Global analysis of ribosome-associated noncoding RNAs unveils new modes of translational regulation

Phosphate Deficiency Impacts the Translational Regulation of a Subset of mRNAs.

A comparative transcriptome and ribosome-footprinting study was performed on the complete root system of seedlings initiated on replete Pi (500 µM Pi, +P) medium for 7 d and then transferred to replete or deficient (12.5 µM Pi, −P) medium for an additional 7 d. The long-term low-Pi treatment triggered the typical adaptative response of the Arabidopsis root system with attenuated growth of the primary root and stimulation of the emergence and elongation of lateral roots (Fig. 1A), a plasticity in development that enhances capture of mineral reserves in the rhizosphere near the soil surface (33). Root tissue was used to isolate total cellular poly(A)⁺ mRNA as well as to determine the position of individual cytoplasmic 80S ribosomes on transcripts (Fig. S1A). For the latter, ribosome–mRNA complexes were captured by translating ribosome affinity purification (TRAP), which takes advantage of a 60S subunit ribosomal protein L18 that is FLAG-epitope tagged and expressed under the control of the near-constitutive CaMV 35S promoter (34). The polyribosome complexes that were purified were treated with RNase I to generate 80S ribosomes, which were processed to obtain 26- to 32-nt RFs for synthesis of a strand-specific ribo-seq library. These were sequenced to a depth of 54–81 million reads. After filtering out fragments of housekeeping RNAs [rRNA, tRNA, small nucleolar RNA (snoRNA)], 81–87% of the RFs mapped uniquely to protein-coding sequences (CDSs), ≤2.7% to 5′-UTRs, and ≤1% to 3′-UTRs. Strand-specific mRNA-seq libraries were prepared with chemically fragmented poly(A)⁺ mRNA and sequenced to a depth of 28–43 million reads. There was a high correlation (r ≥ 0.98) between biological replicates of both library types (Dataset S1a), confirming reproducible affinity purification and processing of mRNA–ribosome complexes.

To evaluate the efficiency and precision of the RNase I digestion, a metagene analysis was performed to map the position of the 5′-end of 28-nt RFs and their number on annotated protein-coding regions (Fig. 1B). The distance between the ribosome peptidyl site and the 5′-end of RFs was 14–15 nt, a signature that commenced at the AUG codon, with only a fraction of RFs mapping to 5′-UTRs (Fig. 1B). The increase in RF coverage at the AUG and decline at the stop codon is consistent with the rate-limiting nature of initiation and termination of translation (35). Finally, over one-half (56%) of the RF 5′-termini mapped to the first nucleotide of individual codons (Fig. S1B), a signature of the trinucleotide periodicity of ribosome decoding.

We first used our data to compare dynamics in transcript abundance and translation in response to Pi starvation. The response to 7-d Pi starvation by 15,297 protein-coding mRNAs was highly correlated between mRNA-seq and ribo-seq datasets (Fig. 1C; r = 0.76; Dataset S1b), and the differentially expressed genes defined by both datasets had similar Gene Ontology (GO) term enrichment (Dataset S1, c and d). This was not unanticipated because sucrose density gradient fractionation of polysomes indicated no difference in global levels of protein synthesis (Fig. S1B). There was an overlap of 84 up-regulated and 163 down-regulated genes between analyses [Fig. S1D and Dataset S1b; |log₂ fold change (FC)| ≥1; false-discovery rate (FDR) ≤ 0.05]. Of the mRNAs previously reported to be differentially regulated in roots by short-term (3–12 h), medium-term (24–48 h), and long-term (12-d) Pi depletion (5 µM Pi) (36), 86 and 84 of these were differentially expressed in our mRNA-seq and ribo-seq data, respectively, including 70 recognized in both studies (Dataset S1b).

Our data enabled the evaluation of differences in translational efficiency (TE) for individual mRNAs, defined as the proportion of ribo-seq to mRNA-seq reads on each protein-coding ORF. In contrast to the strong concordance between changes in mRNA-seq and ribo-seq transcript abundance in response to Pi deficiency, 81 and 184 mRNAs displayed a significant increase or decrease in TE, respectively (FDR < 0.01; Fig. 1D and Fig. S1D and Dataset S1b). These translationally regulated mRNAs were present but not highly responsive to hormonal, nutrient, or abiotic stresses based on public transcriptome datasets (Dataset S1i), indicating that they are not generally regulated at the level of transcript abundance (Fig. S1E). Our data confirm independent modulation of mRNA accumulation and ribosome association by Pi availability, as seen under other environmental responses [photomorphogenesis (20), hypoxia (19), water deficit (21), and heat stress (37)].

mRNAs with a significantly higher TE under Pi depletion were overrepresented for “transporter activity” (P < 7.88E-04) and “kinase activity” (P < 4.19E-05) (Dataset S1e), whereas those with reduced TE were enriched for “transcription” (P < 8.54E-05), with a bias for cellular component “ribosome” (P < 5.34E-06). The reduced translation of ribosomal protein mRNAs under abiotic stress is well established (reviewed by refs. 23 and 38). The decline in TE was more limited under low Pi than under hypoxia (Fig. S1F). Collectively, these data demonstrate that Pi homeostasis selectively influences translation of a subset of cellular mRNAs.

Phosphate Availability Selectively Alters Ribosome Occupancy on Upstream ORF-Containing mRNAs.

The translation of individual mRNAs is often mediated by the presence of uORFs that may be sensitive to cellular metabolites or other factors (23, 38). To discern whether Pi-mediated changes in TE involved the presence and translation of uORFs, we identified mRNAs with nonoverlapping uORFs (>60 nt) located 5′ to the main protein-coding ORF (mORF) (Dataset S1f). Sixty-three of the 4,505 uORF-possessing mRNAs displayed Pi-modulated TE (Fig. S2A), of which 16 had significantly more RFs on the uORF relative to the mORF under Pi-replete conditions (Dataset S1g; repressive uORFs; Fig. S2B). Most of these displayed limited change in mRNA accumulation in response to Pi availability. Of the uORFs identified, 51 were evolutionarily conserved peptide (CP) uORFs (23, 38) (Fig. S2C). Over one-half of the CPuORFs had fewer ribosomes on the downstream mORF under Pi-replete or -deficient conditions, consistent with the propensity of translated uORFs to limit ribosome (re)initiation at a downstream start codon. However, the TE of eight CPuORF mRNAs increased significantly under Pi deficiency, concomitant with a decrease in uORF-to-mORF ratio of RFs and without a change in mRNA abundance (Fig. S2D and Dataset S1h). Two of these encode putative transcriptional regulators, CYTOKININ RESPONSIVE FACTOR 10 (CRF10) (39, 40) and AUXIN RESPONSE FACTOR 4 (ARF4) (41) (Fig. 1E). In both cases, the increase in mORF translation exceeded log₂ 1.3-fold under Pi deficiency. Although CRF10 function is unknown, other CRFs promote lateral root growth (39, 40). The regulation and function of ARF4 is also less known. The uORFs of ARF3, 4, and 5 limit translation in protoplasts (42), with the control of ARF3 and ARF5 mRNAs translation requiring TOR kinase and a specific subunit of the translation initiation factor eIF3 (43). These data provide strong evidence that Pi homeostasis influences uORF-mediated regulation of genes involved in root system architecture.

Our data also revealed that Pi homeostasis influenced the translation of the uORF-containing PHO2 mRNA. mORF-to-uORF ratios for PHO2 suggest similar and significant uORF-mediated translational repression under both conditions (log₂ 2.35 [+Pi], log₂ 2.43 [−Pi]; FDR < 7.0E-9) (Dataset S1b). Consistently, RFs were abundant in the PHO2 5′ leader before the mORF, particularly in the region of three small nonoverlapping uORFs of 15, 9, and 13 codons (Fig. 1E). These precede a cluster of five miR399 binding sites including two sites that are uncleavable (44). Most plant miRNAs guide transcript cleavage, although some mediate bona fide translational repression involving localization to the endoplasmic reticulum (45–47). Based on data from in vitro translation systems and uncapped RNA-sequencing analysis, RNA-induced silencing complex (RISC) bound to cleavable miRNA binding sites does not impair the elongation phase of translation (45–47).

We reasoned that Pi-regulated changes in RISC bound to the cleavable or uncleavable miR399 sites could limit PHO2 mRNA translation. To address this hypothesis, we compared the number of RFs on the full PHO2 5′-leader relative to the mORF. This ratio was greater under Pi-replete conditions (log₂ 2.19 ± 0.26 [+Pi], log₂ 1.56 ± 0.31 [−Pi]). Taking both the decrease in RF coverage on the full PHO2 5′ leader and the twofold decline in PHO2 mRNA under Pi deficiency (Dataset S1b) into consideration, repression of mORF translation may be less under low-Pi conditions and this could be due to elevated miR399-mediated cleavage (48). The increase in the eTMs IPS1/IPS2 (log₂ 5–10; FDR < 3.88E-5; Dataset S1b) and miR399 under Pi deficiency (48) likely modulates the balance between translational inhibition and cleavage to fine-tune PHO2 synthesis according to availability and use of Pi. These results suggest a connection between miRNA cleavage and translational initiation in the control of a key regulator of Pi transporter abundance.

A Subset of Root lncRNAs Are Associated with Ribosomes.

Next, we combined our mRNA-seq and ribo-seq data with sRNA-seq and double-stranded RNA (dsRNA)-seq data obtained with the same root samples to classify lncRNAs expressed under the two conditions. Annotated and de novo predicted lncRNAs were combined as input into a classification pipeline (Fig. 2A). After removal of lncRNAs with insufficient mRNA-seq and ribo-seq reads, 2,382 poly(A)⁺ lncRNAs were identified. Remarkably, over one-half (1,234) were candidate ribosome-associated lncRNAs (ribo-lncRNAs). The dsRNA-sequencing (dsRNA-seq) data were used to filter out highly structured RNAs with regions resistant to RNase I, a step we find beneficial to exclude snoRNAs and other structured RNAs that contaminate RF libraries (19). In this manner, 1,140 ribo-lncRNAs were identified, corresponding to ∼48% of all detected in seedling roots. These were further subclassified as lincRNAs (intergenic), cis-NATs (>50-nt overlap with a protein-coding gene exon), and phasiRNA/siRNA precursors (sRNA clusters enriched for a specific read length and/or showing a phased distribution) (Fig. 2B and Dataset S2a).

The majority of lincRNAs (572 of 710; ∼80%) were associated with RFs. As an example, RFs mapped toward the 5′-terminus and largely overlapped with the first sORF on the Pi-regulated eTMs IPS1 and IPS2 (Fig. 2C and Fig. S3A). Notably, 12 of 15 predicted eTMs were ribo-lncRNAs (Dataset S2a). By contrast, the nuclear lncRNAs ASCO (Fig. 2C) and APOLO (Fig. S3B) lacked RFs. NATs comprised the largest class of lncRNAs (1,676), including 568 with a significant number of RFs. Similar processing of three published ribo-seq datasets from seedlings (20, 24, 49) detected 567 ribo-lncRNAs overall, including 114 ribo-lncRNAs detected in all analyses (81 lincRNAs, 33 NATs), including 8 out of 11 detectable phasiRNA precursors (Fig. S3C and Dataset S2a). Pi-regulated accumulation was evident for 67 lncRNAs (|log₂ FC| ≥ 1, FDR ≤ 0.05), including 56 ribo-lncRNAs (44 lincRNAs, 12 NATs) (Dataset S2a). However, we found no evidence of ribosome-associated miRNA precursors. Taken together, many lincRNAs and NATs may be translated in plants.

Translated sORF Are Hidden in Plant lncRNAs.

All 1,140 ribo-lncRNAs contained at least one sORF encoding ≥10 aa, but only 397 met the threshold of ≥1 rpkM for RFs on at least one sORF of ≥30 nt (Dataset S2b). To seek evidence of sORF translation, we determined the ribosome release score (RRS) (15) (Fig. 2A), which quantifies the decline in RF number (drop-off) seen after the termination codon (Fig. 1C). Comparison of TE and RRS values of protein-coding and ribo-lncRNA ORFs identified 225 putative lncRNA_sORFs based on a depletion of RFs in the 3′-UTR, as seen for annotated protein-coding ORFs (Fig. 3A). Moreover, RRS values of characterized sPEP mRNAs almost completely overlapped with those encoding proteins of >100 aa (Fig. 3A).

As a second metric, we monitored the trinucleotide periodicity of RF 5′-termini on individual transcripts as seen at the global level (Fig. 1B). This reflects the precision in codon decoding and is evident in ribo-seq datasets if RNase I digestion is taken to completion. Recognition of translated ORFs based on codon trinucleotide periodicity quantified using RiboTaper (50) recognized 14 lncRNA_sORFs (P < 0.05; Dataset S2b), including 9 with a significant RSS score (Fig. 3B). This indicates that RiboTaper was more restrictive in sORF discovery for our dataset. Previous RiboTaper analysis of root RFs identified 23 sORFs (24), including 21 that we detected here using RRS or RiboTaper.

Exploration of the evolutionary conservation of the putative lncRNA_sORFs identified 31 present in Brassicaceae outside of the Arabidopsis genus. Of these, 9 were broadly conserved in angiosperms, 13 were recognizable in some eudicots and monocots, and 9 were limited to Brassicaceae (Fig. S4 and Dataset S2b). Three putative lncRNA_sORFs were unannotated members of conserved sPEP families [C-terminally encoded (CEP), CLAVATA-like (CLE), PAMP-induced secreted peptide-like (PIPL)] and one was a new member of an uncharacterized family of sPEPs with a conserved 13-aa C-terminal domain (Fig. S5).

To establish whether lncRNA_sORF-encoded peptides are synthesized and accumulate in planta, we searched 1,653 mass spectrometry files from five previously established proteomic datasets (51–55) from diverse Arabidopsis tissues against a TAIR10 protein database extended with predicted sPEP sequences. Bona fide translation of lncRNA_sORFs was supported by mass spectra of 19 predicted sPEPs supported by RSS (Fig. 3C and Fig. S6 and Dataset S2b). These included sPEP fragments of three conserved sORFs supported by RSS but not RiboTaper. The sPEPs can produce limited tryptic peptides, and therefore it is unsurprising that data mining yielded N-terminal spectra, including three with the initial Met removed and an acetylated penultimate residue (Fig. 3C). Altogether, the validated lncRNA_sORFs ranged from an 11-residue low-Pi–induced micropeptide to a 100-residue polypeptide. An sPEP closely related to CLE26 was detected in all five independent datasets with a total of 32 spectra, providing strong support of its synthesis. Another peptide was from the first of two sORFs encoded by ribo-lncRNA AT2G09795.2 (Fig. 3C and Dataset S2b). The RF coverage and RSS score of the second sORF of this transcript support its translation, indicating this may be a polycistronic mRNA. The proteomic spectra also confirmed synthesis of 14 sPEPs encoded by uORFs (Dataset S2c), including a CPuORF of SUPPRESSOR OF ACAULIS 51 (SAC51) that has uORF-regulated translation mediated by availability of thermospermine (56).

Further support of the biological relevance of ribo-lncRNA_sORFs was obtained with surveys of evolutionary conservation. First, using the PhastCons score derived from whole-genome sequence alignments from 20 angiosperms, a method amenable to evaluation of the conservation of plants coding and noncoding sequences at the nucleotide level (57), we confirmed high conservation near the start codon of predicted ribo-lncRNA_sORFs with high but not low RSS values (Fig. S7A). Second, we monitored the sequence variation in ribo-lncRNA_sORFs and uORFs in 1,135 resequenced Arabidopsis accessions (58). As many of these ORFs had no missense mutations across the accessions, we calculated the proportion of ORFs lacking high-impact nucleotide variants for each class (Fig. S7B) and then determined the ratio of synonymous to missense mutations in the genes with high-impact variants (Fig. S7C). The occurrence of missense mutations was low in the conserved sORFs as well as in sORFs with detected sPEPs. As anticipated, there were few missense mutations in CPuORFs but also in the uORFs that repressed mORF translation under Pi-replete conditions, confirming positive selection on conserved sORFs and uORFs regulating translation.

Our data expand the sORFs repertoire predicted by ribo-seq (24) by nearly 10-fold and confirm the synthesis and detection of 19 sPEPs. Of particular interest was that translation was significantly up-regulated for eight and down-regulated for four lncRNA_sORFs by Pi deficiency based on RSS and RF values (Dataset S2b). One of the up-regulated lncRNA_sORFs encodes an 11-aa sPEP supported by six mass spectra (Fig. 3C). These results indicate lncRNA_sORFs can encode conserved peptides of biological relevance (59) and raise the possibility that sPEP signaling is modulated by Pi homeostasis.

Ribo-cis-NAT Regulation Is Associated with Translational Enhancement.

Many cis-NATs are differentially regulated by environmental conditions in plants, but very few have been functionally characterized (reviewed by refs. 60 and 61). Our lncRNA classification pipeline identified 568 ribo-cis-NATs that are associated with dsRNA but not siRNA production, including 12 with significantly different accumulation under the two Pi conditions (Fig. 2B and Dataset S2 a and d). Among ribo-cis-NATs, 167 had ribosome-bound ORFs including 49 with an RRS score supportive of active translation. To explore whether these cis-NATs might regulate their sense mRNAs under low Pi, we quantified dsRNA-seq reads in overlapping cis-NAT and sense mRNA regions under the two conditions. This identified 143 cis-NATs regulated by Pi availability (dsRNA-seq; log₂FC > 1; FDR < 0.05; Dataset S2d), including 41 ribo-cis-NATs. The corresponding sense mRNAs of five of these displayed significantly increased TE (log₂FC TE > 0.71; FDR ≤ 0.01; Dataset S1a) concomitant with elevation of their cis-NAT under Pi deficiency (Fig. 4A). Three of the five cis-NATs had an sORF with a high RRS score, and all five had RFs that mapped toward their 5′ termini (Fig. 4B), a characteristic subsequently recognized across all ribo-cis-NATs (Fig. S8A). The mRNAs displaying ribo-cis-NAT regulation included two ATP-BINDING CASSETTE SUBFAMILY G transporters (ABCG2, ABCG20), required for endodermal suberization that influences nutrient uptake and lateral root formation (62, 63), and a POLLEN-SPECIFIC RECEPTOR-LIKE KINASE 7 (PRK7) family member associated with root cell elongation (64). PRK7 mRNA is enriched in root-hair cells (65), which elongate under low Pi to increase the surface area for nutrient uptake. Increased translation under Pi deficiency of all five mRNAs was concomitant with a significant elevation in dsRNA production (Fig. 4B and Fig. S8B), indicating sense–antisense interaction. With the exception of the PRK7 cis-NAT, levels of poly(A)⁺ cis-NAT RNA was not influenced by Pi deficiency, indicating regulation occurs at multiple levels in addition to ribosome association.

We propose that translation may stabilize or appropriately target cis-NATs and other ribo-lncRNAs within cells for interaction with their sense target. The five ribo-cis-NATs described above had limited nucleotide variation within their ORF with the highest RSS score across sequenced Arabidopsis accessions (Fig. S7 D and E). Nucleotide variants of significance were overall higher in NAT sORFs than in ORFs of sense transcripts. Nonetheless, point mutations in ribo-cis-NAT 5′-sORFs could impact their abundance or interaction with sense mRNAs, explaining phenotypic variability between Arabidopsis ecotypes (8) and contributing to the diversity of developmental responses to Pi starvation among ecotypes and species (66).

sORF Translation Enhances TAS3a Stability.

The ribo-lncRNAs discovered included known precursors of tasiRNAs (67–69). As an example, RFs mapped just 5′ of the tasiRNA-generating regions of TAS1b, TAS2a, TAS3a, and TAS4 (Fig. 2C and Fig. S9). sRNA-seq data were used to explore the relationship between ribosome occupancy and siRNA biogenesis (Fig. 2A). This yielded 54 and 179 ribo-lncRNAs that coincided with 21- to 22- and 24-nt sRNA clusters, respectively (Dataset S2a). Notably, the RFs on 13 of 14 ribo-lncRNAs that produced 21- to 22-nt phasiRNAs (phase score > 20) were positioned on sORFs located just 5′ of the region that generated the sRNAs (Fig. 2 B and C; AT1G62860, TAS3a; Fig. S9; other tasiRNA loci). Of the 13, 4 had an RRS score highly supportive of translation in seedling roots (Dataset S2c). TAS3a was particularly notable. First, the termination codon of its 52-aa sORF is positioned just 5′ of the noncleavable miR390/ARGONAUTE 7 (AGO7) binding site (67) (Fig. 5A and Fig. S10A). Second, RFs map precisely to the TAS3a sORF in all ribo-seq datasets we generated or reevaluated (19, 20, 24, 49) (Fig. S10B). Third, TAS3s from diverse plants have a similar codon length (50–58 aa) but poorly conserved sORF positioned 8–20 nt 5′ of the miR390 binding site (Fig. S10 C and D). Based on this, we hypothesized that TAS3a sORF translation may promote the production of its phasiRNAs (tasiARF1/2).

To test this, a series of mutations in the TAS3a sORF were made, and levels of TAS3a and tasiARF1/ARF2 were monitored in Nicotiana benthamiana (Fig. 5 A–C). The removal of the first and second in-frame AUG (TAS3a-m1), so that the sORF was reduced to 22 codons commencing just upstream of the miR390 binding site, dramatically reduced TAS3a abundance and tasiARF1/2 production. By contrast, a nonsense mutation at the second in-frame AUG (Met₂₉ → STOP) that reduced the sORF to 28 codons (TAS3a-m2) had no effect. To evaluate the impact of these mutations on TAS3a association with ribosomes, the constructs were tested in N. benthamiana that produces FLAG-tagged AtRPL18B. TRAP was used to isolate ribosomes to avoid contamination with other ribonucleoprotein complexes. The proportion of TAS3a RNA associated with ribosomes ranged from 55% for the nonmutated sORF to just over 20% for TAS3a-m1 (Fig. 5D). This suggests that altering the start site but not necessarily the length of the TAS3a sORF significantly decreased the proportion of RNA that copurified with ribosomes.

These results indicate that translation from the first in-frame AUG coordinately elevates TAS3a abundance, and subsequently phasiRNA accumulation. Monitoring the rate of decline in TAS3a and TAS3a-m1 abundance following inhibition of transcription with the nonfunctional adenosine analog cordycepin revealed that TAS3a-m1 RNA was significantly less stable than TAS3a (Fig. 5 E and F). These data support the conclusion that evolutionarily conserved translation of a nonconserved sORF enhances TAS3 RNA stability and consequently promotes phasiRNA accumulation.

Other studies point to TAS RNA translation as a critical factor in its production of phasiRNAs. First, the mapping of 5′ ends of truncated RNAs identified a signature of ribosome association of TAS3 sORF which was lost in the absence of endomembrane-localized AGO7 (70, 71). Second, TAS RNAs required localization in a subset of polysomes anchored to the rough endoplasmic reticulum to properly trigger phasiRNAs biogenesis (47). Our observation that manipulation of TAS3a sORF position can decrease TAS3a RNA stability indicates that translating ribosomes limit its degradation. The 5′- to 3′-EXORIBONUCLEASE 4 (XRN4) is involved in degradation of numerous mRNAs and competes with RNA-DEPENDENT RNA POLYMERASE 2 and 6 to process miRNA targets after mRNA cleavage (72). However, TAS transcript abundance is not affected in xrn4 mutants, and therefore how they escape XRN4-mediated degradation is unclear (73). As miRNA-mediated cleavage of TAS2 was recently reported to occur in association with ribosomes and to require a translatable sORF (74), it will be of interest to see whether cytosolic lncRNA stability and function is generally linked to the position and translation of an sORF.

Plant Growth and Manipulation.

Arabidopsis thaliana (Col-0) expressing 35S:His₆FLAG-RPL18B (34), grown on Murashige and Skoog (MS/10) medium (Pi-replete, 500 µM), were transferred to Pi-replete (500 µM) or Pi-deficient (12.5 µM) MS/10 for 7 d. Growth was at 23 °C, 16-h light (80 μE⋅m⁻²⋅s⁻¹)/8-h dark cycle. For transient assays, leaves of 2- to 3-wk-old Nicotiana benthamiana expressing 35S:His₆FLAG-AtRPL18B (75) were infiltrated with Agrobacterium tumefaciens (AGL-0) (76) carrying T-DNA plasmid pB7GW2D (77) with TAS3a constructs. To assay mRNA decay, one fully expanded leaf per plant per construct and per bioreplicate was Agro infiltrated; 48 h later, leaf discs were prepared, vacuum infiltrated with 1 mM cordycepin, and sampled after 0, 30, 60, and 120 min.

RNA Analyses.

Total RNA was isolated using TRIzol (Invitrogen), treated with DNase I and reverse transcribed. Real-time (RT)-PCR was used for mRNA quantitation and stem-loop RT-qPCR (78) using two technical replicates each with five leaf discs was used for tasiRNA quantification. Relative differences were calculated by the ΔΔCt method (79) and RNA half-life was determined from the exponential regression of decay (80). To assay mRNAs associated with ribosomes, TRAP was performed (81) using magnetic anti-FLAG–coupled protein G Dynabeads. Results were analyzed using a percent-of-input method and Student’s t tests using three bioreplicates.

Ribo-Seq, mRNA-Seq, dsRNA-Seq, and sRNA-Seq Library Synthesis and Sequencing.

Ribo-seq libraries were prepared with ribosomes obtained by TRAP using root tissue (81) and EZview Red anti-FLAG M2 magnetic beads. RFs were generated by RNase I, purified, size selected, depleted of contaminating rRNA, and processed into libraries (82). This included poly(A)⁺ mRNA-seq libraries that were prepared from DNase I-treated RNA extracted from the clarified supernatant (S-18 fraction) of the TRAP protocol (81) as described (83). sRNA of 18–26 nt from total RNA was prepared for sequencing using the NEXTflex Small RNA Sequencing Kit (BIOO Scientific). dsRNA-seq libraries were prepared as described (84).

Bioinformatic Analyses.

Short read sequencing data were analyzed using a combination of Unix software and R packages from Bioconductor. Read count data (rpkM) for protein-coding gene features were generated as described (19, 85). Statistical identification of differentially expressed genes and feature types was by the generalized linear model, applying FC and FDRs. Enrichment analysis of GO terms was performed with the Classification SuperViewer tool (bar.utoronto.ca). Translational efficiency of protein-coding genes and other ribo-seq statistics were obtained using systemPipeR (85). uORFs were predicted from 5′-UTR sequences from Araport 11 (86) with the predORF function using ATG as start codon and ≥60 nt. RFs on uORFs and mORFs were determined as in ref. 19. For gene feature analysis, we computed coverage from the first nucleotide of each 28-nt RF (87). Data were visualized using normalized and merged datasets. Detailed procedures of the lncRNA pipeline, prediction of sORFs translation with RRS (15) and RiboTaper (50), and analysis of published proteomic data are given in SI Materials and Methods.