Characterization of circRNAs in SDX

CircRNAs arise through a form of AS known as back-splicing. CircRNAs are abundant, stable, conserved, and expressed in a stage-specific manner in plants (Ye et al., 2015; Tan et al., 2017; Zhao et al., 2017; Zeng et al., 2018; Wang et al., 2019). To identify and characterize circRNAs during wood formation on a genome-wide scale, we performed rRNA removal and RNase R treatment to enrich circRNAs from total RNAs, followed by Illumina paired-end sequencing (Figure 1A). We obtained 267,553,488 150-bp paired-end (PE) reads (Table S1). In total, we identified 2,742 circRNAs, including 1,752 circRNAs derived from lariat introns (ciRNAs) and 990 generated from exon circularization (Figure 1B, full list in Table S2). The percentage of ciRNAs (63.89%) was higher than the percentage of circRNAs (Figure 1C), suggesting that ciRNAs are more widespread in P. trichocarpa, which is consistent with findings in soybean (Zhao et al., 2017), human, and mouse (Panda et al., 2017).

To validate the identified circRNAs, we selected 13 circRNAs generated from exon circularization and performed an RNase R resistance experiment. We designed a set of divergent primers (Table S3) for RT-PCR of each circRNA and convergent primers to amplify linear transcripts of EF1α and 18S rRNA as controls. Most circRNAs were resistant to RNase R digestion compared to linear transcripts of 18S rRNA and EF1α (Figure 1D). However, circ-AUX-3 generated from the transcription factor gene AUX (Potri.003G001000) was susceptible to RNase R and did not show enrichment after RNase R treatment, and 18S rRNA transcript was not completely digested, likely because it is the most stable housekeeping gene and is not sensitive to RNase R (Bas et al., 2004). Sanger sequencing of gel-purified bands confirmed the presence of authentic back-splicing junction regions, which is consistent with the finding that some circRNAs are susceptible to RNase R (Jeck et al., 2013; Li et al., 2019).

Among the 990 circRNAs generated from exon circularization, 91 (9.19%) originated from only one annotated exon, and the 899 remaining circRNAs included multiple exons. CircRNA with two circularized exons was the most common type (Figure 1E), which is similar to findings for Phyllostachys edulis (Wang et al., 2019), Poncirus trifoliata (Zeng et al., 2018), Glycine max (Zhao et al., 2017), and A. thaliana (Ye et al., 2015). Interestingly, exons from circRNAs with only one circularized exon were much longer than exons from circRNAs containing multiple circularized exons (Figure 1F). A gradual reduction in the length of circularized exons was observed with increasing number of circularized exons, suggesting that exon length is one condition affecting circRNA biogenesis. Repetitive elements are responsible for most circRNA formation in humans (Dong et al., 2017). However, a few inverted complementary sequences were detected around the circularized exons in P. trichocarpa (Figure 1G). This result indicates that the flanking reverse complementary sequence is not the major factor involved in the biogenesis of circRNAs in P. trichocarpa, suggesting that other mechanisms might be involved, which is consistent with previous findings in Phyllostachys edulis (Wang et al., 2019), Oryza sativa (Lu et al., 2015; Ye et al., 2015), and A. thaliana (Ye et al., 2015; Sun et al., 2016).

To investigate the splicing signal of circRNAs, we extracted and analyzed 11-nt DNA sequences around 5′ splicing sites (5′ SS), 3′ splicing sites (3′ SS), and in regions close to branchpoints for ciRNAs. Motif analysis revealed canonical GT-AG splicing signals involved in the formation of circRNAs (Figure 1H), indicating that circRNA formation depends on the spliceosome machinery. For ciRNAs, the sequence logo for the region close to the branchpoint was enriched for U nucleosides. The mean distance from the branchpoint to the 3′ SS for ciRNAs was approximately 25 bp, which is similar to that of Phyllostachys edulis (Wang et al., 2019). Back-splicing/splicing sites play decisive roles in the origination of circRNA, and can result in the generation of multiple circRNAs from a single gene locus (Jeck et al., 2013; Zhang et al., 2016). Among the 990 circRNAs in SDX, we identified 27 alternative 3′ back-splicing (A3BS) circRNAs and 74 alternative 5′ back-splicing (A5BS) circRNAs (Table S4). In addition, 62 circRNAs were identified with fully overlapping region across the back-splicing junction (Figure 1I). GO enrichment analysis revealed that genes generating ciRNAs are involved in the Golgi apparatus, intracellular protein transport, and ATP-dependent RNA helicase activity. Genes generating circRNAs were enriched for microtubule binding, intracellular protein transport, and microtubule-based movement.

Global survey of post-transcriptional regulation in SDX

PacBio sequencing produces longer reads than short-reads methods, facilitating the generation of a well-annotated transcriptome, including splicing isoforms (Rhoads and Au, 2015; Abdel-Ghany et al., 2016; Wang et al., 2016). We constructed full-length first strand cDNA libraries with insert sizes of 1–2, 2–3, 3–6, and 5–10 kb and sequenced them on the PacBio RSⅡ Single Molecule, Real-Time (SMRT) sequencer, yielding 298,716 reads of inserts (ROIs) (Figure 2A; Table S5). To assess the data quality and obtain FLNC reads, we filtered the polymerase reads, yielding 161,486 FLNC reads (54.1%). We also obtained 120,818,725 paired-end reads by performing dUTP mRNA-seq on the Illumina HiSeq 2500 platform to correct the single-molecule long reads. Finally, 161,334 error-corrected FLNC reads were generated, and 160,622 mapped FLNC reads were obtained (Figure 2A). A total of 7,959 genes containing two or more polyadenylation sites were identified (Figure 2B; Table S6), suggesting that alternative polyadenylation (APA) events are a common phenomenon in P. trichocarpa. Proximal and distal poly(A) cleavage sites showed very similar nucleotide distributions (Figure 2C), indicating that proximal sites are likely common genuine poly(A) cleavage sites in plants. We analyzed the nucleotide composition in the upstream (−50 nt) and downstream (+50 nt) regions of all poly(A) cleavage sites and found that the AAGAAA motif and the UGUA motif were the two most common polyadenylation signals, which were enriched in the regions 5 nt downstream and 40 nt upstream of poly(A) sites, respectively (Figure 2D).

To confirm the expression of the four modes of AS in P. trichocarpa, including alternative 3′ splice site (A3SS), alternative 5′ splice site (A5SS), intron retention (IR), and exon skipping (ES), we analyzed AS events using PacBio long-read sequencing and Illumina short-read sequencing. Using hybrid sequencing, we identified 7,836 (Table S7) and 13,715 (Table S8) AS events on the PacBio Iso-seq and Illumina platforms, respectively (Figure 2E, F). In total, the two sequencing platforms generated 1,581 overlapping AS events, primarily in highly expressed genes (Figure 2F, G). The Illumina platform generated 42.8% (5,869) IR events and 10% (1,367) ES events, whereas the PacBio platform generated 77.9% (6,106) IR events and only 4.7% (365) ES events. Further analysis of the intron lengths of IR events revealed that there were longer introns in the PacBio versus Illumina data (Figure 2H). Thus, the PacBio platform is better able to detect IR in long introns, which is consistent with disadvantage for the Illumina-based short read assembly. However, Illumina is better able to detect low-abundance genes due to the better depth of sequencing.

To validate the results of long-read sequencing, we performed RT-PCR to detect seven predicted AS events. Four of these AS events were confirmed, as two bands of the expected size were produced (Figure 2I). However, the primers for the three remaining AS events failed to amplify the expected bands for long isoforms, perhaps due to the low expression levels of these long isoforms in SDX tissue. The successfully validated AS genes are all homologous to Arabidopsis genes related to cellulose development, such as homologs of FEI1, which functions in cell wall biosynthesis (Xu et al., 2008), and GH9B8, encoding a cell-wall-loosening protein (Xu et al., 2014).

R-loop atlas in SDX

R-loops comprising one single-stranded DNA and a DNA:RNA hybrid are ubiquitous functional structures in Arabidopsis (Xu et al., 2017). However, to date, an atlas of R-loop peaks in P. trichocarpa has not been generated. Therefore, we characterized R-loops in P. trichocarpa by performing DRIPc-seq and DRIP-seq (negative control with RNase H treatment) on SDX samples (Figure 3A). This analysis identified 4,834 R-loop plus (+) peaks and 4,098 R-loop minus (−) peaks (Table S9). Among the 8,932 R-loop peaks, only 785 peaks (8.79%) were also detected in samples subjected to RNase H treatment. The 8,147 remaining R-loop peaks (91.21%) were undetectable in samples treated with RNase H, suggesting that R-loops were enriched in these samples compared to the negative control. Analysis of the distribution of R-loop peaks across the transcripts revealed that R-loop structures were enriched near the transcript start and gene body regions (Figure 3B), which is consistent with previous findings (Xu et al., 2017). Most peaks were 250–500 bp long (Figure 3B, C). In total, 56.55% and 35.81% of the DRIPc signals mapped onto CDSs and UTRs, respectively (Figure 3D). Furthermore, 27.5% of the R-loop peaks contained the sequence motif GARGAAG (R = A or G) (Figure 3E), which is consistent with previous findings in Arabidopsis (Xu et al., 2017). These results confirm the reliability of DRIP-seq in P. trichocarpa and suggest that the GARGAAG motif is conserved in plants.

R-loops associated with circRNAs and AS

To identify R-loops associated with circRNA and AS in SDX, we calculated the number of circRNAs and AS that overlapped with R-loop peaks. In total, 181 circRNAs (Figure 3F) and 672 AS events (Figure 3G) overlapped with R-loop peaks. The AS events that overlapped with R-loops included 3.8% IR, 2.1% ES, 3.5% A5SS, and 2.5% A3SS events. In total, 7.2% circRNAs and 5.5% ciRNA overlapped with AS events. We randomly selected gene regions that included R-loop peaks of the same length and detected circRNAs/AS events that overlapped with these regions. After repeating the process 1,000 times, we calculated the probability that circRNAs/AS events overlapped with R-loops events, including specific mean value (μ) and standard deviation (σ). The probability that R-loop peaks overlapped with circRNAs (Figure 3H) and AS events (Figure 3I) was higher than that of random simulation. Here, we selected seven circRNAs as examples, in which two circRNAs (circRNA-1–2) and four ciRNAs (ciRNA-1–4) overlapped with R-loop peaks (Figure 3J). Among these, four overlapping circRNAs (circRNA-1, ciRNA-1, ciRNA-2, and ciRNA-3) were located exactly where AS had occurred in the parent genes.

CircRNAs associated with R-loops regulate splicing in SDX

We randomly selected gene regions showing AS events of the same length and detected ciRNAs/circRNAs that overlapped with these regions. After repeating the process 1,000 times, we found that the probability that ciRNAs/circRNAs overlapped with AS events was higher than that of random simulation (Figure 4A). Conn et al. (2017) provided the first example of the co-regulation of circRNA-R-loop formation and AS in plants, but whether this is a widespread phenomenon in plants had been unclear. We also performed overlap analysis among the three events (R-loops, circRNAs, and AS). Random simulation revealed that R-loops are generally correlated with circRNAs and AS simultaneously (Figure 4A). In total, we identified 418 R-loop peaks containing regions complementary to circRNAs by cis- or trans-regulation in 234 AS regions (Figure 4B; Table S10). These regions were enriched in GO terms such as ‘microtubule-based process’, ‘hydrolase activity’, and so on (Figure 4C).

Finally, to investigate the effects of circRNAs associated with R-loops on the regulation of AS of their parent genes in SDX, we developed high efficiency transfection of protoplasts for circRNAs. Protoplasts transfected with an empty vector (pUC19-35S-sGFP) were used as a control (Figure 4D). We selected circ-IRX7 from a hemicellulose synthase gene (Potri.009G006500), which presented IR events (Figure 4E). Circ-IRX7 derived from single exon2 and showed enrichment of R-loop. We overexpressed circ-IRX7 in protoplasts derived from SDX using the pUC19-35S-circ-IRX7-35S-sGFP vector (Figure 4F; Table S3). RT-PCR analysis of total RNA from transfected protoplasts using divergent primers of circ-IRX7 confirmed the overexpression of circ-IRX7 (OE) compared to the empty vector (EV) control in protoplasts after 12 h of incubation. First, we measured the levels of R-loops by DRIP-quantitative PCR (DRIP-qPCR), revealing that overexpressing circ-IRX increased the levels of R-loop structures compared to EV (Figure 4F, left panel). Second, we examined the effect of increasing both circRNA and R-loop formation on splicing. The overexpression of circ-IRX7 in P. trichocarpa SDX protoplasts resulted in a reduction in the levels of the long transcript (PtrIRX7-l1) and an increase in the levels of the short transcript (PtrIRX7-s) (Figure 4F, right panel). We also selected circ-GUX1 without R-loop structure to do the same experiment. Circ-GUX1 derived from exon2 and exon3 and overlapped with IR events in linear transcripts (Figure 4G; Table S3). We did not observe obvious alteration of R-loop level in overexpression of circ-GUX1 (OE) sample based on DRIP-qPCR (Figure 4H). Semi-quantitative PCR using convergent primers only detected the up-regulation of the most abundant full spliced transcript, which suggested that circ-GUX1 (OE) might affect the expression of corresponding parental transcripts. The long transcript including retention of first intron was not detected in both EV and circ-GUX1 (OE), which suggested circ-GUX1 without R-loop structure did not affect AS (Figure 4H). CRISPR-Cas9-generated null alleles of circRNA loci were recently obtained, providing the opportunity to investigate the functions of circRNAs with minimal interference from cognate linear transcripts (Zhou et al., 2021). Future technological advances should make it possible to investigate the roles of circRNAs in regulating splicing via R-loop formation in more detail.

Plant Materials

Populus trichocarpa plants were grown in a greenhouse with an average temperature of 22°C under a 16-h light/8-h dark cycle. All plant samples were wrapped in aluminum foil, immediately frozen in liquid nitrogen, and stored at −80°C (Shen et al., 2019). Total RNA was isolated from SDX using an RNAprep Pure Plant Kit (polysaccharides and polyphenolics-rich) (no. DP441; Tiangen, China) and treated with DNase I to remove DNA. Genomic DNA was extracted from SDX using a Plant Genomic DNA Kit (no. DP305; Tiangen). The quality and concentration of RNA and DNA were determined using a NanoDrop 2000 spectrophotometer (Thermo scientific).

Library construction and bioinformatics analysis of circRNA

In total, 2-μg total RNA samples from SDX were used for circRNA library construction. A Ribo-Zero Magnetic Kit (Epicentre) was used to remove ribosomal RNAs (rRNAs), and 1 μl (20 U) RNase R was added to the rRNA-depleted RNA samples to digest linear RNAs at 37°C for one hour. The enriched circRNAs were randomly fragmented and DNA fragments 400-500 bp in size were selected as templates for PCR amplification. A strand-specific RNA-Seq library was then constructed as previously described (Wu et al., 2014; Wang et al., 2017), and the dUTP strand-specific library was sequenced to generate 150-nt paired-end reads.

Raw reads from RNase R-treated RNA sequencing were filtered using ht2-filter (1.92.1) from the HTQC package (Yang et al., 2013) using the default option. These reads were mapped to P. trichocarpa genome v3.0 using TopHat (Trapnell et al., 2009) with the option −r 50 −a 8. Unmapped reads including circRNA junction sequences were remapped to the reference genome using fusion-search with the following option: tophat2 -p 50 --fusion-search --keep-fasta-order --no-coverage-search --bowtie1. Candidate circRNAs were identified using CIRCexplorer (1.1.10) (Zhang et al., 2016). The flanking complementary sequences of the circRNAs were extracted using the CSI method (Dong et al., 2017).

Validation of circRNA by RT-PCR

Divergent primers for the circRNAs were designed to amplify the back-spliced junction using PRAPI (Gao et al., 2018). A 20-μg sample of RNA was dissolved in 54-μL RNase-free water and equally divided into control and RNase R-treated samples. For RT-PCR, 10 μL total RNA (RNase R treated or untreated sample) was reverse transcribed into cDNA in a 20-μL reaction mixture using a PrimeScript^TM II first Strand cDNA Synthesis Kit (Cat# 6210A; TaKaKa) with random primers. PCR products of the predicted sizes were visualized and excised from a 2% agarose gel. Divergent primers for circRNAs and convergent primers for EF1α and 18S rRNA for RT-PCR are listed in Table S3. In addition to RNase R enrichment experiments, Sanger sequencing was performed to validate the selected circRNAs.

Overexpressing circRNA in SDX protoplasts

Genomic DNA of circ-IRX7 or circ-GUX1 generated from the hemicellulose synthase gene was amplified using convergent primers covering the circRNA sequence and two flanking intron sequences with two linkers (attB sides) on the 5′ and 3′ ends (Table S3). In total, approximately 200 ng of genomic DNA in a reaction volume of 50 μL was subjected to PCR (30 cycles: 98°C, 10 s; 50°C, 20 s; 68°C, 1 min) to amplify the target attB-PCR products, which were mixed with the pDONR207 plasmid and attP sides to generate attL-containing entry clones. Finally, the LR recombination reaction was performed with the pUC19-35S-sGFP vector to generate the expression vector pUC19-35S-circ-IRX7-35S-sGFP and pUC19-35S-circ-GUX1-35S-sGFP, which was prepared using an EndoFree Maxi Plasmid Kit (no. DP117; Tiangen) for SDX protoplast transfection.

Protoplast isolation and transfection were performed as previously described (Lin et al., 2013; Lin et al., 2014). In brief, 10-cm de-barked stem segments were submerged into 40 mL of cell wall enzyme digestion solution (1.5% (w/v) Cellulase R-10, 0.4% (w/v) Macerozyme R-10, 20 mM MES (pH 5.7), 0.5 M mannitol, 20 mM KCl, 10 mM CaCl₂ and 0.1% (w/v) BSA) in a 50-mL tube for 3 h in the dark at room temperature. Protoplasts from SDX were released by gently shaking the 50-mL tube for 1–2 min in MMG solution (0.25 M mannitol, 4 mM MES (pH 5.7), and 15 mM MgCl₂). The concentrations of protoplasts were measured with a hemocytometer (no. 3110; Sigma-Aldrich, China) and adjusted to 2 × 10⁵–8 × 10⁵ cells/mL with MMG solution. For protoplast transfection, protoplasts, plasmid DNA (pUC19-35S-circ-IRX7-35S-sGFP and pUC19-35S-circ-GUX1-35S-sGFP), and PEG-4000 solution (40% (w/v) PEG 4000, 0.2 M mannitol and 100 mM CaCl₂) were combined at a volume ratio of 10:1:11. The transformation system was incubated for 10 min at room temperature. Protoplasts from the same batch transfected with empty vector (pUC19-35S-sGFP) were used as a control. Transfected protoplasts were incubated in WI solution (0.25 M mannitol, 4 mM MES (pH 5.7), 20 mM KCl) in a Petri dish coated with 1% (w/v) BSA solution for 12 h at room temperature in the dark. To generate cDNA, each 0.5-μg RNA sample from transfected SDX protoplasts was reverse transcribed using random primers with a PrimeScript^TM II first Strand cDNA Synthesis Kit. The cDNA samples were diluted 5×. Divergent primers (circ-IRX7 and circ-GUX1) and convergent primers (linear-IRX7 and linear-GUX1, Table S3) were used to detect the overexpression of circRNA and the AS of host genes, respectively. The PCR products were visualized on a 1.5% agarose gel stained with GelStain (no. GS101-03; TransGen, China).

PacBio Iso-Seq, bioinformatics analysis, and validation

Total RNA concentrations were measured using an Agilent RNA 6000 Nano Kit (Agilent Technologies, Santa Clara, CA, USA). RNA samples with RNA integrity number (RIN) > 8.5 were reserved for PacBio Iso-Seq library construction. Double-stranded cDNA synthesis was performed using a SMARTer PCR cDNA Synthesis Kit (no. 634925; Clontech). During the cDNA size selection step, we used Blue Pippin (Sage Science) to select cDNA fragments 1-2, 2-3, 3-6, and 5-10 kb in size for downstream library construction. The selected cDNA fragments were further amplified by PCR to obtain sufficient cDNA for subsequent library construction using a Template Prep Kit (no. 100-259-100; PacBio). Adaptor dimer and other contaminating sequences were removed. The small insert libraries (1-2 and 2-3 kb) were purified twice with PB beads. Large insert libraries (3-6 and 5-10 kb) were obtained using Blue Pippin selection.

To pre-process the PacBio Iso-Seq data, we used ConsensusTools.sh from smrtanalysis_2.3.0 to obtain circular consensus sequences (CCS) with the following parameter: --minFullPasses 0 --minPredictedAccuracy 80. Full-length non-chimeric reads (FLNC) were generated by pbtranscript.py classify with the following command: --min_seq_len 300. The FLNC reads were corrected by LSC (Au et al., 2012) with the default option. The FLNC reads were aligned to P. trichocarpa genome v3.0 using GMAP (Wu and Watanabe, 2005) with the following option: --no-chimeras --cross-species --expand-offsets 0 -K 8000 -f 2 -n 1 -t 40. The AS from FLNC reads were identified using rMATS (Shen et al., 2014). Motif analysis was performed by MEME-ChIP (Machanick and Bailey, 2011) with the following option: meme-chip -norc -oc output-meme-minw 6 -meme-maxw 20 -meme-p 40.

Each 1-μg RNA sample from SDX was reverse transcribed into cDNA using Oligo dT as the primer performed using a PrimeScript^TM II first Strand cDNA Synthesis Kit. 5× cDNA was used as a template for the identification of AS (Table S11). RT-PCR was carried out using 15-μL Premix Taq (TaKaRa) in a 30-μL reaction system (1 μL 5× cDNA, 0.5 μL each F/R primer, 13 μL ddH₂O) for 33 cycles (94°C, 45 s; 55°C, 35 s; 72°C, 30 s). The PCR products were visualized on a 1.5% agarose gel stained with GelStain (TransGen).

DRIPc-seq and DRIP-seq to identify R-loop peaks

DRIP was performed as described previously (Xu et al., 2017; Sanz and Chedin, 2019) with minor modifications. In brief, nuclei were isolated from 2.5–3 g of SDX powder using Honda buffer (0.44 M sucrose, 1.25% Ficoll 400, 2.5% Dextran T40, 20 mM HEPES-KOH (pH 7.4), 10 mM MgCl₂, 0.5% Triton X-100, 5 mM β-mercaptoethanol, 1 mM PMSF, and 1% protease inhibitors) (Sun et al., 2013). The isolated nuclei were resuspended in nuclei lysis buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA, 1% SDS) and incubated at 37°C overnight (12–14 h) after adding proteinase K. Genomic DNA was extracted from the samples using the phenol–chloroform method and precipitated with ethanol. The DNA was fragmented to 200–500-bp pieces with an automatic high-throughput DNA sonicator (Bioruptor Pico, Diagenode, BEL). The negative control was treated with RNase H (New England Biolabs) at 37°C overnight, and both the experimental group and negative control were subjected to DRIP using the S9.6 anti-RNA-DNA antibody at 4°C overnight. After antibody-RNA:DNA complexes were bound to Protein A/G Magnetic Beads and thoroughly washed, the DNA pellets were eluted in 30–50 µL RNase-free ddH₂O. After DRIP, the eluted DNA of the experimental group was treated with DNase I (New England Biolabs) at 37°C for 45 min to degrade all DNA. The RNA strands in R-loops were ethanol-precipitated and reverse-transcribed into cDNA using random hexamers. Second-strand synthesis was performed using deoxyuridine triphosphate (dUTP). A USER enzyme treatment was added before the PCR amplification step to ensure strand specificity in the experimental group. Sequencing libraries of the experimental group and control were checked on a StepOne Plus Real-time PCR system (ABI) prior to sequencing on the NovaSeq 6000 platform (Illumina, San Diego, CA, USA). R-loop levels were detected using DRIP-qPCR as described previously (Xu et al., 2017; Sanz and Chedin, 2019).

R-loop data were pre-processed as described previously (Xu et al., 2017) with minor modifications. Reads were aligned to P. trichocarpa genome v3.0 with bowtie2 (v2.2.1) using --local --phred33. Duplicated reads were removed with picard 2.21.6 (http://picard.sourceforge.net/). MACS2 (2.2.6) was used to identify peaks (Zhang, 2008). The aligned reads files were converted to normalized coverage files (bigWig) via Reads Per Genomic Content of 5 bp bins (Ramirez et al., 2014). R-loop peaks were visualized using IGV (Thorvaldsdóttir et al., 2013).