miRNA863-3p sequentially targets negative immune regulator ARLPKs and positive regulator SERRATE upon bacterial infection

Plant small RNAs play important roles in gene regulation during pathogen infection. Here we show that miR863-3p is induced by the bacterial pathogen Pseudomonas syringae carrying various effectors. Early during infection, miR863-3p silences two negative regulators of plant defence, atypical receptor-like pseudokinase1 (ARLPK1) and ARLPK2, both lacking extracellular domains and kinase activity, through mRNA degradation to promote immunity. ARLPK1 associates with, and may function through another negative immune regulator ARLPK1-interacting receptor-like kinase 1 (AKIK1), an active kinase with an extracellular domain. Later during infection, miR863-3p silences SERRATE, which is essential for miRNA accumulation and positively regulates defence, through translational inhibition. This results in decreased miR863-3p levels, thus forming a negative feedback loop to attenuate immune responses after successful defence. This is an example of a miRNA that sequentially targets both negative and positive regulators of immunity through two modes of action to fine-tune the timing and amplitude of defence responses.

D uring pathogen infection, plants undergo reprogramming and fine-tuning of gene expression to activate innate immune responses, and small RNAs (sRNAs) are important regulators in this process [1][2][3][4] . As the first layer of defence, plant pattern-recognition receptors detect conserved pathogen features called pathogen-associated molecular patterns (PAMPs), leading to PAMP-triggered immunity (PTI) 5 . Some pathogens deliver effector proteins to attenuate PTI; however, in resistant hosts, Nod-like receptor (NLR) proteins recognize cognate effectors and activate effector-triggered immunity (ETI), which is more specific, robust and rapid than PTI. ETI is often characterized by the hypersensitive response (HR), a type of cell death, at the site of infection 6,7 . One well-studied effector-NLR model in Arabidopsis is the effector AvrRpt2 from the bacterial pathogen Pseudomonas syringae pv. tomato (Pst) DC3000 and its cognate plant host NLR protein RPS2 (refs 8,9).
sRNAs are critical regulators of plant defence responses and pathogen virulence [1][2][3][4]10 . These microRNAs (miRNAs) and short interfering RNAs (siRNAs) are processed by Dicer or Dicer-like (DCL) proteins and then incorporated into Argonaute (AGO) proteins to form RNA-induced silencing complex to target genes with full or partial complementary sequences. sRNAs can induce gene silencing by guiding post-transcriptional gene silencing via mRNA degradation or translational inhibition, or transcriptional inhibition via DNA methylation or chromatin modification 11 . During infection with Pst DC3000 carrying the effector gene avrRpt2, the natural antisense transcript-associated siRNA natsiRNAATGB2 silences a gene with pentatricopeptide repeats, which acts as a negative regulator of ETI mediated by RPS2 (ref. 12). A long siRNA, AtlsiRNA, also induced by Pst (avrRpt2), silences the AtRAP gene (Arabidopsis thaliana RNA-binding domain abundant in apicomplexans), which has a negative role in defence 13 . miR393 contributes to PTI by targeting auxin signalling receptors upon elicitor flg22 treatment or after infection with virulent Pst DC3000 (ref. 14). The complementary strand of miR393 within the sRNA duplex, miR393*, regulates ETI during infection of Pst (avrRpt2) by targeting a protein trafficking gene MEMB12 (Membrin 12) to promote the secretion of antimicrobial PR (pathogenesis-related) proteins 15 . miR160a enhances PAMP-induced callose deposition by targeting auxin response factors 16 . In addition to these examples of sRNAs that are induced by pathogen infection, some sRNAs are downregulated by infection. For example, miR398b, which targets a cytochrome c oxidase and two copper superoxide dismutases, and miR773, which targets a DNA methyltransferase, MET2, both negatively regulate PAMP-induced callose deposition, as well as PTI against Pst (EV) and Pst (DhrcC) 16 . Collectively, these studies highlight the importance of sRNAs that target critical plant genes involved in immune signalling. However, no sRNAs are reported to target genes with antagonistic roles in plant immunity or any other cellular processes.
We previously identified miRNAs and siRNAs that are differentially expressed on infection with various bacterial Pst strains 17 . One of these miRNAs, miR863-3p, was highly upregulated by the avirulent strain Pst (avrRpt2); thus, we hypothesized that this miRNA may play an important role in ETI, and sought to characterize its targets and their function. Here we show that miR863-3p targets the putative pseudokinases, atypical receptor-like pseudokinase 1 (ARLPK1) and ARLPK2, early on during infection to boost plant defence. ARLPK1 interacts with ARLPK1-interacting receptor-like kinase (AKIK1), an active kinase that also negatively regulates defence. Later during infection, miR863-3p silences SERRATE (SE)-a protein important for miRNA processing and accumulation [18][19][20][21] and for defence-to form a negative feedback loop to regulate the level of miRNA863-3p. Attenuation of immune responses after successful defence is possibly to conserve host resources for growth and reproduction.

miR863-3p is induced by infection with avirulent Pst strains.
To identify miRNAs involved in regulating immune responses to bacterial infection, we previously profiled sRNAs from plants challenged with virulent and avirulent bacterial strains, Pst (EV) and Pst (avrRpt2), respectively 17 . We found that miR863-3p was highly upregulated during Pst (avrRpt2) infection. miR863-3p is a 21-nucleotide (nt) miRNA originating from the 3 0 -arm of the precursor miRNA ( Supplementary Fig. 1a) 22 . Northern blot analysis confirmed that miR863-3p was induced by an avirulent strain Pst (avrRpt2), but not by the virulent strain Pst carrying an empty vector (EV), a strain that has a mutation in its type III secretion system-Pst (DhrcC), or mock solution (10 mM MgCl 2 ; Fig. 1a). miR863-3p levels are also upregulated by avirulent strains Pst (avrRpm1) and Pst (avrRps4), although to a lesser degree compared with Pst (avrRpt2) (Supplementary Fig. 1b). Interestingly, miR863-5p, the complementary strand of miR863-3p, was not induced by any of these Pst strains, and thus is not likely to be involved in defence against Pst (Fig. 1b).
Biogenesis of miRNAs depends on DCL1, which processes pri-miRNAs, and is aided by interacting RNA-binding proteins HYL1 and SE to promote accurate processing 20 . Northern blot analysis revealed that miR863-3p accumulation is dependent on DCL1 and SE, but independent of RDR6, which is involved in the biogenesis and amplification of siRNAs (Fig. 1c) 11 . These results indicate that miR863-3p is indeed a true miRNA instead of a siRNA. Taken together, our results show that miR863-3p is dependent on DCL1 and SE, and it is induced by challenge with avirulent Pst strains, especially Pst (avrRpt2). Its complementary strand was not detected in our experiments, showing that miR863-3p is the stable miRNA strand during pathogen infection.
miR863-3p targets two ARLPKs and SE. We hypothesized that miR863-3p targets genes involved in immune responses, but since miR863-3p has no validated targets in existing databases, we predicted its potential targets (Methods). miR863-3p potentially targets the coding sequences of two atypical receptor-like kinases (RLKs), ARLPK1 and ARLPK2, belonging to the leucine-rich repeat (LRR) III RLK subfamily 23 , as well as the 3 0 -untranslated region (UTR) of SE, which encodes a zinc-finger protein found in both the miRNA biogenesis DCL1 complex 19 , and the nuclear cap-binding protein complex 24 (Fig. 2a). Typical RLKs have an extracellular ligand-binding domain, a single transmembrane region and an intracellular kinase domain 23 . Both ARLPK1 and ARLPK2 have a signal peptide sequence overlapping with their N-terminal transmembrane domain ( Supplementary Fig. 2a,b) and lack extracellular domains 23 . They also have variations in the conserved features of functional kinase domains such as the glycine-rich loop, VAIK motif and HRD motif ( Supplementary  Fig. 3), and are thus predicted pseudokinases 25 . Because ARLPK1 and ARLPK2 are missing conserved key residues in the consensus motifs and are lacking extracellular domains, we named them atypical receptor-like pseudokinases (ARLPK1, At5g61570; and ARLPK2 At5g07620). miR863-3p is also predicted to target a third gene SE, which is involved in the accumulation of miRNAs, including miR863-3p (Fig. 1c). During Pst (avrRpt2) infection, the mRNA levels of the two ARLPKs are downregulated compared with mock-infected plants; however, SE mRNA levels are upregulated. In contrast, the levels of all three transcripts do not change significantly during Pst (EV) or Pst (DhrcC) infection (Fig. 2b).
To confirm that ARLPK1 and ARLPK2 are true targets of miR863-3p, we performed Agrobacterium-mediated transient coexpression assays in Nicotiana benthamiana using wild-type (wt) and mutated (m) kinase genes. The mutated kinases have an altered miR863-3p target site, yet the predicted amino-acid sequence is unchanged (Fig. 2a). MIR863 co-expression with wtARLPK1 or wtARLPK2 resulted in a decrease of the mRNA levels of both kinases, while co-expression with mARLPK1 or mARLPK2 did not affect the mRNA levels of mARLPK1 or mARLPK2 (Fig. 2c). In addition, we performed RNA-ligase mediated 5' rapid amplification of cDNA ends (RLM-RACE) in Pst (avrRpt2)-infected Arabidopsis to map miR863-3p-guided cleavage sites in ARLPK mRNA. As expected, the majority of the 5 0 -RACE cloned products mapped to the predicted target site in ARLPK1 and ARLPK2 mRNAs (Fig. 2d). These data indicate ARLPK1 and ARLPK2 are direct targets of miR863-3p, and they are silenced via mRNA cleavage and degradation.
In plants, most miRNAs silence their targets by directing mRNA cleavage and degradation, while others mediate translational inhibition 26 . In mammals, most miRNA target sites for translational inhibition are located in the 3 0 -UTR, and the target site of miR863-3p in SE is also at the beginning of the 3 0 -UTR. SE mRNA levels are not downregulated by miR863-3p ( Fig. 2b), even though the nucleotide alignment shows very few mismatches (Fig. 2a). Thus, it is possible that miR863-3p silences SE through translational inhibition. In mammals, target mRNAs silenced by RNA-induced silencing complex via translational repression are delivered to and accumulate in processing (P)bodies 27 . Thus, it is possible that we see elevated SE transcripts not only because of continual transcription of SE due to the absence of SE protein product but also because of the accumulation of sequestered SE mRNAs in P-bodies. To test that hypothesis, we generated constructs of green fluorescent protein (GFP) fused to a wt version of truncated SE that includes only the 3 0 -UTR (GFP-wtSE-3 0 -UTR) or a mutated truncated version of SE with an altered miRNA target site (GFP-mSE-3 0 -UTR; Fig. 2a). We co-expressed the fusions with MIR863 in N. benthamiana, and western blotting revealed that GFP-wtSE-3 0 -UTR, but not GFP-mSE-3 0 -UTR, levels were downregulated (Fig. 2e). To confirm this result in vivo, we obtained an anti-SE antibody and detected SE protein levels in Arabidopsis plants challenged with Pst strains. Western blot analysis revealed that SE protein levels are downregulated by Pst (avrRpt2) but not by Pst (DhrcC) or Pst (EV) (Fig. 2f). These results suggest that miR863-3p targets and silences SE by inhibiting its translation. Thus, miR863-3p uses two different modes of action to silence its targets-it suppresses the two ARLPKs via mRNA cleavage and degradation and inhibits SE by translational inhibition.
ARLPKs are negative regulators of disease resistance.
To investigate the role of miR863-3p during plant immune responses, we created transgenic plants overexpressing MIR863 under the control of the cauliflower mosaic virus 35S promoter. Figure 3a shows nine such overexpression (OE) lines. The lines with very high miR863-3p expression (such as MIR863-OE1 and -OE2 lines) resembled the serrate mutant phenotype 28 , further supporting that SE is indeed a miR863-3p target (Fig. 3b). Only a slight increase of miR863-3p was observed after infection with Pst (avrRpt2) in the OE lines, most likely because miR863-3p levels are already very high ( Supplementary Fig. 4). We also detected very low levels of miR863-5p in some of the OE lines (Fig. 3a). We chose two transgenic lines with high miR863-3p levels, MIR863-OE1 and -OE2, and one with lower levels, MIR863-OE3, for functional analyses (Fig. 3a). MIR863-OE1 and -OE2 plants were slightly smaller than Col-0 WT plants (Fig. 3b  panels). In the MIR863 overexpression lines, ARLPK1 and ARLPK2 mRNA levels are downregulated while SE mRNA levels are elevated, as shown by real-time reverse transcription-PCR (RT-PCR) results (Fig. 3c). SE mRNA levels are elevated just as in Pst (avrRpt2)-infected plants (Fig. 2a), most likely due to feedback compensation to increase transcription and also sequestering of transcripts in P-bodies 27 . SE protein levels in the overexpression lines are significantly lower compared with Col-0 WT, especially in MIR863-OE1 and -OE2 (Fig. 3d). Interestingly, even though the target genes were downregulated, the MIR863 overexpression lines showed no altered disease resistance to Pst (EV) or Pst (avrRpt2) compared with Col-0 WT plants (Fig. 3e).
To understand why overexpressing MIR863 did not have an effect on the defence response, we decided to study the function of miR863-3p targets individually. First, we focused on the functional analysis of the two ARLPKs and obtained the null mutants SALK_144635 (arlpk1-1) and SALK_040744C (arlpk2-1) 29 . Each mutant contains a T-DNA insertion in the first exon ( Supplementary Fig. 5a), as confirmed by RT-PCR ( Supplementary Fig. 5b). The single mutants were developmentally similar to Col-0 WT plants (Supplementary   Fig. 5d). This lack of disease phenotype may be due to functional redundancy, so we generated the arlpk1-1 arlpk2-1 double mutant. The arlpk1-1 arlpk2-1 plants were smaller compared with Col-0 WT (Fig. 4a), and when infected with Pst (EV) and Pst (avrRpt2), arlpk1-1 arlpk2-1 plants exhibited less bacterial growth in both cases (Fig. 4b).
To confirm the phenotype of enhanced resistance in the double mutant, we measured the expression of antimicrobial pathogenesis-related protein, PR1, a marker for salicylic acid-dependent disease responses. At 12 h post inoculation (h.p.i.), arlpk1-1 arlpk2-1 showed higher PR1 protein expression compared with Col-0 WT after infection with Pst (EV) and Pst (avrRpt2) (Fig. 4c), indicating that ARLPK1 and ARLPK2 are functionally redundant and negatively regulate plant immune responses. We furthermore generated transgenic Arabidopsis overexpressing ARLPK1 with a mutation in the miR863-target site (mARLPK1-OE, as in Fig. 2a) in the Col-0 WT background. We selected two overexpression lines, mARLPK1-OE1 and -OE2, exhibiting high ARLPK1 transcript levels as detected by real-time RT-PCR for further functional analysis (Fig. 4d). Both lines are more susceptible to Pst (EV) and Pst (avrRpt2) compared with Col-0 WT, further supporting that ARLPKs are negative regulators of plant immunity (Fig. 4e). miR863-3p-mediated silencing of ARLPK1 during bacterial infection is important to promote plant defence responses.
ARLPKs are putative pseudokinases. As noted above, ARLPK1 and ARLPK2 have variations in the conserved motifs of functional kinase domains such as the glycine-rich loop (GXGXXG), VAIK motif and HRD motif ( Supplementary  Fig. 3), and are thus predicted pseudokinases. The glycine loop is located in subdomain I near the catalytic domain and is found in many nucleotide-binding proteins as well as in protein kinases 25 . Both ARLPK1 and ARLPK2 have a serine substitution for the second glycine in the loop. The VAIK, HRD and DFG motifs are all found in the catalytic domain. The lysine in the VAIK motif in subdomain II interacts with ATP 25 . ARLPK1 and ARLPK2 have a VRVL and an IRVL motif, respectively, and are thus missing the lysine residue; it has not been demonstrated that these motifs can bind ATP. The aspartic acid in the HRD motif in subdomain IV is the base acceptor for proton transfer 25 . However, both pseudokinases have a HGN motif containing an uncharged asparagine. Finally, the aspartic acid in the DFG motif, which is found in catalytic domains and binds Mg 2 þ ions that coordinate the phosphates of ATP in the ATP-binding cleft 25 , is conserved in ARLPK1 and ARLPK2 ( Supplementary Fig. 3). As predicted, we were unable to detect autophosphorylation in vitro using recombinant ARLPK1 and ARLPK2 proteins (Fig. 4f). In contrast, we were able to clearly detect kinase activity for the RPM1-induced protein kinase (RIPK), which was used as a positive control 30 . These data support that ARLPK1 and ARLPK2 are most likely pseudokinases.   Fig. 2). We originally thought that the pseudokinases would be localized in the plasma membrane (PM) and we used a PM-localized protein RLK (At5g23740)-cyan fluorescent protein (CFP) as control. To our surprise, we found that both ARLPK1-yellow fluorescent protein (YFP) and ARLPK2-YFP predominantly localize in the ER ( Supplementary Fig. 6). We confirmed this using a mCherrytagged ER organelle marker 31 (ER-mCherry) as a control. ARLPK1-YFP and ARLPK-YFP overlapped with ER-mCherry in merged images, and Hechtian strands were not visible after plasmolysis treatment (Fig. 5a, upper panels). These results only show the steady-state localization of these fluorescently tagged proteins. Trafficking or movement of some RLKs during pathogen challenge is often observed, as in the case of FLS2, which undergoes endocytosis upon flg22 treatment 32 . We also checked the localization of ARLPKs after Pst (avrRpt2) challenge; however, we did not see any changes in the localization after Pst (avrRpt2) infection, both with and without plasmolysis treatment (Fig. 5a, lower panels). Because these two kinases are putative pseudokinases that lack an extracellular domain, we hypothesized that ARLPK1 and ARLPK2 function with another RLK to perceive and/or regulate extracellular signals. We searched for interacting proteins on the Arabidopsis Interactions Viewer 33 (http://bar.utoronto.ca/), and found AT5G59650, a LRR I subfamily RLK, which may interact with ARLPK1. Thus, we named it AKIK1. Owing to noncomplementarity in the nucleotide alignment, AKIK1 is not likely to be targeted by miR863-3p. AKIK1 contains a malectin domain and two LRR domains in its predicted extracellular region ( Supplementary Fig. 7a,b). We found that AKIK1-CFP also predominantly localizes in the ER in untreated and pathogentreated plants, both with and without plasmolysis treatment, as seen in the merged image with ER-mCherry (Fig. 5b, left panels), and it also co-localizes with ARLPK1-YFP (Fig. 5b, right panels).
We confirmed by co-immunoprecipitation (Co-IP) that ARLPK1, but not ARLPK2, interacts with AKIK1 (Fig. 6a). We hypothesized that due to its association with ARLPK1, AKIK1 also has a role in plant defence. We obtained two T-DNA insertion mutant lines, akik1-1 (SALK_022711C) and akik1-2 (CS848612), which have T-DNA insertions in the fourth and ninth introns of AKIK1, respectively ( Supplementary Fig. 8a). These two lines are both null mutants of AKIK1, as confirmed by RT-PCR ( Supplementary Fig. 8b), and neither had any obvious developmental phenotypes compared with Col-0 WT ( Supplementary Fig. 8c). Both akik1-1 and akik1-2 were more resistant to infection with Pst (EV) and Pst (avrRpt2) compared with Col-0 WT plants (Fig. 6b). Because both mutants showed the same disease phenotypes, we used akik1-1 for further analysis. In agreement with the bacterial growth assays, PR1 protein levels were higher in akik1-1 compared with Col-0 WT 12 h.p.i. with Pst (EV) and Pst (avrRpt2), although to a lesser extent with Pst (EV)  ( Fig. 6c). To confirm the roles of ARLPK1, ARLPK2 and AKIK1 in PTI, we performed assays that measured PAMP responses to flg22 treatment. We measured PTI-responsive gene flg22-induced receptor-like kinase 1 (FRK1) by real-time RT-PCR after treatment with flg22. In agreement with the bacterial growth assay results, the double mutant arlpk1-1 arlpk2-1 and the akik1-1 mutant showed highly increased FRK1 expression compared with Col-0 WT (Fig. 6d). Mitogen-activated protein kinase 3 (MPK3) and MPK6 are two kinases involved in the MAP kinase signalling cascade, which act downstream of FLS2 to confer disease resistance 34 . The arlpk1-1 arlpk2-1 double mutant and the akik1-1 mutant showed significantly higher MPK3 and MPK6 levels compared with Col-0 WT after flg22 treatment (Fig. 6e). Unlike ARLPK1 and ARLPK2, AKIK1 does possess conserved catalytic residues in its kinase domain ( Supplementary Fig. 3). The kinase domain of AKIK1 was expressed and purified from Escherichia coli and subjected to an in vitro radiolabelled kinase assay. Consistent with its conserved catalytic residues, we were able to detect autophosphorylation activity for AKIK1, demonstrating that it is an active kinase (Fig. 6f)  SE is a positive regulator of plant immunity. SE was originally identified as a C 2 H 2 -type zinc-finger protein important in leaf and meristem development, inflorescence architecture and phase transition 28,35 . Since then, it has been shown to play a role in a crucial step of miRNA maturation and accumulation [18][19][20][21] , pre-mRNA splicing 36 and alternative splicing 24 . We hypothesized that as a target of miR863, SE should have a role in disease resistance; however, nothing has been uncovered so far regarding its function in antibacterial defence in plants.
We obtained the se-1 mutant (CS3257), which contains an X-ray-induced 7-base deletion, for analysis 21,28,35 . The se-1 mutant is smaller, has serrated leaves and exhibits pleotropic defects in shoot and leaf development 28 . We inoculated the se-1 plants with Pst (EV), Pst (avrRpt2) and a mock solution to determine the role of SE in disease resistance. The se-1 mutant showed increased disease susceptibility manifested as stronger chlorosis, or yellowing of the leaves, 3-4 d.p.i. (days post inoculation) compared with Col-0 WT plants. Infection with pathogen does not stunt growth of se-1 mutants (Fig. 7a). The se-1 mutant also showed delayed HR after infection with Pst (avrRpt2), indicating a weaker or slower RPS2-mediated defence response against Pst (avrRpt2) (Fig. 7b). We included Pst (EV) infection as a control, which did not trigger HR; however, at 48 h.p.i., the se-1 mutant showed chlorosis of the leaves (Fig. 7b) while Col-0 WT did not. We also measured the bacterial titre in se-1 mutant and found that both Pst (EV) and Pst (avrRpt2) growth was greater in the mutants compared with Col-0 WT plants (Fig. 7c). Consistent with the bacterial growth assays, PR1 protein expression is lower in se-1 mutant than in Col-0 WT after infection with Pst (EV) and Pst (avrRpt2) (Fig. 7d). FRK1 transcript level (Fig. 7e) and MPK6 and MPK3 protein levels (Fig. 7f) are lower in se-1 after flg22 treatment than in Col-0 WT. Furthermore, we obtained Arabidopsis transgenic lines overexpressing SE CDS in the se-1 background 21,24 (SE-OE) that have a complemented leaf phenotype and are slightly smaller than Col-0 WT plants (Fig. 7g). We confirmed the overexpression of SE in these lines (Fig. 7h), and found that they show enhanced disease resistance to Pst (EV) and Pst (avrRpt2) (Fig. 7i). Taken together, these data further support SE is a positive regulator of disease resistance. miR863 sequentially silences its targets to regulate defence.
Our results showed that the three confirmed targets of miR863 have opposite roles in plant immunity. While the ARLPKs are negative regulators of defence, SE is a positive regulator, thus explaining why overexpressing miR863 has no clear effect on disease resistance (Fig. 3e). Thus far, known miRNAs involved in immunity target either negative or positive regulators of plant immunity, not both. To find out why miR863-3p silences targets with opposite cellular functions, we performed a time course experiment and examined the levels of the targets during early and late stages after infection with Pst (avrRpt2). Because high inoculum concentration of avirulent Pst can trigger HR and thus affect the expression levels of many genes, we used a lower concentration (2 Â 10 6 colony-forming units (c.f.u.) per ml) that would not trigger macroscopic HR at 15 h.p.i., but could still induce miR863-3p levels. Real-time RT-PCR and western blot analysis revealed the kinases are downregulated early, just at 4 h.p.i. (Fig. 8a)  NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11324 ARTICLE later, around 12 h.p.i. (Fig. 8b). It may be that SE is not targeted by miR863 during earlier stages of infection because a higher level of miR863 is required for translation inhibition of SE, which is not achieved until the late stage of infection. After SE protein levels decrease, the accumulation of miR863-3p also decreases in the later stages of infection (Fig. 8c). We probed miR393 and miR319 as well and found that the levels of both also decrease in latter stages of infection similar to that of miR863-3p, showing that downregulation of SE protein does indeed affect miRNA accumulation ( Supplementary Fig. 9a). We also performed the time course experiment with Pst (EV) infection and found that the levels of ARLPK1 and ARLPK2 mRNA, miR863-3p and SE protein did not significantly change over time ( Supplementary  Fig. 9b). These results show that miR863-3p temporal regulation of its targets is more predominant in ETI, although ARLPK1, ARLPK2 and AKIK1 all have roles in negative regulation of both ETI and PTI.

Discussion
The initiation, timing and amplitude of plant defence responses are tightly controlled. Cellular reprogramming that occurs during pathogen infection is energy consuming; thus, a trade-off exists between plant growth and defence 37 . In recent years, several miRNAs and siRNAs with important roles in plant immunity have been discovered. However, no sRNAs that target genes with antagonistic roles in response to the same environmental or developmental cues have been reported in any cellular processes within any organisms. Here we identified miR863-3p, which is highly induced by Pst (avrRpt2) infection and sequentially silences negative and positive regulators of plant immunity through two different modes of action. miR863-3p first suppresses negative regulators-ARLPK1 and ARLPK2-via mRNA degradation to boost defence responses quickly after infection. ARLPK1 interacts with AKIK1, which also functions as a negative regulator. Then, during later stages of infection, miR863-3p downregulates SE via translational inhibition to attenuate defence signalling by bringing down the level of miR863-3p, of which accumulation is dependent on SE (Fig. 8c,d), thus forming a negative feedback loop to attenuate plant immunity. Thus, miR863-3p plays a critical role in regulating and fine-tuning the timely upregulation of plant defences on pathogen infection, as well as the attenuation of immune responses after successful defence. Just as the timely upregulation of plant defences on pathogen infection is crucial for survival, the attenuation of immune responses after successful defence may also be important to conserve host resources for growth and reproduction. Even though miR863-3p regulates ARLPK1, ARLPK2 and SE during ETI, all targets, as well as AKIK1, are regulators of not only ETI but also PTI. A recent systematic study of network characteristics in ETI and PTI revealed that both share a large subnetwork of genes and interactors that most likely functions to favour plant defence over growth 38 . However, the organization of defence response modules in ETI is independent, rather than  cohesive as it is in PTI, due to the evolutionary demand for a rapid and robust response to pathogen effectors 38 . The RLK/Pelle family has over 600 members and is the largest group of protein kinases in Arabidopsis 23 . Many RLKs are extensively involved in signalling pathways in development and plant immunity [39][40][41] . In the recent years, several RLKs that play a role in negative regulation of plant defence have been discovered, such as growth-promoting phytosulfokine (PSK) receptors PSKR1 and PSY1R (refs 42,43), and BAK1-interactors BIR1 (ref. 44) and BIR2 (ref. 45). Interestingly, the BIR2 kinase domain is missing conserved motifs and exhibits no kinase activity 45 . Nearly 20% of Arabidopsis RLKs are predicted to be pseudokinases and can have phosphorylation-independent mechanisms of mediating signal transduction 46 . Many pseudokinases have important functions in humans 47 , plants [48][49][50][51] and bacteria 52,53 , and still possess kinase catalytic activity or bind nucleotides and/or cations. They can act as scaffolding proteins or regulate the activity of functional kinases 25 . Neither ARLPK1 nor ARLPK2 exhibited detectable autophosphorylation in vitro (Fig. 4f) 45), localize in the plasma membrane, but we found that ARLPK1, ARLPK2 and AKIK1 were all predominantly localized to the ER ( Fig. 5 and Supplementary Fig. 6). ARLPK1 and ARLPK2 both have predicted signal peptides that overlap with their transmembrane domains ( Supplementary Fig. 2); thus, it is possible that the sequences are not true signal peptides. In contrast, AKIK1 has a clear signal peptide (Supplementary Fig. 7). It is possible that AKIK1 also localizes in other cellular organelles. Future experiments would involve obtaining native antibodies specific to AKIK1, ARLPK or ARLPK2, to determine their subcellular localization in vivo. The multigene family of hormone ethylene receptor kinases-including ETR1, ERS1, CTR1 and EIN2-are ER-membrane bound. Ethylene plays a role in many plant growth and development processes, as well as responses to environmental stresses, and its gaseous nature may explain why its receptors can localize in the ER 56 . Arabidopsis hormone cytokinin receptors-AHK2, AHK3 and CRE1/AHK4-which play roles in plant growth and formation, as well as in biotic and abiotic stress responses, are all membrane-bound kinases that mainly localize in the ER. These receptors may be localized in the ER for crosstalk or for close proximity to the nucleus for downstream signalling 57 . On the other hand, Arabidopsis AtIre1-1 and AtIre1-2 are transmembrane receptor kinases located in the ER and most likely function in ER unfolded protein response signalling, like their yeast and mammalian homologue Ire1 (ref. 58). How the roles of ARLPK1, ARLPK2 and AKIK1 are related to their subcellular localization in the ER remains to be determined. Finding their downstream components or other interacting partners may shed light on this question.
The role of SE in antibacterial plant defence has not been previously uncovered. However, dcl1, hyl1 and hen1, the mutants defective in miRNA accumulation, have decreased basal defence 12,14 . One study showed that SE might be involved in the cuticle integrity pathway, and related to that, se-1 mutants are more susceptible to necrotrophic fungal infection 59 . SE has been proposed to regulate gene expression by modifying chromatin structure 35 . As a component of the nuclear cap-binding complex, SE also has a role in pre-mRNA splicing 36 . This distinguishes SE from DCL1 and HYL1, which participate in only miRNA processing 24 . Other SE-dependent miRNAs are also likely to be involved during later infection stages to returning the host immune system to a normal state. Future research focusing on the role of additional SE-related miRNAs as well as ARLPK1 and ARLPK2 targets will enhance our understanding of temporal immune signalling.
Identification of target proteins of miR863. To identify miR863-3p target genes, a set of computational rules for target prediction was adapted from Allen et al. 61 and modified. The miRNA-target duplex must contain four or less unpaired bases, four or less G:U pairs, up to one single-nucleotide bulge and seven or fewer unpaired plus G:U positions. Nucleotides at positions 10 and 11 of the miRNA must be a perfect match with its target. Mismatched pairs or single-nucleotide bulges were given a score of 1; G:U pairs were given a score of 0.5. Mismatched and G:U pair scores were doubled if they were located within the core segment (positions 2-13). A maximum of three continuous mismatches was allowed if the mismatch region contained at least two G:U pairs, and the penalty score of the region was multiplied by 1.5. We used 5.5 as the cutoff score for selecting the miRNA targets. ; and a strain that has a mutation in its type III secretion system (DhrcC) 65 . For bacterial growth assays, 4-week-old Arabidopsis plants were syringe-infiltrated with a 5 Â 10 5 c.f.u. per ml bacterial suspension. Leaves were collected 3 d.p.i. using a cork borer, and bacterial titre was determined by serial dilution, plating and counting the colonies. At least 15 leaf discs were collected for each growth assay. Three biological repeats were performed with similar results. Student's t-test was used to determine the significant differences between mutants and control plants. For northern and western blots and the HR assay, a 1 Â 10 7 c.f.u. per ml bacterial suspension was used. A total of 18 leaves were infected for the HR assay and monitored for the appearance of HR symptoms. For the time course, a 2 Â 10 6 c.f.u. per ml bacterial suspension was used.
Generation of transgenic plants. To generate the construct for the MIR863-OE lines, the miR863 precursor was cloned using a miR319 backbone 66 into a pEarleyGate (pEG) 100 destination vector 67 using LR clonase II (Invitrogen). To generate the mARLPK1-OE lines, the mutated (m) ARLPK1 CDS was cloned into pEG104. Arabidopsis plants were transformed using floral dip method with Agrobacterium tumefaciens strain GV3101 carrying the carrying cloned vectors.
5 0 -RACE of mRNA cleavage products. 5 0 -RACE 69 was performed using FirstChoice RLM-5 0 -RACE kit (Ambion) following the manufacturer's instructions. Briefly, total RNA was extracted from Pst (avrRpt2)-infected tissue 10 h.p.i. and directly ligated to the RNA Oligo adaptor without further modification. Oligo (dT) primer was used to prime cDNA synthesis with reverse transcriptase. Gene-specific 5 0 -RACE reactions were done with the 5 0 Nested Primer and gene-specific primers ARLPK1-R 5 0 -RACE and ARLPK2-R 5 0 -RACE as listed in Supplementary Table 1. The 5 0 -RACE products were gel purified, cloned into pENTR (Invitrogen) vector and sequenced.
Protein kinase assays. ARLPK1 and ARLPK2 were cloned into pMAL-C4X with N-terminal fusions to maltose-binding protein (MBP). MBP-ARLPK1, MBP-ARLPK2 and MBP-RIPK were induced with 0.3 mM isopropyl-b-Dthiogalactoside for 3 h at 28°C and purified by amylose affinity chromatography from E. coli. The kinase domain of AKIK1 (547-892 aa) was cloned into pDEST-15 (Invitrogen) with an N-terminal fusion to glutathione S-transferase (GST). GST-AKIK1 was expressed in E. coli (BL21 strain) at 28°C for 4 h and the recombinant protein purified with Glutathione Sepharose 4B (GE Healthcare). Kinase assays were performed using 3 mg of recombinant protein with [g-32 P]-ATP. The assay was initiated by adding 1 ml (10 mCi) 32 P-ATP and incubated for 40 min at 30°C. The reaction was terminated by the addition of 3 Â laemmli loading buffer and subsequent incubation at 95°C for 5 min. The proteins were separated on a 12% SDS-PAGE gel and signals were visualized by X-ray film exposure 30 .
RNA extraction and analysis. Fresh tissue was ground in liquid nitrogen and RNA was extracted using TRIzol Reagent (Invitrogen) following the manufacturer's instructions. RNA is resolved on a 17% denaturing 8 M urea-PAGE gel and then transferred and chemically crosslinked onto a Hybond N þ membrane (GE Healthcare Life Sciences) using N-(3-Dimethylaminopropyl)-N 0 (3-Dimethylaminopr hydrochloride. miRNAs were detected using end-labelled oligonucleotide probes and exposed to a phosphor imager screen. ImageQuant TL 7.0 analysis software (GE Healthcare Life Sciences) was used to measure relative abundance levels. For quantification of relative gene expression, cDNA was synthesized using Superscript III (Invitrogen), and real-time RT-PCR was performed using SYBR green dye (Bio-Rad) on a MyiQ detection system (Bio-Rad). The primers for all experiments are listed in Supplementary Table 1. MAPK activity assay. Two-week-old seedlings grown on half-strength Murashige and Skoog medium were treated with 10 mM flg22 containing 0.01% Silwet L-77. Samples were collected at 0, 5, 10, 15 and 30 min, and analysed by western blotting using monoclonal rabbit phospho-p44/42 MAPK (Erk1/2) (Thr202/Tyr204) (D13.14.4E) XP antibodies (Cell Signaling Technology, #4370S, 1:2,000 dilution). a-tubulin was used as a loading control using monoclonal mouse anti-a tubulin (Sigma-Aldrich, T6074, 1:4,000 dilution). Subcellular localization. Subcellular localization of fluorescent-tagged proteins was determined using confocal microscopy. Three-week-old N. benthamiana plants were infiltrated with Agrobacterium carrying constructs ARLPK1-YFP, ARLPK2-YFP, AKIK1-CFP, RLK-CFP, ER-mCherry (ER-rk; CD3-959) or empty GFP vector. Untreated tissue was collected 48 h.p.i. Plants were treated with Pst (avrRpt2) 48 h.p.i. after infiltration with Agrobacterium and tissue was collected 12 h.p.i. For the plasmolysis treatment, tissue was treated with 5% NaCl for 5-10 min before visualization ( þ NaCl), or with water as a control (-NaCl). The ER marker is the mCherry fluorescent protein with the signal peptide of Arabidopsis wall-associated kinase 2 (AtWAK2) at the N terminus and the ER retention signal His-Asp-Glu at the C terminus 31 . The plasma membrane marker is an RLK (At4g23740) fused to CFP at the C terminus 70 .
Full uncropped versions of each gel and blot image are included in Supplementary Fig. 10. Results for biological replicates are included in Supplementary Fig. 11.