Comparative TSS-seq analysis among natural variations of Arabidopsis reveals the genetic stability of genic and nongenic promoters Authors: Kazuki Sugekawa, Okechukwu S Ezeh, Yoshiharu Y Yamamoto Published: 2025 ABSTRACT AbstractIn a previous TSS-seq analysis, we identified >300 K promoters in the Arabidopsis genome. The majority were nongenic promoters, composed of intragenic, antisense, and orphan promoters. Although several nongenic promoters for biological processes have been identified, the significance of these nongenic promoters is unclear. In this study, we evaluated several characteristics of these promoters, including their genetic stability by analyzing the promoter sequences, peak positions of the TSS clusters using natural variations of Arabidopsis, the expression levels, and promoter structures. The results indicate that nongenic promoters are not genetically stable in the promoter sequence or the TSS peak position, whereas the canonical, genic promoters are stable in both. Together with the less structured features of nongenic promoters, such as core promoter elements and epigenetic traits, the results strongly suggested that the majority of the nongenic promoters do not have biological significance, except for the genetically stable minorities. Interestingly, the genetic stability of the promoters is positively associated with the expression level. In conclusion, our results raised a possibility that the most nongenic promoters result from leaky transcription or transcriptional noise. Introduction Transcription of gene bodies is achieved by promoters located in the upstream regions. Because each gene requires its promoter, one may expect the same number of promoters in a genome as that of genes, which is ~30 000 for Arabidopsis; however, there are some reports in which multiple promoters are associated with one gene. Some genes have more than one promoter. In Arabidopsis, we found that two promoters for one gene are activated differently depending on light conditions. The upstream promoter produced a longer transcript with less translation efficiency because of an upstream ORF (uORF). This type of promoter switching occurs in HY5 (AT5G11260), HYH (AT3G17609), HPR1 (AT1G68010), and HCF107 (AT3G17040) [1]. Despite the environmental response of the activity ratio among the paired promoters, the biological significance of promoter switching remains unclear. In addition to genic promoters that transcribe the gene body, nongenic promoters have been reported. In Arabidopsis, the presence of intragenic promoters, which are located within the gene body and produce truncated protein products with different intracellular localization has been reported, such as PMDA (AT1G63940) [2], PPDK (AT4G15530) [3], sAPX (AT4G08390) [4], ATX1 (AT2G31650) [5], HO1 (AT2G26670) [6], and GLYK (AT1G80380) [7]. The intragenic promoter of GLYK, the enzyme responsible for the final step of photorespiration, is activated in the absence of the phytochrome signal and produces an enzyme targeting the cytoplasm, which is also indispensable for proper Arabidopsis growth [7]. These examples indicate the biological significance of the intragenic promoters. Natural antisense RNA in Arabidopsis has also been identified and implicated in the regulation of FLC (AT5G10140) [8], NFYA5 (AT1G54160) [9], DOG1 (AT5G45830) [10], MAF4 (AT5G65070) [11], and SERRATE (AT2G27100) [12]. Some of the identified antisense RNAs are transcribed from the downstream region of the corresponding gene body, whereas others are from intragenic antisense promoters, although the promoters for the antisense transcripts are not well characterized. Genome-wide analysis using NGS revealed 4080 natural antisense transcripts, with the majority producing small interfering RNAs (siRNAs) together with the corresponding sense RNA [13]; however, the degree of influence on gene expression and their biological significance are largely unknown. In addition to genes encoding proteins and functional RNA species for translation, including ribosomal RNA and transfer RNA, there are genes for long noncoding RNA (lncRNA) species [14]. These include APOLO, ASCO, ELENA1, IPS1, T5120, and DRIR, all of which function in gene regulation and are transcribed from regions that do not overlap with protein-coding genes. Therefore, these transcripts are encoded in intergenic regions, based on the traditional definition of a gene. Recent deep sequencing discovered 2444 lncRNAs [15]; however, it is unclear whether they play any roles in plant growth and development, or whether these transcribed regions should be considered genes. The promoter region is difficult to determine because it requires a detailed experimental analysis. Therefore, the peak positions of TSS clusters, which can be detected by genome-wide TSS-seq, are referred to as promoters, and they suggest that their surrounding regions support corresponding transcriptional initiation events. They may be used to identify the number and positions of the promoters without experimental data showing the functional promoter regions. In our previous TSS-seq study, 324 K TSS clusters, or promoters, were identified in the Arabidopsis genome [4]. This is ~10-fold more than the number of genes in the genome. The identified promoters included 59 628 genic promoters for 22 211 genes, 193 208 intragenic promoters, 42 070 antisense promoters, and 35 549 orphan promoters that could not be associated with any genes [4]. After identifying this large number of promoters, we wondered if these many promoters have any biological significance. To address this question, we evaluated the genetic stability of these promoters by analyzing the promoter sequences and peak positions of the TSS clusters using natural variations of Arabidopsis. Materials and methods Plant materials Four accessions of Arabidopsis thaliana, Altenb-2, Kly-1, Shigu-2, and Vash-1, whose genome sequences were previously determined [16], were used in this study. They were selected because of a clear difference in cold/high light stress resistance (two strong and two weak). Seeds were surface sterilized, vernalized at 4°C for 2–4 days in the dark, and grown at 22°C under continuous light (40 µE m−2 s−1) for 2 weeks. The seedlings were treated with high light and cold stress conditions with a light and cold stress conditions with a light intensity of 140 µE m−2 s−1 at 4°C for 24 h, or left in the growth conditions without the stressor. Both stress-treated and the control samples were both used for the analysis. RNA extraction and TSS-seq Total RNA was extracted from the aerial parts of 2-week-old seedlings treated with and without stress conditions as described previously [17]. The RNA samples (10 µg) were subjected to CAGE analysis at DNAFORM (Yokohama, Japan) [18]. The resulting FASTQ data had >30 mean quality scores. They were filtered with Cap-Signature [4] and mapped to the Arabidopsis genome sequence (Araport11 [15]) with 2% mismatch allowance by STAR (https://github.com/alexdobin/STAR, [19]). Data for the two conditions (with and without stress treatment) with three biological replicates were combined to prepare 50.5 M, 53.8 M, 64.4 M, and 53.4 M effective TSS tags for Altenb-2, Kly-1, Shigu-2, and Vash-1, respectively (Supplementary Table S1). They were then subjected to cluster analysis to determine the cluster area, peak positions, and tag counts of the TSS clusters for each accession as previously described [4]. Perl tools are available at Github (https://github.com/yyyamamoto/TSS-analysis). Clusters with small TSS tags of <5 were removed from the analysis. The FASTQ data are available at GSE298979 in GEO/NCBI. In silico analysis of the TSS clusters and promoters TSS cluster data [cluster area, peak positions, tag numbers, promoter categories (genic, intragenic, etc.) corresponding gene models, and core promoter info] of the reference accession (Col-0) was previously determined [4]. Of these, the clusters validated by C2 tags and ≥5 TSS tags among them were subjected to this study. The clusters corresponding to multiple gene models were removed from the analysis. Finally, 17 920 genic top, 10 065 genic companion, 34 028 intragenic, 3626 antisense, and 5847 orphan promoters were used. Correspondence of the TSS peaks in the four accessions to the reference peaks in Col-0 was determined based on their chromosomal positions and directions. When no peak was detected in an accession within the reference cluster area, it was considered to have no corresponding cluster. GO analysis was done using clusterProfiler (v 4.6.2, in R package, [20]) along with GO dataset of org.At.tair.db (v 3.16.0, R package, https://bioconductor.org/packages/release/data/annotation/html/org. At.tair.db.html). Genome variation data for 1135 Arabidopsis accessions [21] were used to analyze sequence stability. Nucleotide diversity (π) was calculated from the Variant Call Format (VCF) files of the accessions using VCFtools (v0.1.16, https://vcftools.sourceforge.net, [22]). The trend in nucleotide diversity was visualized with matplotlib.pyplot (v3.8.3, Python). Arabidopsis nucleosome positioning data, determined by sequencing micrococcal nuclease (MNase)-resistant chromosomal fragments, were procured from GEO/NCBI (GSM543296 in https://www.ncbi.nlm.nih.gov/geo/ and PRJNA838070 in https://www.ncbi.nlm.nih.gov/bioproject/) [23, 24]. About the data of PRJNA838070, quality control of the FASTQ data were conducted by fastp (v0.23.4, https://github.com/OpenGene/fastp, [25]) using the "–detect_adapter_for_pe” option. The filtered data were subjected to mapping, removal of PCR-based duplicated reads, and coverage calculation as described by Kawaguchi et al. [26]. ChIP-seq data for H2A.Z and H4K16ac were obtained from SRA/NCBI (DRA015426 in https://www.ncbi.nlm.nih.gov/sra/) [26]. The data processing was conducted as mentioned earlier. Prediction of the local shift of TSS peaks The ratio of each triplet from −1 to +2 relative to the peak TSS among the genic top TSS was transformed into a binary logarithm, and the minimum value among all the triplet ratios was subtracted to set the minimum value at zero. The region from −5 to + 6 relative to a genic top TSS peak was scanned, and the position with the highest score, excluding the +1 position, was predicted as a new peak. The score table is shown in Supplementary Table S8. The prediction results of individual new TSSs and local shifts from Col-0 are shown in Supplementary Tables S9 and S10, respectively. Prediction was done using Excel software (Microsoft Japan, Tokyo). Results Nongenic promoter groups are generally unstable In our previous study with Arabidopsis Col-0, four promoter categories (genic, intragenic, antisense, and orphan) were identified (Fig. 1A). Because one gene has three genic promoters on average, this group was further classified into genic top, which shows the highest activity for each gene, and genic companion, showing relatively less activity. Our analysis revealed that the expression level of a gene is primarily determined by genic top, and the contribution of genic companion is negligible [4]. Intragenic, antisense, and orphan promoters are nongenic. Genic and nongenic promoter categories. (A) Illustration of the genic top, the genic companion, the intragenic, the antisense, and the orphan promoter groups. The number of TSS clusters (=promoters) identified by Tokizawa et al. [4] was ~22k, 37k, 193k, 42k, and 35k, respectively. (B) Expression levels of genic and nongenic promoters. The distribution of expression levels for each promoter group is shown. The horizontal axis shows the tag number of a TSS cluster in Col-0. The vertical axis indicates the ratio of promoters among each promoter group shown in the panel. The data of Tokizawa et al. [4] were modified. Total TSS tags are 26 M. Tpm denotes “tag per million.” Figure 1B shows the expression level of the five promoter categories. Genic top group has the highest expression level, whereas the other groups are less than one-tenth of the genic top based on their median values. However, minorities for each of the four groups showed high expression with comparable levels to the genic top. To estimate the biological significance of these four promoter groups, we analyzed their genetic stability using natural variations of Arabidopsis. First, we evaluated the sequence stability of the local promoter region from −100 to +20 relative to the peak TSSs using 1 135 Arabidopsis sequences [21]. As shown in Fig. 2A, the nucleotide diversity of the genic groups was lower compared with the average of the whole genome, indicating stronger selection pressure for sequence conservation in these regions compared with the average of the whole genome. In contrast, the orphan group showed the highest nucleotide diversity among all of the promoter groups, and it was comparable to that of the whole genome. This shows the weakest selection pressure for sequence conservation in the orphan group among all promoter groups. The nucleotide diversity of the intragenic and antisense groups, both of which were located within the transcribed regions, was significantly lower compared with that in the whole genome. This very low sequence diversity is considered to be an effect of overlap with an additional genomic context as the transcribed region. This interpretation is supported by the observation that the protein-coding region (CDS) contained the lowest appearance ratio of SNPs and InDels among the intron and intergenic regions [16, 27]. Genetic stability of four promoter types. (A) Sequence stability of each promoter type. The average of nucleotide diversity (π) from −100 to +20 bp relative to the TSS peak in 1135 Arabidopsis accessions is shown. The gray bar indicates the average nucleotide diversity of the whole genome. (B) TSS distribution of the genic group. The chromosomal position shown at the bottom of the bottom graph is for the reference genome (Col-0, chr. 4). A rhombus indicates the peak position. Twenty in the vertical axis indicates 220 = 1 048 576. (C) Stability of the TSS peak position for each promoter group. Peak match: all four accessions have the same peak position as in Col-0, resulting in a peak match among the five accessions. Peak shift: at least one accession has a peak shift within the area of the corresponding TSS cluster of Col-0. No peak: at least one accession has no peak within the area of the corresponding cluster of Col-0. Next, we analyzed the stability of the TSS peak positions in each promoter group among the four Arabidopsis accessions based on our reference TSS-seq data from another accession, Col-0. From 50 M to 64 M, validated TSS tags were analyzed for each accession (Supplementary Table S1). Figure 2B shows an example of the TSS distribution for the genic top promoter ELIP2 (At4G14690). For this case, all of the TSS peak positions in the four accessions were the same as those in the reference data. This profile was considered a peak match, in which the peak positions for all four accessions, plus one reference accession, are conserved. If one or more accessions shows a peak shift from Col-0 to a position within the corresponding TSS cluster, it was considered a peak shift. If one or more accessions shows loss of the corresponding TSS peak within the cluster, it is considered no peak. The degree of peak match, peak shift, and no peak of the promoter groups is shown in Fig. 2C The genic top group had the highest peak match ratio (30%), with the lowest ratio in no peak (28%), demonstrating the highest peak conservation among the promoter groups. The genic companion group was next in conservation, with a lower ratio in peak match and a higher ratio for no peak compared with the genic top. Compared with the intragenic, the antisense, and the orphan groups, the genic companion was more conserved. Thus, the combined genic groups remain the most highly conserved with respect to TSS peak position. In contrast, the nongenic groups, which include the intragenic, the antisense, and the orphan groups, were all much less conserved. The ratios of peak match and no peak in the orphan group were 3.68% and 89.9%, respectively. A considerable reduction in peak match and elevation of no peak from the genic top indicated a much lower peak conservation for the orphan compared with the genic top group. These results are consistent with those of sequence conservation, indicating weak conservation of the orphan promoter group (Panel A). The intragenic and the antisense groups also had much lower ratios for peak match compared with the genic groups (0.90% and 2.8% for the intragenic and the antisense, respectively). The ratios for the no peak of the groups were much higher compared with those of the genic top (96% and 91% for the intragenic and the antisense, respectively). Therefore, despite the high sequence conservation shown in Panel A, the comparative TSS analysis revealed that the intragenic and the antisense promoters are genetically much less stable compared with the genic top group. We analyzed the total clusters validated by C2 tags and ≥5 TSS tags in Col-0 (Fig. 2C). We also evaluated weakly expressed clusters with <100 TSS tags. As shown in Supplementary Fig. S1, we confirmed weaker conservation for the peak positions in nongenic promoter groups. Therefore, the high conservation of the peak positions in the genic top group was not due to its high expression (Fig. 1B). We concluded that nongenic promoter groups, which are the intragenic, the antisense, and the orphan, are much less genetically conserved (Fig. 2). These results suggest much less biological significance for the nongenic promoter groups compared with the genic groups. Characteristics of the stable subsets in nongenic promoters We focused on the genetically stable promoters in the nongenic promoter groups. Figure 3 shows the number of promoters that are conserved regarding the peak position in the five accessions, including the reference (Col-0). Among 17 670 genic top promoters for protein-coding genes [genic top (CDS)], 30% were conserved. Interestingly, promoters for pseudogenes showed less conservation [genic top (pseudogene), 12.7%], demonstrating a reduction in conservation based on the loss of biological significance. The genic companion promoters for the protein-coding genes also exhibited a lower conservation rate compared with the genic top (CDS) [Genic Comp (CDS), 9.68%]. Number of conserved promoters for each promoter group. The number of promoters of the peak match in Fig. 2C is shown. Gene lists with the conserved intragenic and antisense groups are shown in Supplementary Tables S4 and S5. As previously observed, nongenic promoters are much less conserved. The conserved promoter numbers were 262, 87, and 215 for the intragenic (CDS), antisense (CDS), and orphan groups, respectively. The genes for the conserved promoters of intragenic (CDS), antisense (CDS), and the conserved orphan promoters are listed in Supplementary Tables S4–S6, respectively, together with genic top (CDS) (Supplementary Table S2) and genic companion (CDS) (Supplementary Table S3). The list of conserved promoters for intragenic (CDS) (Supplementary Table S4) includes the gene encoding glycerate kinase, GLYK (AT1G80380), the final enzyme in the photorespiration pathway. We found that GLYK expression is controlled by genic top and intragenic promoters [7]. The former is dominant under light conditions and produces a chloroplast-targeting enzyme, whereas the intragenic promoter is more active in the dark to yield a truncated and cytosol-targeting enzyme. Because both enzymes are indispensable for the normal growth of Arabidopsis, this indicates that the intragenic promoter of GLYK is indispensable for proper growth of Arabidopsis. The list also includes SRG3 (AT3G02040) and PPDK (AT4G15530), which contain a pair of promoters that also produce full-length and truncated protein products (Supplementary Table S4) [3, 7]. These results suggest that the conserved 262 intragenic and 87 antisense promoters (Fig. 3), which are a minority in each promoter group, are biologically significant nongenic promoters. It should be noted that five other examples of biologically significant intragenic promoters (sAPX/AT4G08390 and MDAR6/AT1G63940 [4], PMDA/AT1G63940 [2], ATX1/AT2G31650 [5], and HO1/ AT2G26670 [6]) are not listed in Supplementary Table S4. Therefore, it is unlikely that all of the biologically significant intragenic promoters have been accounted for. Nevertheless, the conserved list, which is 0.78% of the intragenic promoters, covers three out of seven cases involved in promoter switching, demonstrating high concentration of biologically important intragenic promoters. The results of a GO analysis for the conserved promoters are listed in Supplementary Table S7. The enriched GO terms for the conserved genic top promoters with environmental responses were not constitutive, but regulated genes. This would in part reflect the enrichment of the TATA type in conserved promoters [28, 29]. Regarding the conserved intragenic promoters, the most concentrated GO category associated with biological process was “sulfur compound biosynthetic process,” with less abundance compared with the top GO terms in the conserved genic top. Nucleosome positioning and histone modification are associated with transcription and its regulation [30, 31]. Studies of eukaryotes, including yeast [32], Drosophila [33], human [34] and Arabidopsis [35], have revealed that the promoter region (∼0.2 kb upstream of TSSs) is free from nucleosomes, and they begin to appear just after the TSS toward the downstream, transcribed region. The nucleosome positioning data of Chodavarapu et al. [23] were aligned according to our TSS position data and summarized for the genic and nongenic promoter groups (Fig. 4). Variability of the nucleosome signature is shown in Supplementary Fig. S2. As expected, the genic promoters (genic top + genic companion) exhibited a typical nucleosomal positioning profile with a nucleosome-depleted region just before the TSS peak and high signals following the peak (all, genic). The conserved subset of the genic promoters also showed a similar profile; however, the positions of the nucleosomes downstream of the TSS peak were more synchronous compared with those in the all group (conserved, genic). Nucleosome positioning in all and the conserved nongenic promoter groups. The average read counts from −1000 to +1000 bp relative to the TSS peaks in the promoter groups and subgroups are shown. Data are based on MNase-seq (Chodavarapu et al., 2010). Data were subjected to smoothing with a bin of 21 bp. The number of promoters was: genic all = 27 985, genic conserved = 6312, intragenic all = 34 028, intragenic conserved = 297, antisense all = 3626, antisense conserved = 103, orphan all = 5847, and orphan conserved = 215. In contrast, the intragenic promoters (all, intragenic) lack a nucleosome-depleted region upstream of the TSS peak and an increase in the number of nucleosomes downstream of the peak. Therefore, these results indicate that the intragenic promoters do not contain a promoter structure involving nucleosome positioning. However, the conserved subset of the intragenic promoters does have the characteristics of the genic promoters, which contain a nucleosome-depleted region upstream of the TSS peak and high nucleosome formation downstream of the peak (conserved, intragenic). According to the difference between the conserved subset and the total intragenic promoters, transcription driven by the intragenic promoters is not epigenetically organized compared with that of the genic promoters, whereas the conserved subset is with respect to nucleosome positioning. The results indicate a clear difference between the conserved minority of the intragenic group and the whole group. Similar situations were observed for the antisense and orphan promoters (antisense all and conserved, orphan all and conserved). We also examined the nucleosome positioning in two other conditions, cold-treated and its control, and obtained similar results (Supplementary Fig. S3). These results suggest that the trend of the conserved promoters in nucleosome positioning is independent from physiological conditions. We also analyzed the localization of a histone variant, H2A.Z, using the corresponding ChIP-seq data of Col-0 [26]. Similar to other eukaryotes, Arabidopsis shows a specific occupation of H2A.Z at the 5′-end of the transcribed region [31]. As shown in Fig. 5 (variation of the data is shown in Supplementary Fig. S5), the distribution of H2A.Z in the genic promoters also shows a peak downstream of the peak TSS (+300 bp), and this trend is the same as the conserved subset of the genic promoters (genic, all, and conserved). Histone variant (H2A.Z) in all and conserved promoter groups. The distribution profiles of H2A.Z, a histone variant related to transcriptional regulation. NC represents the negative control (input) and is shown as gray lines in the graphs. The number of promoters was: genic all = 27 985, genic conserved = 6312, intragenic all = 34 026, intragenic conserved = 297, antisense all = 3 626, antisense conserved = 103, orphan all = 5 846, and orphan conserved = 214. Intragenic promoters have a faint peak of H2A.Z distribution (intragenic, all); however, the conserved subset (intragenic, conserved) has a solid peak just after the TSS peak, similar to the genic promoters (genic, all). The antisense and orphan promoters have lower peaks (antisense and orphan, all), but their heights increase in their conserved subsets (antisense and orphan, conserved). The results of H2A.Z distribution are consistent with those of nucleosome positioning (Fig. 3). Similar results were obtained for H4K16ac, a histone modification that is preferentially observed just after the TSS peak (Supplementary Figs S4 and S6). Along with the results of nucleosome positioning, we found that the whole nongenic promoter groups, the intragenic, the antisense, and the orphan, lack the epigenetic structures that are observed in the genic promoters or contain them in limited amounts. However, the conserved subsets of nongenic promoters fully share these epigenetic structures with the genic promoter group. These results indicate that the conserved subsets of the nongenic promoters have different characteristics from the corresponding promoter group as a whole and suggest they are biologically significant. Factors affecting the stability of the peak position We subsequently surveyed the factors affecting the stability of the TSS peaks. A comparison of the stability among the different core promoter types, which are the genic top promoters containing the TATA box, Y patch, CA element, and GA element [28, 36], is shown in Fig. 6. The highest peak match was found for the TATA type. The TATA box coexists with the Y patch [28]; therefore, we also analyzed peak conservation in the promoters containing the TATA box and Y patch. The results indicated higher peak conservation in the promoters compared with those containing the TATA box. This supports the idea that more structured promoters are more conserved in peak position. Core types, expression level, and shape of the TSS cluster affect the stability of the TSS peak position. (A) Stability of the peak positions in the genic top promoters with the core types is shown. The ratio of the three cases of peak shift (see Fig. 2C) is shown. (B) Ratio of the peak match promoters based on the expression level. Promoters with each core type (Coreless, GA, and TATA) were divided into five groups based on their expression level (horizontal axis). (C) The ratio of peak match promoters in five groups divided by the peak ratio for each core promoter (with ≥100 TSS tags). A high peak ratio makes peaky TSS clusters, and a low ratio makes a broad one [28]. The P values in the panels (B) and (C) are calculated with Cochran–Armitage trend test. Reduction of peak match should lead to an elevation of peak shift and no peak (Panel A); however, one interesting feature of the TATA type is the higher value of no peak than peak shift, which is not observed in other promoter types. This suggests that the capacity of the acceptable peak position is narrower in the TATA-type promoters compared with other types. This idea is supported by the observation of a shorter peak shift for the TATA type compared with the promoters with other core types (Supplementary Fig. S7) or the total genic top promoters (Supplementary Fig. S9). Therefore, if not accepted, the total TSS cluster may be cancelled in the TATA type, resulting in no peak. This is also supported by the observation of two phases in the relationship between the distance of the shift and appearance cases in the TATA-type promoters, which was not detected in the other core types (Supplementary Fig. S7). TATA-type promoters exhibit high expression levels [28, 29]. We examined the relationship of expression level to peak match. As shown in Panel B, there is a clear positive relationship between peak match and expression level, regardless of the core promoter type. Therefore, the high ratio of peak match in the TATA-type promoters is caused by high expression compared with promoters containing other core elements and no recognizable core elements. Supplementary Fig. S8 provides supportive data based on six expression data sets determined by TSS-seq analysis [4]. The results indicate that the relationship between peak match and expression level shown in Fig. 6B is conserved with these various tissues and physiological conditions. Another feature of the TATA-type promoters is the peaky TSS clusters (Fig. 3B in [28]). Therefore, there is another possibility for the high peak match in the TATA type in which the peak positions of the peaky TSS clusters are genetically more stable compared with those of the broad TSS clusters. Peaky and broad shapes of the TSS clusters are judged by peak ratio, which is the rate of the peak height in the TSS cluster. In Panel C, analysis of the TATA type and others revealed a positive correlation of the peak ratio to peak match, regardless of the core type. Taken together, the results indicate a clear positive relationship between the expression level and peaky TSS shape to the genetic stability of TSS peaks. The high rate of peak match in the TATA type is the result of these two relationships. Preference of the local sequence around the old and new TSS peaks To understand the characteristics of the new TSS peaks, we analyzed the local sequence around the new and reference TSS peaks. New TSS peaks have an apparent shift from the reference peaks in Col-0 and are also minorities among the four accessions. The frequency of the triplet sequence (−1, +1, and +2, relative to TSS peak) for the genic top TSS peaks in Col0 and new TSS peaks was subjected to frequency analysis (Fig. 7A). As shown, the Col-0 reference peaks prefer T and C at −1 and G and A at +1, which is the YR Rule supported in our previous study (T and C: pyrimidine = Y, A and G: purine = R) [28]. The new peaks also follow the YR Rule, and the overall preference of the new peaks in the triplet is similar to that of the reference peaks. Therefore, the rule determining the peak positions appears to be the same between the reference peaks in Col-0 and the new peaks that are shifted from the reference peaks. Prediction of a new TSS peak based on the local sequence in the genic top group. (A) Minor TSS peak positions that are observed in only one of the four accessions were considered new TSS peaks, and 6311 new TSS peaks were identified. The frequency of the triplet sequences around the TSS peak (−1, +1, and +2 relative to the TSS peak) was compared between the Col-0 genic top TSS peaks (closed bar) and new TSS peaks (open bar). (B) Workflow for the prediction of new peaks. Frequency data for the triplets (−1, +1, +2) (left) were transformed into a common logarithmic expression to make prediction scores (middle). The prediction scores were applied to the sequence in the potential area of a new peak. We focused on local shifts that had a distance within 5 bp of the original position. Therefore, a region from −6 to +7 relative to the genic top TSS peaks in Col-0 was scanned, and the position with the maximum prediction score was predicted as a new peak position (right). (C) Summary of the prediction results for the new genic top peaks and shifts from Col-0. To calculate accuracy, true values (1157) were divided by the number of <6 bp shifts (1668) for the new TSS. The line of “Shift from Col-0” is the result of the prediction of all local shifts in the accessions from Col-0, although it is unclear if the genetic shift is forward (from Col-0) or reverse (to Col-0). The bottom row shows the number of shifts accompanied by SNPs within −5 to +6, which were analyzed in Fig. 8. Based on this similarity, we attempted to predict the new positions based on the rule in the reference peak set in Col-0. Panel B shows the scheme for this prediction. The frequency of all triplets in the reference peak set was counted, which was then transformed into a prediction score by log transformation. The prediction score was applied to the local sequence around a reference TSS, and the position with the highest score was considered the new peak. We applied two data sets for prediction: new TSS peaks representing a minority among the accessions (new TSS), as in Panel A, and shifted peaks from Col-0 in the accession(s) (Shift from Col-0). The latter includes not only genetically newer peaks from the reference but also genetically older peaks compared with the reference. An example of the new TSS is shown in Supplementary Fig. S10A. Three accessions, Altenb-2, Shigu-2, and Vash-1, showed the same position as the reference (Col-0), and only Kly-1 showed a peak at the -3 position (+1 as the peak position of reference TSS). Kly-1 shifted from the reference position for an unidentified reason, and the new position (+5) had the second-highest prediction score after that of the reference. Each data set was divided into two groups: local shifts and more distant ones. The local group had new peaks within 10 bp around the reference peak, whose distance from the reference was 5 bp or less. The distant group shows a new peak with a distance of >5 bp from the reference (>5 bp shift). The case for the local shifts of the New TSS were 1668. Among them, 1157 were the predicted position, yielding a prediction accuracy of 69.4% (Panel C). The remaining cases with more distant shifts were 4443, and they were not subjected to prediction because they were out of the range of local prediction. Similarly, local shifts of shift from Col-0, which were 8737, were also subjected to prediction, and the accuracy was 81.0%. Both accuracies were much higher compared with that of random prediction, which was 10%. These results indicate that the prediction of new TSS peaks is possible with our prediction method under the limited situations. Supplementary Fig. S10B shows the relationship between prediction score and the distance of the shifted peaks from the reference position. The graph shows two drop points at −1 and +1 bp from the reference TSS peak. The other region from −10 to +10 bp remained higher at a value of ∼9. These results in a situation in which the prediction of the shifts to these two points is less successful. In addition, new shifts for the other region in Panel B yielded slightly lower scores compared with the average of the reference TSS peaks shown as no shift. This reduction suggests that the new TSS peaks less stringently follow the preferential local sequence compared with the older peaks. The effect of shifting TSS peaks on the expression level was analyzed in Supplementary Fig. S10C. The results show that local shifts of the TSS peaks tended to cause a reduction in the expression level. SNPs and shift of the TSS peaks Our prediction for a new TSS is dependent on the observation of a shift, and the shift itself was not able to be predicted (Fig. 7). One possible cause for the shift of the TSS peak is the presence of an SNP around the peak, which disturbs the context of the transcription start site or the initiator. Of the total 6111 cases of new shifts, only 200 had an SNP within the area from −6 to +7 (Panel C). Therefore, for the 5911 cases, the causes for the shift of a TSS peak are unclear. SNPs around the TSS peak were analyzed. Figure 8A shows the effects of SNPs on the peak shift. Shift cases accompanied by SNPs from −50 to +50 relative to the reference TSS peak are shown in the graph. SNP occurrence at +1 and +2 relative to the reference peak results in a drop in the rates of peak match and an increase in peak shift. Compared with these points, the other SNPs from −50 to +50 were less effective. Therefore, a tight relationship between SNP and the peak shift was observed only for the SNPs at +1 and +2 when local SNPs were considered. Effect of local SNPs on TSS peak shift in the genic top group. (A) Appearance rate of peak match clusters and peak shift clusters when SNPs are present at the indicated positions. (B) Direction and distance of shifted peaks from Col-0 that have an SNP at +1 (= TSS peak in Col-0, 163 cases) and +2 (163 cases), together with the cases having no SNPs from −1 to +2 (71 152 cases), are shown. (C) Effect of SNPs and local shift of TSS peak on the prediction score. The cases corresponding to the rhombus (SNP at the +1 position & +1 bp shift, B) were subjected to the analysis. The left illustration shows the triplet sequences subjected to the analysis. “Mut” denotes the position of the SNP. I, II, and III represent the triplet sequence in the reference genome around the reference peak position (Col-0), the triplet sequence in the mutated accession genome around the peak position of the reference, and the triplet sequence in the mutated accession around the shifted peak, respectively. Figure 8B shows the relationship of the shifted positions of the TSS peaks to the SNP positions. Three sets of the genic top promoters were analyzed from four accessions, including No SNP (71 152), SNP@+1 (163), and SNP@+2 (163). As shown in the graph, 42% of the No SNP promoters in the region from −1 to +2 bp had the same peak position (0 bp), indicating that 58% of the No SNP promoters shift or lose their peaks. This suggests that there are many cases of peak shift not caused by disturbance in the local context by the introduction of an SNP. Their shifted peak positions are scattered around the reference peak with some preference to the region from +2 to +6. In contrast, the promoters of SNP@+1 and SNP@+2 primarily prefer the +1 bp shift and +2 bp shift from the reference peak, respectively. These shifts occur much more frequently than no shift (0 bp shift). Together with the results in Panel A, SNPs influence the TSS peak positions only at +1 and +2 relative to the original peak. As a result, the peak position shifts 1 and 2 bp downstream from the original peak, respectively. Panel C shows the change in the prediction score in the case of SNP@+1 and a 1-bp shift to the downstream side for the cases labeled with a rhombus in Panel B. The prediction score of the triplet in the reference is labeled “I,” and it was 10.4. With an SNP at −1 (Mut in the left illustration), the score of the same position as a reference, labeled “II,” dropped to 4.6. However, the shifted score with the new TSS position (arrow in the bottom illustration on the left), labeled “III,” was 7.4, which is higher than the median of II. This represents a pattern of cases in which a shift of TSS can be explained by SNPs and the resulting change in the prediction score. The other cases from the one shown in Panel C (left), including the shifting of TSS peaks without any SNPs around the reference TSSs, could not be explained by local SNPs. Discussion The characteristics of the genic and nongenic promoter groups revealed in the present and previous studies are summarized in Table 1. The table lists the sequence stability, stability of the TSS peak positions, epigenetic traits, core promoter ratio, level of the promoter context, and expression level. Table 1.Characteristics of the genic and nongenic promoter groupsSequence stabilityStability of TSS peak positionEpigenetic traits as promoterPresence of core promoter elementsPromoter contextExpression level This work This work This work Tokizawa et al. 2017 Hiratsuka et al. 2022 This work & Tokizawa et al. 2017 Genic Top HighVery highClearHighHighHigh Genic comp HighHighClearHighModerateLow Intragenic Very highVery lowNoneVery lowNoneLow Antisense Very highLowObscureLowNoneLow Orphan LowLowNoneLowLowLowStudies on the core promoter ratio and expression level are reported by Tokizawa et al. [4]. Measurement of the promoter context, which is the evaluation of the spacer and regulatory elements within the range from −200 to −60 bp of the promoter structure, regardless of the core promoter type, was analyzed by Hiratsuka et al. [37]. Characteristics of the genic and nongenic promoter groups Studies on the core promoter ratio and expression level are reported by Tokizawa et al. [4]. Measurement of the promoter context, which is the evaluation of the spacer and regulatory elements within the range from −200 to −60 bp of the promoter structure, regardless of the core promoter type, was analyzed by Hiratsuka et al. [37]. The genic top group had the highest level among the five promoter groups in sequence stability and stability of the TSS peak position. These two features indicate the genetic stability of the group. Epigenetic features, including nucleosome positioning (Fig. 4), H2A.Z (Fig. 5), and H4K16ac (Supplementary Fig. S4), also indicate that the genic top group has the highest level of promoter. The ratio of core element presence (Fig. 4B in [4]) and the level of the promoter context of the group (Fig. 7A in [37]) were also highest. These results indicate that the genic top promoter is the most highly structured among the five groups. These data also indicate that the genic top promoter group is genetically the most stable and structured, which is consistent with its biological significance for gene expression. Interestingly, the expression level was also highest among the groups, suggesting that highly expressed promoters are structured, genetically stable, and biologically significant. The genic companion group was the second to the genic top in terms of sequence stability (Table 1). We estimated that this group was negligible when considering the gene expression level (Fig. 6B in [4]). As mentioned in the introduction, there are several examples in which the truncation of the 5′UTR having a uORF by promoter switching between the genic top and the genic companion results in an increase in translational efficiency [1]; however, the physiological significance is unclear. In eukaryotes, translational regulation through a hairpin structure in the 5′UTR was reported in translational regulation of iron utilization [38]. Considering these reported effects of the 5′UTR in translational regulation, there is a potential for gene regulation by promoter switching between the genic top and the genic companion. Following the identification of 59 551 promoters for 22 211 genes, the question was how many of the 37 340 genic comp promoters were involved in gene regulation. Our results suggest that it is less significant than the genic top and more than the nongenic groups. Therefore, it is not surprising that the majority of the group has no role in gene regulation. One feature of the intragenic and antisense groups is the overlapping of the TSS regulation with the gene body. This feature brings high sequence stability to the surrounding region of the TSS, as shown in Table 1; however, our TSS analysis revealed that their peak positions were genetically unstable. Together with their low structural features, we concluded that these two groups are biologically insignificant, with some exceptions. One interesting feature in Table 1 is the relationship with the expression level. The genetically stable and highly structured promoter group, the genic top, exhibits high expression levels, whereas the low-expressing promoter groups tend to be genetically unstable and poorly structured. This widely observed relationship among the five promoter groups suggests that active promoters with high expression levels require a solid promoter structure, whereas weak promoters do not. This hypothesis further leads to the idea that low-expressing promoters are the result of “leaky” transcription, and this may occur when the event does not cause any negative selection pressure during evolution. The concept of “transcriptional noise” was raised when noncoding RNA was discovered [39]. We identified 193 208 intragenic, 42 070 antisense, and 34 549 orphan promoters in the Arabidopsis genome [4]. It would be surprising if this vast number of promoters were involved in gene regulation. Because the majority of these nongenic promoter groups are much less active than the genic top (Fig. 1B), their existence in the genome may be harmless and thus cause negative selection pressure at a minimum. This idea does not exclude biologically significant promoters in these nongenic groups as a minority, which may be shown in the active fraction on the right side of Fig. 1B. Genetically stable promoters are considered “mature.” Analysis of the genetic stability of the local sequences and TSS peak positions revealed that the genic top group was the most mature among the promoter groups. Interestingly, it has the highest expression level (Table 1). A positive relationship between promoter maturity and expression level was also observed within the genic top group, regardless of the core type (Fig. 6B). Based on these results, we suggest that highly expressing promoters are mature. Among the genic top group, the TATA-type promoters were the most stable in the TSS peak position (Fig. 6). This reflects the characteristics of the TATA-type promoters having peaky TSS clusters [28]. In addition, the TATA-type promoters are rich in regulated genes compared with those that are constitutively expressed [29]. Therefore, the subset of the genic top group with the conserved TSS peak positions is rich in stress-responsive genes (Supplementary Table S7). Based on this point of view, the results of the GO analysis can be an indirect consequence of the enrichment of the TATA type. Before beginning this study, we assumed that the shift of the TSS peak positions was caused by an alteration in the local sequence within the corresponding promoters. However, the cases showing a clear correlation between the SNP and the TSS peak shift are rather limited (Fig. 8B). There are two possibilities for the many shifts observed without corresponding local SNPs. One is the presence of sequence alterations that were not detected by resequencing [16, 21]. Another possibility is that the peak shift is also caused by a mutation in distant positions from the TSS peaks. The latter includes modification of the core promoter elements excluding Inr [40], the middle-range promoter context, which has a distance from tens to hundreds of bps [37], and the long-range promoter context, which is determined by the chromosomal conditions at a distance of tens of kbps. It is not easy to establish a tight relationship between the change in sequence and the peak shift. Easy cases, as shown in Fig. 7B, are, therefore, exceptions. Supplementary Material