Fig 1.
Genome-wide identification of RS sites in humans using 4sUDRB-seq datasets.
(A) A computational pipeline for systematically identifying RS sites. The pipeline contains three steps (Materials and methods). Candidate RS sites result after the first two steps, and sawtooth RS sites result after the third step. (B) A table lists the numbers of putative, candidate, and sawtooth RS sites and RS introns identified in three human cell lines (left panel). Hundreds of candidate RS sites were identified in more than two cell lines (right panel). (C) A schematic to illustrate 4sUDRB-seq captures recursive splicing. 4sU-labeled, newly synthesized, total RNAs from PA1, H9, and H9-differentiated FB cells were collected at different time points (T1, T2, …, Tn) after the commence of synchronized Pol II elongation, and transcripts with or without recursive splicing were captured accordingly (top panel). 4sUDRB-seq was performed on PA1 cells at seven time points, but on H9 and FB cells at the first five time points (bottom panel). (D) Among the 16 sawtooth RS sites with the sawtooth pattern, six sites (38%; light purple) were previously validated, and the remaining 10 (62%; dark purple) are new.
Fig 2.
Validation of identified RS sites.
(A) The branch-point motif is enriched in the 20–50 bp window upstream of RS sites (purple line) and canonical 3′ splice sites (blue line), but not randomly sampled intronic AGGT sites (red line). (B) RS sites reconstitute strong 5′ splice sites. The 5′ splice sites reconstituted from RS sites are significantly stronger than randomly sampled intronic GT sites. The 5′ splice sites reconstituted from sawtooth RS sites are even stronger than the 5′ splice sites of cassette exons and constitutive exons. Median splice site strengths and Wilcoxon rank-sum test p-values are indicated. (C) We validated a novel RS site by RT-PCR. Top, 4sUDRB-seq signal tracks show the sawtooth pattern for an RS site in the first intron of the MAGI1 gene. Bottom, RT-PCR validation of the RS site. This recursive splicing junction was detected with recursive splicing specific primers (lane 1), but not with the primers that mapped to the exons (lane 2). PCR primers are indicated as red arrows in the figure and their sequences are provided in S6 Table.
Fig 3.
Sequence features of recursive splicing in humans.
(A) RS sites tend to reside in long introns. Introns with putative, candidate or sawtooth RS sites have increasing lengths (Wilcoxon rank-sum tests). (B) Most introns with putative, candidate or sawtooth RS sites correspond to the longest introns of their genes. (C) We predict that more than 50% of candidate RS sites are followed by an RS exon supported by at least two unique 4sUDRB-seq junction reads, a significantly higher percentage than the putative RS sites that are not candidate RS sites (Fisher’s exact test). (D) The Percent-Spliced-In (PSI) values of RS exons are significantly lower than those of cassette exons according to steady-state RNA-seq data (Materials and methods; Wilcoxon rank-sum test). (E) Reconstituted 5′ splice sites are significantly stronger than the 5′ splice sites of RS exons (Wilcoxon signed-rank test)—67% of reconstituted 5′ splice sites are stronger than the 5′ splice sites of their corresponding RS exons. Both of these two groups of 5′ splice sites are significantly stronger than randomly chosen intronic sites but weaker than cassette exons with the AGGT motif (Wilcoxon rank-sum test). (F) Left: on average, the exons upstream of recursive 5′ splice sites (purple) have higher GC% than their downstream introns, like canonical exon-intron boundaries (blue), while there is no apparent GC% difference around the exon-intron boundaries of RS exons (orange). Right: reconstituted 5′ splice sites have significantly larger GC% difference than the 5′ splice sites of RS exons (medians are shown along with Wilcoxon signed-rank test p-values).
Fig 4.
Recursive splicing occurs with a delay after transcription.
(A) Percent Recursive Splicing (PRS) is defined as the percentage of junction reads that support the intermediate product of recursive splicing. (B) A schematic diagram to show the definition of co-transcriptional and post-transcriptional recursive splicing. For co-transcriptional recursive splicing, the recursive splicing finished before the entire gene was transcribed (top panel), while in post-transcriptional recursive splicing, recursive splicing started after the gene was transcribed (bottom panel). (C) The landscape of recursive splicing during transcription in PA1 cells. (Left) Each row is an RS event. All RS events are clustered into four classes according to their onset time for recursive splicing and sorted within each class by PRS. (Right) The table lists the counts and percentages of RS intron with post-transcriptional or co-transcriptional recursive splicing. For some RS events, we could not determine whether they were post- or co-transcriptional due to the lack of time resolution. (D) PRS dropped progressively after the onset of recursive splicing. Class II (left panel) and III (right panel) introns are shown and the median PRS time course is shown as purple lines. The boxplots show correlation coefficients between the PRS time course of individual RS events and the median time course.
Fig 5.
Recursive splicing is not constitutive and is cell-type specific.
(A) Pairwise comparison of the occurrence of RS junction reads (R1) and canonical junction reads (R2) in RS introns. RS introns were classified into three group according to the appearance orders of R1 and R2. (B) Identification of full-length intron lariats (Materials and methods) in RS introns. (C) An example in the human GLI3 gene shows that recursive splicing is cell-type specific. Junction reads of recursive splicing (purple) or canonical splicing (black) are depicted as dashed straight lines, and lariats reads are depicted as dashed curved lines.