Fig. 3: Characterization of the nuclear speckle transcriptome by APEX2-seq.

a illustration of APEX2-tagged hiPSCs and biotinylation reaction. b Addition of hydrogen peroxide leads to biotinylation of biomolecules near nuclear speckles (APEX2-SON hiPSCs, left) or in the nucleoplasm (APEX2-NLS hiPSCs, right). Two technical replicates were collected. c APEX2-seq experimental outline. APEX2 hiPSCs were treated with drugs prior to RNA biotinylation. 5 biological replicates were collected per condition. d Pseudocolor images (see methods) of smFISH signal for two gene targets found to differentially localize to nuclear speckles by APEX2 sequencing in the control condition. P-values were calculated using a one-sided Mann–Whitney U test. Box plots show median values and IQR, with whiskers indicating 1.5*IQR. APEX2-seq and smFISH show overall positive correlation (Pearson correlation coefficient = 0.71, p value = 4.98e-06, two sided). Shaded area indicates 95% confidence interval. e Genes enriched in nuclear speckles display a larger proportion of lncRNA and snoRNA/scaRNA/snRNA than nuclear speckle depleted and neither genes. f Network representation of GSEA. Terms associated with ribosome, mitochondria and electron transport chain and the spliceosome are enriched in nuclear speckles. g Overview of feature categories used to build UMAP and train MLR model. Number of features in each category are indicated in brackets. h UMAP of transcript feature space. Nuclear speckle depleted/enriched transcripts occupy distinct regions. Gene sets found to be nuclear speckle enriched by GSEA show similar distributions. i Features from (g) were used to train a MLR model to predict nuclear speckle enrichment of protein-coding genes. Scatterplot shows random samples of 600 transcripts on test sets aggregated over 10-fold CV. j Combinatorial training of MLR model. The mean R² score of ten models is shown (error bars indicate standard deviation, points indicate the score of individual models). k Importance analysis of features used to predict nuclear speckle enrichment. Color indicates Pearson correlation with nuclear speckle enrichment. l TPM and nuclear speckle enrichment of protein-coding transcripts show a negative correlation (Pearson correlation coefficient = -0.38, p value < 2.225e-308, two-sided). m Comparison of APEX2-seq nuclear speckle enrichment and enrichment in different subcellular compartments based on fractionation sequencing. Scale bar corresponds to 10 µm.