Sequencing of complete, chromatin-associated pre-mRNA during inflammatory stimulus reveals differential splicing dynamics among introns of IKBα.

(A) Histogram of reads corresponding to the TNF-induced expression and splicing of IKBα pre-mRNA of BMDMs. RNA-Seq was performed on chromatin associated RNA, enriched for NFkB genes as a function of a TNF stimulation timecourse, time shown in minutes after stimulation. Reads are histogrammed in log10 scale and normalized to each time point’s maximum value. (B) The Coefficient of Splicing (CoSI) metric quantifies extent of splicing as a function of time, expressed as a ratio of reads from each splice junction to total junctional reads. Dynamics of IKBα splicing as a function of each intron’s CoSI is shown (C), where 1=spliced and 0=unspliced, with corresponding introns highlighted in sample timepoint. (D) Differential dynamics of splicing for each Nfikbia intron are further demonstrated in the coverage plot for the transcript.

Heterogeneity of splicing at each intron reveals splicing ‘bottlenecks’.

The Co-SI of each intron per timepoint is shown as a function of the entire inflammatory mRNA dataset as box-whisker plot (A). Each point represents an intron of one of 230 genes, revealing high rates of splicing (median CoSI indicated by bar near 1.0 for each timepoint) for most genes with significant outliers. As an example, Cxcl10 intron 2 (red arrowhead) is represented by the datapoint with arrowhead, and a histogram of reads is shown to demonstrate relative unspliced nature of this intron, which is not involved in alternative splicing. (B) Several similar introns that are relatively unspliced are found throughout the inflammatory transcriptome; shown are bottleneck introns within Cd40, Daxx, and Irf7 as examples in the context of their neighboring introns.

Splicing kinetics of inflammatory introns are heterogeneous, ranging from seconds to minutes.

CoSI of introns representing various splicing rates are measured and fit to half-lives. Cells were treated with Actinomycin D-treated, from which hybrid capture of genes of interest and sequencing was performed on total (unfractionated) RNA. Shown are four representative samples of splicing kinetics.

Bottleneck introns can be repaired, and account for significant alterations to gene expression.

(A) Intron-GFP splicing reporters for each wild-type intron (red) and modified intron (green) are shown as BFP:GFP ratio. (B) Ratio of WT:Fixed slopes is shown; whereas Tfec expression is not altered by improved 5’ sequence, Malt1 intron sequence is significantly impaired owing to its 5’ donor sequence, exhibiting a roughly 5-fold impairment in gene expression due to the 5’ splice donor.

Fine-tuning improves gene expression prediction in macrophages.

(A) Training and test loss curves over 20 epochs of model fine-tuning. (B-C) Predicted versus measured RNA expression values for 230 transcripts induced by TNFα in macrophages, shown for the pretrained model (B) and the fine-tuned model (C).

Putative regulatory sequences are enriched in slow-splicing introns.

(A) Scatterplot showing percent representation of position weight matrices (PWMs) scanned across slow-splicing introns versus all introns genome-wide using FIMO. Sequence logos of the top 5 enriched PWMs are shown to the right. (B–D) Attribution plots highlighting GA-rich sequences (source seqlets for enriched PWMs) and their locations mapped to gene schematics for GA-rich motif (B), and A-rich motif (C). Corresponding RNA-seq tracks are shown for each gene.

Oligos used

Exons of interest for hybrid capture

Hybrid capture strategy for isolating chromatin-associated inflammatory transcripts.

RNA was purified from chromatin-associated bone marrow–derived macrophages (BMDMs), and cDNA was generated using oligo(dT) priming to enrich for polyadenylated transcripts. Biotinylated RNA oligonucleotides complementary to the terminal exons of inflammatory genes were hybridized to the cDNAs, allowing for selective enrichment of these transcripts via streptavidin bead capture.

Hybrid capture enriches NF-κB–responsive transcripts and yields robust intron coverage.

(A) Fraction of sequencing reads corresponding to NF-κB–responsive genes (blue) in hybrid-captured versus poly(A)-selected RNA (orange), showing substantial enrichment after hybrid capture. (B) Histogram of read counts per intron at the 6-min TNF induction time point. Reads were detected for 1,024 introns out of 1,508 targeted introns, with undetected introns largely corresponding to transcripts induced at later time points (>60 min). (C) Distribution of intron fold induction across time points reveals that many NF-κB target genes begin to show induction as early as 4 min post-TNF stimulation, with both the number of induced introns and magnitude of induction increasing markedly by 14 min. (D) Scatterplot comparing induction at 4 min versus 14 min for individual introns demonstrates that most early-induced genes are further upregulated at later time points.

Gene track visualization of Nfkbia induction dynamics.

Genome browser tracks show Nfkbia (IκBα) transcript induction over time in chromatin-associated, hybrid-captured RNA. Signal intensity is displayed on a linear scale (left) normalized to the maximum height at 20 min, and on a log scale (right) normalized to the maximum at each time point.

Splice site strength and expression kinetics among NF-κB–induced genes.

(A) Representative introns from selected genes were scored for 5′ splice donor strength based on similarity to the canonical ‘GTAAG’ motif; introns with weaker matches (e.g., Irf7 intron 3) received lower scores. (B) NF-κB–induced genes display well-characterized variability in expression kinetics (RNA-seq data from Reference 25). Heatmap shows temporal expression profiles of NF-κB target genes following lipid A stimulation, categorized into immediate-early, early, and later expression groups. (C) All introns in the NF-κB transcriptome were scored using MaxEntScan for both 5′ splice donor (top) and 3′ splice acceptor (bottom) sequences, stratified by their gene expression group.

Reporter assay design for assessing intron splicing efficiency.

(A) Schematic of the bidirectional reporter assay. Individual introns were cloned into a bidirectional promoter context together with their flanking exons, positioned upstream of a self-cleaving 2A peptide and GFP reporter. In the opposite transcriptional direction, a BFP reporter served as a transcriptional control. (B) Evolutionarily conserved weak 5′ splice donors were “repaired” to the canonical GTAAG sequence within this reporter construct, and splicing efficiency was quantified by flow cytometry (FACS).

Splicing completion across kinetic gene expression groups.

Box-and-whisker plots depict time-course CoSI values for introns from inflammatory gene cohorts. Each point represents the CoSI value of an individual intron at a given time point. Plots are grouped by expression category: Immediate Early (top left), Early (top right), Intermediate (bottom left), and Late (bottom right).

Relationships between intron length, GC content, transcript position, and splicing kinetics.

(A) Scatterplot of intron length (y-axis) versus GC content (%) (x-axis) for the 150 slowest-splicing introns. Pearson r = −0.44 (p = 2.76 × 10⁻⁴⁵) and Spearman r = −0.44 (p = 1.34 × 10⁻⁴⁶). (B) Histogram of intron length distributions (100-nt bins) for all 1,098 introns (red) and the 150 slowest-splicing introns (blue). Length distributions are largely similar between cohorts. (C) Splice completion (CoSI) trajectories across the time course for the 50 fastest (left, blue) and 50 slowest (right, orange) introns, ranked by area under the CoSI curve (AUC). (D) Same analysis as in (C) but extended to the top 150 fastest and bottom 150 slowest introns. (E) Minimum CoSI values for each intron plotted against transcript abundance (TPM, left), intron length (middle), and position along the transcript (5′ → 3′ %, right) for the bottom 150 and bottom 50 introns. Correlations: TPM — Bottom 150: Pearson r = 0.204 (p = 1.24 × 10⁻²), Spearman r = 0.203 (p = 1.27 × 10⁻²); Bottom 50: Pearson r = 0.276 (p = 5.25 × 10⁻²), Spearman r = 0.316 (p = 2.53 × 10⁻²). Length — Bottom 150: Pearson r = −0.242 (p = 2.88 × 10⁻³), Spearman r = −0.266 (p = 9.92 × 10⁻⁴); Bottom 50: Pearson r = −0.160 (p = 0.268), Spearman r = −0.171 (p = 0.235). Transcript position — Bottom 150: Pearson r = 0.260 (p = 1.33 × 10⁻³), Spearman r = 0.231 (p = 4.44 × 10⁻³); Bottom 50: Pearson r = 0.259 (p = 6.90 × 10⁻²), Spearman r = 0.318 (p = 2.43 × 10⁻²).

Enrichment of sequence motifs near slow-splicing introns.

Scatterplots show the percent representation of position weight matrices (PWMs) scanned across slow-splicing introns compared to all introns genome-wide using FIMO. Sequence logos of the top five enriched PWMs are displayed to the right of each scatterplot. (A) Scatterplot and enriched motifs identified in exons adjacent to the 50 slowest-splicing introns. (B) Scatterplot and enriched motifs identified within the 150 slowest-splicing introns. (C) Scatterplot and enriched motifs identified in exons adjacent to the 150 slowest-splicing introns.