Argonaute 8 (AGO8) Mediates the Elicitation of Direct Defenses against Herbivory

AGO8 forms an important component of induced direct defense machinery as it modulates several regulatory nodes in signaling and response networks during herbivore attack in wild tobacco.

Abstract

In Nicotiana attenuata, specific RNA-directed RNA polymerase (RdR1) and the Dicer-like (DCL3 and DCL4) proteins are recruited during herbivore attack to mediate the regulation of defense responses. However, the identity and role(s) of Argonautes (AGOs) involved in herbivory remain unknown. Of the 11 AGOs in the N. attenuata genome, we silenced the expression of 10. Plants silenced in NaAGO8 expression grew normally but were highly susceptible to herbivore attack. Larvae of Manduca sexta grew faster when consuming inverted-repeat stable transformants (irAGO8) plants but did not differ from the wild type when consuming plants silenced in AGO1 (a, b, and c), AGO2, AGO4 (a and b), AGO7, or AGO10 expression. irAGO8 plants were significantly compromised in herbivore-induced levels of defense metabolites such as nicotine, phenolamides, and diterpenoid glycosides. Time-course analyses revealed extensively altered microRNA profiles and the reduced accumulation of MYB8 transcripts and of the associated genes of the phenolamide and phenylpropanoid pathways as well as the nicotine biosynthetic pathway. A possible AGO8-modulated microRNA-messenger RNA target network was inferred. Furthermore, comparative analysis of domains revealed the diversity of AGO conformations, particularly in the small RNA-binding pocket, which may influence substrate recognition/binding and functional specificity. We infer that AGO8 plays a central role in the induction of direct defenses by modulating several regulatory nodes in the defense signaling network during herbivore response. Thus, our study identifies the effector AGO of the herbivore-induced small RNA machinery, which in N. attenuata now comprises RdR1, DCL3/4, and AGO8.

When plants are attacked by herbivores, they elicit a complex defense response, which includes direct or indirect defenses. Direct defenses often consist of chemicals that function as antifeedants, antinutrients, or toxins for herbivores (Howe and Jander, 2008; War et al., 2012). In Nicotiana attenuata, nicotine, polyamines (Kaur et al., 2010), and diterpenoid glycosides (DTGs; Heiling et al., 2010) often function as direct defenses, as they have been known to be antinutritive and toxic. Attacked plants also emit volatile organic compounds or provide nectar rewards that recruit predators and natural enemies of the attacking insects, and these function as indirect defenses (Howe and Jander, 2008). To elicit these defenses, plants reconfigure their signaling and metabolic networks, and this reconfiguration comprises a complex chain of events involving cell membrane depolarization, calcium flux, and MAPK activation (Maffei et al., 2007; Wu and Baldwin, 2010), increases in reactive oxygen and nitrogen species (Liu et al., 2010; Wu and Baldwin, 2010), and, finally, an intricate interplay of phytohormone signaling involving jasmonic acid (JA; Halitschke and Baldwin, 2003; Howe and Jander, 2008), ethylene (von Dahl et al., 2007), and salicylic acid (SA; Rayapuram and Baldwin, 2007) signaling networks. These complex signaling responses may, in turn, be modulated by small RNAs (smRNAs; Pandey et al., 2008b; Rasmann et al., 2012). However, our understanding of the herbivore-induced smRNA machinery remains incomplete, and the Argonautes (AGOs) involved in this pathway remain unidentified.

In plants, several classes of smRNAs exist (Axtell, 2013). Broadly, smRNAs modulate gene expression either posttranscriptionally (via mRNA degradation or translational blockage) or transcriptionally (such as by heterochromatin formation, paramutation, and other epigenetic mechanisms; Jones-Rhoades et al., 2006; Hutvagner and Simard, 2008; Axtell, 2013; Ma et al., 2013; Rogers and Chen, 2013). Three important components of the pathways for the biogenesis and action of smRNAs in plants are the RNA-directed RNA polymerases (RdRs), the Dicer-like proteins (DCLs), and the AGOs (Brodersen and Voinnet, 2006; Axtell, 2013; Borges and Martienssen, 2015). RdRs use RNA templates to generate double-stranded smRNA precursors by synthesizing the second RNA strand. DCLs have endonuclease activity and process the double-stranded smRNA precursors (or stem-loop single RNAs) into 20- to 24-nucleotide duplexes with two-nucleotide overhangs. These are loaded onto the AGOs that retain one of the strands to form the RNA-induced silencing complex (RISC), thus determining the specificity of target selection based on smRNA-mRNA complementarities. In this way, AGOs are regarded as the effectors of the regulatory smRNA machinery (Axtell, 2013). All three central components of the machinery are coded in multigene families in plants; for instance, in Arabidopsis (Arabidopsis thaliana), six RdRs, four DCLs, and 10 AGOs have been identified (Höck and Meister, 2008; Mallory and Vaucheret, 2010).

Under different physiological conditions, specific members of these three components of the smRNA machinery are recruited to produce a particular functional class of smRNAs. For instance, during the regulation of developmental events of plants, specific microRNAs (miRNAs) are diced by DCL1 and DCL4 and loaded onto AGO1 or AGO10 to form the RISC that mediates the expression of developmentally important genes (Borges and Martienssen, 2015). RdR6 (along with RdR1) are implicated in producing and amplifying exogenous virus-induced small interfering RNAs (siRNAs; Axtell, 2013). During virus defense, the siRNA precursors are diced by DCL2, DCL3, or DCL4 and loaded onto AGO1 or AGO2 to form the RISC, which mediates resistance (Mi et al., 2008; Takeda et al., 2008). The concerted activity of RdR2 and DCL3 help to generate the 24-nucleotide heterochromatin-derived siRNAs that direct the RNA-dependent DNA methylation pathways (Vrbsky et al., 2010). Heterochromatin-derived siRNAs are loaded onto AGO4 and AGO6 to execute transcriptional silencing (Havecker et al., 2010), while AGO1 and AGO7 participate in the trans-acting siRNA-mediated silencing (Baumberger and Baulcombe, 2005; Qi et al., 2005; Brodersen et al., 2008; Montgomery et al., 2008; Mallory et al., 2009). DCL2 participates in the production of 22-nucleotide virus- and transgene-derived siRNAs and 24-nucleotide siRNAs derived from the natural antisense transcripts that help to mediate tolerance to salt stress in Arabidopsis (Xie et al., 2004; Borsani et al., 2005; Bouché et al., 2006).

Nicotiana attenuata is an ecological model plant, native of the southwestern United States, in which the role of smRNAs in herbivore defense is best studied, in part because the mechanisms of herbivore resistance are well known in this plant. N. attenuata’s genome harbors three functionally distinct RdRs (Pandey and Baldwin, 2007, 2008; Pandey et al., 2008a), four DCLs (Bozorov et al., 2012), and 11 AGOs (Singh et al., 2015). Silencing only RdR1 makes N. attenuata plants highly susceptible to herbivory (Pandey and Baldwin, 2007). Furthermore, sequential loss-of-function analysis of the NaDCLs suggested that DCL3 and DCL4 contribute to this plant’s herbivore resistance (Bozorov et al., 2012). Which of the 11 NaAGOs play a role in the herbivore-induced smRNA pathway remained elusive, and that was the subject of this study.

RESULTS

Characterization and Domain Organization of AGOs in N. attenuata

N. attenuata contains 11 unique AGO homologs (Singh et al., 2015). A neighbor-joining phylogenetic tree (Fig. 1) was reconstructed after multiple sequence alignment (MSA) of all 11 NaAGOs (Supplemental Fig. S1). NaAGO2 and NaAGO7 clustered together; similarly, NaAGO4a, NaAGO4b, NaAGO8, and NaAGO9 formed a cluster; while NaAGO1a, NaAGO1b, NaAGO1c, NaAGO5, and NaAGO10 formed independent branches on the tree (Fig. 1). Overall, the topology of the NaAGO phylogenetic tree was similar to that of the AtAGOs (Supplemental Fig. S1).

Figure 1.

Characterization of AGOs of N. attenuata. A, Neighbor-joining phylogenetic tree of N. attenuata AGOs. The MSA of the full-length peptide sequences of NaAGOs was performed using ClustalX 2.1, and phylogeny was created using MEGA 5.2. Clade robustness was assessed with 1,000 bootstrap replicates, and the values are shown on the branches. B, Schematic representation of domain architectures in 11 NaAGOs. Different regions were annotated with the help of the conserved domain search tool and compared with the domains of KpAGO and HsAGO2. The MID and the PIWI domains are fused in all the NaAGOs. C, Diversity in signature residues of the PAZ, the MID, and the PIWI domains in NaAGOs. The height of a residue denotes the probability of occurrence of a residue at the given position in MSA (Supplemental Fig. S1). The positions marked with stars are the previously recognized signature residues. The values on the x axis show the positions of the residues in the MSA of NaAGOs (Supplemental Fig. S1). D, Signature residues in NaAGO8 at the respective positions of C (Supplemental Fig. S1). Specific signature residues marked with red arrows may contribute to the interaction of AGO8 with the substrate.

In order to further characterize these 11 homologs, we performed a comparative analysis of the sequences of the NaAGOs, the Homo sapiens AGO2 (HsAGO2; Protein Data Bank [PDB] code 4F3T; Elkayam et al., 2012), and the Kluyveromyces polysporus AGO (KpAGO; PDB code 4F1N [model for yeast AGO]; Nakanishi et al., 2012). HsAGO2 and KpAGO are the two best biochemically characterized eukaryotic AGOs, as their structures have been determined. We identified the characteristic N, PAZ, MID, and PIWI domains in the NaAGOs (Fig. 1). Furthermore, we identified an additional domain of unidentified function between the L1 (linker 1) and PAZ domains (Fig. 1). The PAZ, MID, and PIWI domains contain R/K/V-F-Y, Y-N-K-K, and D-E-D-H/D signature residues, respectively (Hutvagner and Simard, 2008; Nakanishi et al., 2012; Singh et al., 2015). These signature residues were readily identified in the respective domains of the NaAGOs (Fig. 1; Supplemental Fig. S1). Variation in the R-F-Y signature of the PAZ domains was evident; R (position 600; Fig. 1) had been substituted with K in NaAGO4a, NaAGO8, and NaAGO9 and with V in NaAGO2. NaAGO2 displayed the D-D-D variant in the PIWI signature. Additionally, we discovered five, eight, and 11 functionally important sites in the PAZ, MID, and PIWI domains (Elkayam et al., 2012), respectively, that were variable among NaAGOs. Such variations at important positions in PAZ, MID, and PIWI domains (Fig. 1; Supplemental Fig. S1) may have functional consequences related to the mechanism by which AGOs interact with RNA substrates (Singh and Pandey, 2015).

Changes in Transcript Abundance of NaAGOs in Response to Herbivory

We quantified transcript abundances of the NaAGOs that are elicited in leaves or roots during the attack of Manduca sexta larvae (Fig. 2; Supplemental Fig. S2). Immediate application of distilled water-diluted M. sexta oral secretions (W+OS) to puncture wounds in leaves recapitulates the responses elicited during attack from this insect (Halitschke et al., 2001). Although no significant differences in transcript profiles in roots were noticed (Supplemental Fig. S2), in leaves 18 h after W+OS treatment (Fig. 2), the accumulation of NaAGO8 transcripts increased 3-fold over the wounding controls (similarly wounded plants in which the puncture wounds were treated with distilled water). Such an increase in transcript levels suggested that AGO8 contributes to the smRNA-mediated regulation of defenses. Additionally, we noticed an increase of nearly 2-fold in NaAGO5 transcripts 1 h after oral secretions (OS) elicitation. A second 6-fold peak in the accumulation of NaAGO5 transcripts over untreated and wounding controls also was observed 18 h after OS elicitation (Fig. 2). No significant effects were observed in the abundance of transcripts of the other AGOs in response to the W+OS treatment (Fig. 2). This analysis was duplicated, and similar results were obtained. W+OS-induced differences in AGO5 and AGO8 transcripts at the 18-h time point were validated in independent experiments with at least three (independent) biological replicates. Compared with control plants, a significant increase of 3.5-fold or greater in AGO8 transcripts (Student’s t test, t = 5.937, P < 0.05) and a 5-fold or greater increase in AGO5 transcripts (Student’s t test, t = 8.261, P < 0.01) were recorded. This indicates that AGO5 and AGO8 may be involved in modulating the induced defense responses of N. attenuata against herbivore attack. It is plausible that AGO5 may modulate certain defense-signaling events independent of AGO8 or that AGO5 and AGO8 also may have overlapping targets.

Figure 2.

Abundance of NaAGO transcripts in the leaves during time-course experiments. Rosette leaves of 3- to 4-week-old wild-type N. attenuata plants were wounded with pattern wheels, and 20 μL of water (W+W) or M. sexta OS (W+OS; diluted 5-fold in water) were immediately applied to these wounds. Leaf samples from control and treated plants were collected at the time intervals depicted on the x axis. SYBR Green assays were performed, and the ΔΔCT method was used to determine relative transcript accumulations. Dark, When lights were off; Cut-off range, relative abundance of transcript, above or below which was considered up- or down-regulated, respectively.

Characterization of Transgenic NaAGO Lines

The inverted-repeat stable transformants (irAGO) of NaAGO1 (a, b, and c), NaAGO2, NaAGO4 (a and b), NaAGO5, NaAGO7, NaAGO8, and NaAGO10 with complete T-DNA insertions were screened for homozygosity in the T2 generation, as detailed earlier (possibilities of incomplete T-DNA insertions were ruled out as described by Gase et al. [2011]). After testing the transformant’s seedling survival rate on hygromycin plates and conducting diagnostic PCRs, we identified two independently transformed lines of NaAGO2, NaAGO4, NaAGO5, NaAGO7, NaAGO8, and NaAGO10 and one for NaAGO1. These were subjected to DNA gel-blot analysis, which revealed single insertions in both the independent lines of NaAGO2, NaAGO4, NaAGO7, NaAGO8, and NaAGO10 (Fig. 3; Supplemental Fig. S3). For NaAGO1 (866-1-5) and NaAGO5 (909-6-2), the analysis revealed that the lines were heterozygous, harboring double insertions (Fig. 3). Upon evaluating the silencing efficiency of the targeted genes in all the irAGO lines, a minimum decrease of 50% transcript levels as compared with the wild type was observed for irAGO4a. Stronger reductions were recorded for all the other genotypes, including a nearly complete silencing of AGO2, AGO5, and AGO10 (Fig. 3). The irAGOs with two independent homozygous single insertion lines with significantly reduced transcript levels as compared with the wild type (for NaAGO2, NaAGO4, NaAGO7, NaAGO8, and NaAGO10) were selected for further experiments.

Figure 3.

Characterization of irAGO lines. A, Southern-blot analysis of transgenic genotypes to determine lines harboring a single insertion; genomic DNA was digested with XbaI (left) and EcoRV (right). B, Evaluation of the silencing efficiency of irAGO lines. Wild type (WT) and irAGO genotypes were W+OS elicited, and samples were collected after 18 h. Transcript levels, quantified by qPCR, of AGOs were normalized to wild-type levels, which were set to 1. ND, Transcripts not detected. Significant differences from the wild type are indicated with asterisks (ANOVA [Fisher’s lsd]: *, P ≤ 0.05; **, P ≤ 0.01; and ***, P ≤ 0.001).

Silencing NaAGO8 Improves the Performance of the Specialist Herbivore M. sexta

To identify if any of the AGOs are involved in modulating the defense response of N. attenuata plants against M. sexta, caterpillar performance assays were conducted with all irAGO lines and the wild type. We observed no significant differences in the larval biomass from larvae feeding on lines silenced in NaAGO1 (a, b, and c), NaAGO2, NaAGO4 (a and b), NaAGO7, and NaAGO10 expression compared with those feeding on wild-type plants (Fig. 4). Due to the absence of two independent transformants with single T-DNA insertions for irAGO5, irAGO5 plants were not included in the caterpillar assays. In contrast, when fed on the two independent homozygous lines of irAGO8 (A-12-871-8-8 and A-12-873-5-2, which did not show any developmental or growth abnormalities compared with wild-type plants [Supplemental Fig. S4]; silencing AGO8 had no effect on the transcript accumulations of any other AGOs [Supplemental Fig. S5]), caterpillars were significantly larger at the 6-, 9-, and 11-d measures. Caterpillar mass was 63% larger on irAGO8 lines in comparison with those on wild-type plants after 11 d of feeding (Fig. 4; repeated-measures ANOVA, P ≤ 0.001). These results were verified in an independent caterpillar bioassay (Supplemental Fig. S6). In order to further evaluate off-target effects other than on the other NaAGOs in the N. attenuata genome, we used the AGO8 inverted-repeat sequence present in the pRESC8AGO8 construct, both with the SGN VIGS tool (Fernandez-Pozo et al., 2015) and a BLAST search against the complete transcriptome of N. attenuata. We did not find any off-target effects related to herbivore resistance (Supplemental Fig. S5; detailed further in “Materials and Methods”). From these results, we conclude that silencing NaAGO8 expression specifically increases caterpillar performance, suggesting that this AGO plays a key regulatory role in resistance to insect attack.

Figure 4.

Silencing of AGO8 makes N. attenuata susceptible to M. sexta larval attack. Caterpillars (one neonate per plant) were allowed to feed continuously; their mass was recorded at 3, 6, 9, and 12 d of feeding. Significant differences from the wild type (WT) are given (repeated-measures ANOVA: for irAGO8 line 873-5-2, F = 16.71, P ≤ 0.001; and for line 871-8-8, F = 23.57, P ≤ 0.001). ***, Significant difference at P ≤ 0.001.

Silencing AGO8 Reprograms the N. attenuata smRNA Population

We determined how AGO8 silencing influenced the smRNA profiles of N. attenuata before and after simulated herbivory (OS elicitation). We deep sequenced the smRNome (in the range of 15–30 nucleotides) of wild-type and irAGO8 plants before (0 h) as well as 0.75 and 18 h after OS elicitation. An average of 18.1 ± 1.6 million reads were generated in each of the samples; after quality-control measures and removing the structural RNAs and mRNAs (Pandey et al., 2008b), we obtained, on average, 4.8 ± 0.9 million unique smRNA reads in each sample in the range of 15 to 30 nucleotides (Fig. 5A). We determined the composition of miRNAs influenced by AGO8 silencing in this pool using the miRDeep pipeline (Friedländer et al., 2012). We focused our analysis on miRNAs that are conserved across plant species, as these miRNAs may have profound roles for a plant’s physiology. A total of 135 miRNA reads corresponding to 125 unique miRNAs (Fig. 5A; Supplemental Table S1), conserved in 32 plant species, were annotated using the miRDeep pipeline (Fig. 5G; Supplemental Table S2). Secondary hairpin structures of 166 putative novel miRNA candidates were produced and evaluated by miRDeep. Overall, the length distributions of small RNA reads indicate that the libraries are highly enriched in miRNAs in the 21- to 24-nucleotide lengths (Supplemental Table S1).

Figure 5.

Herbivore-induced irAGO8-dependent smRNA profiling of N. attenuata. A, Pipeline used for N. attenuata miRNA identification. Counts for reads, unique sequences, genomic locations, and precursor miRNA detected are given at each step. Details on filtering steps are provided in “Materials and Methods.” B to E, Pie charts represent the distribution of precursor miRNA genomic loci on the N. attenuata genome for the conserved (B) and predicted (D) miRNAs; C and E show the number of mature and star sequences identified for conserved and predicted miRNAs, respectively. CDS, Coding sequence; UTR, untranslated region. F, Histogram shows the size distribution of the precursor miRNAs for the conserved (black bars) and novel (gray bars) miRNAs. The number of miRNAs present in each size class is shown. G, Distribution of mature miRNAs in different species. Mature sequences were found widespread in 32 plant species (analysis performed on miRNAs from 72 plant species obtained from miRBase). Bars represent the number of miRNAs mapping to each species.

Results of mapping the clean reads to different publicly available databases produced 301 mature and 551 precursor miRNA reads (Fig. 5A): mapping results showed 135 known reads (corresponding to125 miRNAs) and 166 novel miRNA reads (Supplemental Table S1). Complete lists and the main characteristics of miRNAs and their precursors, length, coordinates, and annotation in the genome, are provided in Supplemental Table S1. Also, details for miRNAs found in intergenic and intragenic regions are given in Supplemental Table S1. As expected for miRNA genes (conserved in Fig. 5, B and C, and novel miRNA reads in Fig. 5, D and E), most were found in intergenic or intronic regions (conserved, 119 reads [88%]; novel, 134 reads [81%]). However, two conserved and two novel reads mapped to untranslated regions, whereas two conserved and five novel reads mapped to the predicted coding sequence. Furthermore, 12 conserved and 25 novel reads mapped to exonic regions. Precursor miRNA length varied from 45 to 111 nucleotides, with an average of 71 nucleotides; and 50% of conserved and 46% of novel precursor miRNA reads comprised between 71 and 90 nucleotides (Fig. 5F).

Expression Profiles of Conserved and Novel miRNAs

Mature miRNAs from a total of 32 conserved families were detected (Fig. 5G; their expression counts are tabulated in Supplemental Tables S1 and S2). Silencing AGO8 profoundly influenced the composition of the miRNAs of N. attenuata in a complex manner. Of the 125 conserved miRNAs, 71, 74, and 63 miRNAs were detected in irAGO8, whereas 68, 72, and 77 miRNAs were captured in the wild type (presence of AGO8), at 0, 0.75, and 18 h after OS elicitation. Similarly, 82, 90, and 64 novel miRNAs were expressed in irAGO8, whereas 75, 67, and 77 novel miRNAs were captured in the wild type, at 0, 0.75, and 18 h after OS elicitation. Wild-type and irAGO8 plants shared nearly 57% of conserved and 44% of novel miRNAs, of which a large number were differentially accumulated in irAGO8 compared with the wild type (Fig. 6; Supplemental Table S3). The expression level of conserved miRNA families can be classified into three categories, namely as high (reads per million [RPM] > 10,000; e.g. miR167-5p, miR165a-3p, miR156b-5p, miR168a-5p, miR6149a, and miR6021), moderate (RPM = 100–10,000; e.g. miR156a, miR172a-3p, miR1846e, miR403a-3p, miR5386, miR5789, miR6020b, miR7122a, miR8015-3p, and miR8764), and low (RPM < 100; e.g. miR1066, miR1169-3p, miR160-5p, miR394, miR477a-3p, miR6218-5p, miR7504e, and miR9496).

Figure 6.

Transcript abundance patterns for conserved and novel miRNAs. Hierarchical clustering of heat maps was performed on the miRNA abundance values using the Pearson correlation method. Higher abundance is represented by dark color, whereas dark blue shows very low abundance. Line graphs are used to highlight changes in abundance for a few miRNAs (explained in the text). Black and green line graphs show patterns for miRNAs present in irAGO8 and the wild type (WT), respectively. A, Heat map representing the abundance pattern for 125 conserved miRNAs. Fifty-five miRNAs are differentially abundant in both genotypes across all three time points (0, 0.75, and 18 h). On the other hand, 56 miRNAs (31 in the wild type and 25 in irAGO8) were captured only at one or two time points in one of the two genotypes. For example, miRNAs 5062b-5p and 5198 were captured only in the wild type and were repressed at 0.75 h compared with 0 and 18 h. miR159b was accumulated only at 0.75 h in the wild type, whereas nine miRNAs were captured only in irAGO8 at 0.75 h. B, Heat map for 166 predicted novel miRNAs. Ninety novel miRNAs (33 in the wild type and 57 in irAGO8) were accumulated at one or two time points in either the wild type or irAGO8. At 0.75 h, 10 miRNAs accumulated only in the wild type and 22 only in irAGO8.

Identification of irAGO8-Specific Conserved and Novel miRNAs

miRNA profiles of wild-type and irAGO8 plants, in triplicate, at 0-, 0.75-, and 18-h time points were compared. Thirty-one conserved miRNAs are expressed only after silencing of AGO8 (Fig. 6A, black line graphs). Of these, miR8703a, miR6484, miR6226-3p, miR5497, miR4358, miR319-3p, miR1066, and miR169j-3p were expressed only at 0 h (before OS elicitation); miR7808, miR7756-3p, miR6295, miR6147, miR5069, miR482c-5p, miR395, miR170-3p, and miR3701 were expressed only at 0.75 h after OS elicitation; and miR9496, miR8769, miR7486h, miR5049-3p, miR5780c, and miR6206 were expressed only at 18 h; whereas miR2596, miR394, miR8129-3p, miR6020b, miR6158a, miR1536, and miR5027 were expressed at both 0 and 0.75 h, and miR6218-5p was expressed at all the three time points only in the irAGO8 samples.

Twenty-five conserved miRNAs are expressed only when AGO8 is present (and, thus, may be regarded as specific to AGO8; Fig. 6A, green line graphs). miRNAs such as miR5781, miR1846e, and miR7725b-3p.1 (and others) are expressed only at 0 h; miRNAs such as miR7980a, miR5163a-3p, and miR159b are expressed only at 0.75 h; whereas miRNAs like miR6483, miR5820, and miR6444 are expressed at 18 h specifically.

Fifty-seven novel miRNAs are expressed only in irAGO8: Nat-AGO8-PN119, Nat-AGO8-PN112, and Nat-AGO8-PN26 are expressed only at 0 h; Nat-AGO8-PN72, Nat-AGO8-PN58, and Nat-AGO8-PN39 are expressed at 0.75 h; and miRNAs such as Nat-AGO8-PN96, Nat-AGO8-PN78, and Nat-AGO8-PN20 are expressed at 18 h (Fig. 6B, black line graphs). Thirty-three novel miRNAs are not captured in the absence of AGO8 (and, thus, are considered dependent on AGO8). For example, AGO8-dependent novel miRNAs such as Nat-AGO8-PN154, Nat-AGO8-PN115, and Nat-AGO8-PN32 are expressed only at 0 h; Nat-AGO8-PN117, Nat-AGO8-PN68, and Nat-AGO8-PN50 are expressed only at 0.75 h; and Nat-AGO8-PN101, Nat-AGO8-PN76, and Nat-AGO8-PN17 are expressed only at 18 h (Fig. 6B, green line graphs).

Differential Expression of miRNAs

Complex patterns of differential accumulation of miRNAs (commonly expressed in the wild type and irAGO8) were observed across the three time points, 0, 0.75, and 18 h, due to silencing of AGO8. Information on these patterns of differentially expressed miRNAs (DEMs) in both genotypes at the three time points is summarized in Table I and Supplemental Table S4. Most of the miRNAs have higher expression values in the presence of AGO8 at 0 h. Twenty-one miRNAs accumulated more in the wild type compared with irAGO8 (Table I). Only 29 miRNAs were differentially expressed within irAGO8 before and after OS elicitation, whereas a large number of miRNAs (66 DEMs) were reprogrammed in the wild type after OS elicitation as compared with the 0-h control (Supplemental Fig. S7). Fifty-one and 32 miRNAs were significantly down-regulated at 0.75 h (compared with 0 and 18 h, respectively) in the wild type, whereas in irAGO8, only 11 and 16 miRNAs were repressed (Table I). For example, miRNAs 162, 1163.2, 4414b, 7504c, and 5789 were down-regulated by more than 1.5-fold in the wild type, whereas only miR4414b was down-regulated in irAGO8 at 0.75 h (as compared with 0 h; Supplemental Table S4). We found that miR408-5p and miR8688 were significantly down- and up-regulated, respectively, in the absence of AGO8 (Supplemental Table S4). miRNAs 72a-3p, 4379, 8015-3p, 8165, and 6218-5p were up-regulated in irAGO8 at 0.75 and 18 h when compared with the wild type. miR1846e was significantly highly expressed at the baseline (0 h) in the presence of AGO8, whereas very few reads were obtained in irAGO8 samples, indicating that its expression might be dependent on AGO8. These results suggest that (1) AGO8 is needed for the modulation of miRNA levels, as shown by the larger number of DEMs in the wild type compared with irAGO8, and (2) 0.75 h showed the maximum change in AGO8-dependent miRNA expression (Table I).

Table I. Differentially expressed miRNAs at P < 0.05 in six comparisons.

P values were computed using χ². C, Conserved; N, novel.

Comparison	Total miRNA Expressed in irAGO8 and the Wild Type	DEMs	Up	Down
irAGO8 and the wild type at 0 h	95	28 (C = 20; N = 8)	7 (C = 6; N = 1)	21 (C = 14; N = 7)
irAGO8 and the wild type at 0.75 h	89	37 (C = 21; N = 16)	29 (C = 15; N = 14)	8 (C = 6; N = 2)
irAGO8 and the wild type at 18 h	104	18 (C = 12; N = 6)	11 (C = 8; N = 3)	7 (C = 4; N = 3)
irAGO8 at 0.75 and 0 h	115	20 (C = 14; N = 6)	9 (C = 8; N = 1)	11 (C = 6; N = 5)
irAGO8 at 18 and 0 h	93	18 (C = 14; N = 4)	16 (C = 12; N = 4)	2 (C = 2; N = 0)
irAGO8 at 18 and 0.75 h	87	22 (C = 18; N = 4)	16 (C = 12; N = 4)	6 (C = 6; N = 0)
The wild type at 0.75 and 0 h	105	57 (C = 32; N = 25)	6 (C = 6; N = 0)	51 (C = 26; N = 25)
The wild type at 18 and 0 h	114	35 (C = 20; N = 15)	10 (C = 6; N = 4)	25 (C = 14; N = 11)
The wild type at 18 and 0.75 h	105	38 (C = 20; N = 18)	32 (C = 14; N = 18)	6 (C = 6; N = 0)

NaAGO8 Modulates the Accumulation of Key Defense Metabolites

After determining the impact of AGO8 silencing on miRNA accumulation during herbivory, we determined which physiological processes were impacted by such profound changes. Moreover, specific susceptibility of irAGO8 plants to herbivore attack indicated that the complex reprogramming in miRNA accumulation during the course of OS elicitation may lead to compromised inducible defenses. We used an unbiased metabolomics approach to help us identify the changes in metabolites in caterpillar-fed irAGO8 plants as compared with the wild type that might be responsible for the changes in resistance (Wu and Baldwin, 2010; Gaquerel et al., 2014). The leaf extracts from 12-d caterpillar-fed wild-type and irAGO8 plants were analyzed using liquid chromatography-electrospray ionization-time of flight-mass spectrometry (n = 15 for the wild type and n = 10 for irAGO8). When comparing the peak matrices of the two genotypes, we found mass spectra with reduced peak areas in irAGO8 plants that corresponded to several defense-related metabolites, such as nicotine, phenolamides, and HGL-DTG (Fig. 7). The important molecular features identified from loading plots were annotated with elemental formulas and summarized (Fig. 7; Supplemental Table S5). Projection (VIP) scores based on partial least squares-discriminate analysis confirmed the reduced levels of these defense metabolites in irAGO8 compared with their levels in the wild type (Fig. 7; P < 0.05). No primary metabolites were detected among the key molecular features of the VIP plot, and no significant differences in acyl sugars were observed (Supplemental Table S5). Highly significant silencing of AGO8 transcripts was observed under both basal (n = 4, ANOVA, F = 28.02, P < 0.05) as well as induced (n = 3, ANOVA, P < 0.001; Fig. 3) states. Yet, the growth and morphology of the irAGO8 plants were fully comparable to those of wild-type plants (Supplemental Fig. S4); no differences in rosette diameter (n = 5, ANOVA, Fisher’s lsd, F = 2.12, P > 0.05), number of rosette leaves (n = 5, ANOVA, Fisher’s lsd, F = 0.22, P > 0.05), plant height (n = 5, ANOVA, Fisher’s lsd, F = 2.58, P > 0.05), number of flowers (n = 5, ANOVA, Fisher’s lsd, F = 0.222, P > 0.05), or chlorophyll content (n = 5, ANOVA, Fisher’s lsd, F = 0.3686, P > 0.05) of irAGO8 and wild-type plants were observed under unstressed conditions (Supplemental Fig. S4). From these results, we infer that silencing AGO8 did not affect the primary metabolism of N. attenuata plants.

Figure 7.

Profiling of metabolomes of M. sexta caterpillar-attacked wild-type (WT) and irAGO8 leaves. A, Untargeted metabolite profile (from A–C) showing representative tandem mass spectrometry ion chromatograms of wild-type and irAGO8 (871-8-8) genotypes exhibiting alterations in major defense metabolites (nicotine, phenolamides, and HGL-DTGs) in response to herbivory. B, Loading plot for the selected principal components of wild-type and irAGO8 genotypes, discriminating the major herbivore-responsive defense metabolites, nicotine and HGL-DTGs. C, Key molecular features identified by partial least squares-discriminant analysis among the genotypes. The colored boxes on the right indicate the relative concentrations of the corresponding metabolite in each genotype. D, Targeted metabolite analysis. Silencing of the expression of NaAGO8 significantly reduces the accumulation of nicotine, DTGs, and rutin in response to herbivore attack. Compromised levels of nicotine, HGL-DTGs, and rutin were validated in herbivore-fed leaves of irAGO8 plants, as compared with levels in the wild type. Values presented are means ± se. All pairwise multiple comparison procedure values are as follows: nicotine (ANOVA, F = 24.72, P ≤ 0.001), DTGs (ANOVA, F = 8.921, P ≤ 0.001), and rutin (ANOVA, F = 4.074, P ≤ 0.05). Asterisks indicate significant differences as follows: *, P ≤ 0.05; **, P ≤ 0.01; and ***, P ≤ 0.001. FW, Fresh weight.

We performed targeted metabolite analysis of caterpillar-fed wild-type and irAGO8 leaf samples to reexamine the accumulation patterns of nicotine, rutin, and HGL-DTGs (Fig. 7; Supplemental Fig. S8). The nicotine levels were significantly lower in both irAGO8 lines (Fig. 7; ANOVA, P ≤ 0.01). Similarly, HGL-DTG levels were reduced significantly in both the irAGO8 lines (Fig. 7; ANOVA, P ≤ 0.05 [871-8] and P ≤ 0.001 [873-5]). As previous work found that insect feeding increases rutin levels in N. attenuata (Yang et al., 2013), we examined the concentrations of these metabolites with a targeted analysis, even though rutin had not been found in the unbiased analysis. Rutin levels also were significantly lower in the leaves of irAGO8 plants that had been attacked by caterpillars as compared with wild-type fed leaves (ANOVA, P ≤ 0.05; Fig. 7).

NaAGO8 Modulates MYB8-Mediated Defense Responses

Metabolomic analyses revealed that silencing AGO8 reduced the concentrations of key defense metabolites such as nicotine, phenolamides, and DTGs. MYB8 is a transcription factor in N. attenuata known to regulate the induction of nicotine, phenylpropanoid, and phenolamide pathways (Kaur et al., 2010). We conducted time-course experiments with OS-elicited wild-type and irAGO8 plants in the fully expanded rosette stage of growth. The transcript levels of MYB8 were strongly reduced in irAGO8 (Fig. 8). As NaMYB8 regulates the expression of genes for phenolamide metabolism, we investigated how the transcription of this MYB8 network of genes was affected: transcripts of AT1, DH29, and CV86, which are elicited 6 h after the wild-type plants are OS elicited (Onkokesung et al., 2012), were reduced in irAGO8 (Fig. 8). As phenylpropanoid metabolism is regulated by MYB8 during herbivory (Onkokesung et al., 2012), we also examined the accumulation of transcripts of key genes in the pathway: PAL1, PAL2, and C4H; lower levels of transcripts of these genes were recorded in the irAGO8 lines (Fig. 8). This implies that AGO8 has a role in regulating the expression of MYB8 and its associated genes in phenolamide and phenylpropanoid metabolism after herbivore attack.

Figure 8.

Silencing AGO8 in N. attenuata perturbs the regulation of genes of the MYB8-dependent defense network and nicotine pathway in response to herbivore attack. Rosette leaves of the N. attenuata wild type (WT) and irAGO8 (871-8-8) were subjected to OS elicitation. Samples were collected at time intervals of 0, 0.5, 0.75, 1, 6, 18, and 24 h. Transcript abundance was determined for MYB8 and key genes of phenolamide metabolism (AT1, DH29, and CV86), the phenylpropanoid pathway (PAL1, PAL2, and C4H), HGL-DTG biosynthesis (GGPPS), FPPS, and the nicotine biosynthesis pathway (ODC and PMT). PMT levels were evaluated in wild-type and irAGO8 genotypes 24 h after OS elicitation. Values are means ± pooled variances. Transcript levels in irAGO8 (873-5-2) were significantly different from those in the wild type in the time-course experiment (two-way repeated-measures ANOVA, Fisher’s lsd: *, P ≤ 0.05; **, P ≤ 0.01; and ***, P ≤ 0.001). Significant differences in PMT transcripts were evaluated with ANOVA: F = 61.51, P ≤ 0.05 (*).

In congruence with the metabolomic profiles, the transcripts of the key genes involved in the biosynthesis of HGL-DTGs (NaGGPPS) and nicotine (NaODC and NaPMT) were clearly less abundant in the irAGO8 lines after simulated herbivory (Fig. 8). These results are consistent with the hypothesis that AGO8 has a regulatory role in the biosynthesis of defense-responsive metabolites during herbivory.

Silencing AGO8 Partially Affects Phytohormone Signaling

JA signaling plays a central role in activating robust defenses (that are described above) in N. attenuata against herbivore attack (Halitschke and Baldwin, 2003). Therefore, we analyzed JA and JA-Ile levels in the leaves of 12-d caterpillar-fed wild-type and irAGO8 plants (Supplemental Fig. S8), which indicated that herbivory-induced JA levels might be lower in irAGO8 lines. OS elicitation provides a means of rigorously quantifying JA dynamics in a manner that is not obfuscated by variations in caterpillar feeding behavior; therefore, we examined JA levels in W+OS-treated wild-type and irAGO8 plants. JA levels were partly, yet significantly, lower in irAGO8 in comparison with the wild type 30 min after OS elicitation (38.83%; ANOVA, P ≤ 0.05; Supplemental Fig. S8; no differences in JA-Ile levels were observed). JA-Ile levels are typically one-tenth of the JA levels, and the amounts of JA present in irAGO8 plants may be sufficient to supply the JA-Ile burst, as the genes responsible for such conversion may not be under AGO8 control. WRKY3 and WRKY6 are the major transcription factors in N. attenuata that regulate JA signaling (Skibbe et al., 2008). Consistent with a significant reduction in JA levels, at early time points (45 and 60 min), the transcript levels of the WRKY3 and WRKY6 transcription factors and many JA biosynthetic pathway genes (of those tested, MYC2, LOX3, AOS, AOC, and OPR3) were reduced in OS-elicited irAGO8 lines (Supplemental Fig. S10). The modulatory action of the smRNA pathway on defense signaling is complex and not fully understood. Loss of such modulation in cells may result in the misregulation of gene expression networks, which could result in incoherent patterns of accumulation of mRNAs of genes of the JA signaling pathway.

In addition, abscisic acid (ABA) levels, which also are increased strongly by OS elicitation and herbivore attack in the wild type (Dinh et al., 2013), were reduced in irAGO8 lines, while SA and JA-Ile levels remained unchanged (Supplemental Fig. S9). Genes involved in ethylene production were either unchanged (ACS3) or reduced (ACO3) in irAGO8 lines as compared with the wild type after OS elicitation (Supplemental Fig. S10).

Taken together, these results are consistent with the hypothesis that AGO8 has a regulatory role in inducing the biosynthesis of defense-responsive metabolites during herbivory. In a complex manner, AGO8 appears to modulate several regulatory nodes in the pathway eliciting direct inducible defenses in N. attenuata, for example, by moderately regulating WRKY-mediated responses upstream of JA, by partially regulating phytohormone signaling, and by modulating the elicitation of MYB8-dependent gene networks downstream for secondary metabolism (Supplemental Fig. S11).

Structural Insights into Plausible Interactions of NaAGO8 with smRNA Substrates

Variability in residues at important sites (Fig. 1) may affect the structural conformations of NaAGOs during their interaction with a variety of smRNA substrates, which, in turn, may have implications for their functional specificities. To gain mechanistic insights into the functional diversity of the AGOs, we used a comparative structural analysis approach. By modeling the interactions of each NaAGO with two bona fide smRNA substrates (see “Materials and Methods”), we found significant differences (greater than 3 Å) among the three-dimensional structures of the substrate-bound NaAGOs (Fig. 9A, i). The structural conformation of NaAGO8 was significantly different (greater than 3 Å) than those of NaAGO4a, NaAGO7, and NaAGO10. We also found significant differences (3.97 Å) in the overall structure of NaAGO8 and AtAGO8. NaAGOs also showed diversity in surface area and volume of the largest binding pocket as well as in the properties of the largest positive electrostatic patches on the surface of the protein (Fig. 9A, ii). NaAGO8 possesses the largest area and volume of the binding pocket, whereas it contains the fewest residues in the predicted largest positive electrostatic patch (Fig. 9A, ii). Also, the distance between the nucleotides at the 5′ end of the RNA substrate and the interacting residues in the MID-PIWI lobe varied across the AGOs (Fig. 9B). For instance, the distance between the residue in the PIWI domain corresponding to MSA position 1,185 (Fig. 1C) and the first nucleotide at the 5′ end of the smRNA with which it interacts is 19.4 Å in NaAGO8, while it is 13.1 and 14.9 Å in NaAGO7 and NaAGO10, respectively (Fig. 9B). The physicochemical properties of the residue corresponding to MSA position 1,130 (Fig. 1C), which may interact directly with the second nucleotide at the 5′ end of the smRNA substrate, are different between NaAGO7 (Leu-894), NaAGO8 (Ile-803), and NaAGO10 (Gln-885; Fig. 9B). These findings suggest that the structural conformations of the smRNA-binding and catalytic domains of NaAGO8 differ from those of other NaAGOs and that these differences likely affect substrate and functional specificity. Differences in structural conformations of the smRNA-binding and catalytic domains of the AGOs may have arisen from varying evolutionary constraints on the important residues as well as due to their genetic correlations with residues in other positions (Singh et al., 2015). Revisiting our earlier study (Singh et al., 2015), we found that residue Glu-563 in the nucleotide specificity loop of the MID domain in NaAGO8 (corresponding to MSA position 870; Fig. 1C), which may interact directly with a nucleotide at the 5′ end of the smRNA (Fig. 9B, ii), is evolutionarily correlated with four other residues (Arg-564 and Ile-627 of the MID domain and Arg-850 and His-853 of the PIWI domain; Fig. 9C). We did not observe any such evolutionary correlations at corresponding positions in other NaAGOs.

Figure 9.

Structural analysis of the interaction of NaAGOs and the smRNAs provides mechanistic insights. A, Root mean square deviations (Å) between NaAGOs bound to smRNAs, 4F3T:R (top diagonal) and 4W5O:B (bottom diagonal), are shown in (i). Details of the largest binding pocket and the positive electrostatic patch on the surface among NaAGOs bound to smRNA, 4F3T:R, are shown in (ii). B, The modeled structures of NaAGO7 (i), NaAGO8 (ii), and NaAGO10 (iii) docked with smRNA (4F3T:R) show variation in the identities of interacting residues as well as the distance between 5′ nucleotide U1 and the interacting residues corresponding to MSA positions 872 and 877 of the MID domain and MSA position 1,185 of the PIWI domain (Fig. 1C). In NaAGO7, distances are 6.3, 8.3, and 13.1 Å; in NaAGO8, distances are 4.8, 7.4, and 19.4 Å; and in NaAGO10, distances are 5.5, 4.9, and 14.9 Å. Residues interacting with the second nucleotide (A2) of the smRNA substrate that differs in NaAGO7, NaAGO8, and NaAGO10 also are shown. C, The modeled structure of NaAGO8 shows residues coevolving only in NaAGO8.

Mapping of AGO8-Dependent miRNA-mRNA Interactions

Reprogramming of miRNAs as well as defense-related pathways indicated intensive interaction of AGO8-dependent miRNAs and the mRNAs of the disregualted genes. We generated miRNA-mRNA interaction networks by mapping of AGO8-dependent miRNA-binding sites into the mRNAs of the 18 differentially regulated genes of the defense-signaling pathway studied here. Twenty-six differentially changed miRNAs (between two genotypes at 0.75 and 18 h) showed binding to 18 genes (Supplemental Table S6). Of these 26 differentially expressed miRNAs, 16 may target differentially expressed genes at 0.75 h inversely. In other words, these miRNAs were up-regulated in irAGO8 at 0.75 h compared with the wild type, and their target genes were down-regulated in the same comparison, as shown in Figure 10A. miR1429-5p and miR6450b and their target genes, ODC and C4H, both were down-regulated in irAGO8 at 0.75 h. Similarly, we found that 16 differentially expressed miRNAs (not the same set) were up-regulated at 18 h in irAGO8 compared with the wild type and their target genes were down-regulated in the same comparison (Fig. 10B). miR6450b and its target gene, C4H, were down-regulated in irAGO8 at 18 h when compared with the wild type. For the majority of miRNAs, more than one gene has binding sites and was identified as a possible target. The miRNA-target interaction data set presented here (Fig. 10) gives a unique possibility to find the miRNA-mediated modulation of important biological processes, such as the MYB8-dependent defense network.

Figure 10.

AGO8-dependent inversely related miRNA and their target genes are presented as networks. Differentially accumulated miRNAs in irAGO8 compared with the wild type at 0.75 and 18 h showed binding sites in genes involved in the defense-signaling pathway. A, Network representing inversely related differentially accumulated miRNAs in irAGO8 compared with the wild type and their target genes at 0.75 h. Fourteen miRNAs showed binding sites in more than one gene. B, Many-to-many miRNA-mRNA interactions are evident in the network. These miRNAs were differentially more accumulated in irAGO8 compared with the wild type at 18 h, whereas their target genes were down-regulated in the same comparison.

For example, differentially more accumulated miRNAs, in irAGO8 as compared with the wild type, such as miRNAs 395 (at 0.75 h), 6218-5p (at 0.75 and 18 h), 5049-3p (at 18 h), 477a-3p (at 18 h), and 8015-3p (at both 0.75 and 18 h), were predicted to target key genes of phenolamide metabolism that gets perturbed during herbivory attack (Fig. 10). Similarly, differentially more accumulated miRNAs in irAGO8 as compared with the wild type, miRNAs 8769 (at 18 h), 8015-3p (at both 0.75 and 18 h), 395 (at 0.75 h), 8586 (at 0.75 h), 8129-3p (at 0.75 h), 6218-5p (at both 0.75 and 18 h), 7504e (at 0.75 h), 8165 (at both 0.75 and 18 h), and 8764 (at 0.7 5h), targeted key genes in the JA signaling pathway and defense pathways such as nicotine biosynthesis, HGL-DTG biosynthesis, and phenylpropanoid and phenolamide metabolism, indicating that these also might be regulated by miRNAs.

DISCUSSION

Regulatory smRNAs modulate the complex responses of plants that are elicited during attack by insect herbivores (Pandey et al., 2008b). Yet, the central component of the RISC machinery, the AGOs, of the herbivore-induced smRNA pathway had remained unknown, and this was addressed in this investigation. Previously, we had shown that RdR1 and DCL3/4 mediate herbivore defense in N. attenuata (Pandey and Baldwin, 2007; Pandey et al., 2008b; Bozorov et al., 2012). Here, we advance this understanding and propose that, when plants are attacked, they recruit a specific pathway to elicit herbivory-induced smRNAs. The herbivore-induced precursor miRNAs and the RdR1-dependent precursors of other smRNAs are diced by DCL3/4 and loaded onto the RISC that may be largely composed of AGO8. The silencing signal may further be amplified with the help of RdR1. These herbivory-induced smRNAs modulate defense responses by regulating the expression of genes related to signaling and defense directly or by regulating the expression of major transcription factors. This analysis suggests that the smRNA machinery has specialized components devoted to regulating complex herbivore responses. Further mechanistic insights related to this hypothesis may be tested by the AGO HITS-CLIP technique (Chi et al., 2009), which has been demonstrated to work with virus-cell culture interactions, but not yet with intact plants, and hence is beyond the scope of this article.

Comparative sequence and structure analyses of domains indicated conformational diversity, specifically in the smRNA-binding pocket. NaAGO8 has specific residues that may recognize smRNAs with different preferences. Indeed, the analysis of various domains of NaAGOs (Fig. 1, C and D) showed variability in the physicochemical properties of residues at functionally important sites. NaAGO8 has significantly different area and volume of the smRNA-binding pocket and the distribution of electrostatic patches that may help in recognizing/differentiating the incoming, herbivory-induced RNA substrate (as evident in Fig. 9). Specific biochemical studies with site-directed mutagenesis, and dedicated crystallization efforts using the herbivore-induced smRNAs as substrates, may further improve the mechanistic understanding of the specificity of AGO8.

In N. attenuata, an array of metabolites, including nicotine, phenolics, diterpenoids, and volatiles, are produced to counter herbivore attack at many different levels (Wu and Baldwin, 2010; Gaquerel et al., 2014). These induced defenses are largely downstream of JA (Halitschke and Baldwin, 2003). Interestingly, both in the OS-elicited and the caterpillar-fed leaf samples of NaAGO8-silenced lines, JA concentrations were partially reduced, which phenocopies the JA accumulations in attacked irRdR1 and irDCL4 (Pandey et al., 2008b; Bozorov et al., 2012). One of the most important defenses of N. attenuata is nicotine (Steppuhn et al., 2004), and in irAGO8 plants, nicotine levels are reduced significantly, a result likely explained by reduced transcript accumulations of NaODC and NaPMT, important genes in nicotine biosynthesis (Steppuhn et al., 2004). Again, lower nicotine levels of irAGO8 plants after herbivory phenocopy those of attacked irRdR1 and irDCL3/irDCL4 plants (Pandey and Baldwin, 2007; Bozorov et al., 2012). Furthermore, irAGO8 plants accumulated very low levels of HGL-DTGs and phenylamine defense compounds. Abundant herbivore-induced HGL-DTGs, such as lyciumoside IV, attenoside/lyciumoside II, and malonylated nicotianoside Ia, accumulated little in irAGO8, as evident from their VIP scores. Silencing NaGGPPS, which supplies geranylgeranyl diphosphates for the production of HGL-DTGs, significantly lowers the accumulation levels of HGL-DTGs (Jassbi et al., 2008; Heiling et al., 2010). Transcripts of two prenyltransferase genes, NaGGPPS and NaFPPS, were reduced significantly in response to NaAGO8 silencing. These results are consistent with the hypothesis that the NaAGO8-dependent smRNA pathway modulates the biosynthesis of secondary metabolites required for herbivore defenses.

In N. attenuata, NaMYB8 is a key transcription factor that regulates the accumulation of major phenolamides, caffeoylputrescine and dicaffeoylspermidine, upon herbivore attack (Kaur et al., 2010; Onkokesung et al., 2012). The key biosynthetic genes of shikimate and polyamine metabolism are regulated in a MYB8-dependent manner (Kaur et al., 2010): AT1 mediates caffeoylputrescine biosynthesis, while DH29 and CV86 mediate the dicaffeoylspermidine biosynthesis (Supplemental Fig. S10; Onkokesung et al., 2012). Silencing NaAGO8 reduces the accumulation of transcripts of these key downstream genes, which is consistent with the observed reduced levels of MYB8 and the increased susceptibility of irAGO8 plants to herbivores.

It is often observed that herbivore-induced defense signaling results in a positive feedback on the expression of genes of the signaling pathway. A classical example of such a feedback is commonly seen in the JA signaling pathway itself, where W+OS treatment results in a rapid increase of JA and JA-Ile within 15 to 30 min, whereas the maximum transcript levels are observed after 45 min to 1 h (Halitschke and Baldwin, 2003; Paschold et al., 2008; Gulati et al., 2013; Hettenhausen et al., 2013). As AGO8 is downstream in the smRNA pathway, it is plausible that basal levels of AGO8 may be sufficient to bind to herbivory-induced smRNAs that may help in overcoming the suppression of defense signaling under basal conditions. We also noticed that several miRNAs were repressed after OS elicitation in wild-type plants. We postulate that such down-regulation of miRNAs may help in eliciting defense signaling in the wild type (such reprograming is altered in irAGO8 plants). Once the defense signaling is induced (resulting in active biosynthesis of, for example, MYB8-dependent defenses as well as nicotine), signals may exert a positive feedback for further modulation and fine-tuning of expression levels. Additionally, an initial suppression of AGO8 transcript may result in additive induction of positive regulators of the defense-signaling network. It is evident that the smRNA-mediated regulation of defense signaling is highly complex and may exert roles that are modulatory in nature to optimize the induction of costly defense responses.

Because of the negative regulatory nature of the AGO-mediated RISC, and the complex nature of defense signaling, we also postulate the presence of negative regulators or suppressors of defense-signaling responses. It is plausible that, under constitutive conditions (absence of herbivore attack), such a suppressor(s) of defense signaling is expressed in the cells to prevent the synthesis of costly defense responses. During herbivore attack, plants reprogram their smRNome profiles (Figs. 5 and 6), and AGO8-dependent RISC may target such repressors of defense signaling, leading to the induction of defense responses, which are largely regulated by the MYB8 transcription factor.

In conclusion, we propose that AGO8, DCL3/4, and RdR1 participate in the herbivore-induced smRNA pathway to modulate the reprogramming of plant defenses in N. attenuata. Plants have diverse populations of smRNAs that are recruited differentially in response to various biotic and abiotic stress conditions and have specific regulatory roles to combat these stresses (Pandey and Baldwin, 2008; Pandey et al., 2008a, 2008b; Khraiwesh et al., 2012; Kruszka et al., 2012). Herbivore attack results in changes in expression of smRNAs that are produced in an RdR1- and DCL3/4-dependent manner. These smRNAs may be channeled to modulate targets specific to herbivore resistance in an AGO8-dependent manner. In the 11 AGOs of N. attenuata (Singh et al., 2015), the functionally important nucleotide specificity loop between the MID and PIWI domains shows diversity (Fig. 1; Supplemental Fig. S1), as do the largest binding pocket and the positive electrostatic patch involved in interaction with substrate RNAs (Fig. 9). Importantly, positions in the MID and PIWI domains (Figs. 1 and 9), which are thought to be involved in the interaction and stabilization of smRNA, have residues that differ in their physicochemical properties in NaAGO8 and may help AGO8 in sorting and recruiting smRNAs that are specific to herbivore resistance processes.

MATERIALS AND METHODS

Plant Growth

Seeds of the 31st inbred generation of Nicotiana attenuata, originally collected from the native population in Utah, were used. Seeds from wild-type and transformed plants were germinated, and plant populations were grown under a day/night cycle of 16 h (26°C–28°C)/8 h (22°C–24°C) as described earlier (Krügel et al., 2002; Halitschke and Baldwin, 2003; Onkokesung et al., 2012).

Sequence Characterization of N. attenuata AGOs

The recently reported 11 N. attenuata AGOs (Singh et al., 2015) were aligned, and distance values were calculated with 1,000 bootstrap replicates. Various domains in NaAGOs were annotated by performing a comparative sequence analysis of NaAGOs against the two well-studied eukaryotic AGOs whose structures have been solved: HsAGO2 (PDB code 4F3T; Elkayam et al., 2012) and KpAGO (PDB code 4F1N [model for yeast AGO]; Nakanishi et al., 2012). The NaAGO sequences were then subjected to the conserved domain search tool (http://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi; Marchler-Bauer et al., 2011). MSA and neighbor-joining phylogeny of NaAGOs was performed with the help of ClustalX 2.1 (Larkin et al., 2007) and MEGA 5.2 (Tamura et al., 2011), respectively. WebLogo 3 (Crooks et al., 2004) was used to display the position-specific residues at the functionally important sites.

Transcript Accumulation by Quantitative Real-Time PCR

To determine the changes in transcript accumulation of AGOs in leaves and roots after herbivore attack, the leaves of three to four replicates of sand-grown rosette-stage N. attenuata plants were wounded with a fabric pattern wheel, and the punctured wounds were immediately treated with 20 µL of W+OS. Leaves and roots were harvested at 0.25, 0.5, 0.75, 1, 1.5, 2, 3, 4, 6, 12, 14, 18, 26, 36, and 50 h after elicitation (Bozorov et al., 2012). Total RNA was isolated using the TRIZOL method, and DNA traces were removed with DNase I (DNA-free kit; Ambion).

To analyze the transcript accumulation of genes of phytohormone signaling and defense pathways, time-course experiments were conducted. Fully expanded rosette leaf samples were harvested at 0, 0.5, 0.75, 1, 6, 18, and 24 h after W+OS elicitation of wild-type and AGO8-silenced (irAGO8) lines. All samples were obtained from four independent biological replicates for each time point, genotype, and treatment.

cDNA synthesis (SuperScript II Reverse Transcriptase; Invitrogen) and qPCR (qPCR Core Kit for SYBR Green I; Eurogentec) were done following each manufacturer’s protocol. qPCR assays were performed using gene-specific primers (designed with the help of Primer Express software version 3.0.1; http://www.appliedbiosystems.com) on cDNA corresponding to 100 ng of total RNA before transcription (Supplemental Table S7). The N. attenuata sulfite reductase (ECI), a housekeeping gene, was used as an endogenous reference (Bubner et al., 2004). The 2^−∆∆CT method was used for data analysis. For data interpretation, the expression level in control plants (time 0) was fixed to 1, and relative expression levels for other time points were calculated with respect to this reference value (Bubner et al., 2004; Pandey and Baldwin, 2007; Bozorov et al., 2012).

Generation and Characterization of Stable Transgenic Lines Silenced for AGO Expression

Partial cDNA fragments of NaAGO1 (common to AGO1a, AGO1b, and AGO1c), NaAGO2, NaAGO4 (common to AGO4a and AGO4b), NaAGO5, NaAGO7, NaAGO8, and NaAGO10 were inserted into a derivative of pRESC800 (GenBank accession no. JQ354897) binary transformation vector backbone as inverted repeats as described earlier (Steppuhn et al., 2004). Primers for cloning are indicated in Supplemental Table S7. Wild-type N. attenuata plants from the 31st inbred generation were used for Agrobacterium tumefaciens-mediated plant transformation (Krügel et al., 2002; Gase et al., 2011). In the subsequent T1 generation, plants were tested for homozygocity by segregation analysis. Characterization of stable transformants and screening of individual transgenic lines were conducted as described earlier (Gase et al., 2011). Transgenic lines with complete T-DNA insertions were tested for homozygosity through the T2 generation. Transgenic seeds were germinated on agar plates containing 35 mg L⁻¹ hygromycin B. Ten healthy seeds of each transgenic (T1) line were inbred through the T2 generation to obtain homozygous lines. Complete T-DNA insertion was determined by diagnostic PCR (Gase et al., 2011). Total genomic DNA was isolated by a modified cetyl-trimethyl-ammonium bromide method (Bubner et al., 2004). Positive lines with complete insertion of T-DNA were assayed further to determine single insertion by using Southern hybridization (Bubner et al., 2004), which was performed as described earlier using a ³²P-labeled fragment of the hptII gene as a probe (Jassbi et al., 2008). Seven micrograms of genomic DNA, digested with XbaI and EcoRV (New England Biolabs), was blotted onto a nylon membrane (Gene-ScreenPlus; PerkinElmer) according to the manufacturer’s protocol. Homozygous T2 plants with single T-DNA insertions were used for further experiments. To determine the silencing efficiency of the transgenic lines, plants were grown in a controlled environment in a glasshouse as described above. At fully expanded rosette stage, leaves were W+OS elicited, and leaf samples were collected after 18 h. Transcript abundance was measured to determine the silencing efficiency using qPCR assays as described above. The silencing efficiency of AGO8 in irAGO8 lines was determined by qPCR assays, both before elicitation (basal levels; Supplemental Fig. S4C) as well as after simulated herbivory (Fig. 3). Possibilities of off-target effects in the N. attenuata genome were evaluated by exposing the AGO8 silencing sequence (inverted repeat sequence present in the pRESC8AGO8 construct) to both the SGN VIGS tool (Fernandez-Pozo et al., 2015) and a BLAST search against the complete transcriptome of N. attenuata. The only identical stretch longer than 19 bp in the entire transcriptome was a 22-nucleotide fragment from the xylem Cys proteinase1-like transcript (XCP1). This single short stretch is less likely to cause significant silencing in off-target effects; furthermore, the xylem XCP1 is not involved in the herbivore defense response (Thomas et al., 2001; Gulati et al., 2013; Zhirnov et al., 2015). Still, to completely rule out any off-target effects, we evaluated the transcript levels of XCP1: no differences in levels were recorded between the wild-type and irAGO8 genotypes (Supplemental Fig. S5). Thus, it is safe to conclude that silencing of AGO8 did not generate any off-target effects in irAGO8 plants.

Herbivore Performance Assay

Manduca sexta performance assays were conducted on stably transformed lines as described earlier (Pandey and Baldwin, 2007; Oh et al., 2013). Fifteen to 25 replicates from each irAGO genotype and wild type were used. Plants were grown in a glasshouse as described above in a completely randomized design. Neonates were placed on the lower surface of the +2 leaf of rosette stage plants, and larvae were allowed to feed for 11 to 13 d. Caterpillar biomass was recorded every 2 to 3 d. Caterpillar performance on irAGO8 lines was verified in an independent biological experiment with a similar setup.

Evaluation of the AGO8-Dependent smRNome

In order to evaluate the expression and composition of AGO8-dependent smRNAs, deep sequencing was performed in triplicates on +2 leaf samples of rosette stage wild-type and irAGO8 plants before (0 h) as well as 0.75 and 18 h after OS elicitation. Clean reads were populated after filtering out low-quality reads such as reads with N (unidentified nucleotides) as well as reads with single nucleotide stretches of more than five nucleotides, trimming the adaptors, and discarding reads with short stretches of six nucleotides; clean reads in the range of 15 to 30 nucleotides were finally retained (Pandey et al., 2008b). Next, the structural RNAs, such as tRNA, rRNA, snoRNA, and snRNA, were annotated (and removed) by aligning the clean reads to the Rfam 11.0 database (ftp://ftp.sanger.ac.uk/pub/databases/Rfam). Reads mapped to the N. attenuata transcriptome were discarded using bowtie with the -norc parameter (no reverse complement; Langmead et al., 2009). For the remaining reads, replicates were merged for each condition/time point, and reads present in at least two of the three replicates were retained for further analysis. These are referred to as miRNA-mappable reads. These were aligned to the N. attenuata genome sequence using bowtie with a maximum of two mismatches. Conserved miRNAs were identified using the mirDeep2 software version 2.0.0 (Friedländer et al., 2012) with default parameters and Viridiplantae miRNAs deposited in miRBase 21.0 (www.mirbase.org). Sequences with perfect matches to known miRNAs were regarded as bona fide conserved miRNAs. Furthermore, counts were normalized using RPM. Median values were calculated, and differential accumulation of miRNAs and other smRNAs was evaluated using the χ² test with a stringent P ≤ 0.05 after Benjamini-Hochberg multiple correction (Benjamini and Hochberg, 1995). Reads with counts greater than 2 in at least one comparison were considered for this analysis.

Differential Expression of miRNAs

The raw counts for each miRNA, both conserved and novel, were normalized using the RPM method (Jia et al., 2014). To evaluate the expression of conserved and novel miRNAs between the two genotypes and at the three time points, pairwise comparison was performed on the RPM value for each miRNA. Furthermore, to identify significantly differentially expressed miRNAs, χ² (Li et al., 2016) was performed and P < 0.05 was set as the cutoff.

Untargeted Metabolomic Analysis

A total of 100 mg of leaf tissue was homogenized to fine powder with a GenoGrinder (SPEX SamplePrep), and 1 mL of extraction buffer (40% methanol/water and 50 mm acetate buffer, pH 4.8) was added and mixed well. The samples were centrifuged at 16,100g for 20 min at 4°C, and the supernatant were transferred to a new 1.5-mL microcentrifuge tube (Eppendorf); samples were recentrifuged, and the supernatant was transferred carefully into glass vials. The samples were injected into a C18 Acclaim column (2.2-μm particle size, 150 × 2.1 mm i.d.; Dionex) and separated using an RSLC system (Dionex). As mobile phase, solvent A (0.1% [v/v] acetonitrile and 0.05% [v/v] formic acid in deionized water) and solvent B (acetonitrile and 0.05% [v/v] formic acid) were used with the following gradient conditions: 0 to 0.5 min, 10% B; 0.5 to 6.5 min, linear gradient to 80% B; 6.5 to 10 min, 80% B; and reequilibration at 10% B for 3 min with a flow rate of 300 μL min⁻¹. A MicroToF mass spectrometer (Bruker Daltonics) coupled with an electrospray ionization source in positive ion mode was used to detect the eluted compounds with the following settings: capillary voltage, 4,500 V; capillary exit, 130 V; dry gas temperature, 200°C; dry gas flow, 8 L min⁻¹. Mass calibration was done with the help of sodium formate clusters (10 mm solution of NaOH in 50%:50% [v/v] isopropanol:water containing 0.2% formic acid). The Data Analysis version 4.0 software (Bruker Daltonics) was used to convert raw data files to netCDF format and processed using the XCMS package and the R package CAMERA (http://www.bioconductor.org/biocLite.R). The Metaboanalyst software (Xia et al., 2009; Xia and Wishart, 2011) was used to perform multivariate analysis. The data were filtered using coefficient of variation, and Pareto scaling was used for normalization (Xia et al., 2009; Gaquerel et al., 2010). Insilico viewer was used to view the peaks.

Targeted Secondary Metabolite Analysis

Approximately 100 mg of frozen leaf tissue was ground to a fine powder and homogenized with extraction buffer (40% methanol with acetate buffer [4:6], 10 μL mg⁻¹ tissue) in a GenoGrinder for 1 min at 1,200 strokes min⁻¹. The samples were centrifuged at 16,100g for 20 min at 4°C, and the supernatants were carefully transferred into new 1.5-mL tubes, recentrifuged to remove any particles, transferred into 1.5-mL glass vials, and analyzed by ultra-HPLC as described previously (Keinänen et al., 2001; Jassbi et al., 2008). One microliter of the sample was injected into a Chromolith Fast Gradient (RP18e, 50 × 2 mm; Merck) column connected to a precolumn (Gemini NX RP18, 3 μm, 2 × 4.6 mm). As mobile phase, solvent A (0.1% formic acid and 0.1% ammonium hydroxide solution in water) and solvent B (methanol) were used in a gradient mode with the following conditions: 0% B for 0.5 min; to 80% B in 6.5 min; and isocratic for 3 min with a flow rate of 0.8 mL min⁻¹. External standard calibration was done with a dilution series of nicotine, chlorogenic acid, and rutin in 40% methanol, and samples were normalized by gram fresh weight. Secondary metabolites were detected on diode array detector channels (260, 254, 210, 320, and 360 nm; ELS1A) and data were analyzed using Chrom software. Representative runs of standards are presented in Supplemental Figure S8.

Phytohormone Extraction

For the analysis of phytohormones (JA, JA-Ile, ABA, and SA), ∼100 mg of frozen, fine-powdered leaf tissue was homogenized with 1 mL of extraction buffer (ethyl acetate containing 2 µL of the internal standard mix [D₂-dihydro-JA and D₅-JA, 50 ng µL⁻¹; D₆-ABA, D₄-SA, and ¹³C₆-JA-Ile, 10 ng µL⁻¹]) in a GenoGrinder for 1 min at 1,200 strokes min⁻¹. Samples were centrifuged at 16,100g for 20 min at 4°C, and the supernatant was transferred into fresh 2-mL microcentrifuge tubes. To the residue, 0.5 mL of ethyl acetate (without internal standards) was added and ground using a GenoGrinder for 1 min at 1,200 strokes min⁻¹ (setting: 1× at 200). After centrifugation of samples at 16,100g for 20 min at 4°C, the supernatants were pooled. The samples were placed in a vacuum concentrator (Speed-Vac) at a temperature of 30°C, and the dried residues were dissolved in 0.5 mL of 70% (v/v) methanol for analysis on a liquid chromatography-electrospray ionization-tandem mass spectrometry instrument (1200 Triple-Quadrupole-LC-MS system; Varian) as described previously (Oh et al., 2012). The samples were injected in a ProntoSIL column (C18-ace-EPS, 50 × 2 mm, 5 μm, 120 Å; Bischoff) connected to a precolumn (C18, 4 × 2 mm; Phenomenex). Gradient mode was used with mobile phases 0.05% formic acid and 0.1% acetonitrile in water (solvent A) and methanol (solvent B) with the following conditions (time/%B/flow [mL min⁻¹]): 0/15/0.4, 1/15/0.4, 1.5/15/0.2, 2.5/15/0.2, and 4.5/98/0.2. Compounds were detected in the electrospray ionization negative mode with the help of the multiple reaction monitoring method (Bonaventure et al., 2011).

Structural Analysis

HMM profile-based homology modeling of structure was used to predict the secondary structure of NaAGOs using the HHpred server (http://toolkit.tuebingen.mpg.de/hhpred/; Söding et al., 2005). NaAGO peptide sequences were used as input in the HHpred server to search similarities in secondary structure with default parameters. HHpred produces pairwise query-template alignment to generate HMM profiles that search through various databases, such as PDB, SCOP, Pfam, SMART, COGs, and CDD. The final three-dimensional structural models were calculated by MODELER. The ClusPro server (https://cluspro.bu.edu/login.php; Comeau et al., 2004) was used for the docking of smRNAs (PDB codes 4F3T:R and 4W5O:B) with each of the modeled NaAGOs, and their potential energies were calculated. These two smRNAs are bona fide substrates for AGOs and have been used widely in crystallographic studies (Elkayam et al., 2012; Schirle et al., 2014). Root mean square deviations (Å) between NaAGOs were calculated using the TMalign server (http://zhanglab.ccmb.med.umich.edu/TM-align/; Zhang and Skolnick, 2005). The area and volume of the largest binding pocket on the each docked NaAGO were calculated using the CASTp server (http://sts.bioe.uic.edu/castp/calculation.php; Dundas et al., 2006). CASTp uses the weighted Delaunay triangulation and the alpha complex to measure the area and volume of solvent-accessible structural pockets and cavities (Dundas et al., 2006). The distribution of positive electrostatic patches on the surface of docked NaAGOs was computed using the Patch Finder plus server (http://pfp.technion.ac.il/index.html; Shazman et al., 2007). Using the Poisson Boltzmann equation, it first calculates the electrostatic potential (using APBS software) of the protein on a three-dimensional grid and then extracts all the three-dimensional patches of adjacent grid points that meet the defined cutoff of 2kT/e. PyMOL (DeLano, 2002) was used for the visualization and figure preparation of the modeled structures.

Target Prediction and Mapping of miRNAs on mRNAs

With the aim to better understand the biological role of conserved and AGO8-specific miRNAs, we searched for putative binding sites and identified miRNAs in the sequences of previously reported genes that are involved in defense mechanisms against herbivory attack. We adopted approaches used previously (Pandey et al., 2008b) for binding site identification.

Statistical Analysis

Data were analyzed using StatView software. Caterpillar assay, phytohormone, secondary metabolite, and gene expression data were analyzed by ANOVA (repeated measures wherever applicable) at a significance level of P ≤ 0.05.

Accession Numbers

The smRNA sequencing data have been deposited in the National Center for Biotechnology Information under accession number PRJNA335958.

Supplemental Data

The following supplemental materials are available.

• Supplemental Figure S1. MSA of NaAGOs corresponding to Figure 1A, and phylogenetic tree of Arabidopsis AGOs.

• Supplemental Figure S2. Transcript abundance of NaAGOs in roots in a time-course experiment.

• Supplemental Figure S3. Complete southern blots of all the transgenic genotypes.

• Supplemental Figure S4. Silencing of AGO8 expression does not affect the development and growth of N. attenuata.

• Supplemental Figure S5. Evaluation of the silencing efficiency of nine NaAGOs and XCP1 in irAGO8 samples.

• Supplemental Figure S6. Validation of the susceptibility of irAGO8 plants to herbivore attack.

• Supplemental Figure S7. DEMs in the wild type and irAGO8.

• Supplemental Figure S8. Chromatograms of the standards used for targeted analysis of nicotine, DTGs, and rutin.

• Supplemental Figure S9. Evaluation of phytohormones in the wild type and irAGO8 upon herbivory attack.

• Supplemental Figure S10. Dynamics of transcript accumulation in phytohormone signaling genes in response to herbivory.

• Supplemental Figure S11. Simplified scheme for the synthesis of major components of defense-related metabolites in N. attenuata.

• Supplemental Table S1. Conserved and predicted novel miRNAs identified in the wild type and irAGO8 across the three time points: 0, 0.75, and 18 h.

• Supplemental Table S2. Mature miRNAs conserved to other species when mapped against 72 plant species in miRBase.

• Supplemental Table S3. Accumulation of miRNAs at each time point (0, 0.75, and 18 h) for irAGO8 and the wild type.

• Supplemental Table S4. DEMs for irAGO8 and the wild type.

• Supplemental Table S5. Differentially accumulated metabolites on silencing NaAGO8.

• Supplemental Table S6. DEMs showed binding sites in defense-signaling pathway genes.

• Supplemental Table S7. List of primers used in this study.

Glossary

DTG

diterpenoid glycoside

JA

jasmonic acid

SA

salicylic acid

smRNA

small RNA

RISC

RNA-induced silencing complex

miRNA

microRNA

siRNA

small interfering RNA

MSA

multiple sequence alignment

PDB

Protein Data Bank

OS

oral secretions

W+OS

distilled water-diluted oral secretions

RPM

reads per million

DEMs

differentially expressed microRNAs

ABA

abscisic acid