Identification of PASR and TASR peaks on protein-coding genes

As summarized in Fig 1 (also see details in Materials and Methods), we started a transcriptome-wide search for the PASR and TASR peaks on the protein-coding genes of Arabidopsis. After manual screening, PASR peaks were identified on the sense strands of 233 protein-coding genes [including 18 and four genes located on the chloroplast genome (named as chloroplast genes hereafter) and the mitochondrial genome (named as mitochondrial genes hereafter) respectively] (S1 Fig), and on the antisense strands of 231 protein-coding genes (including four chloroplast genes) (S2 Fig). TASR peaks were identified on the sense strands of 287 protein-coding genes (including six chloroplast genes and 15 mitochondrial genes) (S4 Fig), and on the antisense strands of 265 protein-coding genes (including one mitochondrial gene) (S5 Fig). Among the above identified genes, we observed that a total of 65 genes possessed paired PASR peaks symmetrically distributed on both sense and antisense strands (S3 Fig), and 132 genes possessed paired TASR peaks on both strands (S6 Fig). Besides, we also noticed that for a large portion of chloroplast genes possessing PASR or TASR peaks (such as ATCG00140, ATCG00270 and ATCG01120), the PASRs and TASRs were prominently accumulated in leaves and young seedlings in which the chloroplasts were highly enriched (Fig 2 and S1, S2, S3 and S4 Figs).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Analytical workflow for the transcriptome-wide identification of PASR (promoter-associated small RNA) and TASR (terminus-associated small RNA) peaks on protein-coding genes of Arabidopsis.

Four small RNA (sRNA) high-throughput sequencing data sets including GSM707678 (the flowers of wild type plants), GSM707679 (the leaves of wild type plants), GSM707680 (the roots of wild type plants) and GSM707681 (the seedlings of wild type plants) were utilized for this analysis. Each step including the parameters for PASR and TASR identification was provided. TSS: transcription start site. Terminus: transcription terminus. RPM: reads per million. See detailed description in “Materials and Methods”.

https://doi.org/10.1371/journal.pone.0169212.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Examples of the chloroplast genes generating PASRs (promoter-associated small RNAs) or TASRs (terminus-associated small RNAs) dominantly in green organ (leaves) of Arabidopsis.

(A) The PASR peak was identified on the sense strand of ATCG00540 (encoding photosynthetic electron transfer A). The x axis measures the genomic positions surrounding the TSS (transcription start site, marked by a black vertical bar) of this gene. The y axis measures the abundance of the small RNAs perfectly mapped onto the genomic region surrounding the TSS, which is also applied to the other figure panels. (B) The PASR peak was identified on both strands of ATCG01120 (chloroplast ribosomal protein S15). The x axis measures the genomic positions surrounding the TSS of this gene. (C) The TASR peak was identified on the sense strand of ATCG00270 (photosystem II reaction center protein D). The x axis measures the genomic positions surrounding the transcription terminus (marked by a black vertical bar) of this gene.

https://doi.org/10.1371/journal.pone.0169212.g002

Next, we set out to analyze the sequence characteristics of the PASRs and TASRs. The sRNAs located within or partially resided within the previously defined regions for 50-nt window scrolling (see Fig 1) were included in this analysis. We realized that these sRNAs were not only constituted by PASRs or TASRs, but also included the sRNAs surrounding the PASR or TASR peaks. However, this analysis is performed based on the following considerations: (1) The identification of the PASR and TASR peaks was assisted by computer programming combined with manual screening. Thus, it will be a hard task to extract the exact genomic region for each PASR or TASR peak since hundreds of the identified peaks have distinct locations. In contrast, it will be easier to extract the sRNAs located within or partially overlapped with the uniformly defined regions for 50-nt window scrolling. (2) Although the sRNAs regarded as “noise” were included in this analysis, a dominant portion of the analyzed sRNAs were constituted by those located within the PASR or TASR peaks. (3) The “noise” sRNAs distribute surrounding the TSSs or the transcription termini of the protein-coding genes. To some extent, these sRNAs outside the peaks could also be regarded as TSS- or transcription terminus-associated sRNAs. Thus, a total of 10,952 PASRs and 18,566 TASRs on the sense strands, and 8,891 PASRs and 16,762 TASRs on the antisense strands were included for the analysis, and were termed as “PASR_plus”, “PASR_RC”, “TASR_plus” and “TASR_RC”. Notably, for the four categories, 23- to 24-nt sRNAs occupied a dominant portion (47.16% of “PASR_plus”, 52.09% of “PASR_RC”, 56.77% of “TASR_plus” and 61.41% of “TASR_RC”) followed by 21- to 22-nt ones (23.50% of “PASR_plus”, 21.89% of “PASR_RC”, 21.00% of “TASR_plus” and 19.81% of “TASR_RC”). Moreover, the 5’ A-started sRNAs occupied a dominant portion of the four categories (35.98% of “PASR_plus”, 35.77% of “PASR_RC”, 36.39% of “TASR_plus” and 41.75% of “TASR_RC”), followed by the 5’ U-started ones (29.24% of “PASR_plus”, 26.61% of “PASR_RC”, 29.95% of “TASR_plus” and 25.40% of “TASR_RC”) (Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Sequence characteristics of the PASRs (promoter-associated small RNAs) and the TASRs (terminus-associated small RNAs).

(A) Sequence length distribution of the PASRs identified on the sense strands of the protein-coding genes of Arabidopsis. (B) Sequence length distribution of the PASRs identified on the antisense strands of the protein-coding genes of Arabidopsis. (C) Sequence length distribution of the TASRs identified on the sense strands of the protein-coding genes of Arabidopsis. (D) Sequence length distribution of the TASRs identified on the antisense strands of the protein-coding genes of Arabidopsis. (E) 5’ terminal nucleotide compositions of the PASRs identified on the sense strands of the protein-coding genes of Arabidopsis. (F) 5’ terminal nucleotide compositions of the PASRs identified on the antisense strands of the protein-coding genes of Arabidopsis. (G) 5’ terminal nucleotide compositions of the TASRs identified on the sense strands of the protein-coding genes of Arabidopsis. (H) 5’ terminal nucleotide compositions of the antisense strands of the protein-coding genes of Arabidopsis.

https://doi.org/10.1371/journal.pone.0169212.g003

AGO loading preference of the PASRs and the TASRs

As introduced in the above section, sRNAs with different sequence features will be selectively recruited by distinct AGO-associated silencing complexes [11], leading to divergent pathways for sRNA actions [3–5]. Thus, it is prioritized to see the AGO loading preference of the PASRs and the TASRs before investigating their biological functions. According to the sequence features, a large portion of the PASRs and the TASRs are 5’ A-started and 23- to 24-nt in length, followed by 5’ U-started, 21- to 22-nt ones (Fig 3). The above characteristics are close to the sRNAs preferentially associated with AGO4 and AGO1 respectively [11]. In this consideration, four sRNA HTS data sets prepared from the sRNAs associated with AGO1 (GSM707682, GSM707683, GSM707684 and GSM707685; see detail in Materials and Methods), and four data sets from the sRNAs associated with AGO4 (GSM707686, GSM707687, GSM707688 and GSM707689; see detail in Materials and Methods) were included in this analysis. To investigate whether the AGO-associated sRNAs could also form peaks, the sRNA sequencing reads of the eight HTS data sets were mapped onto the genomic sequences containing TSSs or transcription terminus (the sequence range was described in Fig 1) of the protein-coding genes with PASR or TASR peaks (listed in S1, S2, S4 and S5 Figs), and the perfectly mapped reads were retained for “50-nt window” screening (refer to Materials and Methods). As a result, AGO-associated PASR peaks were identified on the sense strands of 145 protein-coding genes (62.23% of the 233 genes identified above) (S7 Fig), and on the antisense strands of 134 protein-coding genes (58% of the 231 genes identified above) (S8 Fig). AGO-associated TASR peaks were identified on the sense strands of 142 protein-coding genes (49.48% of the 287 genes identified above) (S10 Fig), and on the antisense strands of 128 protein-coding genes (48.30% of the 265 genes identified above) (S11 Fig). We also observed that a total of 56 genes (86.15% of the 65 genes with previously identified paired PASR peaks) possessed paired AGO-associated PASR peaks (S9 Fig), and 130 genes (98.48% of the 132 genes with previously identified paired TASR peaks) possessed paired AGO-associated TASR peaks (S12 Fig). To determine which AGO the sRNAs are preferentially loaded into, an accumulation level-based comparison between AGO1-associated and AGO4-associated sRNAs resided within the same peaked regions on specific genes was performed. The result showed that many sRNA peaks on certain genes exhibited specific AGO preference. For examples, the PASRs within the peaked regions on the genes AT1G01073, AT1G08030, AT1G08035, AT1G09060, AT1G10820, AT2G23950, AT4G16460 and AT5G35526 are preferentially incorporated into AGO4 (S7 and S8 Figs). The TASRs within the peaked regions on the genes AT1G01180, AT1G20830, AT1G26400, AT1G28140, AT1G36950, AT2G14247, AT2G18465 and AT5G26770 are also preferentially recruited by AGO4 (S10 and S11 Figs). On the other hand, the PASRs identified on the genes AT3G06110, AT4G24026, AT5G07140, AT5G48830, ATCG00050 and ATCG00065 are preferentially incorporated into AGO1 (S7 and S8 Figs). The TASRs identified on AT1G16460, AT2G07771, AT2G28725, AT4G24030, ATMG00270 and ATMG00570 are preferentially loaded into AGO1 (S10 and S11 Figs). Among the sRNAs preferentially loaded into one of the AGO proteins, we noticed that some of these sRNAs also showed organ-specific accumulation patterns. For examples, some of the PASRs preferentially associated with AGO1 were observed to be specifically accumulated in roots (for the PASRs identified on the genes AT1G53542, AT1G64130, AT2G38025, AT3G18820, AT3G52570, AT5G02820 and AT5G05950), seedlings (ATCG00280), or flowers (AT1G61030, AT3G16230 and AT4G14810). Some of the PASRs preferentially associated with AGO4 were specifically detected in roots (AT5G48280) or flowers (AT2G24670 and AT3G42550) (S7 and S8 Figs). Moreover, some of the TASRs preferentially associated with AGO1 were specifically accumulated in roots (AT1G18420, AT1G50110, AT4G25580, AT4G25590, AT4G30993, AT4G34030, AT5G06130, AT5G52360, AT5G60280 and ATMG00070) or seedlings (ATCG00270, ATCG00420, ATCG00840, ATCG01300 and ATMG00900). Some of the TASRs preferentially associated with AGO4 were specifically detected in roots (AT1G22720) or seedlings (AT2G02520) (S10 and S11 Figs).

PASR and TASR peaks covered by dsRNA-seq reads

To date, several sRNA species such as nat-siRNAs, ta-siRNAs, ra-siRNAs (or heterochromatic siRNAs) have been reported to be produced through RDR-dependent pathways [3,5–10]. In other words, the biogenesis of most of the currently known siRNAs requires RDR for dsRNA precursor synthesis. Thus, it is intriguing to see whether the PASRs and the TASRs are originated from double-stranded precursors. Recently, Zheng and his colleagues developed an approach called dsRNA-seq by marrying the classical nuclease-based structure mapping technique with HTS [31]. Before sequencing library construction, the ribosomal RNA (rRNA)-depleted RNA samples were treated with the single-strand-specific ribonuclease, thus enabling transcriptome-wide identification of the dsRNAs in vivo. In this consideration, dsRNA-seq data sets GSM575243 and GSM575244, a gift of Zheng et al.’s study, were recruited for this study. The dsRNA-seq reads were mapped onto the genomic sequences containing TSSs or transcription termini (described in Fig 1) of the protein-coding genes with the above identified PASR or TASR peaks, and the perfectly mapped reads were retained. As a result, PASR peaks on the sense strands of 30 protein-coding genes (S13 Fig) and on the antisense strands of 13 genes (S14 Fig), and TASR peaks on the sense strands of 29 protein-coding genes (S16 Fig) and on the antisense strands of nine genes (S17 Fig) were identified to be resided within the continuous regions of 100 nt or longer which were covered by dsRNA-seq reads. Notably, pairs of PASR peaks (on both sense and antisense strands), covered by dsRNA-seq reads, were identified on seven genes (AT1G16820, AT1G53265, AT3G13857, AT3G43270, AT3G57770, AT4G16640 and AT5G48000) (S15 Fig), and pairs of TASR peaks were identified on 13 genes (AT1G28304, AT1G66620, AT3G25130, AT3G41762, AT3G52830, AT4G04030, AT4G08160, AT4G14365, AT5G43525, AT5G50480, AT5G54410, AT5G54700 and AT5G65005) (S18 Fig). Summarily, the result indicates that a portion of PASRs and TASRs are likely to be processed from dsRNA precursors.

DCL-, RDR- and Pol IV-dependent biogenesis pathway of certain PASRs and TASRs

The result that certain PASRs and TASRs were potentially produced from double-stranded precursors prompted us to interrogate the particular biogenesis pathway(s) that they were involved in. The dependence of the sRNAs on DCLs and RDRs was analyzed by using sRNA HTS data prepared from wild type plants, and dcl and rdr mutants (see details in Materials and Methods). According to the biogenesis model of the heterochromatic siRNAs which contribute to a great portion of the plant endogenous siRNAs, Pol IV is required for the transcription of the siRNA precursors. Relying on RDR2, the transcripts are converted to long dsRNAs which are then subjected to DCL3-mediated processing for siRNA production. A portion of the heterochromatic siRNAs was loaded into AGO4 silencing complexes to guide site-specific chromatin modifications [3]. In this study, the sRNA HTS data sets prepared from the pol iv mutants (including nrpd1a and nrpd1b) were also included. The sequencing reads of the above mentioned sRNA HTS data were mapped onto the genomic sequences containing TSSs or transcription termini of the protein-coding genes with PASR or TASR peaks, and the perfectly mapped reads were retained for “50-nt window” screening. After screening, we made a comparison of the signal intensity between the “wild type” peaks and the “mutant” peaks, in order to detect the dependence of PASRs and TASRs on DCL(s), RDR(s) and Pol IV. Based on the comparison, we could also see which DCL(s) and RDR(s) are specifically implicated in the biogenesis of the sRNAs forming a PASR or TASR peak. As a result, many PASR and TASR peaks show a clear dependence on specific DCLs, RDRs and Pol IV (S19 Fig). For examples, the accumulation level of the PASR peak on the sense strand of AT1G53265 is greatly repressed in the mutants dcl3, dcl234, rdr2 and rdr6. The abundance of the PASR peak on the antisense strand of AT4G16640 is greatly reduced in dcl234, rdr2, nrpd1a and nrpd1b. The level of the TASR peak on the sense strand of AT3G41762 is significantly repressed in dcl2, dcl234, rdr2 and nrpd1a. The accumulation level of the TASR peak on the antisense strand of AT3G41762 is significantly repressed in dcl234, rdr2 and nrpd1a. Notably, the DCL/RDR/Pol IV dependence was detected for the paired PASR peaks on AT1G53265, AT4G16640 and AT5G48000, and the paired TASR peaks on AT1G28304, AT3G25130, AT3G41762, AT3G52830, AT4G04030, AT4G08160, AT4G14365, AT5G43525, AT5G50480, AT5G54700 and AT5G65005 (S20 Fig).

Site-specific DNA methylation signals detected on the genes with PASR or TASR peaks

According to the above results, interesting features were observed for several paired sRNA peaks. For example, PASR peaks surrounding the TSS were detected on both strands of AT5G48000. The paired peaks are covered by the dsRNA-seq reads (Fig 4A), and the PASRs are preferentially incorporated into AGO4 (Fig 4B). Moreover, the signal intensity of the PASR peaks is greatly repressed in rdr2, dcl234, nrpd1a and nrpd1b (Fig 4C and 4D). Another example was observed for AT3G41762, TASRs peaked surrounding its transcription terminus on both strands. The paired peaks are also covered by the dsRNA-seq reads (Fig 5A), and the TASRs on the antisense strand are preferentially incorporated into AGO4 (Fig 5B). Moreover, the signal intensity of the TASR peaks is greatly reduced in rdr2, dcl2, dcl234 and nrpd1a (Fig 5C and 5D). The features of these PASRs and TASRs, including the association with AGO4 and the dependence on RDR2, DCL2/3/4 and Pol IV, led us to investigate whether these sRNAs could participate in chromatin modifications just as the heterochromatic siRNAs. In this regard, all of the protein-coding genes with PASR or TASR peaks (see gene lists in S1, S2, S3, S4, S5 and S6 Figs) were sent to the web server Arabidopsis epigenome maps (neomorph.salk.edu/epigenome/epigenome.html) [32] to discover site-specific DNA methylation signals. Fortunately, several methylated regions were observed, whose genomic positions correlated well with those of the PASR or TASR peaks on the host genes. Briefly, seven genes (AT1G48660, AT1G67450, AT2G36180, AT3G17490, AT3G22730, AT3G56450 and AT5G40320) with PASR peaks on the sense strands and one gene (AT2G29460) with a PASR peak on the antisense strand were site-specifically methylated (S21 and S22 Figs). Three genes (AT1G52180, AT5G36220 and AT5G42203) with TASR peaks on the sense strands and five genes (AT1G59835, AT1G66540, AT2G04090, AT2G04830 and AT5G48605) with TASR peaks on the antisense strands were also site-specifically methylated (S24 and S25 Figs). Additionally, 13 genes with paired PASR peaks and 32 genes with paired TASR peaks were site-specifically methylated (S23 and S26 Figs). For example, PASR peaks were identified on both strands of AT1G53265. Accordingly, strong DNA methylation signals surrounding the TSS of AT1G53265 were present on the Arabidopsis epigenome maps (Fig 6A). TASR peaks were identified on both strands of AT5G54700. Accordingly, DNA methylation signals surrounding the transcription terminus of AT1G53265 were observed on the Arabidopsis epigenome maps (Fig 6B). Examples of paired PASR and TASR peaks potentially mediating site-specific DNA methylation are listed in Table 1 and S1 Table which also shows the potential pathways for their biogenesis and action.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. PASRs (promoter-associated small RNAs) identified on both strands of AT5G48000, and small RNA (sRNA) and double-stranded RNA (dsRNA) high-throughput sequencing (HTS)-based evidences showing potential Argonaute (AGO) loading preference and biogenesis pathways of PASRs.

(A) Total: Initially, four sRNA HTS data sets belonging to GEO (Gene Expression Omnibus; www.ncbi.nlm.nih.gov/geo) accession ID GSE28591 were utilized to identify PASR peak near the TSS (transcription start site, marked by a red vertical bar) of the gene. Refer to Fig 1 for the analytical workflow. (B) AGO: Eight sRNA HTS data sets belonging to GSE28591 were divided into AGO1 data group (GSM707682, GSM707683, GSM707684 and GSM707685) and AGO4 data group (GSM707686, GSM707687, GSM707688 and GSM707689). To analyze the AGO loading preference of the PASRs, an accumulation level-based comparison was performed between the two AGO-associated sRNA HTS data groups. The higher accumulation levels of the PASRs were detected in the AGO4 data group, and were denoted by black lines. (C) RDR-, DCL-dependence: sRNA HTS data sets from different mutants (including dcl and rdr mutants) involved in sRNA biogenesis pathways were recruited for this analysis. Prominently repressed abundances of PASRs observed in specific mutants were denoted by black lines. (D) Pol IV-dependence: sRNA HTS data sets from two mutants (nrpd1a and nrpd1b, and both were denoted by black lines) of RNA polymerase (Pol) IV were used to analyze the dependence of PASR biogenesis on Pol IV. For all the figure panels, the dsRNA sequencing read-covered regions (the positions were provided on the top right) were highlighted in semitransparent red (for sense strand) and green (for antisense strand) background color. For the detailed explanation of the HTS data sets, please refer to “Data sources” within the “Materials and Methods” section.

https://doi.org/10.1371/journal.pone.0169212.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. TASRs (terminus-associated small RNAs) identified on both strands of AT3G41762, and small RNA (sRNA) and double-stranded RNA (dsRNA) high-throughput sequencing (HTS)-based evidences showing potential Argonaute (AGO) loading preference and biogenesis pathways of TASRs.

(A) Total: Initially, four sRNA HTS data sets belonging to GEO (Gene Expression Omnibus; www.ncbi.nlm.nih.gov/geo) accession ID GSE28591 were utilized to identify TASR peak near the transcription terminus (marked by a red vertical bar) of the gene. Refer to Fig 1 for the analytical workflow. (B) AGO: Eight sRNA HTS data sets belonging to GSE28591 were divided into AGO1 data group (GSM707682, GSM707683, GSM707684 and GSM707685) and AGO4 data group (GSM707686, GSM707687, GSM707688 and GSM707689). To analyze the AGO loading preference of the TASRs, an accumulation level-based comparison was performed between the two AGO-associated sRNA HTS data groups. The higher accumulation levels of the TASRs were detected in the AGO4 data group, and were denoted by black lines. (C) RDR-, DCL-dependence: sRNA HTS data sets from different mutants (including dcl and rdr mutants) involved in sRNA biogenesis pathways were recruited for this analysis. Prominently repressed abundances of TASRs observed in specific mutants were denoted by black lines. (D) Pol IV-dependence: sRNA HTS data sets from two mutants (nrpd1a and nrpd1b, and nrpd1a was denoted by black lines) of RNA polymerase (Pol) IV were used to analyze the dependence of TASR biogenesis on Pol IV. For all the figure panels, the dsRNA sequencing read-covered regions (the positions were provided on the top right) were highlighted in semitransparent red (for sense strand) and green (for antisense strand) background color. For the detailed explanation of the HTS data sets, please refer to “Data sources” within the “Materials and Methods” section.

https://doi.org/10.1371/journal.pone.0169212.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Examples of site-specific DNA methylation potentially mediated by PASRs (promoter-associated small RNAs) and TASRs (terminus-associated small RNAs) in Arabidopsis.

(A) Site-specific DNA methylation signals were observed within the genomic region surrounding the TSS (transcription start site; marked by a vertical dashed line) of AT1G53265. Accordingly, abundant small RNAs (i.e. PASRs) were mapped onto this region. (B) Site-specific DNA methylation signals were detected within the genomic region surrounding the transcription terminus of AT5G54700. Accordingly, abundant small RNAs (i.e. TASRs) were mapped onto this region. Arabidopsis epigenome maps (neomorph.salk.edu/epigenome/epigenome.html) [32] were employed for this analysis.

https://doi.org/10.1371/journal.pone.0169212.g006

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Examples of paired PASR (promoter-associated small RNA) and TASR (terminus-associated small RNA) peaks potentially mediating site-specific DNA methylation, and analysis of their biogenesis and action pathways.

https://doi.org/10.1371/journal.pone.0169212.t001

A biogenesis and action model proposed for certain PASRs and TASRs

In this study, by using sRNA HTS data, dsRNA-seq data and DNA methylation profiling data, we reported the following interesting discoveries: (1) For hundreds of the protein-coding genes in Arabidopsis, sRNAs peaked within the genomic regions neighboring to or covering the TSSs or the transcription termini, which were termed as PASR and TASR peaks respectively. (2) Some of the PASRs and TASRs might be processed from dsRNA precursors, the biogenesis of which relies on RNA Pol IV and specific RDR(s). (3) The processing of these PASRs and TASRs from the double-stranded precursors depends on the activity of specific DCL(s). (4) A portion of the PASRs and TASRs are preferentially loaded into AGO1 or AGO4. (5) For some of the sRNAs associated with AGO4, site-specific DNA methylation signals were detected within the genomic regions covering the PASR or TASR peaks.

Based on the above results, a biogenesis and action model was proposed for certain PASRs and TASRs (Fig 7). However, several key points should be discussed here for this model. First, in this study, the sRNA HTS data from the pol iv mutant was used to investigate the dependence of the transcription of the PASR and TASR precursors. Whether the other RNA polymerases, such as Pol II and Pol V are also involved in the transcription needs further investigations. Second, for different genes, PASRs and TASRs peak at distinct positions related to their TSSs and transcription termini respectively. Some peak upstream and some peak downstream of the TSSs or the transcription termini, while the other peaks cover the TSSs or the transcription termini (S1, S2, S4 and S5 Figs). It will be interesting to see whether these sRNA peaks with different locations are produced through distinct pathways. Third, only AGO1 and AGO4 were analyzed in the AGO association study. In fact, a portion of the sRNA peaks are not associated with AGO1 or AGO4. Whether the PASRs and the TASRs are associated with other AGOs for their actions remains to be investigated. Fourth, the levels of certain PASRs and TASRs were intensively repressed in the triple mutant dcl234, but not in the single mutant dcl2, dcl3 and dcl4 (see Fig 4C for example). Thus, the functional redundancy of the three DCLs in the processing of the PASRs and the TASRs needs to be further analyzed. On the other hand, in addition to RDR2 and RDR6, whether other RDRs are implicated in the production of the PASRs and the TASRs requires further investigations. A recent study in murine embryonic stem cells uncovered a class of sRNAs peaked surrounding TSSs. These sRNAs called TSSa-RNAs (transcription start site-associated RNAs) range from 20 to 90 nt in length [15]. In human, chicken and Drosophila, Taft and his colleagues identified 18-nt tiny RNAs, which reside within the genomic regions ranging from 60 nt upstream to 120 nt downstream of the TSSs [17]. In bacteria, sRNAs associated with TSSs were also discovered. But, these sRNAs have an average size of 45 nt [19]. In our study, both the PASRs and the TASRs range from 19 to 28 nt (Fig 3A, 3B, 3C and 3D). Thus, in the following studies, it is necessary to clarify whether the observed length range of the PASRs and the TASRs is resulted from the sRNA HTS data used in our analysis (since all the sRNA reads of the HTS data sets are shorter than 40 nt), or is the fact resulting from the nature of the PASRs and the TASRs specifically existed in plants.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Proposed biogenesis pathways, action modes of the PASRs (promoter-associated small RNAs) and the TASRs (terminus-associated small RNAs) identified in Arabidopsis.

For a protein-coding gene possessing a bidirectional promoter, the precursors of the PASRs could be transcribed either upstream or downstream of the TSS (transcription start site), which needs experimental investigations. However, the underlying mechanism of the transcription of the TASR precursors could not be deduced. Although several pieces of high-throughput sequencing (HTS) data-based evidences pointed to the dependence of the biogenesis of the PASRs and the TASRs on RNA polymerase (Pol) IV, RDR2 (RNA-dependent RNA polymerase 2), RDR6, DCL2 (Dicer-like 2), DCL3 and DCL4, other key factors implicated in the accumulation of the two small RNA (sRNA) species might have not been uncovered owing to the limited sRNA HTS data utilized in this study. Based on the HTS data from the Argonaute (AGO)-associated sRNAs, many Pol IV-, RDR- and DCL-dependent PASRs and TASRs were preferentially loaded into AGO4 silencing complexes. And, site-specific DNA methylation was observed on the genomic regions well corresponding to the origins of the PASRs and the TASRs. Thus, the AGO4-associated PASRs and TASRs were proposed to mediate cis-methylation. Besides, the AGO1-associated PASRs and TASRs were proposed to mediate target cleavages similar to the plant microRNAs.

https://doi.org/10.1371/journal.pone.0169212.g007

In order to clearly depict the biogenesis and action pathway(s) of certain transcription boundary-associated sRNAs, some fine-scale experimental approaches are proposed here. First, the dependence of the PASR or TASR production on specific RNA polymerase, RDR(s) and DCL(s) should be detected in the corresponding pol, rdr and dcl mutants by Northern blotting, taking wild type plants as a control. Second, to investigate the action mode of AGO4-associated sRNAs, DNA methylation sequencing should be carried out for both wild type plants and the ago4 mutant, to compare the methylation patterns on the specific genomic regions (e.g. the sRNA-coding regions if cis-DNA methylation is mediated by the sRNAs) between the two plant materials. Third, target prediction and degradome sequencing could be performed for the AGO1-associated sRNAs, in order to identify the downstream target genes of specific sRNAs. Besides, transgenic experiments could be carried out for both the AGO1-associated sRNAs (over-expression) and their downstream target genes (knock down), in order to see the biological outputs of the sRNA-mediated regulatory pathways.

What are the biological outputs of the PASRs and the TASRs?

Our study shows that certain AGO4-associated PASRs and TASRs are likely to mediate site-specific DNA methylation. However, further in-depth analyses are needed to elucidate the influence of the PASR- or TASR-mediated DNA methylation on the expression of their host genes. As described above, some of the AGO1- or AGO4-associated PASRs and TASRs were observed to be accumulated with organ-specific patterns (S7, S8, S10 and S11 Figs). Thus, whether the host genes could be regulated by specific PASRs or TASRs in an organ-specific way remains to be an open question. In further studies, degradome sequencing data could be employed for identification of the target genes regulated by AGO1-associated PASRs and TASRs.

Another interesting phenomenon is that the PASRs and the TASRs from certain chloroplast genes were abundantly accumulated in the green organ such as leaves and the young seedlings (see Fig 2 for example). We questioned whether these sRNAs were implicated in the regulation of the chloroplast genes. However, no methylation signal was detected surrounding the TSSs or the transcription termini of the chloroplast genes. Perhaps, other regulatory mechanisms underlie the regulation of the chloroplast genes encoding PASRs or TASRs. Or, the accumulation of PASRs and TASRs might be the result of the transcription of their host genes in chloroplasts. Or, a feedback regulatory circuit might exist between PASRs/TASRs and their host genes.

A recent report by Kapranov and his colleagues revealed an intriguing regulatory function for the termini-associated sRNAs in human cells. These novel sRNAs antisense to the 3’ ends of the annotated transcripts could increase the RNA copy numbers through a yet uncharacterized endogenous biochemical pathway [29]. In our study, a large portion of TASRs were discovered on the antisense strands of certain protein-coding genes. Thus, whether these antisense TASRs play a role in increasing the transcript copy numbers of their host genes is worth investigating in depth.

Data sources

Based on the genomic information provided by TAIR (The Arabidopsis Information Resource; release 10; www.arabidopsis.org/index.jsp) [33], 500 bp upstream and 499 bp (if the length of the gene body is shorter than 500 nt, then the whole gene body from the TSS to the transcription terminus was collected) downstream of the TSSs (a total of 1000 bp, or 500 bp plus the length of the gene body) of the protein-coding genes, and 499 bp upstream (if the length of the gene body is shorter than 500 nt, then the whole gene body from the TSS to the transcription terminus was collected) and 500 bp downstream of the transcription terminus (a total of 1000 bp, or 500 bp plus the length of the gene body) of the protein-coding genes were collected for this study. For the protein-coding genes with multiple transcription models, only the TSSs and the transcription termini of the longest models were used for sequence collection.

Small RNA HTS data sets and dsRNA-seq data sets utilized in this study were retrieved from GEO (Gene Expression Omnibus; www.ncbi.nlm.nih.gov/geo) [34], and their accession IDs with brief introductions are provided in S2 Table.

To investigate site-specific DNA methylation potentially mediated by PASRs or TASRs, the web server Arabidopsis epigenome maps (neomorph.salk.edu/epigenome/epigenome.html) [32] were utilized.

Transcriptome-wide identification of PASR and TASR peaks

Four sRNA HTS data sets (GSM707678, GSM707679, GSM707680 and GSM707681) were utilized to identify the PASR and TASR peaks. To allow accumulation level-based comparison among different data sets, normalized read counts (in RPM, reads per million) of the sRNAs were calculated for each data set as follows: the raw read count of each sRNA was divided by the total raw counts of this data set, and then multiplied by 10⁶.

Firstly, the sequences surrounding TSSs and transcription termini (include sense and antisense strands) were collected from TAIR (see “Data sources”). And, the sRNAs from the above four HTS data sets were mapped onto the collected sequences by using in-house Perl script. Only the perfectly mapped sRNAs were retained. For each sequence encompassing TSSs or transcription termini, “A₁” (average sRNA count of this sequence) was calculated by dividing the “S₁” (summed sRNA count of this sequence) by “L” (the length of the sequence) (see Fig 1). Secondly, we defined a specific region for 50-nt window scrolling on this sequence (please refer to Fig 1 for the parameters for defining scrolling regions). Subsequently, a 50-nt window was scrolled within the defined region. And, for each 1-nt scrolling step, “A₂” (average sRNA count of the scrolling window at this step) was calculated by dividing the “S₂” (summed sRNA count of the scrolling window at this step) by 50 nt. Thirdly, computer-assisted screening was performed to search for the sequences containing 50-nt windows accommodating highly abundant sRNAs by employing the following criterion: “A₂” should be ten times or more than “A₁”, and “A₂” should be 1 RPM/nt or higher. Finally, the retained 50-nt windows on specific sequences were subjected to manual screening to identify PASR or TASR peaks by adopting the following criterion: sRNAs should peaked within the regions defined for 50-nt window scrolling, and the peaks should not be submerged in the surrounding sRNA “noise”.