAs. We looked for GUS-mapping siRNAs downstream of the hpRNA target region, which can only occur through secondary siRNA production. Stable transgenic N. tabacum plants containing a traditional GUS hpRNA transgene can produce secondary siRNAs from the downstream region (9,18). Consistent with these reports, we observed varying levels of siRNAs downstream of the target site in hpGUS[WT] lines (Supplementary Figure S3B). However, compared to the hpGUS[WT] lines, the hpGUS[Δ22nt] lines had higher levels of 21nt siRNAs downstream of the hpRNA target region, indicating strongly enhanced production of secondary siRNAs (Figure 2B, C, Supplementary Figure S3). Interestingly, the hpGUS[Δ23nt/[Δ24nt] plants also showed increased amounts of 21nt siRNAs in this downstream GUS region (Figure 2C, Supplementary Figure 4). Thus, all the asymmetric hpRNA constructs could induce secondary siRNA production from the GUS target gene.

Asymmetric hpRNA transgenes induce effective silencing against endogenous genes

We next examined whether asymmetric hpRNA transgenes can silence endogenous genes and induce secondary siRNA production from the endogenous target. To do this we targeted 200 bp regions of two endogenous genes ETHYLENE INSENSITIVE 2 (EIN2) and CHALCONE SYNTHASE (CHS) in Arabidopsis thaliana. Silencing of EIN2 can be scored via hypocotyl length of seedlings grown in the dark on 1-aminocyclopropane-1-carboxylic acid (ACC) medium (32). Silencing of chalcone synthase (CHS) can be characterized through a loss of pigmentation in seed coats.

Approximately 20 independent T2 lines for each hpEIN2 transgene were grown for four days in the dark in the presence of the selective agent kanamycin and scored for EIN2 silencing based on hypocotyl length (Figure 3, Supplementary Figure S5). hpEIN2[WT], hpEIN2[Δ22nt], hpEIN2[Δ23nt], hpEIN2[Δ24nt-1] and hpEIN2[Δ24nt-3] all had multiple lines with extended hypocotyl length compared to the wild-type Arabidopsis plant, indicating strong EIN2 silencing. However, the hpEIN2[Δ24nt-2] lines showed no phenotypic difference from the wild-type control. A comparison of the top 5 performing lines from each construct demonstrated a 10% increase in hypocotyl length (P< 0.01; two-tailed Student's t-test) in the hpEIN2[Δ22nt] lines compared to the hpEIN2[WT] lines (Figure 3, Supplementary Figure S5). Real-time RT-PCR analysis showed strong suppression of EIN2 in all but one of the hpEIN2 constructs used (Figure 4A).

Figure 3.

EIN2 knockdown using asymmetric bulge molecules results in elongated hypocotyls. 4-day old seedlings were grown in the dark in the presence of ACC. Five different T2 lines per construct were arranged next to Arabidopsis thaliana Columbia wildtype plants. At least 10 individuals per line were measured for hypocotyl length. Bar = 5 mm. Error bars = S.E.M.

Figure 4.

Asymmetric bulge transgenes can produce altered siRNA size classes, trigger secondary siRNAs and silence EIN2 expression. (A) Real-time PCR of EIN2 expression in 4-day seedlings grown in the dark (Error bars = Standard deviation). sRNA sequencing reads aligned to the hpRNA target region (B) and the downstream region (C) of the EIN2 gene. (D) Graph representing normalized reads (reads per million) over EIN2 cDNA. Darker colour represents reads mapping to the hpRNA target region while lighter colours representing reads mapping either upstream or downstream of target region. Sense reads mapping to asymmetric bulge stems are not presented on these graphs. Reads mapped to Col WT and hpEIN2[Δ24–2nt] line represent background levels of sRNAs. (E) Bisulphite PCR results looking at DNA methylation levels over part of the hpRNA target region and downstream of the target region. Black bar represents hpRNA target region while the green represents the region bisulphite converted. Detailed base pair methylation levels can be found in Supplementary Table S2. Error bars = S.E.M

For the endogenous CHS target gene, we tested a hpCHS[WT] and hpCHS[Δ22nt] transgene in Arabidopsis thaliana. A comparison of 19 randomly selected T2 lines demonstrated a generally stronger loss of seed coat pigmentation in the hpCHS[Δ22nt] lines (8/19) than the hpCHS[WT] lines (Supplementary Figure S6). These results demonstrate that the 22nt asymmetric hpRNA transgenes can produce stronger phenotypic changes than the traditional hpRNA transgenes.

To examine if the predicted siRNA size classes were produced by the asymmetric hpRNA constructs and whether these siRNAs could induce secondary siRNA production from the target endogenous genes, we performed sRNA deep sequencing analysis on the hpEIN2 plant lines. Like the hpGUS constructs, the sense siRNAs showed similar size distributions among the different transgenes, with 21nt siRNAs being the dominant size class (Figure 4B). However, antisense siRNAs produced by the asymmetric hpEIN2 transgenes varied in size depending on the number and size of asymmetric bulges, with hpEIN2[Δ22nt], hpEIN2[Δ23nt] and hpEIN2[Δ24–3nt] generating 22, 23 and 24nt antisense siRNAs as the dominant size classes, respectively. Consistent with the lack of EIN2 silencing phenotypes, the hpEIN2[Δ24–2nt] plants showed no siRNA accumulation (Figure 4B). In association with these altered siRNA size classes, plants containing the asymmetric hpEIN2 transgenes accumulated increased amounts of EIN2 siRNAs downstream of the hpRNA target region compared to the hpEIN2[WT] plants, albeit at lower levels than that observed for the GUS transgene (Figure 4C, D, Supplementary Figure S4). This indicated that the asymmetric hpRNA transgenes could induce secondary siRNA production from endogenous target genes. Differences in the amount of secondary siRNAs between individual lines may be a result of different transgene integration patterns or transgene copy numbers.

The 24nt size class of siRNAs are known to induce RNA-directed DNA methylation that can lead to transcriptional gene silencing (12). However, plants containing the asymmetric hpRNA transgenes, including hpEIN2[Δ24nt], showed no increase in DNA methylation around the target region of the EIN2 gene when compared to the hpEIN2[WT] plants (Figure 4E, Supplementary Figure S7, Supplementary Table S2). In fact, the strong CG methylation detected for all lines was largely due to the inherent CG methylation of the target locus (Supplementary Figure S7). This result suggests that the observed EIN2 silencing induced by the different constructs occurs through the post-transcriptional gene silencing pathway independent of DNA methylation.

hpRNA[Δ22nt] enhance plant resistance to viral pathogens

DCL2-processed 22nt vsiRNAs contribute positively to antiviral defence in plants by inducing secondary siRNA production from viral RNA template (43). Traditional hpRNA transgenes produce relatively small amounts of 22nt siRNAs, which would limit the involvement of 22nt siRNAs in hpRNA-mediated virus resistance. As the asymmetric hpRNA[Δ22nt] construct expressed 22nt siRNAs as the predominant size class and induced secondary siRNA production, we tested whether this hpRNA design would confer more effective virus resistance in plants than the traditional hpRNA. An asymmetric hpRNA construct (hpCMV:PVY[Δ22nt]) and a traditional hpRNA construct (hpCMV:PVY[WT]), targeting both the Cucumber Mosaic Virus (CMV) and the Potato Virus Y (PVY), were transformed into N. tabacum (cv. W38) and N. benthamiana. As negative controls, two equivalent hpRNA constructs targeting the mGFP5 version of the green fluorescence protein gene (hpGFP[WT] and hpGFP[Δ22nt]), were also transformed into N. tabacum and N. benthamiana.

Non-transgenic (WT) and primary (T0) transgenic N. tabacum plants containing the various hpRNA constructs were inoculated with a high dose of CMV. The severity of viral infection, indicated by mottled symptoms in systemic leaves, was scored once a week over 3 weeks (Figure 5A, B). After approximately one week, all CMV-inoculated plants of WT, hpGFP[WT], hpGFP[Δ22nt] and hpCMV:PVY[WT] displayed varying degrees of symptoms (Figure 5B). While the majority of hpCMV:PVY[Δ22nt] T0 plants showed some levels of CMV symptoms, the severity of symptoms was generally reduced compared to the hpCMV:PVY[WT] plants and the hpGFP controls. Importantly, ∼30% of the CMV-inoculated hpCMV:PVY[Δ22nt] plants did not show any viral symptoms throughout the assay period, in contrast to the other plants that all developed symptoms after two weeks of infection (Figure 5A, B). Northern blot analysis confirmed that these symptomless hpCMV:PVY[Δ22nt] lines lacked any virus, while all other infected lines showed the presence of viral RNA (Figure 5C). sRNA northern blot hybridization verified that the symptomless hpCMV:PVY[Δ22nt] lines all accumulated siRNAs with a dominant size class of 22nt, in contrast to the hpCMV:PVY[WT] lines that accumulated a dominant 21nt siRNA size class plus a 22nt size class (Figure 5C).

Figure 5.

hpCMV:PVY[Δ22nt] transgene provides improved viral resistance compared to wildtype and hpCMV:PVY[WT] plants. (A) Photos of plants and leaves of T0 transgenic plants 9 days after viral inoculation. (B) Scoring of CMV symptoms over 3 weeks. Severity of the symptoms was scored from symptomless (–) to severe (strong mottled phenotypes +++++). (C) Northern blot analysis of individual T0 plants, including 4 hpCMV:PVY[Δ22nt] lines which were resistant to CMV infection. From top to bottom: Large RNA northern of viral titre; Gel image of RNA loading; small RNA northern with a probe showing small RNAs derived from the transgene. U6 loading image of the same small RNA gel.

To examine if the virus resistance observed in the T0 lines were stably inherited, ∼15 T1 sibling plants from 2 hpCMV:PVY[WT] and 2 hpCMV:PVY[Δ22nt] lines were inoculated with a high dose of CMV. After one week, the hpCMV:PVY[Δ22nt] population displayed a clear improvement in viral resistance, with an average of 77% of plants across the two lines showing no symptoms compared to an average of 20% for the 2 hpCMV:PVY[WT] lines (Supplementary Figure S8A, B). After 3 weeks the majority of hpCMV:PVY[WT] (>95%) plants displayed viral symptoms while 77% of hpCMV:PVY[Δ22nt] plants remained symptomless (Supplementary Figure S8B). Northern blot analysis confirmed the absence of CMV RNA in the symptomless hpCMV:PVY[Δ22nt] plants (Supplementary Figure S8C).

We also evaluated PVY viral resistance in primary (T0) transgenic N. benthamiana plants containing the various hpRNA constructs. Plants were infected with PVY and scored for viral symptoms such as vein clearing over a 3-week period (Supplementary Figure S9A). Both hpCMV:PVY[WT] and hpCMV:PVY[Δ22nt] transgenic lines displayed improved PVY resistance at 35% and 48% respectively over a 3 week period. Again hpCMV:PVY[Δ22nt] displayed less severe symptoms compared to hpCMV:PVY[WT] with viral RNA titre consistent with the observed phenotypes (Supplementary Figure S9B,C). Taken together this data demonstrates that enhancing the production of 22nt siRNAs targeting viral pathogens can enhance overall plant viral resistance when compared to traditional hairpin structures.

Discussion

Plants encode four different Dicer-like enzymes which process dsRNA into distinct size or functional classes of sRNAs. miRNAs, processed by DCL1 in the nuclei from short hairpin precursors, are mostly 21nt in size and induce post-transcriptional gene silencing to regulate plant development. siRNAs are comprised of three main size classes, 21, 22 and 24nts, which are processed from long dsRNA by DCL4, DCL2 and DCL3, respectively. The 21 and 22nt siRNAs induce post-transcriptional gene silencing or RNA interference, whereas the 24nt siRNAs direct de novo DNA methylation which can lead to transcriptional gene silencing. The 22nt size class of sRNAs has the unique ability to initiate the production of secondary siRNAs amplifying the silencing signals (20,22,29), but this size class is a rarity in plants compared to the 21 and 24nt size classes. This is due to an antagonistic relationship between DCL4 and DCL2 with DCL4 inhibiting DCL2 expression levels and outcompeting DCL2 for dsRNA substrates (10,33). In dcl4 mutants, DCL2 can increase the production of 22nt siRNAs and transitive silencing (9,18). However, this can have a detrimental effect due to indiscriminate secondary siRNA biogenesis causing several developmental defects (17,33).

Production of 22nt miRNAs depends on the rare occurrence of an asymmetric bulge in the miRNA precursor. The present study was aimed to utilize the endogenous siRNA biogenesis pathway to produce high amounts of 22nt siRNAs as well as other desired siRNA size classes from modified hpRNA designs. The traditional hpRNA, with a perfect dsRNA stem, is processed primarily by DCL4 into the dominant 21nt siRNAs and sometimes also by DCL3 to generate significant amounts of 24nt siRNAs, with DCL2-processed 22nt siRNAs being less represented (8). Adding an asymmetric bulge to the dsRNA stem of hpRNA every 22 nucleotides successfully enabled the production of a dominant 22nt size class of siRNAs from the bulged strand. miRNA precursors have been modified to express artificial 22nt miRNAs (15), with each vector producing a single 22nt miRNA. These 22nt artificial miRNA precursors are likely processed by the miRNA biogenesis machineries including DCL1. The different siRNA size classes from the asymmetric bulge constructs could be processed by the different DCL proteins (eg. 24nt siRNAs from hp[Δ24nt] by DCL3). However, siRNAs from the sense strand of these asymmetric bulge hpRNAs such as hpGUS[Δ22nt] is predominantly 21nt in size, in contrast to the increased 22, 23 and 24 nt sizes of the antisense siRNAs. This suggests that the asymmetric hpRNAs, like the traditional hpRNAs, are processed primarily by DCL4, although involvement of DCL1 cannot be ruled out. Experiments characterizing these different asymmetric bulge molecules in different DCL mutants would help to address this question.

The asymmetric hpRNA[Δ22nt] transgenes generate a mixed population of 22nt siRNAs from the whole length of dsRNA stem. Increasing the size and frequency of asymmetric bulges in the hpRNA stem resulted in the shift of dominant siRNA sizes to 23 and 24nt, making it possible to produce any desired size classes of siRNAs in plants using the asymmetric hpRNAs. However, the increased frequency and number of asymmetric bulges along the dsRNA stem of hpRNA could have consequences on dsRNA stability and DCL processing of siRNA as suggested by the relatively low levels of 24nt siRNAs from hpRNA[Δ24nt] construct.

Our results show that the asymmetric hpRNA transgenes induced effective silencing against both the GUS reporter transgene and two endogenous genes. The hpRNA[Δ22nt] transgenes induced stronger silencing of the endogenous genes tested, than either the traditional hpRNA[WT] molecules. Surprisingly, the reduced amount of siRNAs with the hpEIN2[Δ24nt] transgenes did not impact their ability to silence the endogenous EIN2 gene. The 24nt size class of siRNAs are typically involved in RNA-directed DNA methylation (RdDM) where they recruit DNA methyltransferase enzymes to the cognate DNA for methylation causing transcriptional silencing (12). However, the hpRNA[Δ24nt] transgenic plants showed no increase in DNA methylation at the target gene DNA when compared to plants containing the traditional hpRNA transgene, suggesting that the observed silencing of EIN2 and GUS in the hpRNA[Δ24nt] plants does not involve RdDM.

A key characteristic of 22nt sRNAs is their ability to initiate the biogenesis of secondary siRNAs. Consistently, the hpGUS[Δ22nt] and hpEIN2[Δ22nt] plants accumulated increased amounts of secondary siRNAs compared to the corresponding hpRNA[WT] plants, which are likely to account for the stronger gene silencing phenotypes in the hpRNA[Δ22nt] plants. Interestingly, the hpRNA[Δ23nt] and hpRNA[Δ24nt] transgenes were also effective at inducing secondary siRNA production, making it possible that either the increased siRNA size or the presence of asymmetric bulges in the dsRNA precursor contributes to the amplification process (27). The number of secondary siRNAs induced by the asymmetric hpRNA transgenes from the endogenous EIN2 gene, a lowly expressed gene, was relatively low compared to that from the highly expressed GUS reporter gene. It is possible that increasing the target transcript or targeting a highly expressed endogenous gene would increase the number of secondary siRNAs induced by the asymmetric bulge constructs (28). One potential drawback of the asymmetric hpRNA design is the potential to increase off-target affects via gene silencing by secondary siRNAs. It would therefore be important to not only test the targeted region for off-target effects but also the downstream region.

Viral pathogens represent a significant challenge to the agricultural industry where they impact several economically important crops such as beans (Bean Golden Mosaic Virus), Cacao (Cacao swollen-shoot virus) and Cassava (Cassava Mosaic Virus) (43–48). The gene silencing pathway represents the primary defence mechanism used by plants to overcome viral infection via viral siRNA (vsiRNA)-mediated viral RNA degradation (7). As a result, hpRNA technologies have been used to enhance plant resistance against viruses, including agriculturally relevant viruses such as Cassava mosaic virus and green bean mosaic virus (44,48). In this system DCL4 plays a leading role via the production of 21nt siRNAs, while DCL2 produces 22nt siRNAs to induce systemic silencing. Under normal viral infection DCL4 activity out-competes the DCL2 pathway limiting 22nt siRNA production and systemic silencing (10,33). We believe that the hpCMV:PVY[Δ22nt] construct produced abundant 22nt siRNAs through a DCL2-independent manner using DCL1 or more likely DCL4 which is known to process 21nt siRNAs from hpRNA transgenes (8). It is possible that the 22nt siRNAs from the asymmetric hpRNA transgenes enhances secondary vsiRNA production from the target viral RNA, promoting plant viral resistance. Indeed, transgenic N. tabacum plants expressing hpCMV:PVY[Δ22nt] showed enhanced resistance to CMV under high viral inoculum. Under this infection condition all plants carrying a traditional hpRNA developed disease symptoms whereas a proportion of the asymmetric hpRNA plants remained symptomless with viral resistance stably inherited into the T1 generation.

A key characteristic of 22nt sRNAs is their ability to initiate transitive and systemic silencing throughout the plant (9,16,18). In dcl2 mutants this systemic silencing is inhibited while in dcl4 mutants secondary siRNA accumulation and systemic silencing are enhanced. A 22nt miRNA from the parasitic plant Cuscuta campestris allows systemic silencing of host Arabidopsis plant genes via secondary siRNAs, which facilitates their cross-kingdom interactions (26). Viral infection also demonstrates systemic silencing through secondary vsiRNAs which move from source to sink initiating viral recovery in plants (49). The results provided here along with those present in the literature suggest that a combination of DCL4-generated primary vsiRNAs and hpRNA[Δ22nt]-induced secondary vsiRNAs enhance systemic silencing and viral resistance. The 22nt size class of siRNAs are also shown to be more effective at inducing systemic silencing via topical applications (50). Future work will examine whether the molecules described here provide stronger silencing via topical applications compared to traditional double stranded RNA molecules.

In conclusion, we have demonstrated that modifications of hpRNA transgenes via the introduction of asymmetric bulges can manipulate siRNA size classes in plants. A similar strategy could be applied to manipulate siRNA sizes in other organisms. These modified hpRNA transgenes and the resulting siRNA size classes are effective in silencing the target gene and in amplifying the silencing signal through the production of secondary siRNAs. We believe that the molecules described here can improve plant resistance against viruses and potentially other pathogens enhancing global food security.

Data availability

Sequencing data has been submitted under GEO database super series GSE243297.

Supplementary data

Supplementary Data are available at NAR Online.

Funding

D.J. was supported by the China Scholarship Council (202108500081).

Conflict of interest statement

None declared.