Pathogen-associated Molecular Pattern-triggered Immunity Involves Proteolytic Degradation of Core Nonsense-mediated mRNA Decay Factors During the Early Defense Response

One-sentence summary: Nonsense-mediated mRNA decay is a major regulatory process by which pattern recognition receptors fine-tune resistance gene transcript levels to reduce fitness costs and achieve effective immunity. The author responsible for distribution of materials integral to the findings presented in this article in accordance with the policy described in the instructions for authors (www.plant cell.org) ABSTRACT Nonsense-mediated mRNA decay (NMD), an mRNA quality control process, is thought to function in plant immunity. A subset of fully spliced (FS) transcripts of Arabidopsis thaliana resistance ( R ) genes are upregulated during bacterial infection. Here, we report that 81.2% and 65.1% of FS natural TIR-NBS-LRR (TNL) and CC-NBS-LRR


INTRODUCTION
Plant immunity, a counterattack mechanism against microbial infection, is exquisitely controlled by two different immune receptors known as extracellular immune receptors (pattern recognition receptors, PRRs) and intracellular immune receptors (resistance [R] proteins; nucleotide binding oligomerization domainlike receptors [NOD-like receptors or NLRs]; or nucleotide binding site-leucine-rich repeat [NBS-LRR] proteins) that recognize microbe-associated molecular patterns (MAMPs) and pathogenderived effectors, respectively. In general, pattern-triggered immunity (PTI), which is controlled by PRRs, confers moderate disease resistance to a broad spectrum of pathogens, and effectortriggered immunity (ETI), which is controlled by R proteins, is responsible for resistance to a specific pathogen carrying a cognate avirulence gene (Chisholm et al., 2006;Jones and Dangl, 2006). In addition to their predominance, R genes are also indispensable for establishing basal immunity to virulent pathogen infection and maintaining the balance between growth and defense (Li et al., 2001;Shirano et al., 2002;Palma et al., 2010;Maekawa et al., 2011;Karasov et al., 2017). The maintenance of the switched-off state and posttranslational modification of R proteins by other cellular components are crucial for maintaining the proper homeostasis of R activity (DeYoung and Innes, 2006;Li et al., 2015;Karasov et al., 2017). Transcriptional controls, such as histone modification and DNA methylation, are fundamental for fine-tuning R transcript levels (Halter and Navarro, 2015;Lai and Eulgem, 2018). Alternative splicing (AS) can secure the accumulation of diverse transcript isoforms for full immunity (Dinesh-Kumar and Baker, 2000;Zhang and Gassmann, 2007;Xu et al., 2012;Yang et al., 2014). A few fully spliced (FS) and AS versions of R transcripts are targets of miRNAs and nonsense-mediated mRNA decay (NMD; Zhai et al., 2011;Shivaprasad et al., 2012;Gloggnitzer et al., 2014;Zhang et al., 2016), indicating that R gene expression is also under posttranscriptional control in plants.
NMD is a translation-coupled mRNA surveillance process in eukaryotes that prevents mRNAs containing premature termination codons (PTCs) from being translated. PTC-containing transcripts can arise from mutations or AS, and NMD is usually triggered by the presence of a downstream splice junction or long 39 untranslated region (UTR; Rebbapragada and Lykke-Andersen, 2009;Peccarelli and Kebaara, 2014). FS natural transcripts can also be subject to NMD by virtue of containing NMD-triggering characteristics such as one or more introns in the 39UTR, long 39UTRs, or upstream open reading frames (uORFs) encoding no fewer than 35 amino acids (Kertész et al., 2006;Kerényi et al., 2008;Nyikó et al., 2009;Kalyna et al., 2012;Le Hir et al., 2016).
The canonical NMD process occurs when a translating ribosome encounters a PTC and fails to properly terminate translation. This leads to the phosphorylation and activation of the core NMD factor UP-FRAMESHIFT1 (UPF1), a process promoted by the interaction of UPF1 with UPF2 and UPF3, which associate with a downstream exon-junction complex (EJC; Kim et al., 2001;Banihashemi et al., 2006;Schweingruber et al., 2013). In a transient wild tobacco (Nicotiana benthamiana) assay, NMD factor SUPPRESSOR FOR MORPHOLOGICAL DEFECTS OF GENITALIA7 (SMG7) bound to the phosphorylated C-terminal region of UPF1 for subsequent recruitment of the decapping and deadenylation complexes before the decay of the target mRNA . A recent study provided evidence that the dephosphorylation of UPF1 involves an SMG7 function (Kesarwani et al., 2019). NMD is involved in flowering, Suc metabolism, and stress responses in plants (Yoine et al., 2006;Jeong et al., 2011;Rayson et al., 2012;Riehs-Kearnan et al., 2012;Shi et al., 2012;Garcia et al., 2014;Gloggnitzer et al., 2014;Filichkin et al., 2015;Sureshkumar et al., 2016).
The most noteworthy function of NMD is its participation in the regulation of plant immunity. A null mutation in UPF1 or UPF3 confers disease resistance via increased levels of salicylic acid (SA) and elevated expression of defense genes, but this mutation also causes seedling lethality and late flowering in Arabidopsis (Arabidopsis thaliana; Jeong et al., 2011;Rayson et al., 2012;Riehs-Kearnan et al., 2012;Shi et al., 2012). A subset of FS natural TIR-NBS-LRR (TNL)-type R gene transcripts carrying NMD-eliciting features accumulated with known AS-coupled NMD (AS-NMD) transcripts in both smg7-1, an NMD-compromised Arabidopsis mutant showing autoimmunity, and a wild-type plant infected with Pseudomonas syringae pv tomato DC3000 (PstDC3000; Gloggnitzer et al., 2014). In addition, the autoimmune phenotype of smg7-1 requiring RESISTANT TO P. SYRINGAE6 (RPS6) was suppressed by the simultaneous mutation of PHYTOALEXIN DEFICIENT4 (PAD4) or ENHANCED DISEASE SUSCEPTIBILITY1 (EDS1), which are important for TNL-dependent signaling (Gloggnitzer et al., 2014). These findings suggest that autoimmunity in NMDcompromised mutants results from the activation of TNL-type R-dependent immunity and that bacterial infection might immediately suppress the NMD machinery in Arabidopsis. However, the pathway and timing of the decrease in NMD efficiency during Pseudomonas infection are still unclear.
Here, we analyzed the architecture of R genes expressed in NMD-compromised Arabidopsis mutants and found that most R genes produce transcripts that are targets of NMD. The recognition of bacterial infection by PRRs leads to the accumulation of a group of TNL-and CC-NBS-LRR (CNL)-type transcripts by triggering the decay of UPF proteins through the ubiquitin-26S proteasome system (UPS). These findings demonstrate that the expression of a subset of R genes is controlled by NMD in the face of pathogen infection.

Simultaneous Mutation of UPF1 and UPF3 Causes Severe Autoimmune Responses
Arabidopsis mutants with a defect in any of the genes encoding NMD core components (e.g., UPF1, UPF3, and SMG7) accumulate NMD-sensitive transcripts and exhibit autoimmunity and developmental arrest (Jeong et al., 2011;Drechsel et al., 2013;Gloggnitzer et al., 2014). When we performed reciprocal crosses of upf1-5 and upf3-1, the resulting double mutants upf1-5 upf3-1 and upf3-1 upf1-5 exhibited distinctly dwarfed, curly leaves and seedling lethal phenotypes when grown under typical growth conditions (22°C, 16-h day/8-h night) but not when grown at a high temperature (28°C,Figures 1A and 1B). When the plants were grown at 22°C under a 12-h day/12-h night photoperiod, the growth-arrested phenotypes were partially rescued in the double mutants (Supplemental Figure 1A). Thus, we used Arabidopsis plants grown under this neutral-day condition for pathophysiological studies.
To determine whether the simultaneous mutation of UPF1 and UPF3 would also reinforce resistance responses against bacterial infection, we analyzed the responses of wild-type, upf1-5, upf3-1, upf1-5 upf3-1, and upf3-1 upf1-5 plants to infection with PstDC3000 and Pseudomonas cannabina pv alisalensis ES4326 (PcaES4326; formerly P. syringae pv maculicola ES4326; Bull et al., 2010). Although the number of bacteria in upf1-5 upf3-1 and upf3-1 upf1-5 leaves was comparable to that in upf1-5 and upf3-1, these double mutants, but not any single mutant, exhibited severe necrotic symptoms 3 d after Pseudomonas infection ( Figure 1C). Considering the atypical symptom development observed in upf1-5 upf3-1 and upf3-1 upf1-5 after virulent Pseudomonas infection, we tested whether basal defense responses are boosted in these double mutants after infection. The expression of FLG22-INDUCED RECEPTOR-LIKE KINASE1, NDR1-HIN1-LIKE10 (MAMPresponsive genes), and PATHOGENESIS-RELATED PROTEIN1 (PR1) was not significantly altered in the double mutants after PstDC3000 infection compared with the single mutant or wildtype plants (Supplemental Figures 1B and 1C). Before infection, however, the expression levels of these genes were higher in the upf3-1 upf1-5 mutant than in the other plants. These results suggest that immune-related genes are positively regulated in the double mutant. In addition, accelerated SA accumulation was observed in the double mutant both before and after PstDC3000 infection (Supplemental Figure 1C). Next, we examined the level of electrolytic leakage in Arabidopsis leaves after PstDC3000 infection. Electrolyte leakage was more severe in the leaf disks of upf3-1 upf1-5 than in the wild type, upf1-5, and upf3-1 grown at 22°C during infection, while the level in every single mutant was not significantly different from that of wild-type plants (Supplemental Figure 1D, left representation). (A) and (B) Five-week-old Arabidopsis ecotype Columbia-0 (wild type [WT]), upf1-5, upf3-1, upf1-5 upf3-1, and upf3-1 upf1-5 plants were grown at 22 6 1°C (A) or 28 6 1°C (B) with a 16-h d/8-h night photoperiod. Simultaneous mutation of UPF1 and UPF3 arrested Arabidopsis growth at 22°C, while the growth defect was moderately recovered by a temperature shift to 28°C. Scale bars 5 5 cm. (C) Bacterial counts and disease symptoms in the wild type (WT) and NMD-compromised mutants 3 d after PstDC3000 and PcaES4326 infection. Different letters (a and b) indicate statistically significant differences. P < 0.01; one-way ANOVA. Error bars indicate SE (n 5 8).
(D) Venn diagrams of TNL-and CNL-type R genes illustrating the number of genes carrying different NMD features. Numbers in parentheses are the total number of expressed genes (*) and the total number of genes bearing each NMD event (**). (E) to (G) Stability of NMD reporters (E), representative TNL transcripts (F), and representative CNL transcripts (G). RT-qPCR analysis was performed using total RNA extracted from ActD-treated leaves of the wild type (WT) and upf3-1 upf1-5 (u3u1) mutants that had been collected at 0, 1, 2, and 4 hpt. The However, an increase in electrolyte leakage was not observed in the double mutant grown at 28°C (Supplemental Figure 1D, right representation). The transcripts of 39 TNL-type R genes were more stable in smg7 pad4 than in pad4, 20 of which showed NMDsensitive features (Gloggnitzer et al., 2014). To determine whether the accumulation of TNLs is a representative physiological characteristic in these upf mutants, we measured the mRNA levels of selected TNLs in plants under both NMD-compromised conditions (upf1-5, upf3-1, upf1-5 upf3-1, upf3-1 upf1-5, and cycloheximide [CHX]-treated wild type) and wild-type plants grown at two different temperatures. We analyzed well-characterized FS R gene transcripts showing NMD sensitivity (Gloggnitzer et al., 2014). We treated wild-type and mutant plants with actinomycin D (ActD) for 4 h to inhibit transcriptional elongation (Sobell, 1985), as described in a previous analysis of upf3-1, in which the 4-h treatment was found to be optimal without damaging plant tissues (Hori and Watanabe, 2005). This blockage of de novo transcription enables the NMD sensitivity of individual transcripts to be evaluated at the post-transcriptional level (Kurosaki et al., 2014). The known NMD-sensitive TNL transcripts (Gloggnitzer et al., 2014) accumulated under potent NMD-compromised conditions (in the double mutants and CHX-treated wild-type plants) regardless of temperature (Supplemental Figure 2). These results suggest that double mutants possessing an excessive amount of TNL transcripts remain under the control of temperaturedependent EDS1-mediated immune responses, as described for smg7-1 Alcázar and Parker, 2011;Heidrich et al., 2013;Carstens et al., 2014;Gloggnitzer et al., 2014;Stuttmann et al., 2016). Notably, the differences in transcript levels among the wild type and upf mutants were reduced when de novo transcription continued to occur compared with ActD-treated plants (DMSO in Supplemental Figure 2), indicating that continuous transcription is necessary for maintaining a steady-state level of R gene transcripts in Arabidopsis (Drechsel et al., 2013). Altogether, we propose that the accumulation of a variety of NMDsensitive R transcripts leads to enhanced basal immunity in these mutants.

A Significant Number of R Gene Transcripts Retain NMD-Sensitive Characteristics
To comprehensively examine the extent to which the expressed R transcripts differ both quantitatively and qualitatively across wildtype and single-and double-mutant plants, we analyzed the FS natural transcripts and the AS variants of TNLs and CNLs based on our RNA-sequencing (RNA-seq) data of ActD-treated plants (Supplemental Data Set 1) and determined whether each transcript carried NMD-eliciting features (Supplemental Data Set 2; Supplemental Files 1 to 7). Among the 85 TNL and 43 CNL genes expressed, 19 and 8 of the TNL (22.3%) and CNL genes (18.6%) analyzed, respectively, contained at least one intron in the 39UTR.
Next, we tested the stability of these NMD-susceptible FS R transcripts in upf3-1 upf1-5 mutants by measuring the half-lives of R transcripts from leaf samples taken from wild-type and mutant plants at 0, 1, 2, and 4 h after blocking de novo transcription. We used At2g29210 and LYSOPHOSPHATIDYLETHANOLAMINE ACYLTRANSFERASE2 (LPEAT2) AS variants (PTC1) as NMD controls (Gloggnitzer et al. 2014). The half-lives of these NMD markers and R transcripts carrying the NMD-sensitive features were elevated in upf3-1 upf1-5 compared with normal-looking transcripts showing similar levels of stability between wild type and upf3-1 upf1-5 (Figures 1E to 1G; Supplemental Figure 3), indicating that NMD is one of the main processes controlling the expression of TNL and CNL genes. Other transcripts bearing NMD-sensitive features, however, showed no differences in RNA stability between wild type and upf3-1 upf1-5, e.g., RPP5 (Supplemental Figure 3). This result suggests that not all R genes bearing NMD-sensitive features are genuine NMD targets, as widely observed in eukaryotic genes (Rebbapragada and Lykke-Andersen, 2009;Kurosaki et al., 2014;Peccarelli and Kebaara, 2014).
To determine whether FS natural TNL and CNL transcripts are post-transcriptionally regulated, we measured the transcription rates of the randomly selected genes by performing a chromatin immunoprecipitation (ChIP) assay of wild-type and upf3-1 upf1-5 plants using an a-RNA polymerase II (a-RNAPII) C-terminal domain antibody, and primer sets specific to regions between the transcription and translation initiation sites. The ChIP signals in the TNL genes were largely constant between the wild type and upf3-1 upf1-5 (Supplemental Figure 4A), indicating that most TNL-type R genes with NMD-sensitive features are constitutively transcribed in Arabidopsis leaves and that the resulting transcripts are subject to immediate turnover by NMD in the wild type. Unlike TNLs, half of half-lives were calculated by nonlinear least-squares regression analysis (average 6 SD, n 5 3; three biological replicates with three technical repeats). Stability analyses of other R transcripts are provided in Supplemental Figure 3. Sequences of the individual primers used in this study are presented in Supplemental Data Set 3. the CNL genes tested here showed higher enrichment of the RNAPII elongation complex in upf3-1 upf1-5 than in wild type, e.g., RPP7 and RPM1 (Supplemental Figure 4B), suggesting that these genes are transcriptionally upregulated in the upf3-1 upf1-5 mutant. Taken together, these findings support the notion that NMD is one of the regulatory pathways controlling the steadystate levels of both FS and AS R transcripts carrying NMDsensitive features (Gloggnitzer et al., 2014).

Bacterial Infection Results in Turnover of UPF Proteins
Because bacterial infection upregulates a subset of NMDsensitive transcripts (Gloggnitzer et al., 2014), we monitored the transcript levels of core NMD factors in PstDC3000-infected wildtype leaves to detect any decrease in transcript levels upon bacterial infection. The FS natural transcripts encoding Arabidopsis UPF1, UPF3, and SMG7 are known NMD substrates Gloggnitzer et al., 2014;Degtiar et al., 2015;Kesarwani et al., 2019). Our RNA-seq data also revealed that the UPF1 transcript, a possible NMD target, carries a 508 nucleotides-long 39UTR (Supplemental Figure 5A). These transcripts were upregulated after PstDC3000 infection, while the level of the FS UPF2 transcript remained constant (Supplemental Figure 5A). The upper products, which accumulated during infection, were identified as variants of UPF1 and UPF2 produced by unspliced introns 26 and 14, respectively, and they eventually retained NMD-triggering features owing to the long 39UTR containing two introns (single and double asterisks, Supplemental Figures 5A and 5B). The results indicate that the plant NMD system becomes impaired at an early stage of infection, despite the increasing levels of transcripts of NMD factors (Kerényi et al., 2008;Nyikó et al., 2013;Gloggnitzer et al., 2014;Degtiar et al., 2015;Shaul, 2015).
The inconsistency between the increased transcript levels of UPF genes and the reduction in NMD activity after infection (Gloggnitzer et al., 2014) suggests that bacterial infection might disrupt the plant NMD system by affecting the integrity of the machinery at the protein level. To address this hypothesis, we prepared monoclonal antibodies against UPF1 (a-UPF1), UPF2, and UPF3 and used them for immunoblot analysis (Supplemental Figures 6B and 6C). a-UPF1 detected a faint signal of UPF1 in the knockdown mutant upf1-5 and in the putative knockdown mutant upf1-4. However, the protein was undetectable in the amino acid substitution mutant upf1-1 for unknown reasons (Supplemental Figure 6B). We examined the levels of UPF1, UPF2, and UPF3 in wild-type leaves infected with PstDC3000 using a-UPF1, a-UPF2, and a-UPF3. The levels of UPF1 and UPF3 were reduced to <40% (UPF1) to 50% (UPF3) within 0.5 h post inoculation (hpi) and became undetectable from 10 to 30 hpi, whereas the level of UPF2 was maintained for up to 6 hpi. All three NMD factors were undetectable by 20 hpi (Figure 2A). The turnover of UPF1 occurred more quickly than that of UPF3, i.e., UPF3 was still detectable at 10 hpi when UPF1 was absent ( Figure 2A, upper representation).
Because UPF1 decay began as early as 30 min post inoculation (mpi; Figure 2A, upper representation), we wondered how early the turnover of UPF proteins was initiated. We monitored the stability of UPF proteins at various time points beginning at 1 mpi with PstDC3000. Surprisingly, the reduction in UPF1 levels was clearly observed at 5 mpi, and a sharp decrease occurred between 15 and 20 mpi, after which only ;50% UPF1 remained. A drastic decrease in UPF3 levels was observed between 15 and 25 mpi, during which >90% of UPF2 remained ( Figure 2A, lower representation). These findings suggest that the turnover of UPF proteins is an early response to PstDC3000 infection.
To determine whether other protein components involved in NMD (Kerényi et al., 2008) are also unstable upon bacterial infection, we generated transgenic Arabidopsis plants overexpressing the SMG7 and EJC proteins. SMG7 and the other EJC components-Y14, MAGO, eIF4AIII, and CASC3/Barentsz (BTZs)-were stably maintained during infection compared with the overexpressed UPF proteins, although the overexpressed versions of UPF proteins (the first to third rows of the left representation in Figure 2B) were destabilized later than the native proteins ( Figure 2B). Thus, we conclude that bacterial infection initiates the specific destruction of UPF proteins.
To determine whether the decay of UPF proteins is initiated by only virulent Pseudomonas infection, we inoculated wild-type plants with an attenuated strain, PstDC3000 hrcC 2 , and two avirulent derivatives of PstDC3000 carrying AvrRpm1 or AvrRps4 and subjected them to immunoblot analysis using an anti-UPF1 polyclonal antibody. Like the turnover pattern of UPF1 in PstDC3000-infected plants, infection with these different strains also triggered the decay of UPF1 proteins in Arabidopsis (Supplemental Figures 7A and 7B), indicating that the sensing of P. syringae is sufficient to initiate the decay of the core NMD factors. Unlike the immunoblotting results obtained using monoclonal a-UPF1 antibody, the UPF1 level at 6 hpi was somehow recovered, which was more pronounced after infection with the PstDC3000 hrcCstrain. Furthermore, a putative UPF1 degradation product (single asterisk in Supplemental Figure 7) was also observed in parallel with UPF1 dynamics and was not detected in the blots using monoclonal a-UPF1 antibody. We speculate that UPF1 might undergo a conformational change during the decay process and that the monoclonal antibody is unable to cross react with this transitional form of UPF1. The detection of a small product by the polyclonal antibody even before infection (single asterisk in Supplemental Figure 7) suggests that the UPF1 level may be controlled posttranslationally under normal conditions.

UPF1 and UPF3 Are Destroyed Via Ubiquitination at an Early Stage of Infection
To examine whether UPF proteins are post-translationally degraded during an early stage of infection, MG132, an inhibitor of the proteasome, was coinfiltrated into wild-type Arabidopsis leaves with PstDC3000. While the coapplication of the solvent DMSO alone with PstDC3000 did not affect the protein dynamics of UPF1, UPF2, or UPF3, MG132 coinfiltrated with PstDC3000 interrupted the infection-induced destruction of these proteins ( Figure 3A). These results support our observation that the reduction and subsequent disappearance of UPF proteins resulting from Pseudomonas infection ( Figure 2) is due to the decay of UPF proteins rather than the translational inhibition of mRNAs encoding these NMD factors.
To confirm that the decay of the UPF proteins occurs via the UPS during bacterial infection, we monitored the levels of ubiquitinated UPF proteins in the a-UPF1, a-UPF2, and a-UPF3 immunoprecipitates by immunoblot analysis using an a-ubiquitin (a-UBQ) antibody. The ubiquitination of UPF1 and UPF3 was strongly induced at 30 mpi and 1 hpi, respectively ( Figure 3B; Supplemental Figure  8A), which is consistent with the dynamics of these proteins (A) Dynamics of UPF1, UPF2, and UPF3 proteins up to 30 hpi (top) and 50 mpi (bottom). Immunoblot analyses (left) were performed for leaf samples taken from wild-type Col-0 plants that had been infected with Pseudomonas and collected at the indicated time points using an anti-UPF1 monoclonal antibody (a-UPF1), a-UPF2, or a-UPF3. The right shows UPF protein levels in infected leaves at the indicated time points (average 6 SD, n 5 4). Different letters above the bars (i.e., a, b, c, d, ab, and bc) indicate statistically significant differences (P < 0.05, one-way ANOVA). Successful infection was verified by examining the levels of PR1. Both experiments were performed with four biological replicates. (B) The NMD factor SMG7 and the EJC core protein components, but not UPF1, UPF2, or UPF3, remained stable during PstDC3000 infection. The leaves of transgenic Arabidopsis plants (T 3 ) stably expressing protein A-fused UPF1, UPF2, UPF3, SMG7, Y14, MAGO, elF4A-III, BTZ1, and BTZ2 were infected with PstDC3000 and collected at the indicated time points after infection for immunoblot analysis. GFP-protein A was used as a representative stable protein.
The recombinant proteins were detected using an a-PAP antibody. M, prestained protein ladder.
( Figure 2A). These results suggest that the ubiquitination of UPF1 and UPF3 is a temporally controlled early response in plant immunity. UPF1 was already ubiquitinated before infection ( Figure  3B), which likely explains the previous findings that the turnover of UPF1 occurs within 4 h in ActD-treated upf1-5 plants (Supplemental Figure 6C) and that the putative degradation product was detected by polyclonal a-UPF1 antibody even before infection (single asterisks in Supplemental Figure 7). These results suggest that, in Arabidopsis, UPF1 levels are also regulated by ubiquitination under normal growth conditions. The induced ubiquitination signals of UPF1 and UPF3 were maintained for up to 6 and 10 hpi ( Figure 3B), respectively, when the proteins were still present ( Figure 2A). The ubiquitinated UPF1 and UPF3 signals, however, were no longer evident at 10 hpi and 20 hpi, respectively, and at 20 hpi, both were nearly or entirely undetectable ( Figure 3B), indicating that the induction of UPF1/UPF3 ubiquitination is responsible for the decay of these proteins. To validate this result, we treated plants with MG132 and PstDC3000 infection and found that the induced level of UPF1 and UPF3 ubiquitination signals was maintained through 20 hpi ( Figure 3C; Supplemental Figure 8B), supporting the suggestion from Figure 3B that the disappearance of the induced ubiquitination signals was due to the absence of both proteins at this stage. We did not detect multiple bands larger than UPF1 and UPF3 (estimated molecular masses of 136.9 and 54.0 kD, respectively), suggesting that these proteins are degraded by the 26S proteasome via monoubiquitination or multi-monoubiquitination rather than polyubiquitination (Shabek et al., 2012;Ciechanover and Stanhill, 2014;Braten et al., 2016).
UPF2, which is also decayed during PstDC3000 infection (Figures 2A and 3A), was not ubiquitinated after infection ( Figures  3B and 3C). Because UPF2 and UPF3 are present in complexes from the nucleolus to the cytoplasm (Kerényi et al., 2008;Kim et al., 2009), we reasoned that UPF2 might enter the 26S proteasome (A) Effects of the 26S proteasome inhibitor MG132 on the decay of UPF1, UPF2, and UPF3 during infection. DMSO alone or MG132 dissolved in DMSO was coinfiltrated into wild-type leaves with PstDC3000. Immunoblot analyses were performed using a-UPF1, a-UPF2, or a-UPF3. Successful infection was verified by examining the levels of PR1. (B) Ubiquitination of UPF1 and UPF3, but not UPF2, in wild-type leaves after PstDC3000 infection. Immunocomplexes with the a-UPF1, a-UPF2, or a-UPF3 antibody were subjected to immunoblotting with a-UBQ. (C) IP assay with proteins extracted from MG1321PstDC3000-treated wild-type plants. The levels of UPF proteins in IP complexes in (B) and (C) are shown in Supplemental Figure 8A and 8B, respectively. (D) Degradation of UPF proteins in N. benthamiana leaves infected with P. syringae pv syringae B728a (PssB728a). (E) Induction of UPF1 and UPF3 ubiquitination in N. benthamiana leaves during PssB728a infection. Levels of ubiquitinated UPF1 and UPF3 were determined by IP with a-UPF1 or a-UPF3 and subsequent immunoblot analysis with a-UPF1, a-UPF3, and a-UBQ. complex with UPF3 during infection. Indeed, UPF2 coimmunoprecipitated with UPF3 under both uninfected and PstDC3000infected conditions (Supplemental Figure 9). To test whether UPF1/UPF3 decay upon bacterial challenge is universal among plants, we subjected N. benthamiana plants to infection with the known bacterial pathogen P. syringae pv syringae B728a (PssB728a). The UPF proteins of the N. benthamiana plants were also destroyed by 30 hpi ( Figure 3D). The induction of UPF1 and UPF3 ubiquitination was detected at 2 and 6 hpi, respectively ( Figure 3E), suggesting that plant immunity involves the suppression of NMD via the control of UPF protein levels in members of the Solanaceae family.
Because TNL genes are known to be deregulated upon Pseudomonas infection due to decreased NMD efficiency (Gloggnitzer et al., 2014), we investigated whether the kinetics of the NMDsensitive TNL and CNL transcripts reflect the decay of UPF proteins upon PstDC3000 infection. We monitored the stability of TNL and CNL transcripts from ActD-treated infected leaves taken at 0, 1, 2, and 4 hpi by RT-qPCR analysis. The transcripts of NMD markers as controls, namely, At2g29210 and LPEAT2 retaining PTC (At2g29210 PTC1 and LPEAT2 PTC1), were more stably maintained in infected wild-type leaves than in noninfected leaves ( Figure 4A). Similarly, the TNL and CNL transcripts bearing NMD-sensitive features, e.g., SIKIC3, RPS6, RPM1, and RPP7, were more long-lived in infected leaves than in noninfected leaves (Figures 4B and 4C;Supplemental Figure 10). The half-lives of SNC1 and RPP5, which are representative targets of small-interfering RNA (Yi and Richards, 2007), and non-NMD targets, namely, SUMM2 and RPS4, were similar between infected and noninfected leaves (Figures 4B and 4C; Supplemental Figure 10). The gaps between the NMD target R transcripts in infected and noninfected plants were less distinct than those between upf3-1 upf1-5 and wild type ( Figures 1F, 1G, 4B, and C4C; Supplemental Figures 3 and 10), suggesting that NMD may have still occurred by 4 hpi, when UPF1 is still present (Figures 2A and 3B).
In parallel, we analyzed the accumulation of R transcripts in infected leaf samples taken at the same time points (Figures 2 and 3) used for protein analysis. As a result, we were able to sort the R genes bearing NMD-sensitive features into three groups. The early-response R genes, including RPM1 and At3g44400, were upregulated as early as 30 mpi of PstDC3000 (Figure 4; Supplemental Figure 11), suggesting that the transcripts of these genes are instantly stabilized when the UPF1 level is reduced to <40%. The second group, the so-called late-response genes, e.g., SOC3, At3g14470, At4g14610, and At5g38340, appeared to be upregulated later than 10 hpi, when UPF1 is completely decayed (Supplemental Figure 10). The nonresponsive genes, including RRS1 and At5g35450, were not regulated by Pseudomonas infection (Supplemental Figure 10).
The miRNA target CNL-type R genes At1g50180 and RSG2 (Boccara et al., 2015;Cai et al. 2018), carrying uORF/long 39UTR and long 39UTR, respectively, are thought to also be targets of NMD, because their transcripts are stabilized upon Pseudomonas infection (Supplemental Figure 10B). By contrast, the half-lives of other known miRNA targets including RPS5, At1g12290, and At1g61180 were not altered or shortened by the mutation of UPF1/ UPF3 or Pseudomonas infection (Supplemental Figure 11A), suggesting that the steady-state level of a subset of CNL transcripts bearing NMD-sensitive features may be under the control of miRNA-mediated gene silencing rather than NMD. Moreover, three TNL-type R transcripts, At3g44400, At4g14370, and At5g41740, which were upregulated as early as 30 mpi by Pseudomonas infection, were more stable in infected versus noninfected plants. By contrast, the stability of these transcripts was reduced in upf3-1 upf1-5 versus wild-type plants (Supplemental Figure 11B). These results suggest that susceptibility to NMD varies across FS R transcripts carrying NMDsensitive features (Rebbapragada and Lykke-Andersen, 2009;Kurosaki et al., 2014;Peccarelli and Kebaara, 2014) and that the control of the expression of these genes might involve other processes, e.g., transcriptional regulation, under a variety of conditions. Based on these results, we propose that NMD is one of the main posttranscriptional control pathways involved in maintaining R transcript levels in Arabidopsis during the immunity response.

The Induction of UPF3 Ubiquitination Partially Depends on UPF1
The Cys-and His-rich domain of UPF1, which serves as a Really Interesting New Gene-related E3 ligase, is responsible for the UPF3-dependent self-ubiquitination required for activation of NMD in yeast (Takahashi et al., 2008). To determine whether UPF3 is required for UPF1 ubiquitination and vice versa in Arabidopsis, we monitored the ubiquitination and turnover of UPF proteins in upf1-1, upf2-12, and upf3-1 plants at 1 h after PstDC3000 infection. The ubiquitination level of UPF1 was similar across the wild type, upf2-12, and upf3-1 ( Figure 5A; second row in Figure 5B), suggesting that the induction of UPF1 ubiquitination upon PstDC3000 infection occurs independently of UPF2 or UPF3. By contrast, the level of ubiquitinated UPF3 decreased 1.24 6 0.042-fold in upf1-1 compared with the wild type (fourth row in Figure 5B; sixth row in Figure 5C). The UPF3 level stabilized from 1 to 10 hpi in upf1-1 compared with wild type (third and sixth rows in Figure 5C), most likely due to decreased ubiquitination. These results are consistent with the observation that the decay of UPF1 occurs before that of UPF3 (Figures 2A and 3B and 3C), and that UPF1 protein itself or ubiquitinated UPF1 modulates the ubiquitination of UPF3 upon bacterial infection. Bacterial infection induces the ubiquitination of UPF1/ UPF3 for the decay of three UPF proteins, which differs from the UPF3-dependent self-ubiquitination of UPF1 required for NMD in yeast (Takahashi et al., 2008).

Microbial Perception by PRRs Initiates the Decay of UPF Proteins
NMD-sensitive transcript levels also increased in wild-type plants treated with 1 mM of the bacterial elicitor flg22, while the expression of these transcripts was suppressed in flagellin sensitive2 (fls2) mutants (Gloggnitzer et al., 2014), suggesting that MAMP recognition can also affect the integrity of UPF proteins. Compared with mock treatment, in which the levels of UPF proteins were unchanged, both UPF1 and UPF3 started to degrade at 1 and 2 h post treatment (hpt) with flg22 in wild-type plants, respectively, (A) Stability of NMD markers At2g29210 PTC1 and LPEAT2 PTC1 in infected wild-type leaves. (B) and (C) Transcript levels and half-lives of representative TNL transcripts, the non-NMD target RPS4, and SIKIC3 and RPS6 bearing NMD-sensitive features (B), and CNL transcripts, the non-NMD target SUMM2, and the NMD-targets RPM1 and RPP7 (C). Left, steady-state levels of selected R transcripts were measured in infected wild-type leaves during PstDC3000 infection without ActD treatment. Data points are the average 6 SD of three biological replicates with two technical repeats (n 5 3). Different letters (i.e., a, b, c, ab, and bc) above the bars indicate statistically significant differences (P < 0.01, one-way ANOVA). Right, either ActD alone (cyan) or ActD together with PstDC3000 (magenta) was infiltrated into wild-type leaves, and the treated leaves were collected at the indicated time points. Data points are the average 6 SD of three biological replicates with three technical repeats (n 5 3). Sequences of the gene-specific primers are presented in Supplemental Data Set 3. while UPF2 levels were significantly reduced at 6 h ( Figure 6A). flg22 treatment did not trigger the decay of UPF proteins in fls2 or the co-receptor mutant bri1-associated receptor kinase1-5 (bak1-5), although UPF1 in bak1-5 disappeared within 30 h ( Figure 6B). These findings suggest that the flg22-FLS2 association is pivotal for the effect of flg22 on UPF proteins. Consistent with UPF degradation via the UPS during Pseudomonas infection, flg22 treatment also caused the ubiquitination of UPF1 and UPF3 but not UPF2, and induced ubiquitination was observed at 1 and 2 hpt for UPF1 and UPF3, respectively ( Figure 6C; Supplemental Figures 8C and 8D), when these proteins began to be destroyed ( Figure 6A). Ubiquitination of UPF1 and UPF3 was uninducible in fls2 or delayed to 20 hpt in bak1-5 compared with 1 to 2 hpt in wild-type plants ( Figure 6D). These results indicate that the perception of MAMPs is sufficient for initiation of the ubiquitination of UPF1 and UPF3 during infection. The decay of UPFs was still triggered in fls2 and bak1-5 after PstDC3000 infection, although the proteins were more stable in these mutants than in the wild type ( Figure 6E). The relatively high dose of flg22 described above did not induce as strong a response in UPF proteins as PstDC3000 infection (compare Figures 2A and 3B with Figures 6A and 6C, respectively). These findings suggest that microbial sensing by PRRs leads to the decay of UPF proteins, that other pathogenderived molecules might simultaneously act during this early event, and that unidentified bacterial effectors might also contribute to the decay process. (A) Levels of UPF1, UPF2, and UPF3 in the leaves of the wild type (UPF1 UPF3 and UPF2) and the mutant (upf1-1 UPF3, UPF1 upf3-1, and upf2-12) plants before and after PstDC3000 infection (1 hpi). M, prestained protein ladder. (B) Induction of UPF1 and UPF3 ubiquitination in wild-type and mutant plants. Protein complexes with the a-UPF1 or a-UPF3 antibody were subjected to immunoblot analyses with a-UBQ. Notably, depletion of UPF3 did not affect the induction of UPF1 ubiquitination (second row), while the induced ubiquitination signal of UPF3 was reduced in the upf1-1 allele (fourth row). The experiment was performed with three biological replicates, which showed similar results. M, prestained protein ladder (C) Depletion of UPF1 reduces the ubiquitination level of UPF3 during PstDC3000 infection. Total proteins were extracted from the wild type (WT) and upf1-1 that had been infected with PstD3000 and immunoprecipitated with a-UPF3 to examine whether UPF3 is differentially ubiquitinated between the wild type and upf1-1 mutant at the indicated time points. The immunocomplexes were subjected to immunoblotting with a-UPF3 or a-UBQ. M, prestained protein ladder.

MITOGEN-ACTIVATED PROTEIN KINASE3 and MITOGEN-ACTIVATED PROTEIN KINASE6 Regulate the Induction of UPF1 and UPF3 Ubiquitination
The perception of microbial infection rapidly triggers an oxidative burst and dual phosphorylation of MITOGEN-ACTIVATED PRO-TEIN KINASE3 (MPK3) and MPK6 in plants (Nicaise et al., 2009). To examine whether the degradation of UPF proteins is associated with these early events, we observed the decay of these proteins in wild-type and mutant plants defective in early signaling. UPF1 turnover occurred by 20 hpi after PstDC3000 infection in mpk3 and mpk6 compared with 10 hpi in the wild type. UPF1 was still detectable until 20 hpt after flg22 treatment in mpk6 cells ( Figure 7A). To address whether the induction of UPF1/UPF3 ubiquitination is affected by mutations in MPK3 and MPK6, we monitored the a-UPF1 and a-UPF3 immunoprecipitates from flg22-treated mpk3 and mpk6 leaves by immunoblotting using an a-UBQ antibody. The induction of UPF1 ubiquitination was initiated at 2 and 6 hpt in mpk3 and mpk6, respectively, compared with 1 hpt in wild type ( Figures 6C, 6D, and 7B). The ubiquitination of UPF3 was unaffected in mpk3, but it was delayed to 6 hpt in mpk6 from 2 hpt in wild type ( Figures 6C, 6D, and 7B). By contrast, the induction of UPF1/UPF3 ubiquitination was unaffected by mutations in the RESPIRATORY BURST OXIDASE HOMOLOGUE D (RBOHD) and RBOHF genes, although UPF2 was not degraded in the flg22treated rbohD rbohF mutant compared with the responses in wild type ( Figures 7A and 7B; Supplemental Figure 12A). These findings indicate that the induction of UPF1/UPF3 ubiquitination requires the mitogen-activated protein kinase (MAPK) signaling cascade and the crucial early response in PTI, and that reactive oxygen species might stimulate the turnover of UPF2. Indeed, transgenic plants expressing constitutively active MPK6 (MPK6CA) showed a low level of UPF1 compared with the wild type (Supplemental Figure 12B), supporting the idea that MAPK signaling participates in the decay of UPF proteins after sensing microbial infection.
We examined the effects of SA-related defense responses on the decay event. The levels of UPF proteins were not significantly altered in the SA signaling mutants eds1-1, sa induction deficient2-1 (sid2-1) and nonexpresser of pr genes1 (npr1-1) compared to the wild type after PstDC3000 infection and flg22 treatment ( Figure 7C). Exogenous treatment with BTH, a SA agonist, did not have any effect on the stability of UPF proteins in wild-type plants (Supplemental Figure 12C). The induction of UPF1/UPF3 ubiquitination upon flg22 treatment was also unaltered in these mutants (Supplemental Figure 12D). However, UPF2 persisted in sid2-1 and npr1-1 leaves for up to 20 h. These results suggest that the complete decay of UPF2 might involve SA-related events during a late stage of bacterial infection, which would explain the stability of UPF2 during early stages of PstDC3000 infection, i.e., 0.5 to 6 hpi (Figures 2A and 7C). Based on these results, we suggest that the decay of UPF1 and UPF3 by the UPS involves as-yet unidentified events downstream of the MAPK pathway triggered by bacterial recognition rather than subsequent SA-related events.

Unannotated Fused TNL-type Transcripts Accumulate in Wild-Type Plants after Bacterial Infection
Finally, we aimed to establish a model explaining why the suppression of NMD and the resulting upregulation of NMD-susceptible TNL transcripts lead to enhanced net immunity. The best way to illustrate this concept would be to produce a plant overexpressing the whole body of NMD-susceptible TNLs without suppressing NMD, which is impossible. As an alternative, we took previously uncharacterized fusion transcripts from RNA-seq data that were overrepresented in upf1-5 upf3-1 and upf3-1 upf1-5 (Supplemental Data Sets 1 and 2) and comprised four pairs of adjacent genes annotated as independent genes: At1g57630-A t1g57650, At2g17055-At2g17050, At5g38344-At5g38350, and At5g46490-At5g46500 (Supplemental Figures 13 and 14). The individual genes in pairs encoded N-and C-terminal fragments of TNLs, such that the fusion transcripts effectively encoded fulllength TNLs (Supplemental Figure 13; Supplemental File 6). All the fusion transcripts appeared to become upregulated in wild-type plants within 30 mpi with PstDC3000 except At2g17055-A t2g17050, although the levels were not significantly different from those under noninfection condition. Later, the fusion transcripts were over-represented within 30 hpi ( Figure 8A). Except at the loci of At5g38344 and At5g38350, RNAPII enrichment at the other three loci was not significantly altered by the simultaneous mutation of UPF1 and UPF3: The ChIP signals were similar in both upf3-1 upf1-5 and wild-type plants ( Figure 8B). Again, these results demonstrate that the accumulation of TNL-type transcripts is regulated by NMD and/or transcription.
Finally, to test whether these fusion transcripts were functional, we produced transgenic Arabidopsis plants expressing the At1g57630-At1g57650 and At5g38344-At5g38350 cDNAs under the control of the CaMV 35S promoter and terminator. Stable overexpression of these fusion transcripts, which do not retain NMD-eliciting features, conferred greater resistance to virulent Pseudomonas infection compared with wild-type plants ( Figure 8C, left), showing that these transcripts are involved in (A) Dynamics of fusion TNL-type R transcript accumulation in wild type upon PstDC3000 infection. Data points are the average 6 SD with three biological replicates with two technical repeats (n 5 3). Statistically significant differences are shown using different letters (i.e., a, b, c, and ab) above the bars (P < 0.01, one-way ANOVA). (B) RNAPII enrichment at the genomic loci containing the split genes in the wild type (WT; white bars) and upf3-1 upf1-5 (gray bars). Four independent biological experiments were performed in technical triplicates (average 6 SE, ** P < 0.01, two-tailed Student's t test, n 5 4). (C) Bacterial growth in the wild type (WT; white bars) and in two independent transgenic Arabidopsis Col-0 plants expressing either At1g57630-At1g57650 or At5g38344-At5g38350 (light and dark gray bars, respectively) 3 d after PstDC3000 and PstDC3000 hrcC 2 infection. Average 6 SE values are plotted. Different letters (a and b) above the bars indicate statistically significant differences (P < 0.01, one-way ANOVA, n 5 8). The numbers of PstDC3000 in wildtype plants were compared with those in independent transgenic lines expressing the same fusion transcript, as different transgenic plants had been grown separately.
(D) PR1 mRNA expression in noninfected wild-type (WT) and transgenic Arabidopsis plants used in (C). Different letters (a and b) above the bars indicate statistically significant differences among the genotypes (average 6 SD, P < 0.05, one-way ANOVA, n 5 3). Three biological experiments were performed with three technical repeats. basal resistance to bacterial infection. However, the bacterial numbers of PstDC3000 hrcC 2 from these transgenic lines were comparable to those from wild-type plants ( Figure 8C, right). In contrast to upf3-1 upf1-5, which produces a high level of the PR1 transcript even before pathogen infection (Supplemental Figure 1C), the PR1 mRNA levels were reduced in the transgenic plants in the absence of pathogen infection ( Figure 8D). The conformational changes in translated NLRs crucial for NLR activation may not occur under infection by the disarmed strain and noninfection conditions (Sukarta et al., 2016). Overall, we suggest that the suppression of NMD via the degradation of UPFs and the resulting stabilization of TNL transcript levels from an early stage of bacterial infection contribute to the basal immunity of plants to virulent pathogen infection.

DISCUSSION
Microbial perception by PRRs and R proteins occurs at the forefront of the battle between plants and microbes. In this study, we demonstrated that MAMP recognition by PRRs controls the levels of R transcripts carrying NMD-sensitive features via the induction of UPF1/UPF3 ubiquitination, followed by their 26Sproteasome-dependent degradation at an early stage of infection.
After their export from the nucleus, virtually all mRNAs undergo mRNA surveillance processes, including NMD and miRNAmediated gene silencing, by which only those certified as "normal" are deemed competent at translation (Eulalio et al., 2007;Lai and Eulgem, 2018). Because most FS R transcripts retain one of the NMD-triggering characteristics, they are thought to be either prevented from undergoing translation or immediately decayed by the NMD machinery, which may be essential for avoiding the unnecessary costs of immunity under normal growth conditions (Maekawa et al., 2011;Karasov et al., 2017). The observation that a majority of TNL and CNL genes carry NMD-sensitive features suggests that the NMD machinery and R genes coevolved in plants. The coexistence of both normal genes and those with NMD-regulated features in a few single R gene clusters (Supplemental Figure 15) suggests that gene duplication events including unequal crossing-over (Meyers et al., 2003) may have occurred before genetic drift, followed by subsequent positive selection. Such gene clusters might have evolved in land plants to facilitate the switching of R transcript levels (Raxwal and Riha, 2016).
Most FS R transcripts of the genes bearing NMD-eliciting features that we tested were highly stable in upf3-1 upf1-5, except for those of a few genes, supporting previous observations (Gloggnitzer et al., 2014). However, the gaps in the stability of these FS R transcripts between noninfected and Pseudomonasinfected plants were smaller than those between wild type and upf3-1 upf1-5 under our experimental conditions. In addition, the steady-state levels of a few R transcripts carrying typical NMD features might also be regulated by the putative miRNA-mediated control system during Pseudomonas infection. Based on these findings, we speculate that NMD in Pseudomonas-infected plants still occurs until the UPF proteins are entirely eliminated. It is also likely that other mRNA control processes, including miRNAmediated gene silencing, are deregulated in Arabidopsis during infection, because the intergenic pri-mRNAs may contain uORFs or long 39UTRs due to their weak protein-encoding capability as noncoding mRNAs or primary transcripts of small nucleolar RNAs and are thus NMD targets (Kurihara et al., 2009;Drechsel et al., 2013). The predicted NMD-sensitivity of R transcripts did not fully explain their actual biological NMD susceptibility, as the R transcripts were upregulated, unregulated, or even downregulated at a variety of time points after Pseudomonas infection, although most were upregulated or highly stable in the upf3-1 upf1-5 mutant.
In addition to the FS R transcripts, their AS variants also predominantly accumulated in upf mutants (Supplemental Files 6 and 7), which suggests that the atypical cell death-like symptoms observed in upf1-5 upf3-1 and upf3-1 upf1-5 after virulent Pseudomonas infection resulted from the increased heterogenicity of the NMD-sensitive R transcript repertoire as well as RPS6 (Gloggnitzer et al., 2014). Therefore, we suggest that the diversified expression of R genes, not their canonical functions in ETI, confers basal immunity in NMD-compromised mutants. NMD-sensitive R transcripts also accumulate in wild-type leaves after virulent P. syringae infection by suppressing NMD. How do virulent P. syringae strains overcome the induced barriers resulting from the accumulated R proteins? One possible explanation is that wild-type Col plants do not produce any functional R proteins capable of recognizing effectors delivered from PstDC3000 and PcaES4326 and are unable to trigger ETI in response to infection with these virulent pathogens, even when they accumulate high levels of NMD-sensitive R transcripts. If this is the case, these R transcripts may not function in the susceptible response in plants. During ETI, the accumulation of various sensor and helper NLRs is necessary for full immunity against avirulent pathogen infections (Jones et al., 2016;Wu et al., 2017). To achieve full immunity against a countless number of different pathogens, plants may have evolved diverse NLRs, known as the NLRome and pan-NLRome, via an arms race against a variety of pathogens (Meyers et al., 2003;van de Weyer et al., 2019). Unlike miRNAs that directly regulate the levels of their target R transcripts, NMD covers both AS and FS R transcripts as long as they carry NMD-triggering characteristics. Thus, NMD-mediated posttranscriptional regulation can diversify the expressed R repertoire to maximize the chances to achieve full resistance to unexpected pathogen infection. Our data do not provide details on every upregulated FS R transcript and its AS variants during pathogen infection, although we described their RNA-seq coverage data extracted from upf mutants (Supplemental Files 6 and 7). It would be interesting to monitor the dynamic changes in AS variants in real time after infection, because most splicing factor genes have AS forms, and many are also NMD-sensitive (Palusa and Reddy, 2010).
The UPS regulates diverse fundamental processes in plants, including hormonal signaling and immune responses (Kelley and Estelle, 2012;Marino et al., 2012;Dudler, 2013). The latter has been referred to as a double-edged sword, in which E3 ligases function as either positive or negative regulators of plant immunity (Marino et al., 2012). The best examples are the E3 ligases PLANT U-BOX12 (PUB12)/PUB13 and PUB22, which are involved in the decay of flagellin-induced FLS2 and EXOCYST SUBUNIT EXO70 FAMILY PROTEIN B2, respectively, resulting in attenuated plant immunity (Lu et al., 2011;Stegmann et al., 2012). The homologous triplet PUB22/PUB23/PUB24 can dampen early immune responses stimulated by MAMPs (Trujillo et al., 2008). By contrast, the transcription of ARABIDOPSIS TOXICOS EN LEVADURA9, encoding a Really Interesting New Gene zinc-finger E3 ligase, is induced by a variety of external stimuli, including chitin, and the mutant shows enhanced susceptibility to Golovinomyces cichoracearum and reduced reactive oxygen species production after infection (Ramonell et al., 2005;Berrocal-Lobo et al., 2010). Thus, the early immune response is either suppressed or activated by ubiquitination (Marino et al., 2012). We report here that the decay of UPF1 and UPF3 via the UPS is linked to the suppression of NMD. These processes are governed by the roles of PRRs as sensors, reporting dire circumstances to the downstream components of PTI, leading to the suppression of NMD and involving the decay of even nonubiquitinated UPF2.
Unlike the well-characterized target proteins of the UPS, e.g., receptors and transcription factors, UPF proteins participate in diverse cellular processes in plants by controlling the transcriptome (Kurihara et al., 2009;Drechsel et al., 2013). Why must UPF proteins decay upon bacterial infection at the expense of such pleiotropic effects? The most likely answer is that a considerable variety of R proteins must accumulate during an early stage of infection to protect the host plants from attack by unpredictable pathogens before they successfully colonize the plants. When R proteins monitor their cognate avirulence proteins derived from the pathogen, plants may only have to pay a reasonable cost for subsequent immunity. If not, the accumulated R proteins can confer a defense response to plants. Thus, we suggest that the consecutive events between MAMP perception by PRRs and the accumulation of R transcripts are well programmed in the "black box." The decay of UPF proteins and the accumulation of R transcripts occur in infected leaves at an early phase after bacterial infection. Interestingly, the significant delay in the induction of ubiquitination and subsequent decay of UPF1 in mpk3 and mpk6 suggests that the MAPK cascade is associated with these events. Because the mpk3 mpk6 double mutant was lethal, we could not quantify the ubiquitination of UPF1 and UPF3 in these plants. However, the results sufficiently indicate that the MAPK cascade not only modulates transcription activators/coactivators that regulate target genes involved in plant immunity (Rasmussen et al., 2012), it also links bacterial sensing and the maintenance of R transcripts by participating in the induction of the ubiquitination of proteins, including UPF1 and UPF3. Therefore, our findings point to an additional layer of plasticity between the PTI and ETI by NMD.
The ORFs of UPF1, UPF2, UPF3, SMG7, Y14, MAGO, elF4AIII, BTZ1, and BTZ2 were cloned into CTAPi, a plant transformation vector (Rohila et al., 2004), for stable expression in Arabidopsis. The oligonucleotide sequences used to amplify the cDNA fragments are listed in Supplemental Data Set 3. Transgenic plants were produced by the floral dip method (Clough and Bent, 1998). Seeds were screened on Murashige and Skoog (MS) medium (Duchefa) containing 50 mg mL 21 of DL-phosphinothricin (Duchefa). T 3 homozygotes were isolated and used for analysis.
Freshly incubated bacterial strains PstDC3000, PcaES4326, PstDC3000/AvrRpm1, PstDC3000/AvrRps4, and PstDC3000 hrcC 2 in King's B medium supplemented with the appropriate antibiotics were diluted in 10 mM of MgSO 4 to different concentrations as follows: OD 600 5 0.05 to measure ion leakage, OD 600 5 0.01 to evaluate pathophysiological responses, and OD 600 5 0.0001 (PstDC3000 and PcaES4326), or OD 600 5 0.001 (PstDC3000 hrcC -) to count the number of bacteria in leaf disks. The number of bacteria in infected leaves was determined 3 d after inoculation after a typical serial dilution method, and the infected leaves were photographed on the same day.

RNA-Seq
Three-week-old Arabidopsis plants were vacuum-infiltrated for 5 min in 0.53 MS medium containing 80 mM ActD and/or 20 mM CHX and incubated in the dark for 4 h before RNA isolation (Ambion; Hori and Watanabe, 2005). After confirming the RNA purity with a bioanalyzer using the RNA 6000 Pico Kit (Agilent), the total RNA was processed for mRNA sequencing library preparation using a TruSeq Stranded mRNA Sample Preparation Kit according to the manufacturer's instructions (Illumina). In brief, mRNAs were isolated from 400 ng of total RNA with RNA purification beads using poly(A) capture followed by enzyme shearing. After firstand second-strand cDNA synthesis, A-tailing and end repair were performed for the ligation of proprietary primers incorporating unique sequencing adaptors with an index for tracking Illumina reads from multiplexed samples run on a single sequencing lane. All samples were processed in two biological replicates per genotype, each from individual plants. For each library, an insert size of ;270 bp was confirmed with a bioanalyzer, and the library was quantified by RT-PCR using the CFX96 System (BioRad). Each library was sequenced for 75-bp paired-end reads with a NextSeq500 High Output 150 Cycle Kit on the NextSeq500 platform (Illumina).
The raw image data were transformed by base-calling into sequence data and stored in FASTQ format. The RNA-seq data were aligned by the softwares TopHat2 and BowTie2 to detect the canonical and noncanonical intron motifs (Langmead and Salzberg, 2012;Kim et al., 2013). Unaligned reads using the above platforms were mapped by the software BLAST (Trapnell et al., 2010). AS events were extracted from The Arabidopsis Information Resource 10 (TAIR10) gene models (www.arabidopsis.org). To evaluate NMD-triggering features (see "Evaluation of R Transcripts for NMD-Sensitive Features"), the data sets were mapped in the Integrative Genomics Viewer browser (Robinson et al., 2011;Thorvaldsdóttir et al., 2013). To quantify the total transcript mass in fragments per kilobase of transcript per million mapped reads, biological replicates of ActD-treated wild-type Col-0, ActD/CHX-treated wild-type Col-0, upf1-5, upf3-1, and upf3-1 upf1-5 were separately aligned with the Arabidopsis TAIR10 gene model using the Cufflinks package (Saeed et al., 2003).

Evaluation of R Transcripts for NMD-Sensitive Features
Although many normal-looking mRNAs can be regulated by NMD via low translation efficiency of the main ORF or out-of-frame translation due to ribosome read-through (Celik et al., 2017), we ruled out these possibilities due to the absence of sufficient evidence in Arabidopsis genes. We examined three NMD-triggering features to judge whether the individual TNL or CNL transcripts are putatively NMD-regulated:
uORFs Although uORF-driven NMD has also been considered an EJC type, exactly how it can trigger NMD is still unclear, because the translation of uORFs is often leaky. Approximately 20% of Arabidopsis genes contain uORFs (Kochetov et al., 2002) of which only up to 2% are putative NMD targets that contain uORFs encoding $ 35-amino acid peptides for stable expression. (Nyikó et al., 2009. The MAGNESIUM/PROTON EX-CHANGER (AtMHX) uORF, encoding a 13-amino-acid peptide, induces NMD in a downstream intron-dependent manner (Saul et al., 2009). Here, however, we considered CNL and TNL transcripts containing only uORFs encoding peptide(s) $ 35 amino acids as NMD targets based on general observations (Nyikó et al., 2009). The transcripts carrying uORFs that overlap with the authentic initiation codons were also considered to have putative NMD-sensitive features because this event showed strong correlations with sensitivity to NMD, even in the transcripts with uORFs encoding # 35 amino acids (Kalyna et al., 2012).

RNA Analysis
Leaves of 4-week-old Arabidopsis (;0.1 g) infiltrated with the PstDC3000 strain or 80 mM of ActD/20 mM CHX were harvested at the indicated time points after treatment. RNA was extracted from the leaves following the manufacturer's instructions (TRIzol reagent; Thermo Fisher Scientific). TURBO DNase (2 units; Thermo Fisher Scientific) was used to remove contaminating genomic DNA. After phenol/chloroform/isoamyl alcohol extraction, the purified RNA was dissolved in diethyl pyrocarbonatetreated water. Five micrograms of total RNA were used to synthesize first-strand cDNA by SuperScript-II RTase (Thermo Fisher Scientific), and the first-strand cDNA was diluted 10-fold for standard RT-PCR or RT-qPCR. RT-qPCR was performed with SYBR Premix Ex Taq (TaKaRa Bio) as described previously by Jung et al. (2009). All experiments were performed in three biological replicates with two or three technical repeats. The oligonucleotide sequences used in this study are presented in Supplemental Data Set 3.

Quantification of Cell-Death-Like Responses
Electrolyte leakage measurements were performed to quantify the hypersensitive response in plants after infection (Sohn et al., 2012). After infiltrating PstDC3000 into the leaves of wild-type and mutant plants, 10 leaf disks per biological replicate (n 5 4) were collected from the infected leaves of different plants, rinsed with distilled water for 30 min, and submerged in 2 mM of MgSO 4 . Conductivity was measured from 1 to 36 hpi with a LAQUAtwin Compact Water Quality Meter (HORIBA Life Science Solutions).
flg22 Treatment and SA Measurement flg22 (1 mM; Peptron) was infiltrated into Arabidopsis leaves, and the treated plants were transferred to a growth chamber (22 6 1°C, 12-h day/ 12-h night). The leaves were collected at the indicated time points after treatment to examine the levels of UPF proteins. Free-SA levels were also quantified using HPLC coupled with a fluorescence detector (Agilent) as described previously by Seskar et al. (1998) and Jung et al. (2009).

Monoclonal and Polyclonal Antibodies against UPF Proteins, IP, and Immunoblot Analysis
After analysis of the UPF1, UPF2, and UPF3 primary sequences using the software package SEAL (http://www.ab-mart.com), four peptide fragments per protein were designed to yield monoclonal antibodies (Supplemental Figure 16). After vigorous tests for 3 to 4 antibodies raised from each peptide using enzyme-linked immunosorbent assay and immunoblot analysis, one epitope each from the four peptides against each protein appeared to generate a specific antibody: 41-GSPTAWPTPSDS-52 (UPF1), UPF2 (933-ENGEAHGEESDS-944) and UPF3 . These antibodies were used for further immunoblot and IP analyses. The specificity and sensitivity of the selected monoclonal antibodies were verified by immunoblotting using different upf1, upf2, and upf3 mutant alleles (Supplemental Figure 6).
GST-UPF1(1-232), covering the N-terminal and Cys-and His-rich domains, was bacterially expressed and purified by glutathione-agarose chromatography (Green and Sambrook, 2012), followed by extraction from SDS-polyacrylamide gels, and used to raise an anti-UPF1 rabbit polyclonal antibody (Youngin Biotech).
For IP, leaf samples were ground in extraction buffer containing 20 mM of Tris-Cl at pH 8.0, 1 mM of EDTA, 100 mM of NaCl ( Figure 5; Supplemental Figure 9) or 150 mM of NaCl (Figures 3,6,and 7;Supplemental Figures 8 and 12), 1 mM of PMSF, and 13 proteinase inhibitor, and clarified at 15,000g for 10 min at 4°C. The homogenate was incubated with protein G-magnetic beads (Invitrogen) that were coated with the a-UPF1, a-UPF2, or a-UPF3 monoclonal antibody for 1 h at 4°C. The immune complex on the magnetic beads was washed five times in buffer containing 20 mM of Tris-Cl at pH 8.0, 1 mM of EDTA, 100 mM (or 150 mM for high stringency) of NaCl, and 0.5% (v/v) Triton X-100, resuspended in SDS sample buffer, and loaded on SDS-polyacrylamide gels for subsequent immunoblot analysis.
Finely ground leaf powder was homogenized in an equal volume of protein extraction buffer containing 20 mM of Tris-HCl at pH 7.5, 1 mM of EDTA, 150 mM of NaCl, 0.1% (v/v) Triton X-100, 0.1% (w/v) SDS, 5 mM of dithiothreitol, and 13 proteinase inhibitor (Roche Applied Science) and loaded onto SDS-polyacrylamide gels after clarification for immunoblot analysis as described by Green and Sambrook (2012). The antibodies and manufacturers were as follows: a-UBQ (AS08 307, Agrisera), a-PR1 (AS10 687, Agrisera), and a-PAP (P1291; Sigma-Aldrich). The dilution of primary antibodies used in this study was 1:5,000. The signal was visualized with a chemiluminescent substrate (Intron).

ChIP-qPCR
Nuclei from the leaves of 4-week-old wild-type and upf3-1 upf1-5 plants were extracted and purified as described in Jaskiewicz et al. (2011). The purified nuclei were resuspended in nucleus lysis buffer (50 mM of Tris-Cl at pH 8.0, 10 mM of EDTA, 1% [w/v] SDS, and 13 proteinase inhibitor [Roche Applied Science]), and the resulting chromatin was sheared by sonication to obtain fragment sizes ranging from 200 to 800 bp (Biorupter; Diagenode). The ChIP assay was performed using an anti-C-terminal domain (phospho-S5) antibody (ab5131; Abcam) as described by the manufacturer (Pierce Agarose ChIP Kit; Thermo Fisher Scientific). RNAPII enrichment between the transcription start sites and translation initiation sites of representative TNL-and CNL-type R genes was determined by RT-qPCR and calculated by the percent input method (Lin et al., 2012). The primers used in this study are listed in Supplemental Data Set 3.

Statistical Analysis
One-way ANOVA with a post hoc Tukey honest significant difference test was used for multiple comparisons. To analyze significant differences between the control and treatment groups, a two-tailed Student's t test was employed in this study. The sample size (n) and type of error bar are indicated in the figure legends. All pathophysiological experiments were performed in at least triplicate. The results of all statistical analyses performed in this study are presented in Supplemental Data Sets 4 and 5.

Accession Numbers
The accession numbers for the genes examined and discussed in this article are as follows: ACTIN2 (At3g18780), BTZ1 (At1g80000), BTZ2 Supplemental Figure 2. Representative R transcripts that accumulated in the leaves of NMD-compromised mutants.
Supplemental Figure 3. Stability of the mRNAs of TNLs and CNLs in wild-type and upf3-1 upf1-5 plants.
Supplemental Figure 5. Upregulation of the FS natural UPF1, UPF3, and SMG7 transcripts and AS versions of UPF1 and UPF2 in wild-type leaves during infection.
Supplemental Figure 6. Morphology of the wild-type and NMDcompromised mutants and the specificities of monoclonal antibodies used in this study. Figure 7. Stability of UPF1 in wild-type leaves infected with different P. syringae strains.

Supplemental
Supplemental Figure 8. Levels of UPF proteins in the immune complexes shown in Figures 3B, 3C, and 6C.
Supplemental Figure 9. UPF2 physically associates with UPF3 regardless of bacterial infection.
Supplemental Figure 10. Steady-state levels and stability of the mRNAs of TNLs and CNLs in wild-type leaves infected with a virulent PstDC3000 strain.
Supplemental Figure 11. Steady-state levels and stability of putative miRNA-target R transcripts.
Supplemental Figure 12. Induction of UPF1 and UPF3 ubiquitination occurs independently of reactive oxygen species burst or SA signaling.
Supplemental Figure 13. Structures of the fusion transcripts encoded by two separately annotated genes and their protein products.