The reference genome sequence and transcriptome annotation GTF files of Arabidopsis thaliana were downloaded from The Arabidopsis Information Resource (TAIR10.47). The previously published strand-specific total RNA-seq and small RNA (sRNA)-seq data derived from the leaves of Arabidopsis thaliana ecotype Columbia (Col-0) (Woo et al., 2016) were used in this study.

The TruSeq adapter sequences were removed using Trimmomatic to obtain clean reads (average Phred quality score ≥ 20), which were then aligned to the Arabidopsis reference genome (TAIR10.47) using TopHat2, with a default parameter (mismatch ≤ 2 nt) (Kim et al., 2006a; Bolger et al., 2014). To predict novel transcripts, mapped reads from each bam file were assembled using Cufflinks with the -M parameter (Trapnell et al., 2010). The resultant GTF file was merged with the annotated lncRNAs from TAIR10.47 using cuffmerge. Read counts were then generated using HTseq-count, and expression levels were estimated as transcripts per million (TPM) (Anders et al., 2015). To select Arabidopsis lncRNAs, only transcripts with Cufflinks class codes ‘u’ (intergenic transcripts), ‘x’ (exonic overlap with reference sequence on the opposite strand), and ‘i’ (transcripts entirely within the intron) were retained. Then, short transcripts (<150 nt) and low-abundance transcripts (maximum TPM [TPMmax] < 1) were removed. Unannotated transcripts were named using the Cufflinks annotation (XLOC_).

Small RNA and their precursors were predicted based on the previously published small RNA-seq data of aging leaves (Axtell, 2013b; Woo et al., 2016) using Shortstack with default parameters. Then, the ribosome footprint sequencing (Ribo-seq) data of the leaves of 3-week-old plants (Lukoszek et al., 2016) were used to predict the ribosome-associated lncRNAs (ribo-lncRNAs). Subsequently, putative translated sORFs were predicted by calculating the ribosome release score (RRS), which indicates whether the ribosome footprint decreases after the termination codon of sORFs (≥30 nt, 10 amino acids) (Guttman et al., 2014; Bazin et al., 2017).

Differentially expressed transcripts were identified using the DESeq2 method (Love et al., 2014). Enrichment analysis of Gene Ontology (GO) terms was performed using DAVID v6.8 (Huang et al., 2008).

Interaction energies between AR-lncRNAs and mRNAs were calculated from the FASTA file of TAIR10.47 using RIBLAST 1.1.1 (Fukunaga and Hamada, 2017a). with the following thresholds: interaction energy < -16 kcal/mol, and interaction length ≥ 15 nt. Pearson correlation coefficients of the predicted AR-lncRNA and mRNA pairs detected throughout the leaf lifespan were calculated using the average TPM values.

Arabidopsis thaliana ecotype Columbia (Col-0) is the wildtype strain for all mutants. Plants were grown in an environmentally controlled growth room (Korea Instruments, Korea) at 22℃ under 16h light:8h dark photoperiod and photosynthetic photon flux density of 130μmol m^-2s^-1. The Arabidopsis transfer DNA (T-DNA) insertion lines, SALK_100875 (at5g01595), SALK_151843 (atfer1) SALK_124431C (at1g33415), and SALK_135316 (at2g14878) were obtained from Arabidopsis Biological Resource Center. The genotype of each mutant line was confirmed by PCR-based genotyping.

For developmental leaf senescence, the third and fourth leaves of each plant, at the indicated age, were harvested at 4 to 5h after light-on. For dark-induced senescence, the 14-d-old third and fourth leaves of each plant were detached and floated upside down on 3mM MES buffer (pH 5.7) in 24-well plates, which were completely wrapped with aluminum foil. The photochemical efficiency (F_v/F_m) ratio of photosystem II was measured by a Walz IMAGING-PAM machine.

Total RNA was extracted from third and fourth rosette leaves were extracted with TRIzol and treated with DNase I (Ambion). cDNA was synthesized using the ImPromII™ system (Promega) reverse transcription kit following the manufacturer’s instruction. The extracted cDNA was used for semi-quantitative and quantitative real time Polymerase chain reaction (RT-PCR and qRT-PCR). qRT-PCR analysis was carried out to determine the gene expression levels (CFX96 system, Bio-Rad). Transcript abundances of target genes were analyzed by the comparative threshold method, with ACTIN2 (AT3G18780) as the internal control. For visualization of amplified cDNA band in RT-PCR, 10µl of the amplified product was run in 0.8% agarose gel containing ethidium bromide.

Previously, we generated multidimensional transcriptome data of 4-d- to 30-d-old Arabidopsis leaves, and characterized the regulatory features of leaf senescence (Woo et al., 2016) ( Supplementary Data 1 ). In this study, we re-analyzed our previously generated RNA-seq dataset to assess the role of lncRNAs in the development of leaf organs. A total of 700 million reads were aligned against TAIR10 using TopHat2 (Kim et al., 2006a), and the mapped reads were assembled using Cufflinks (Trapnell et al., 2010). This led to the identification of 59,368 unique transcripts, corresponding to 31,405 gene loci. Among the 59,368 unique transcripts, 50,465 were previously annotated as protein-coding transcripts or as noncoding RNAs such as housekeeping RNAs (e.g., tRNAs, rRNAs, small nuclear RNAs, and small nucleolar RNAs) and miRNA precursors. In addition to these 50,565 transcripts, 4,815 short transcripts (length < 150 nt) and 3,262 low-abundance transcripts (TPMmax < 1) were excluded from the dataset of unique transcripts. Subsequently, 54 transcripts showing protein-coding potential, as determined by the Coding Potential Calculator (CPC) (Kang et al., 2017), were further eliminated. Thus, 771 lncRNAs were identified, of which 539 were previously annotated and 232 were novel ( Figure 1A , Supplementary Data 2 ).

Based on their predicted functions, the 771 lncRNAs were classified into the following three categories: 1) ribo-lncRNAs, lncRNAs that potentially encode an sORF or are associated with the stability and translation of their cognate mRNAs in trans; 2) sRNA precursors, lncRNAs that generate precursors of small RNAs such as miRNAs and trans-acting or phased small interfering RNAs (tasiRNAs/phasiRNAs); and 3) other canonical lncRNAs not included in the first two categories ( Figure 1A ). Ribo-lncRNAs (145/771 [18.78%]) were predicted using ribo-seq data generated from the leaves of 3-week-old Arabidopsis plants (Lukoszek et al., 2016). Translated sORFs are often hidden among ribo-lncRNAs (93/145 [64.14%]). Potential sORFs encoding > 10 amino acids from ribo-lncRNAs were recognized using two analytics: RRS, which evaluates the decrease in ribosome footprint number after termination codons (Bazin et al., 2017); and getorf, which finds and outputs the sequence of ORFs in nucleotide sequences (Rice et al., 2000). This analysis allows to identify putative lncRNA-encoded sORFs with RRS ≥ 0.9 (29). The remaining ribo-lncRNAs might be involved in the stabilization and translation of their cognate mRNAs or in the trans-regulation of mRNAs. LncRNAs capable of generating the precursors of 21–22-nt long sRNAs (34/771 [4.40%]) were predicted based on the small RNA-seq data (Axtell, 2013a; Woo et al., 2016) ( Figure 1A ). Furthermore, we confirmed the RNA-seq read coverage of representative lncRNAs in each category (ribo-lncRNAs, sRNA precursors, and canonical lncRNAs) using gene viewer ( Figure 1B ).

In addition to the classification of 771 lncRNAs based on their predicted functions, we further classified these lncRNAs according to their genomic locations. Six genomic location-based categories of lncRNAs were identified: intergenic (325 out of 771 lncRNAs, 42.15%), antisense (305 [39.55%]), divergent (74 [9.60%]), convergent (96 [12.45%]), divergent & convergent (31[4.02%]), and intronic (3 [0.39%]) ( Figure 1C ). We then characterized the features of lncRNAs, such as average length, exon number, isoform number, and expression level, in each category, and compared the results with the features of protein-coding transcripts. The lncRNAs were shorter (average length = 878.757 bp) and contained fewer exons (average exon number = 1.72) than coding transcripts (2215.58 bp and 4.6 exons, respectively) (p < 0.0001, Mann-Whitney U-test, two-tailed) ( Figures 1D, E ). On the other hand, the number of isoforms of lncRNAs (1.34) was comparable with that of coding RNAs (1.6) ( Figure 1F ). The median expression levels of lncRNAs (average TPM along leaf age) were significantly (11-fold) lower than those of coding transcripts (p < 0.0001, Mann-Whitney U-test, two-tailed) ( Figure 1G ). These results are consistent with those of previous studies, which identified lncRNAs involved in other biological processes (Di et al., 2014; Tsai et al., 2022). The RNA-seq read coverage of the novel lncRNAs identified in this study was confirmed using gene viewer ( Supplementary Figure 1A ), and their expression levels were validated through the RT-PCR analysis of eight randomly-selected transcripts ( Supplementary Figures 1B, C ).

The functional transition of leaves during aging inspired us to examine the dynamic landscapes of lncRNAs. Of the 771 lncRNAs identified in this study, 446 (57.8%) were differentially expressed during leaf aging, as examined by DEseq2 (Love et al., 2014) (|log2(fold change)| ≥ 1, adjusted p-value [p_adj ] ≤ 0.05) ( Figure 2A ). Among the AR-lncRNAs, 192 and 292 lncRNAs showed dynamic changes of their expressions during the early biogenesis period (from growth [G] to maturation [M], 4–18 d) and late degeneration period (from M to senescence [S], 16–30 d), respectively, indicating that lncRNAs play important roles in leaf development. The majority of AR-lncRNAs were intergenic (44.39%), followed by antisense (37.44%), divergent (9.86%), convergent (12.11%), divergent & convergent (4.04%), and intronic (0.22%). The proportions of genomic location of AR-lncRNAs were similar to that of detected lncRNAs, and not significantly different between the G→M and M→S transitions ( Figure 2B ). Notably, the number of senescence-associated AR-lncRNAs (M→S) was greater than that of biogenesis associated-lncRNAs (G→M), suggesting that lncRNAs are more relevant to the leaf senescence process than to the leaf biogenesis process.

Differentially expressed lncRNAs were categorized into six major clusters, including three upregulated (U1–U3) and three downregulated (D1–D3) clusters, based on their expression kinetics over time, which represents 97% of the AR-lncRNAs ( Figures 2C, D ). We also examined the temporal expression profiles of lncRNAs showing age-dependent changes in transcript levels during the G→M and M→S transitions. During the G→M transition, the numbers of upregulated and downregulated AR-lncRNAs were similar (54% and 46%, respectively) ( Figure 2E ). Intriguingly, the majority of AR-lncRNAs (69%) were upregulated during the M→S transition. AR-lncRNA were preferentially upregulated, regardless of their genomic location ( Figure 2E ). Canonical lncRNAs were also preferentially upregulated, but rather similar numbers of ribo-lncRNAs and sRNA-precursor lncRNAs were up- and downregulated. The expression patterns of four randomly-selected AR-lncRNAs were verified by RT-PCR ( Figure 2F ).

Studies show that lncRNAs localize to various subcellular organelles, and regulate gene expression at various levels (transcriptional, post-transcriptional, and translational) (Bridges et al., 2021). Therefore, knowledge of the subcellular localization patterns of lncRNAs would provide information for inferring their gene regulation mode. We determined the subcellular localization of each AR-lncRNA by analyzing the publicly available transcriptome data of the cytosolic and nuclear fractions of 2-week-old Arabidopsis seedlings (Zhao et al., 2018). Of the 287 AR-lncRNAs identified in these transcriptomes, 211 (73.52%) were predominantly present in the nuclear fraction, whereas only 76 (26.48%) were enriched in the cytosolic fraction ( Supplementary Figures 2A, B and Supplementary Data 5 ). This pattern was robust, regardless of the functional categories of AR-lncRNAs. For instance, both sRNA-precursor lncRNAs (19/23 [81.61%]) and sORF-encoding lncRNAs (56/93 [60.22%]) were more abundantly localized in the nuclear fraction. To confirm this result, we performed qRT-PCR on six randomly-selected lncRNAs using cDNA isolated from the nuclear and cytosolic fractions of 2-week-old seedlings, which validated the subcellular localization of all, but one, lncRNAs ( Supplementary Figure 2C ).

Leaf development is an integrated response of plants to an innate developmental program and environmental stress responses. Thus, some of the genes involved in leaf senescence are expected to control environmental responses. We therefore investigated whether AR-lncRNAs are regulated by stress responses using the previously published transcriptome data of ABA-, drought-, and cold-treated Arabidopsis (Zhao et al., 2018). Among the lncRNAs differentially expressed by the ABA, drought, or cold treatment, a large proportion (68/102 [66.7%], 72/112 [64.3%], and 70/127 [55%], respectively) was also affected by the developmental age, suggesting an extensive overlap between leaf senescence and stress responses. This result is consistent with that of the previous study on protein coding mRNAs (Zhao et al., 2018) ( Supplementary Figures 3A, B and Supplementary Data 4 ).

Expression levels of some lncRNAs are significantly correlated with those of their neighboring protein-coding genes. Several lncRNAs are also known to control the expression of nearby genes (Statello et al., 2021), suggesting that these lncRNAs act as cis-regulators of genes. Thus, the positional relationship between lncRNAs and mRNAs in the genome would be important for predicting the lncRNA-controlled regulation of nearby genes. To infer the effect of lncRNAs on the expression of neighboring genes in aging leaves, we estimated the degree of co-expression between AR-lncRNAs and their adjacent protein-coding genes by calculating the Pearson correlation coefficients (PCCs). The co-expression of all pairs of age-related protein-coding genes was also analyzed for comparison. Notably, the PCC between pairs of antisense AR-lncRNAs and overlapping protein-coding genes was significantly higher than that between overlapping protein-coding gene pairs (p < 0.05, Mann-Whitney U-test, two-tailed) ( Figure 3A ). This result implies that antisense AR-lncRNAs act as cis-regulators of adjacent genes during leaf aging. Of the 168 antisense AR-lncRNAs, the TPM fractional density of 72 antisense AR-lncRNAs and their overlapping protein-coding genes pairs (PCC > 0.7) was visualized as a heatmap ( Figure 3B ). Gene ontology biological process (GOBP) enrichment analysis revealed that the neighboring protein-coding genes of antisense lncRNAs were significantly enriched by cytokinin catabolic/metabolic process, flavonoid glucuronidation, defense response, and oxidation-reduction processes ( Figure 3C ). Cytokinin is a representative hormone that negatively regulates leaf senescence in plants. The expression of cytokinin biosynthetic genes decreases, while that of cytokinin degradation genes increases during leaf senescence in Arabidopsis (Buchanan-wollaston et al., 2003; Breeze et al., 2011; Statello et al., 2021). Genes encoding two SOB five-like (SOFL) genes, AtSOFL1 and AtSOFL2, which act as positive regulator of cytokinin levels and cytokinin-mediated development including longevity (Zhang et al., 2009), are found to be overlapped with AT1G26210 and AT1G26208 AR-lncRNAs, respectively. Antisense AR-lncRNA AT3G63445 is overlapped with CYTOKININ OXIDASE (CKX) 6 that catalyzes the degradation of cytokinin.

Defense responses are also one of the typical age-associated biological processes (Kus et al., 2002; Mao et al., 2017). FERRITIN1 (FER1), overlapped with antisense AR-lncRNA AT5G01595, plays a role in iron hemostasis and accumulates upon exposure to oxidative stress or to pathogen attack, as well as developmental factor. Mutation of FER1 causes earlier onset of leaf senescence (Murgia et al., 2007). So, it is likely that AT5G01595 lncRNA might play a role in leaf senescence possibly through modulating FER1. To validate the functional role of AT5G01595, the senescence phenotype of the third and fourth leaves of knockout line (at5g01595) and wild-type (Col-0) plants during age-dependent natural senescence was compared ( Figure 3D ). Initiation of leaf yellowing, which is an indicator of chloroplast senescence in mesophyll cells, occurred earlier in at5g01595 than in Col-0. The photochemical efficiency (F_v/F_m), a representative physiological marker of leaf senescence, also declined rapidly. The early leaf senescence phenotype of at5g01595 was further confirmed by analyzing the expression of the SENESCENCE-ASSOCIATED GENE 12 (SAG12) and CHLOROPHYLL A/B-BINDING PROTEIN 2 (CAB2), the molecular markers of leaf senescence ( Supplementary Figures 4B, C ). We then tested the effect of AT5G01595 antisense lncRNA on the expression of FER1. The result showed that FER1 transcript level was significantly lower in at5g01595 than in Col-0 leaves ( Figure 3D ). These results indicate that AT5G01595 potentially contributes to leaf senescence by modulating FER1.

Together, these findings suggest that leaves might utilize antisense lncRNAs as a regulatory program for controlling biological processes, particularly cytokinin-related processes and defense responses, throughout its lifespan.

The lncRNAs can regulate mRNAs by sequestering specific miRNAs and mimicking their target recognition sequence in organisms. We searched for potential competitive endogenous RNAs (ceRNAs) involved in lncRNA–miRNA–mRNA interactions. We first integrated the TarDB (Liu et al., 2021) and TarBase (Karagkouni et al., 2018) datasets to search for mRNA–miRNA interactions, and StarBase (Li et al., 2014) to search for lncRNA–miRNA interactions in Arabidopsis. Using the hypergeometric test, potential ceRNA sets (lncRNA–miRNA–mRNA) were identified ( Figure 4A ) by evaluating the significance of interacting miRNAs shared by both mRNAs and lncRNAs (p < 0.05); these shared miRNAs were used as a junction. Among the potential ceRNA sets, we further narrowed down high-confidence ceRNA sets by calculating the PCCs of lncRNAs and their cognate mRNAs. The selection of lncRNAs and mRNAs with PCC > 0.7 led to the identification of 602 positively-correlated ceRNA sets ( Figure 4A and Supplementary Data 6 ). Eleven of the identified lncRNAs that paired with ceRNAs would likely compete with miRNAs involved in leaf development or leaf senescence, such as miR156, miR164, and miR169 ( Figure 4B ). Among these miRNAs, miRNA164, a negative regulator of leaf senescence, is known to mediate the cleavage of ORESARA1 (ORE1), which induces cell death and leaf senescence (Kim et al., 2009). One of the AR-lncRNAs, AT4G36648, was identified as a target of miRNA164 in this analysis, which involves in the ceRNA set linking AT4G36648-miRNA164-ORE1. Both AT4G36648 and ORE1 showed a rapid change in expression at the late degeneration stage. A strong positive correlation between AT4G36648 and ORE1 (PCC = 0.99) supports the presence of ceRNA that anchors these transcripts through miRNA164. Moreover, AT4G36648 was expressed during the M→S transition, which suggests the regulatory role of this ceRNA in leaf senescence. We also identified AT5G23410-miR169-NUCLEAR FACTOR Y (NF-Y) as ceRNA set. The expression level of the AR-lncRNA AT5G23410, followed by that of miR169-targeted NFYA5, which is known to modulate ABA-dependent stress responses (Li et al., 2008), was induced during aging. In this ceRNA set, the expression level of AT5G23410 was highly correlated with that of NFYA5 (PCC = 0.90). ABA is one of the hormones accelerating leaf senescence. Thus, this module is likely to be involved in mediating crosstalk between leaf senescence and stress responses. The AR-lncRNA AT1G26208 was identified as an interacting partner of miRNA156, which inhibits the action of SQUAMOSA PROMOTER BINDING-LIKE (SPL). Both miRNA156 and SPL form a regulatory module to control age-dependent developmental transition as well as abiotic and biotic stress responses (Wang et al., 2009; Ma et al., 2021b). Similar to the above two ceRNA sets, the age-dependent expression levels of AR-lncRNA AT1G26208 and SPL10 were highly correlated (PCC = 0.98), implying that the AR-lncRNA AT1G26208 modulates age-dependent pathways or integrates aging cues with stress responses.

Next, we reconstructed the ceRNA (lncRNA–miRNA–mRNA) interaction network. This network is composed of 106 mRNAs differentially expressed during aging, and 27 AR-lncRNAs which are commonly targeted by 38 miRNAs ( Figures 4C, D and Supplementary Data 6 ). To predict the potential function of AR-lncRNAs comprising the ceRNA network, the interacting mRNAs in the ceRNA network were subjected to the GOBP enrichment analysis. These genes were over-represented by the regulation of transcription and leaf senescence, suggesting that AR-lncRNAs perform an important regulatory role during leaf aging by participating in the ceRNA network ( Figure 4E ).

Numerous lncRNAs regulate gene expression by directly interacting with target mRNAs and affecting their stability or processing (Bardou et al., 2014; Sebastian-delaCruz et al., 2021). To identify putative mRNA-lncRNA pairs, we utilized RIBLAST, a computational tool used to predict comprehensive lncRNA–RNA interactions based on the seed-and-extension approach, where a target prediction was experimentally validated (Kretz et al., 2013; Fukunaga and Hamada, 2017b). We first calculated the interaction energy between the AR-lncRNA and mRNA sequences in the TAIR10 database, and then selected RNA segment pairs with < -16 kcal/mol interaction energy in 15 ≥ nt. This analysis led to the identification of 316,475 AR-lncRNA-mRNA pairs. These interacting pairs were further narrowed down based on the PCC values, resulting in the identification of highly co-expressed AR-lncRNA and target mRNA pairs (|PCC| ≥ 0.9) during the leaf lifespan. Through this analysis, we obtained 2,220 putative interactions among 446 AR-lncRNAs ( Figure 5A and Supplementary Data 7 ). To explore AR-lncRNAs associated with leaf senescence, we searched for pairs of AR-lncRNAs and mRNAs whose genes are known to be involved in the regulation of leaf senescence ( Supplementary Data 8 ) (Li et al., 2020). The predicted regulatory modules were further validated through the characterization of loss-of-function mutants. In this study, we focused on two AR-lncRNAs: AT1G33415 and AT2G14878 ( Figures 5B, C ).

Autophagy is required for nutrient recycling during leaf senescence. In Arabidopsis, Autophagy9 (APG9) is known to regulate the formation of autophagosomes, which are crucial for autophagy process, from the endoplasmic reticulum (Zhuang et al., 2017). The expression of APG9 is markedly upregulated during leaf senescence, and the apg9 mutant exhibits precocious senescence phenotypes, indicating that APG9 plays a negative role in leaf senescence (Zhuang et al., 2017). Given that AT1G33415 interacts with APG9 and the corresponding genes are highly co-expressed ( Figure 5B ), we decided to evaluate the involvement of AT1G33415 in the regulation of leaf senescence.

Firstly, we compared the yellowing phenotype of the third and fourth leaves of mutant (at1g33415) and wild-type (Col-0) plants during age-dependent natural senescence. The progression of leaf yellowing occurred more rapidly in the mutant than in the wild type ( Figure 5D ). The early leaf senescence phenotype of at1g33415 was confirmed by measuring the photochemical efficiency (F_v/F_m), which declined rapidly in at1g33415 ( Figure 5E ). We also conducted a dark-induced leaf senescence assay using detached leaves. Similar to the leaf phenotype observed during age-dependent senescence, dark-induced senescence symptoms (leaf yellowing and F_v/F_m decline) were also accelerated in the leaves of at1g33415 ( Figures 5G, H ). SAG12 expression was also analyzed for confirming the senescence phenotype ( Supplementary Figures 5A, B ). We then examined the expression level of APG9 in mature Col-0 and at1g33415 leaves. The results showed that APG9 transcript levels were significantly lower in at1g33415 than in Col-0 leaves ( Figure 5F ). These results imply that AT1G33415 potentially contributes to the stabilization of APG9 mRNA level in trans through RNA–RNA interaction, thereby playing a role as a negative regulator of leaf senescence.

AT2G14878-ADENINE PHOSPHORIBOSYL TRANSFERASE 1 (APT1) was predicted as a component of another regulatory module. APT1 inactivates cytokinin by catalyzing its conversion from free bases to nucleotides, and the loss-of-function apt1 mutant exhibits delayed leaf senescence in darkness (Zhang et al., 2013). The results of interaction energy analysis and the co-expression pattern of APT1 and AT2G14878 raised the possibility that AT2G14878 regulates dark-induced leaf senescence involving cytokinin metabolism (Figure 5C). The at2g14878 knockout mutant did not exhibit altered leaf senescence phenotype during aging ( Figures 5D, E ). However, similar to the extended longevity of atp1 leaves under dark-induced leaf senescence, the at2g14878 leaves showed a delayed senescence phenotype when leaf yellowing symptom was monitored in darkness ( Figure 5G ). The F_v/F_m of at2g14878 leaves was also maintained for a longer duration during dark incubation, supporting the positive role of AT2G14878 in dark-induced leaf senescence (Figure 5H). The senescence phenotype was further confirmed by examining expression of SAG12 ( Supplementary Figures 5A, B ). To determine whether delayed senescence in the at2g14878 mutant is caused by the suppression of APT1 expression, the transcript level of APT1 was analyzed in Col-0 and at2g14878 leaves. To rule out the possibility that APT1 expression was altered by delayed senescence, leaves at the early maturation stage were utilized for this experiment. As expected, the transcript level of APT1 was significantly suppressed in the mutant ( Figure 5I ). This suggests that AT2G14878 might promote cytokinin metabolism during leaf senescence by interacting with APT1, which decreases the cytokinin content of aged leaves.

Overall, the AR-lncRNA-mRNA pairs identified in this study provide valuable information for elucidating the biological function of lncRNAs, and will help to explore new lncRNA-mediated regulatory pathways involved in leaf senescence.