The feedback control of UPF3 is crucial for RNA surveillance in plants

Nonsense-mediated-decay (NMD) is a eukaryotic RNA surveillance mechanism that controls the levels of both aberrant and normal transcripts. The regulation of this process is not well understood. The Arabidopsis NMD factor UPF3 is regulated by a negative feedback-loop that targets its own transcript for NMD. We investigated the functional significance of this control for the overall regulation of NMD in Arabidopsis. For this, we tested the ability of NMD-sensitive and -insensitive forms of UPF3, expressed under the control of UPF3 promoter, to complement NMD functionality in NMD-mutant plants and investigated their impact in wild-type (WT) plants. The sensitivity of UPF3 transcript to NMD was essential for efficient complementation of NMD in upf3 mutants. Upregulated UPF3 expression in WT plants resulted in over-degradation of certain transcripts and inhibited degradation of other transcripts. Our results demonstrate that, in contrast to mammalian cells, a delicate balance of UPF3 transcript levels by its feedback loop and by restriction of its transcription, are crucial for proper NMD regulation in Arabidopsis. Interestingly, the levels of many small-nucleolar-RNAs (snoRNAs) were decreased in upf1 and upf3 mutants and increased upon enhanced UPF3 expression. This suggests that proper snoRNA homeostasis in Arabidopsis depends on the integrity of the NMD pathway.

The NMD mechanism was studied in detail (18,19). When the translating ribosome reaches a termination codon (TC), it binds the eukaryotic release factors eRF1 and eRF3. Interaction of the poly(A)-binding protein (PABP) with both eRF3 and the eukaryotic initiation factor 4G (eIF4G) leads to normal translation termination and ribosome recycling, which prohibits NMD (20). In transcripts having long 3 untranslated regions (3 UTRs), PABP interaction with eRF3 is decreased due to the large physical distance, preventing efficient ribosome recycling by eIF4G (20). Consequently, eRF3 is indicated to interact with the NMD factor UPF1, followed by UPF1 phosphorylation by SMG1 and NMD activation (20,21). It was shown that UPF1 can also bind the mRNA before translation and be displaced to the 3' UTR by the translating ribosome, possibly providing another means of sensing 3 UTR length (22)(23)(24) and references therein). NMD-susceptible transcripts have long 3' UTRs either naturally or due to the appearance of a PTC. The presence of introns >50-55 nt downstream of TCs can also facilitate NMD (25,26), because NMD factors may be present at EJCs. The EJC is a complex of proteins that are deposited on the mRNA 20-24 nucleotides upstream of exon-exon junctions. The NMD factor UPF3, which is a nucleo-cytoplasmic shuttling protein, interacts with the EJC before export of the mRNA to the cytoplasm (27,28). UPF3 also interacts with UPF2, which functions as a bridge between UPF3 and UPF1 (29). UPF1 ability to bind the mRNA also before translation (22)(23)(24) explains why, although introns >50-55 nt downstream of TCs increase NMD efficiency, they are not essential for it, and NMD can be activated by long 3 UTRs alone. NMD can also be activated by upstream open reading frames (uORFs) which, if translated, may result in the prevalence of a long 3 UTR and potentially downstream introns (30). As in other eukaryotes, plant transcripts were shown to be targeted to NMD by PTCs (1,(4)(5), long (>300-350 nt) 3 UTRs (13)(14)(31)(32), uORFs (16,17) and introns >50-55 nt downstream of TCs (8,14,31).
Compared to the substantial knowledge about NMD mechanism, the regulation of this process is less well understood. We and others showed that expression of several eukaryotic NMD factors increases upon depletion of other factors necessary for this process (8,12,17,(33)(34)(35). This indicates that the affected factors are themselves NMD substrates that are regulated by negative feedback loops. We showed that expression of the UPF3 gene of the model plant Arabidopsis thaliana is higher in upf1 mutant plants than in wild-type (WT) plants (17). This indicates that UPF3 expression is regulated by a negative feedback loop that restricts its expression in WT Arabidopsis plants, in which NMD functions effectively. It was shown that human UPF3B (encoding the main active isoform among the two mammalian UPF3 isoforms) is also regulated by a negative feedback loop that restricts its expression under normal conditions (33). However, this restriction is apparently not critical for the overall regulation of mammalian NMD, since 15-to 20-fold elevated expression of UPF3B did not affect NMD in human cells (33). Here we studied the functional significance of the feedback loop dominating the expression of the Arabidopsis UPF3 gene for the overall regulation of plant NMD. We first identified that the cis-element that targets UPF3 transcript to NMD is its long 3 UTR. We then tested the ability of NMD-sensitive and -insensitive forms of UPF3, expressed under the control of the native UPF3 promoter, to complement NMD functionality in NMD-mutant plants and investigated their impact in WT plants. We found that the maintenance of NMD functionality in Arabidopsis requires that UPF3 expression will be delicately balanced by the sensitivity of UPF3 transcript to NMD, as well as by proper restriction of UPF3 transcription. Microarray and model gene analyses indicated that a moderate (1.5-to 2-fold) increase in UPF3 transcript levels can lead to NMD impairment for certain transcripts, while the endogenous level of UPF3 expression is apparently rate-limiting for NMD of other transcripts. This suggests that there are differences in the way NMD is regulated or executed in plant and mammalian cells. As mentioned, an intense (15-to 20-fold) elevation in the transcript levels of UPF3B did not impair NMD in mammalian cells (33). This elevated expression also did not enhance NMD, indicating the UPF3B is also not rate-limiting for mammalian NMD (33). Interestingly, we also found that a functional NMD pathway is essential for proper homeostasis of small nucleolar RNAs (snoRNAs) in Arabidopsis. Our finding that a non-functional NMD pathway results in snoRNA depletion in Arabidopsis is in contrast to a recent observation in human cells, in which NMD impairment does not alter snoRNA content (although it affects the expression of genes that host snoRNAs in their introns) (36,37). Our work uncovers the crucial role of fine-balancing UPF3 expression for RNA surveillance in Arabidopsis and highlights differ-ences in NMD regulation and cellular impact between plant and mammalian cells.

Mutant lines, plant transformation and expression analysis
The WT A. thaliana (L.) plants were of the Col-0 accession and the mutant plants were homozygous, and in the Col-0 background. The selection of homozygous progenies of the Arabidopsis T-DNA insertion mutants (38) SALK 112922 (upf1-5) and SALK 025175 (upf3-1) was described (17). The plants were transformed using the floral dip technique. Following selection on 20 g/ml hygromycin B, about 25 independently-transformed T1 plants were obtained for each construct analyzed in this study in each type of transformed plant. The T2 plants used for expression analysis were germinated on MS (Duchefa Biochemie BV) plates containing 20 g/ml hygromycin B (for elimination of non-transformed progenies) and grown for 2 weeks in a climate-controlled growth room in a photoperiod of 16 h light and 8 h dark.

Cycloheximide treatment
For cycloheximide (CHX) treatment in seedlings, seeds were surface sterilized and grown in liquid B5 (Duchefa Biochemie BV) medium for 12 days. The seedlings were vacuum-infiltrated during 3 min with B5 medium including 10 g/ml CHX (or the ethanol solvent as a control) and then grown with the same medium for 4 h before harvesting. For CHX treatment in cell cultures, 7-day-old Arabidopsis cell suspensions [(39), in the Erecta background] were treated with B5 medium including 10 g/ml CHX (or the ethanol solvent as a control) for 2 h before harvesting.

Generation of constructs
Sequence information about the UPF3 gene (Supplementary Figure S1) was obtained from The Arabidopsis Information Resource (TAIR). Construct +uORF was similar to construct W+I described previously (17) except that the 5 UTR of AtMHX was replaced with that of UPF3. For this, the native sequence of the 5 UTR of UPF3 (Supplementary Figure S1) was synthesized by GenScript flanked by the restriction sites of BamHI and NcoI at its 5 and 3 ends, respectively. The −uORF construct was created in a similar way, except that an ATG to AAG mutation eliminated the uORF. In both the +uORF and −uORF constructs, the coding sequence of the Escherichia coli ␤-glucuronidase (GUS) reporter was derived from the pCAMBIA1301 vector (Cambia, Australia) and included an intron. This intron was located 53 nt downstream the TC of the uORF of UPF3. A similar construct, as well as an intronless GUS, were previously successfully used to demonstrate that the uORF of another gene (AtMHX) can expose its transcript to NMD (17). To create constructs N-ter, U-ter and U-ter+I, we used plasmid pJD330 (a kind gift of DR Gallie) that included the CaMV 35S promoter fused to the omega ( ) 5 UTR (derived from the coat protein of tobacco mosaic virus), the coding sequence of GUS (that did not include an intron) and the terminator of the Agrobacterium tumefacience gene encoding nopaline synthase (NOS). This construct was called N-ter. In the two other constructs, the NOS terminator was replaced with the UPF3 terminator, with or without the last intron of this gene. The terminator of UPF3 (Supplementary Figure S1) included the 545 nt long 3 UTR (with or without the 12th intron) followed by the 300 nt located downstream this 3 UTR in the Arabidopsis genome (to ensure proper termination of transcription). The UPF3 terminator, flanked with appropriate restriction sites, was synthesized by GenScript. To create construct 35S::NR (NR stands for Non-Regulated; see 'Results' section), the coding sequence of UPF3, including the first intron of this gene, was synthesized by GenScript and used to replace the GUS coding sequence in construct N-ter. To create construct U3::NR (U3 stands for UPF3 promoter), a fragment including the promoter and 5 UTR of UPF3 (with the point mutation that eliminated the uORF) was synthesized by GenScript and used to replace the promoter and 5 UTR of construct 35S::NR. The UPF3 promoter included the 321 bp located upstream of the transcription initiation site of UPF3 (Supplementary Figure S1). These 321 bp comprised all the untranscribed sequence separating UPF3 from the presumed TATA box of AT1G33970, the gene located upstream UPF3 in the Arabidopsis genome. AT1G33970 is transcribed in reverse orientation relative to UPF3. Thus, no sequence downstream its presumed TATA box was utilized in order to prevent transcription at reverse orientation. The first intron of UPF3 was maintained in the UPF3 coding sequence utilized in this work. In construct U3::R (R stands for Regulated; see 'Results' section), the native 5 UTR (including the uORF) and terminator of UPF3 (including the 12th intron), which were synthesized by GenScript as described above, were used to replace the mutated 5 UTR and NOS terminator of U3::NR. All parts of construct U3::R were fused exactly as in the native UPF3 gene and the sequence of this construct is essentially as shown in Supplementary Figure S1B. The three constructs U3::R, U3::NR and 35S::NR had an identical UPF3 coding sequence, which included the first intron of UPF3. All these constructs were cloned into a binary vector that included the coding sequence of E. coli hygromycin phosphotransferase, which confers hygromycin B resistance, immobilized into A. tumefacience strain EHA105, and used for transformation of Arabidopsis plants.

Cloning of gene probes
Total RNA was extracted with the Tri-Reagent (Sigma). Preparation of cDNA was carried out using the Verso cDNA Kit (No. AB-1453/A) according to the manufacturer's instructions and was followed with DNase I treatment. The primer sets used for cloning of partial cDNA sequences and for probe preparation for UPF3, EF1a and the two model genes AT3G53400 and AT5G45430 are presented in Supplementary Table S1. The probe of UPF3 was derived from its coding sequence only (and not from its UTRs).

RNA extraction and northern blot analysis
Total RNA was extracted with the Tri-Reagent (Sigma). RNA samples were denatured with glyoxal (Sigma) and fractionated on 1% agarose gels. Gel preparation and fractionation were carried out with 10 mM NaPi buffer, pH 7.0. The gels were blotted onto a Zeta-Probe GT membrane (Bio-Rad) with 25 mM NaPi buffer, pH 7.0. RNA was fixed by UV. The membranes were stained by 0.02% methylene blue in 0.3 M sodium acetate (pH 5.5) to visualize the ribosomal RNA and then rinsed in H 2 O. Hybridization was carried out using the DIG-labeling system (Roche Diagnostics GmbH) according to the manufacturer's instructions. Quantification of band densities on gels was performed with the ImageJ program (NIH).

Quantitative GUS analysis
Quantitative measurement of GUS activity was carried out using the fluorometric assay (40). Plant material was ground in liquid nitrogen and extracted in a buffer containing 50 mM NaPO 4 , pH 7.2, 1 mM Na 2 EDTA, 10 mM ␤-mercaptoethanol and 10% (v/v) Triton X-100. Following centrifugation (5 min, 14 000 g, 4 • C), the supernatant was collected and the concentration of proteins was determined using the Bradford reagent (Sigma). Samples including equal amounts of protein were suspended in 250 l extraction buffer including 1 mM (final concentration) of the fluorescent GUS substrate 4-methylumbelliferyl-␤-Dglucuronide (MUG) (Duchefa Biochemie BV). GUS activity was assayed on a 96-well fluorescent plate-reader (Fluoroscan II, Lab Systems) with the excitation wavelength set at 350 nm and the emission wavelength at 460 nm. GUS activity (milli units.mg protein −1 ) was calculated from the slope of the line generated from measures taken at three-minute intervals during two hours, with respect to the slope of commercial pure GUS enzyme (Roche Diagnostics GmbH).

Microarray analysis
U3::NR-transformed WT plants, as well as empty vector (EV)-transformed upf1 and upf3 mutants, were grown two weeks in plates including MS medium containing 20 g/ml hygromycin B and 1% sucrose. Each of the three plant types was grown in two separate 15 mm perti dish plates. Each plate was divided to two halves: one including one of the three indicated plant types and the second including EV-transformed WT (EV-W) plants. Total RNA was extracted with the Tri-Reagent (Sigma). RNA quality was determined using the Plant Total RNA Nano 6000 assay kit (Agilent Technologies) on the Agilent Technologies 2100 Bioanalyzer. To generate the fluorescently labeled cRNA, total RNA was labeled using the Low Input Quick Amp Labeling kit (Agilent Technologies). Arabidopsis microarrays having experimentally validated gene-specific oligonucleotide probes, designed to differentiate between different models of the same gene (based on TAIR10), were designed by OakLabs GmbH and produced by Agilent Technologies. The microarrays were hybridized using the Gene Expression hybridization kit (Agilent Technologies) and scanned using the SureScan Microarray Scanner (Agilent Technologies). The data from all arrays were first subjected to background correction and LOESS within-array normalization using Agilent Feature Extraction software version 9.5.1.1 (Agilent Technologies). The remaining analyses were performed in Partek R Genomics Suite software version 6.6 (Partek Inc.). The log expression ratios produced during the normalization step were analyzed. Data from two biological replicates of each of the three plant types (U3::NRtransformed WT plants, and EV-transformed upf1 or upf3 mutants) were compared to eight biological replicates of the EV-W plants and the P-values were determined by Student's t-tests. The normalized data were analyzed to identify genes with significantly up-or downregulated expression in each of the three plant types compared to EV-W plants, with P < 0.05 and a cutoff of 1.5-or 1.2-fold change (FC), as indicated in the text. All P-values presented are FDR corrected. The heat map was generated using Euclidean distance as a similarity measure by using Partek R Genomics Suite TM software, version 6.6 ( C 2012 Partek Inc., St Louis, MO, USA). Information about the properties of the genes analyzed in the microarray, including the sequences of their 5 -and 3 UTRs and the locations of introns, were obtained from TAIR V10.

Statistical analysis
One-way analysis of variance (ANOVA) tests were done on log-transformed data, followed by Tukey's Honestly Significant Difference (HSD) post-test. Chi-square goodness-offit tests were done with the Yates' continuity correction. The Benjamini and Hochberg (BH) procedure was used to adjust the P-values of the Gene Ontology (GO) terms enriched in the microarray analysis of U3::NR plants.

The long 3 UTR of UPF3 exposes its transcript to NMD
We showed that the transcript levels of UPF3 are higher in mature leaves of upf1 mutant plants, having depleted UPF1 levels, than in WT Arabidopsis plants (17). A similar observation was made in whole 2-week-old seedlings (Supplementary Figure S2). To further support the conclusion that UPF3 expression is governed by NMD, we used the translation inhibitor CHX. CHX is often used as an NMD inhibitor, since termination of translation is essential for NMD activation (41). Treatment with CHX increased UPF3 transcript levels in seedlings and cell cultures of WT Arabidopsis plants ( Figure 1). The increased levels of UPF3 transcript in upf1 mutants and in CHX-treated WT plants, i.e. upon NMD inhibition, supported the conclusion that UPF3 transcript is controlled by a negative feedback loop that restricts its levels when NMD functions properly.
To understand the significance of UPF3 feedback control for the overall regulation of NMD in plants, it was necessary to identify the elements in UPF3 transcript that expose it to NMD and to study the consequences of eliminating this regulation. As detailed in the Introduction, elements that can expose plant transcripts to NMD include uORFs, long (>300-350 nt) 3 UTRs and introns >50-55 nt downstream of TCs. A schematic illustration of the Arabidopsis UPF3 gene is presented in Supplementary Figure S1A. UPF3 includes a 48 nt long uORF, a long (545 nt) 3 UTR and its last intron is localized 19 nt downstream its TC. We investigated the ability of all three elements to expose transcripts including the coding sequence of the GUS reporter gene to NMD in stably transformed Arabidopsis plants. Plants expressing reporter constructs that included the 5 UTR of UPF3 with (+uORF) or without (−uORF) its uORF ( Figure 2A) had similar levels of GUS mRNA ( Figure 2B and D) and GUS activity ( Figure 2C). The fact that GUS transcript levels and GUS activity (which is proportional to the amount of the GUS protein) remained unaltered upon uORF elimination indicated that the uORF of UPF3 does not affect the expression of downstream coding sequences at neither the level of transcript stability nor at the level of translation. The upstream AUG (uAUG) codon of UPF3 has a very weak Kozak context (U residues at the −3, −2 and +1 positions), making it very likely that this uAUG is not recognized by the ribosome. To conclude, UPF3 is not targeted for NMD by its uORF (and is also not affected by this uORF at the translational level).
In plants, NMD is induced by 3 UTRs whose lengths exceed 300-350 nt (13)(14)(31)(32). Long 3 UTRs were identified in the transcripts of most NMD factors regulated by negative feedback loops, including the Arabidopsis SMG7 gene and most human NMD factors (8,33,35). The addition of the long 3 UTRs of the mammalian UPF1 or SMG7 genes to reporter gene constructs greatly reduced the transcript levels of these reporter constructs, indicating that the above factors are indeed exposed to NMD by their long 3 UTRs (33,35). Using a similar approach, we transformed plants with reporter constructs including either the short NOS terminator or the native terminator (and long 3 UTR) of UPF3, with or without its last intron located downstream the TC ( Figure 2E). The ability of the two latter constructs having the long UPF3 3 UTR to equally decrease GUS transcript levels ( Figure 2F and G) suggested that the long 3 UTR of UPF3, but not the 3 UTR intron (which is located only 19 nt downstream the TC) is the main structural feature that exposes UPF3 transcript to NMD. To further explore the significance of the length of the 3 UTR of UPF3, two internal deletions were made within this 3 UTR in the context of the U-ter construct ( Supplementary Figure S3). An internal deletion that reduced the 3 UTR length from 545 to 253 nt resulted in a two-fold elevated expression of the reporter gene construct. This supported the idea that UPF3 transcript is subjected to NMD by its long 3 UTR.

Expression of NMD-sensitive and -insensitive forms of UPF3 in WT and NMD mutant plants
We aimed to understand the significance of the sensitivity of UPF3 transcript to NMD for the overall regulation The constructs used to determine the impact of the long 3 UTR and its intron. The N-ter construct had the short NOS terminator. Construct U-ter+I included the native terminator as well as the 12th intron of UPF3 (indicated by the letters In12 on gray background). Construct U-ter was similar to U-ter+I but lacked the 3 UTR intron. 5 UTR--the 5 UTR of the coat protein of tobacco mosaic virus, called omega. (F) GUS transcript levels (normalized to that of the N-ter construct) of WT plants expressing the constructs illustrated in (E). Each column presents the mean value ±SEM of GUS transcript levels of 2-week-old seedlings grown in two experiments in 10 plates (which constituted 10 biological replicates) that included in total ∼500 plants. These 500 plants were composed of ∼20 T2 progeny plants from each of the 25 independent T1 transformants that were regenerated for each construct in each plant. An asterisk indicates a statistically significant difference (P < 0.05) between the N-ter and the other constructs, as determined by Student's t-test. (G) Two of the 10 biological replicates that were used for determination of GUS transcript levels in (F). The corresponding methylene blue staining of the rRNA is shown below the blot. The U-ter and U-ter+I transcripts, which have longer 3 UTRs, appear in the blot as higher bands compared to the N-ter transcript.
of this surveillance mechanism in Arabidopsis. For this, we tested the ability of NMD-sensitive and -insensitive forms of UPF3 to complement NMD functionality in NMDmutant plants and investigated their impact in WT plants.
To this end, we created three constructs that had an identical UPF3 coding sequence (which also included the first intron of UPF3), but differed in the presence of elements that could expose the transcript to NMD (Figure 3). The first construct included the native promoter, 5 UTR (including the uORF) and long terminator (including the last intron) of UPF3 (Figure 3). The native promoter of UPF3 was used in order to keep the level and sites of expression as similar as possible to those of the endogenous UPF3 gene. The resulting construct was named U3::R (U3 stands for UPF3 promoter and R for Regulated form). Notably, this construct was almost identical to the native UPF3 gene and included all the elements that could target UPF3 transcript for NMD.
To test the significance of the sensitivity of UPF3 transcript to NMD, we created a second construct, in which UPF3 transcript could not be targeted by NMD. This construct, called U3::NR (NR stands for non-regulated form), was similar to U3::R except that all elements that could expose UPF3 transcript to NMD were eliminated. The long terminator of UPF3 was replaced with that of NOS, and, to be on the safe side, the uORF was also eliminated by a point mutation (ATG to AAG) ( Figure 3). This excluded the possibility that this uORF is targeted for NMD under some unpredicted conditions (however, all our analyses were carried out under the conditions utilized in the analyses presented in Figure 2, in which this uORF was shown not to affect expression).
We also wanted to understand the impact of strong overexpression of UPF3. For this, the coding sequence of UPF3 was expressed under the control of the strong, constitutive CaMV 35S promoter, a 5 UTR called omega ( ; derived from the coat protein of tobacco mosaic virus) that was shown to mediate very efficient translation (42) and the NOS terminator. This construct, called 35S::NR, not only lacked elements that could expose UPF3 transcript to NMD (and was, therefore, also termed NR) but was also expected to lead to strong overexpression of UPF3.
Each of the constructs shown in Figure 3, as well as the EV, was stably transformed into plants having mutated UPF3 or UPF1 genes and also into WT Arabidopsis plants. The upf3 mutants are homozygous for the upf3-1 allele and do not have intact UPF3 mRNA (5). The upf1 mutants are homozygous for the upf1-5 allele and have severely reduced, but not completely eliminated expression of UPF1 (a null mutation in UPF1 is lethal in Arabidopsis) (4,10). About 25 independently-transformed T1 plants were obtained for each construct in each type of transformed plant. The plants used for expression analysis were germinated from mixtures including an equal number of T2 seeds from each of the 25 independent T1 transformants of each construct. In our experience, this approach efficiently compensated for the position effect. The suitability of our approach was validated by the observation that the U3::R construct, which is almost identical to the native UPF3 gene, efficiently lowered the elevated levels of NMD substrate transcripts in upf3 mutants (see below) and also restored the mutant phenotype of upf3 plants into an apparently normal one (supplementary Figure S4).
The levels of UPF3 transcript in plants expressing each construct were determined by northern blot analysis (Figure 4). UPF3 transcript levels in upf3 mutants expressing the U3::R or U3::NR constructs were somewhat lower or higher, respectively, than those of WT plants (or WT plants expressing the EV, named EV-W) ( Figure 4A and B). There was a significant (P < 0.05 in Student's t-test) increase of ∼1.5-fold in UPF3 transcript levels in upf3 mutants expressing the U3::NR as compared to the U3::R construct (Figure 4A). The NR form, which has a shorter 3' UTR, appears in the blot as a lower band compared to the R form ( Figure 4B). Naturally, no UPF3 transcript was detected in upf3 mutants transformed with the EV (EV-U3 plants). As expected, very high levels of the NR form of UPF3 tran-script were observed in upf3 plants expressing the 35S::NR construct ( Figure 4A and B).
The changes in UPF3 transcript levels in WT plants expressing the three constructs were in line with those seen in the transformed upf3 mutants, but now the R form was the sum of the endogenous and transformed transcripts (Figure 4C and D). The fact that UPF3 transcript levels were similar in plants expressing the U3::R and U3::NR constructs may result from NMD impairment in the U3::R plants (see below). The R form could not be quantified in plants expressing the 35S::NR construct due to the high intensity of the NR band.
In upf1 mutants expressing the EV (EV-U1 plants), UPF3 transcript levels were higher than those of WT or EV-W plants ( Figure 4E and F), as expected based on our previous findings [(17) and Supplementary Figure S2]. Almost no further increase in UPF3 transcript levels was observed in plants expressing the U3::R construct, possibly due to ability of the upf1 mutants to still balance UPF3 expression. Increased UPF3 transcript levels were seen in U3::NR transformed plants. Expression of the 35S::NR construct was higher than that of the two former constructs, but the extent of this difference was much lower compared to that observed in the transformed upf3 or WT plants. This was confirmed by direct comparison of UPF3 transcript levels in transformed WT (Col-0), upf1 and upf3 plants expressing the 35S::NR construct (Supplementary Figure S5). Examination of reporter gene constructs that did not include any element of the UPF3 gene indicated that expression of the 35S promoter is much weaker in upf1 mutants compared to WT plants or upf3 mutants (Supplementary Figure S6). As discussed by Moreno et al. (43), this finding can be explained by silencing of transgenes including the 35S promoter in the SALK mutant line upf1-5. The T-DNA insert in SALK mutants contains a 35S promoter, which resulted in silencing of transgenes having the same promoter in about half of the SALK mutants analyzed (44).

The sensitivity of UPF3 transcript to NMD is important for efficient NMD complementation in upf3 mutants
To determine the impact of the NMD-sensitive andinsensitive forms of UPF3 on NMD functionality, we selected two model genes: AT3G53400 and AT5G45430 (named hereafter M1 and M2, respectively). There are several indications that these genes are genuine NMD substrates. The transcripts of both M1 and M2 include an element that is highly likely to directly expose them to NMD: an uORF with an evolutionarily conserved peptide (CpuORF). Genes having CpuORFs were found to be hugely over-represented among the common substrates of several NMD mutants in Arabidopsis (45). The evolutionally conservation of the CpuORF peptides strongly suggests that these peptides are indeed translated. Hence, their TCs, which are followed by very long 3 UTRs, can cause NMD (45). The CpuORFs of both M1 and M2 encode rather long peptides (∼40 amino acids) and have strong Kozak-context uAUG codons, which increase the chance that these uAUGs are recognized by the ribosome. Several studies showed that the transcript levels of M1 and M2 are indeed significantly higher in upf1, upf3 and smg7 mutants,   Figure 3 or the empty vector (EV). Each column presents the mean value ±SEM of UPF3 transcript levels of 2-and in UPF2-silenced plants, as compared to WT Arabidopsis plants (2,9,(45)(46). Increased expression of M1 and M2 in NMD mutants was also detected in our experimental conditions (Supplementary Figure S7). Overall, these data suggest that the two model genes are bona fide NMD substrates.
Before testing the expression of these model genes in the transformed plants, we examined the transcript levels of the control EF1a gene. EF1a expression was previously shown to be similar in WT, upf1 and upf3 plants (4,17), suggesting that it is not affected by NMD. It was found that the transcript levels of EF1a were similar in all the transformed plants (Supplementary Figure S8).
In upf3 mutants expressing the U3::R construct, the transcript levels of the two model genes were much lower compared to EV-transformed upf3 mutants, and similar to, or only slightly higher than those of WT or EV-W plants (Figure 5). This demonstrated the ability of the regulated form of UPF3 transcript expressed from the U3::R construct to almost completely restore NMD in upf3 mutants (at least for the two model genes examined).
Interestingly, expression of U3::NR complemented NMD to a much lower extent compared to U3::R (for M2) or not at all (for M1) ( Figure 5). As indicated above, UPF3 transcript levels were only about 1.5-fold higher in upf3 plants expressing the U3::NR compared to the U3::R construct ( Figure 4A). This indicated that even a moderate increase in UPF3 transcript levels could result in NMD inhibition in Arabidopsis. This conclusion was strengthened by the observations in transformed WT plants that had a moderate increase in the regulated (U3::R) form of UPF3 transcript (see below). The fact that only expression of the NMD-sensitive form of UPF3 transcript resulted in efficient complementation of NMD in upf3 mutants indicated that the sensitivity of UPF3 transcript to NMD is very important for NMD control in Arabidopsis.
In upf3 mutants that expressed the 35S::NR construct, in which UPF3 transcript levels were particularly high (Figure 4A and B), expression of the NMD target genes examined was either similar to that of upf3 mutants expressing the EV (for M2) or even higher than that (for M1) ( Figure  5). This supports the conclusion that restriction of UPF3 transcript accumulation is essential for prevention of NMD inhibition. Altogether, the observations in the transformed upf3 mutants indicated that the NMD feedback loop restricting UPF3 transcript levels is important for the maintenance of NMD functionality in plants.

The maintenance of NMD functionality depends not only on post-transcriptional but also on transcriptional control of UPF3 expression
In WT plants expressing the three constructs, the exogenous UPF3 transcript accumulated on top of the endogenous one, resulting in elevated levels of UPF3 transcript in all three types of transformed plants as compared to WT plants ( Figure 4C). The transcript levels of the two model genes were higher in all transformed plants compared to WT or EV-transformed plants ( Figure 6). Thus, NMD was inhibited not only by the NMD-insensitive form of UPF3 but also by the moderate increase in UPF3 transcript derived from the U3::R construct, which included all the native regulatory elements of the UPF3 gene. As shown above, the U3::R construct efficiently complemented NMD in upf3 mutants. Yet, increased expression of the same regulated form in WT plants inhibited NMD. This indicates that in general, the maintenance of NMD functionality in Arabidopsis requires not only post-transcriptional control of UPF3 by NMD, but also proper restriction of this gene transcription.
Altogether, the observations in transformed WT and upf3 plants demonstrate the importance of the mechanisms that restrict the accumulation of UPF3 transcript for the overall regulation of plant NMD. This is in contrast to the situation in mammalian cells, in which even an intense (15-to 20-fold) increase in the transcript levels of human UPF3B, encoding the main active mammalian UPF3 isoform, did not affect NMD (33).
Although the total levels of UPF3 transcript were much higher in WT plants expressing the 35S::NR construct compared to the two other constructs ( Figure 4C and D), the levels of the model genes increased to similar extents upon expression of the three constructs ( Figure 6). At least in animals, NMD efficiency was shown to vary in different tissues and cell types (47). It is reasonable to assume that the UPF3 promoter is preferentially expressed in cells whose NMD process is particularly active. At the same time, since the 35S promoter is constitutive, the high amounts of UPF3 transcript expressed under the control of the 35S promoter are probably not limited to cells that are exceptionally active in NMD. Thus, a moderate increase in UPF3 transcript levels in the endogenous sites of this gene expression is apparently more effective in inhibiting NMD than excessive UPF3 transcript accumulation in other sites. In upf3 mutants, expression of the two model genes was higher following expression of 35S::NR as compared to the two other constructs ( Figure 5). This is presumably because in upf3 mutants, which lack endogenous UPF3 expression, the week-old seedlings grown in 5-7 plates (which constituted 5-7 biological replicates) that included in total ∼300 plants. These 300 plants were composed of ∼12 T2 progeny plants from each of the 25 independent T1 transformants that were regenerated for each construct in each plant. The internal graphs show the constructs with lower expression on a smaller scale. The plant genotypes into which the constructs were transformed (Col-0, upf3 or upf1) are indicated below the graphs. WT--non transformed, wild-type Col-0 plants. The names EV-W, EV-U3 or EV-U1 refer to EV-transformed Col-0, upf3 or upf1 plants, respectively. The R and NR forms of UPF3 transcript are presented by black and gray columns, respectively. When both forms were present in a specific lane, the two forms were also quantified relative to each other and then the NR form was normalized to the R form. All data were also normalized to the intensity of the rRNA staining. The probe of UPF3 was derived only from its coding sequence (and not from its UTRs). An asterisk indicates a statistically significant difference (P < 0.05) in the total levels of UPF3 transcript between the WT and the indicated plant, as determined by Student's t-test. (B, D and F) Each blot presents one of the 5-7 biological replicates that were used for determination of UPF3 transcript levels in (A, C and E), respectively. The corresponding methylene blue staining of the rRNA is shown below each blot. The plant genotypes into which the constructs were transformed are indicated above the blots. Each column presents the mean value ±SEM of M1 or M2 transcript levels of 2-week-old seedlings that were grown in two independent experiments in 10 plates (which constituted 10 biological replicates) that included in total ∼500 plants. These 500 plants were composed of ∼20 T2 progeny plants from each of the 25 independent T1 transformants that were regenerated for each construct in each plant. The plant genotypes into which the constructs were transformed are indicated below the graphs and by different column colors--gray (Col-0) or black (upf3). EV-W and EV-U3 are EV-transformed Col-0 or upf3 plants, respectively. ANOVA test demonstrated that there were significant differences between the different plants (for M1: P < 0.001, F = 64.591, df U3::R-and some of the U3::NR-mediated increase in UPF3 transcript contributed to NMD complementation.

A delicate balance of UPF3 expression is also essential under UPF1 limitation
Crosses with plants having strong or weak alleles for UPF3 deprivation showed that in homozygous upf1-5 plants it was not possible to reduce UPF3 levels to the same extent as that enabled in WT plants [ (4); see also 'Discussion' section in (17)]. This observation, together with the natural increase in UPF3 transcript levels observed in upf1 mutants [(17) and Supplementary Figure S2], supported the assump-tion that excess expression of UPF3 can compensate, to a certain extent, for UPF1 limitation (17). It was, therefore, interesting to learn whether further increase in UPF3 expression could boost NMD complementation under UPF1 limitation. As shown in Figure 4E, almost no increase in UPF3 transcript levels was seen in upf1 mutants expressing the U3::R construct. In accord, U3::R expression did not alter the expression of the two model genes compared to upf1 mutants expressing the EV (EV-U1 plants) (Figure 7). In upf1 mutants expressing the U3::NR construct, UPF3 transcript levels were only moderately increased compared to the EV-or U3::R-transformed upf1 plants ( Figure 4E). However, this was accompanied by a significant increase in M1 and M2 transcript levels in the U3::NR transformed upf1 mutants (Figure 7). Thus, even a small elevation of UPF3 expression on top of its naturally raised expression in upf1 mutants could be detrimental to NMD. This suggests that UPF3 expression must be delicately balanced also under UPF1 limitation and that the sensitivity of UPF3 transcript to NMD is very important for this balancing [as mentioned, the Arabidopsis upf1-5 mutants used here still have residual levels of UPF1 expression (4) and, hence, residual NMD activity].
The total content of UPF3 transcript was higher in upf1 plants expressing the 35S::NR as compared to the U3::NR construct ( Figure 4E). Nevertheless, the average levels of M1 and M2 transcripts were lower in plants expressing the 35S::NR compared to the U3::NR construct ( Figure 7). When the differences in model gene expression were analyzed by Student's t-test only between U3::NRand 35S::NR-expressing upf1 mutants, they were found to be significant (P = 0.001 and 0.01 for M1 and M2, respectively). This supports the previous suggestion that excessive expression of UPF3 is more effective in inhibiting NMD in the endogenous sites of UPF3 expression than in other sites. The deleterious impact of the 35S::NR construct was similar to, or higher than that of U3::NR only in the transformed WT or upf3 plants, in which the extent of overexpression mediated by the 35S promoter was much higher compared to the upf1 mutants.

Transcriptome analysis suggests that UPF3 transcript is delicately balanced between adequate and excessive levels
The observations with the two model genes indicated that the mechanisms restricting UPF3 transcript accumulation are important for NMD functionality in Arabidopsis. To further examine this conclusion, we aimed to determine the impact of de-regulated UPF3 expression on the whole Arabidopsis transcriptome. For this, the transcriptomes of WT plants expressing the U3::NR construct, and of EVtransformed upf1 and upf3 mutants, were compared to EVtransformed WT (EV-W) plants using microarray analysis. The analysis was carried out in transformed WT (and not NMD mutant) plants in order to exclude any possibility that the observations for certain NMD substrates might result from inefficient expression of the transformed construct in certain cells (and, hence, from deficient and not increased expression of UPF3).
We first examined whether the group of genes upregulated in U3::NR plants (as compared to EV-W plants) is enriched with genes whose expression is also upregulated in EV-transformed upf1 or upf3 mutants grown under the same conditions (as compared to EV-W plants). Figure 8A shows that the group of genes upregulated in U3::NR plants had an about 2.5-fold elevated proportion of genes upregulated in upf1 or upf3 mutants, as compared to the proportion of the latter genes among all array genes. This difference was highly significant (P = 4.3E-49 in chi-square test; Supplementary Dataset S1). The two model genes showed small but highly significant increased expression in U3::NR plants also in the microarray analysis (Supplementary Dataset S2, model genes). Genes whose expression in- creases in upf1 or upf3 mutants are defined as NMD substrates. The microarray data indicated that, as pre-validated by the northern blot analysis of the M1 and M2 model genes, de-regulated expression of UPF3 can increase the transcript levels of certain NMD substrates.
Interestingly, an increased proportion of NMD substrates was observed not only among genes upregulated in U3::NR plants, but also among genes downregulated in the latter plants. The group of genes downregulated in U3::NR plants had an about 2-fold elevated proportion of genes up-regulated in upf1 or upf3 mutants, as compared to the proportion of the latter genes among all array genes. This difference was highly significant (P = 5.5E-14 in chi-square test; Supplementary Dataset S1). These data indicated that increased expression of UPF3 could also decrease the levels of some NMD substrate transcripts, i.e. apparently enhance their NMD.
The analyses presented in Figure 8A were carried out for genes whose expression was altered by an arbitrary cutoff of at least 1.5-fold (with FDR P < 0.05) in the indicated plant types. However, Figure 6 indicated that the changes in expression of the model genes M1 and M2 in WT plants expressing the different constructs were apparently small. When the cutoff for analysis of the microarray data was lowered to 1.2-fold (with FDR P < 0.05), the results were qualitatively similar to those obtained using the 1.5-fold cutoff, with significant chi-square test values (Supplementary Dataset S2).
The overlap between genes whose expression was altered (either up-or downregulated >1.5-fold) in U3::NR plants and the genes upregulated >1.5-fold in at least one upf mutant is schematically illustrated by a Venn diagram (Figure 8B). This overlap was even more apparent using the 1.2fold cutoff ( Figure 8C). The highly significant chi-square tests values detailed above indicated that these overlaps represented highly significant enrichments of the genes altered in U3::NR plants with NMD substrates (i.e. that these overlaps were much larger than those expected to occur between the genes upregulated in upf mutants and a random group of array genes).
We then examined the prevalence of potential NMD eliciting features among genes whose expression was altered in the different plants. The features considered were an uORF, a 3 UTR >300 nt, an intron ≥50 nt downstream the TC or the gene model being a pseudogene. Supplementary Dataset S3 presents this analysis for genes whose expression was altered by at least 1.5-fold (with P < 0.05) and for which either the 5 -or 3 UTR sequence information was available in TAIR (or were pseudogenes). Among these genes, the proportion of those having at least one potential NMD eliciting feature was 57 and 72% for genes upregulated in upf1 and upf3 mutants, respectively, and 56 and 59% for genes up-or downregulated in U3::NR plants, respectively. Thus, the proportion of potential NMD features among the genes altered in U3::NR plants was similar to that of upf1 mutants, although lower than that of upf3 mutants. The potential NMD features considered did not include transposable elements and potential natural antisense RNAs, for which UTR sequence information was not available, but such transcripts may also be targeted to NMD (2).
The finding that the transcript levels of some NMD substrates were reduced by increased UPF3 expression could be anticipated based on the known function of UPF3 in NMD activation. Apparently, the endogenous level of UPF3 expression in WT plants is rate-limiting for NMD of these substrates. The unexpected conclusion that a moderate increase in UPF3 transcript levels could also inhibit NMD for certain substrates was supported by the combination of the microarray data and the northern blot analysis of the M1 and M2 model genes. The fact that deregulated UPF3 expression resulted in enhanced NMD of certain substrates and decreased NMD of other substrates suggests that UPF3 transcript is delicately balanced between adequate and excessive levels (see 'Discussion' section).

A functional NMD pathway is essential for proper homeostasis of Arabidopsis snoRNAs
The DAVID functional annotation tool (48) was used to search for GO terms that were over-represented among the transcripts upregulated by at least 1.5-fold (with P < 0.05) in U3::NR-transformed WT plants. The highest enrichment score, of 19-fold (with BH adjusted P-value of 1E-35), was displayed by a group of 35 genes (Supplementary Dataset S4). Strikingly, all these 35 genes were small nucleolar RNAs (snoRNAs) (Supplementary Dataset S4). We therefore examined the fate of all 64 snoRNAs (all of which having a gene-specific probe) that were represented on the microarray. The expression of most snoRNA genes was increased in U3::NR plants as compared to EV-W plants ( Figure 8D and Supplementary Dataset S4). These snoR-NAs were apparently not direct NMD substrates, because their levels were not increased in upf1 or upf3 mutants grown under the same conditions. On the contrary, there was a strong tendency for decreased levels of these snoRNAs in upf1 and upf3 mutants as compared to WT plants (Figure 8D and Supplementary Dataset S4). The reduced expression in upf1 and upf3 mutants indicates that a functional NMD pathway is essential for proper homeostasis of Arabidopsis snoRNAs.
It was reported that the common genes upregulated in mutants of plant NMD factors have over-representation of proteins involved in the response to pathogens, such as pathogen-related proteins and WRKY transcription factors (45,49). Most of these defense response genes are likely to be indirectly affected by NMD (46). The DAVID analysis of transcripts upregulated in U3::NR plants showed that a group of 43 genes with GO terms related to plant defense was overrepresented by 1.6-fold (with unadjusted P = 0.0017) as compared to WT plants (Supplementary Dataset S4). Within these 43 defense genes, analysis of only those for which either the 5 -or 3 UTR sequence was available in TAIR showed that at least one NMD feature could be identified in only 29% of them (compared to 56% among all U3::NR upregulated genes; see above). Thus, both previously characterized NMD mutants (45,49) and U3::NR transformed WT plants have increased proportion of defense genes, most of which apparently being indirect NMD substrates.

DISCUSSION
The idea that UPF3 is regulated by a negative feedback loop was based on the increase in UPF3 transcript levels in upf1 mutants and in WT plants treated with the NMD inhibitor CHX. It was found that the main structural element that exposes UPF3 transcript to NMD is its long 3 UTR. The functional significance of UPF3 feedback regulation was studied by testing the ability of NMD-sensitive and -insensitive forms of UPF3 to complement NMD functionality in NMD-mutant plants and investigating their impact in WT plants. The native promoter of UPF3 was utilized in order to maintain the intensity and spatial control of the expression as similar as possible to those of the endogenous UPF3 gene. The NMD-sensitive form of UPF3 almost completely restored NMD in upf3 mutants for two model genes having features of bona fide NMD substrates. However, the NMD-insensitive form of UPF3 (the U3::NR construct), which mediated only a moderate (about 1.5-fold) increase in UPF3 transcript levels compared to the U3::R construct, had impaired ability to complement NMD in upf3 mutants. This demonstrated the significance of the sensi-tivity of UPF3 transcript to NMD for the overall regulation of this RNA surveillance mechanism in Arabidopsis. The observations in WT plants expressing the U3::R construct showed that NMD impairment occurred even when the transcript levels of the NMD-sensitive form were moderately (2-fold) increased. This indicates that a delicate balance of UPF3 transcript levels by its feedback loop, as well as by restriction of its transcription, are crucial for proper NMD regulation in Arabidopsis.
Transcriptome analysis showed that genes that were either up-or downregulated in U3::NR plants had a significantly higher proportion of genes upregulated in upf mutants as compared to the proportion of the latter genes among all array genes ( Figure 8A, and chi-square tests in Supplementary Datasets S1 and S2). Genes upregulated in upf mutants are defined as NMD substrates. Such substrates can be either direct or indirect targets of NMD. It is difficult to definitely determine based on the presence of specific features which genes are direct targets, because not every gene having a potential NMD features is subjected to NMD and NMD-inducing features are often undefined (50). Yet, the percentages of genes having potential NMD features in genes up-or downregulated in U3::NR plants (56 and 59%, respectively) were similar to that of upf1 mutants (57%), although lower than that of upf3 mutants (72%). In addition, although it is difficult to undoubtedly determine for any specific gene whether it is a direct substrate, the fact that the enrichment of genes altered in U3::NR plants with genes upregulated in upf mutants is highly significant at a global statistical level (the chi tests), supports the deduction that some of the common genes are direct substrates, while others are similarly affected due to similar changes in direct substrates. As discussed in detail in the 'Results' section, the two model genes used in this study are highly likely to be direct NMD substrates.
The fact that a moderate increase in UPF3 expression reduced the levels of some NMD substrates is in agreement with the known function of UPF3 in NMD enhancement. The unexpected finding that a moderate increase in UPF3 transcript levels could also inhibit NMD of certain substrates was supported by the combination of the microarray data and the northern blot analysis of the model genes. The distinctive observations for different NMD substrates may result from expression of these substrate genes in different cells, having a priori dissimilar levels of expression of the endogenous UPF3 gene. A model that is coherent with our observations is illustrated in Figure 9. This is a simplified model presenting only UPF3 impact, but NMD efficiency obviously depends not only on UPF3 expression but on the interaction between all NMD factors. According to this model, in cells or under conditions in which the endogenous level of UPF3 expression is still rate limiting for NMD (point A in Figure 9), an increase in UPF3 expression can enhance NMD efficiency. Restriction of UPF3 expression at both the transcriptional and post-transcriptional levels is therefore essential for preventing undue, over-degradation of potentially affected transcripts. At the same time, in cells or under conditions in which UPF3 transcript levels reach a certain threshold (point B in Figure 9), further increase in UPF3 levels can lead to NMD inhibition (point C in Figure 9). To conclude, limited UPF3 expression (e.g. in upf3 mutants) inhibits NMD, while a moderate increase in UPF3 expression can lead to over-degradation of certain transcripts and inhibited degradation of other transcripts. These findings indicate that a delicate balance of UPF3 transcript levels is crucial for RNA surveillance in Arabidopsis.
The downregulation of certain transcripts due to increased expression of UPF3 suggests that UPF3 is rate limiting for NMD of certain Arabidopsis genes. Figure 8A-C show that the degree of overlap with genes upregulated in upf mutants was somewhat lower for genes downregulated versus genes upregulated in U3::NR plants. At the same time, the proportion of genes having NMD features was not lower in U3::NR down-versus upregulated genes (59 and 56%, respectively). This raises the possibility that some of the genes downregulated in U3::NR plants are expressed in cells in which UPF3 expression is normally too low to support efficient NMD (point A in the model presented in Figure 9). Consequently, these genes are not efficiently suppressed in WT plants and are therefore not upregulated in upf mutants. The fact that increasing UPF3 expression is sufficient for reducing the transcript levels of these genes suggests that UPF3 is rate limiting for NMD of certain Arabidopsis genes and that modulating UPF3 expression is apparently sufficient for controlling NMD efficiency for at least certain transcripts. The findings that the endogenous level of UPF3 expression is rate-limiting for NMD of certain substrates, whereas a moderate (1.5-to 2-fold) increase in UPF3 transcript levels can lead to NMD impairment for other substrates suggest that there are differences in the way NMD is regulated or executed in plant and mammalian cells. In human HeLa cells, even a 15-to 20-fold increase in the transcript levels of UPF3B, which encodes the main active human UPF3 isoform, did not inhibit NMD (33). This increased expression also did not enhance NMD, indicating that UPF3B Nucleic Acids Research, 2015, Vol. 43, No. 8 4233 levels are also not rate-limiting for human NMD (33). Increased expression of non-plant NMD factors was shown to inhibit NMD only in the case of human SMG7, which dephosphorylates and hence deactivates UPF1 (51), or the secondary isoform of human UPF3, called UPF3A, which out-competes the stronger NMD activator UPF3B for participation in NMD complexes (52). An intense elevation in the transcript levels of other human NMD factors either enhanced NMD efficiency (for SMG1, SMG5 and SMG6) or did not alter it (for UPF1, UPF2 and UPF3B) (33).
Interestingly, the group of transcripts that showed the highest enrichment score in U3::NR transformed WT plants was entirely composed of snoRNAs. Most snoRNAs are involved in modifying ribosomal RNAs and spliceosomal small nuclear RNAs (53). The change in snoRNA content apparently did not reflect some pleiotropic impact of extensive UPF3 overexpression because the extent of increase in UPF3 transcript levels was moderate, in the range of that naturally occurring in NMD impaired plants (Supplementary Figure S2), i.e. in the range of physiological relevance. Furthermore, the expression of UPF3 was carried out under the control of its own promoter, i.e. in regions of physiological relevance. The phenotype of the U3::NR transformed WT plants was apparently similar to that of WT plants (data not shown). Moreover, there was a strong tendency for reduced expression of the same snoRNAs in upf3 and upf1 mutant plants ( Figure 8D). Thus, the correlation between snoRNAs content and the integrity of the NMD pathway was detected not only in the U3::NR plants, which have increased UPF3 expression, but also in upf mutants. The UPF3 protein was previously shown to be enriched in the Arabidopsis nucleolus (6). Yet, the changes we observed in snoRNA content are apparently not a simple function of UPF3 levels, because UPF3 expression is higher compared to WT plants in both U3::NR plants and upf1 mutants (Supplementary Figure S2). The reduction in snoRNA levels in both upf3 and upf1 mutants hints that, while snoRNAs are not direct NMD substrates (otherwise their levels in the mutants should have been increased), snoRNA levels may be collectively affected by putative regulatory or structural protein(s) whose expression is affected by NMD. The expression of these putative protein(s) could be inversely affected in upf mutants as compared to U3::NR plants, presumably due to enhanced NMD of some primary target(s) in the latter plants. Specifying such presumed primary target(s) is currently not possible due to the lack of sufficient knowledge about the regulation of snoRNA expression. Nevertheless, our observations in upf mutants indicate that a functional NMD pathway is essential for proper homeostasis of Arabidopsis snoRNAs.
While a non-functional NMD pathway leads to snoRNA depletion in Arabidopsis, it was recently shown that NMD impairment does not alter snoRNA content in human cells. Most vertebrate snoRNA coding units are intronic, nested within introns of protein-coding host genes (53). Interestingly, NMD impairment increases the RNA levels of human snoRNA host genes, but does not affect the levels of their nested snoRNAs (36,37). Regulation of those human host genes by alternative splicing and NMD was suggested to uncouple their expression from that of their nested snoR-NAs (37). In contrast to vertebrates, most (except few) Ara-bidopsis snoRNAs are not intronic (53). Many Arabidopsis snoRNAs are found in polycistronic clusters (53). Examination of all 64 snoRNA genes on the array showed that the extent of their modification in U3::NR and upf mutant plants did not correlate with a specific genomic organization of these genes, i.e. their being intronic or non-intronic or being organized in clusters or as single genes (Supplementary Dataset S4). This suggests that the process that affects snoRNA content upon NMD impairment in Arabidopsis does not depend on the genomic organization of the snoRNAs. Further knowledge about the regulation of snoRNA expression will be necessary for identification of the affected process.
Overall, this work demonstrates that the NMD feedback loop dominating UPF3 transcript levels, as well as fine restriction of UPF3 transcription, are crucial for NMD homeostasis in plants. As mentioned, an intense elevation in the transcript levels of UPF3B did not affect NMD in mammalian cells (33). Further study will be necessary to determine how can an increase in UPF3 expression inhibit plant NMD and what underlies the contrasting findings in plant and mammalian cells. Yet, some hypotheses can be raised based on the current knowledge about NMD. One possible explanation for our findings is that, when in excess, UPF3 acts in a dominant negative manner, by binding other NMD factor(s) with which it can interact in inappropriate timing or locations, thereby depleting the levels of these factors that can be targeted to other locations or undergo different interactions. It was shown that human UPF3B can interact with UPF2 and with EJC proteins [(54,55); see 'Introduction' section]. Similarly, plant UPF3 was shown to bind UPF2 (12). There were some indications that yeast and animal UPF3 proteins could undergo direct interaction with UPF1 (55-57), but plant UPF3 failed to immunoprecipitate with UPF1 in the absence of UPF2 (12). Depletion of plant EJC proteins impaired only intron-based NMD, but did not impair NMD that was dependent on long 3 UTRs (12). Among the two model genes utilized here, M2 contains many introns, but M1 transcript does not contain any intron. Therefore, it is unlikely that the ability of excess UPF3 to inhibit NMD of the intronless M1 transcript resulted from depletion of EJC proteins. This leaves UPF2 as a possible interacting candidate that could be depleted by excess UPF3. Although the nucleolus is not the site were NMD occurs (13,58), it was shown that the Arabidopsis UPF3 is predominantly localized in the nucleolus, UPF2 is found in both the nucleolus and the cytoplasm, and UPF1 is largely cytoplasmic (6). The fact that both UPF3 and UPF2 are enriched in the nucleolus raises the hypothesis that excess UPF3 can inhibit NMD in plants by binding UPF2 in the nucleolus and depleting the amounts of this factor that remain available for shuttling into the cytosol, were NMD occurs. It should also be considered that in contrast to human cells (33), NMD impairment in plants does not lead to a compensatory increase in UPF2 expression, since our microarray analysis indicated that expression of UPF2 remained unaltered in upf1, upf3 or U3::NR plants (data not shown). In addition, the transcript of the Arabidopsis UPF2 gene (AT2G39260) does not include any potential NMDinducing feature. The inability of plant cells to elevate UPF2 expression upon NMD inhibition may enhance the poten-tial of excess UPF3 to deplete UPF2. According to our hypothesis, excess UPF3B does not inhibit NMD in human cells because expression of human UPF2 can increase upon NMD impairment (33) and/or because human UPF3B and UPF2 are not enriched in the same cellular compartment (27,59). In human cells, UPF3B was predominantly localized in the nucleus (and was not concentrated in the nucleolus), UPF2 was predominantly cytoplasmic or perinuclear, and UPF1 was cytoplasmic (27,59). Further study will be necessary to test this hypothesis.

SUPPLEMENTARY DATA
Supplementary Data are available at NAR Online.