Article Figures & Data
Figures
- Figure 1. Overview of experimental design.
Fibroblast cells were isolated and cultured from the adult F1 hybrid mice between C57BL/6J and SPRET/EiJ. Two replicates of RNAs collected at three different time points following transcriptional arrest were sequenced.
- Figure S1. Reproducibility of mRNA sequencing data.
(A) Scatterplot comparing the abundance of cellular mRNA (log2-transformed sum of both alleles) between two biological replicates at 0 h. Each dot represents one gene. (B) Scatterplot comparing the log2-transformed fold change of the two alleles between two biological replicates at 0 h. (C) Scatterplot comparing the abundance of cellular mRNA (log2-transformed sum of both alleles) between two biological replicates at 0.5 h. Each dot represents one gene. (D) Scatterplot comparing the log2-transformed fold change of the two alleles between two biological replicates at 0.5 h. (E) Scatterplot comparing the abundance of cellular mRNA (log2-transformed sum of both alleles) between two biological replicates at 1.5 h. Each dot represents one gene. (F) Scatterplot comparing the log2-transformed fold change of the two alleles between two biological replicates at 1.5 h.
- Figure S2. Filtering of SNP loci with potential allelic mapping and assignment biases.
(A) Scatterplot comparing log2-transformed fold change of gene expression between parental strains and the two alleles in mock F1 hybrid created by mixing parental strain sequencing reads (see the Materials and Methods section for details). (B) Scatterplot comparing log2-transformed fold change of the two alleles between using uniquely mapped reads only and those including also multiple mapped reads.
- Figure S3. FDR of Δλ estimation.
FDR (y-axis) was plotted against different Δλ threshold (x-axis) in identifying genes with significant ASD. See the Materials and Methods section for details.
- Figure 2. Identification of genes with significant ASD.
(A) Scatterplot showing the bootstrap means (x-axis) and standard deviations (y-axis) of estimated ASD. Dashed blue lines indicate the Benjamini–Hochberg–adjusted P-value of 0.05 and dashed purple lines indicate a minimum decay rate difference of 0.06. Out of 8,815 genes (black), 621 (red) exhibited significant ASD. (B) Bar plots showing the number of sequencing reads assigned to BL6 (red) or SPRET (blue) alleles (y-axis) at different SNP loci (x-axis) of three time points (0, 0.5, and 1.5 h). BL6 and SPRET allele degraded faster in Armc7 and Rbak genes, respectively. (C) Scatterplot comparing allelic decay rate difference (Δλ) estimated based on Illumina sequencing data (y-axis) to that based on PacBio sequencing (x-axis) for the 25 randomly selected genes. Δλ estimated based on the two technologies was significantly correlated
- Figure 3. Sequence features that were correlated with ASD.
(A) The cumulative distribution function (CDF) of SNP density (number of SNPs per kb) for genes with significant ASD (red) and without (control genes, blue). Compared with the control genes, the genes with significant ASD showed significantly higher SNP density (P-value ˂ 2.2 × 10–16, two-sided Kolmogorov–Smirnov test). (B) Box plots and scatterplots showing the distribution of miRNA-binding site number difference between the stable and unstable alleles for genes with significant ASD and controls. For controls, the difference centered around zero (P-value = 0.86, two-sided Mann–Whitney U test), whereas in ASD genes, unstable alleles tend to possess more miRNA target sites than the stable alleles (P-value = 1.0 × 10–4, two-sided Mann–Whitney U test). Only the genes with ≥10 miRNA-binding sites combining the two alleles together and ≥1 different sites between the two alleles were used. (C) Violin plots and scatterplots comparing the distribution of the absolute MFE difference (|ΔMFE|) between ASD genes and controls. The horizontal lines indicate the median. Compared with controls, ASD genes exhibited larger allelic differences (P-value = 4.4 × 10–3 two-sided Mann–Whitney U test).
- Figure S4. SNP density comparison.
The cumulative distribution function (CDF) of SNP density (number of SNPs per kb) in 5′ UTR (A), CDS (B), and 3′ UTR (C) for genes with significant ASD (red) and without (control genes, blue). Compared with the control genes, the genes with significant ASD always showed significantly higher SNP density.
- Figure S5. Selection of control genes with similar density of sequence variants.
The cumulative distribution function (CDF) of SNP density (number of SNPs per kb) in whole gene, 5′ UTR, CDS, and 3′ UTR for genes with significant ASD (red) and a group of selected control genes with similar density of sequence variants.
- Figure S6. Comparison of miRNA-binding sites using top 100 highly expressed miRNAs.
Box plots and scatterplots showing the distribution of top 100 highly expressed miRNA-binding site number difference between the stable and unstable alleles for genes with significant ASD and controls. For controls, the difference centered around zero, whereas in ASD genes, unstable alleles tend to possess more miRNA target sites than the stable alleles
- Figure S7. Box plots and scatterplots showing the distribution of miRNA-binding site number difference between the stable and unstable alleles for genes with significant ASD and controls estimated using miRanda.
To validate our findings for miRNA-binding sites using TargetScan (Fig 3B), a similar analysis was performed using a different miRNA-binding site prediction algorithm miRanda (version 3.3a), with parameters miranda mirna.fa target.fa -sc 180 -en 1 -scale 4, in which mirna.fa was downloaded from miRBase (http://www.mirbase.org/). For controls, the difference centered around zero (P-value = 0.77, two-sided Mann–Whitney U test), whereas in ASD genes, unstable alleles tend to possess more miRNA target sites than the stable alleles (P-value = 0.027, two-sided Mann–Whitney U test). Only the genes with ≥10 miRNA-binding sites combining the two alleles together and ≥1 different sites were used.
- Figure S8. Comparison of miRNA-binding sites in different regions.
Box plots and scatterplots showing the distribution of miRNA-binding site number difference between the stable and unstable alleles for genes with significant ASD and controls for top 50 highly expressed miRNAs in 5′ UTR, CDS, and 3′ UTR, respectively.
- Figure S9. RNA secondary structure comparison using different window sizes.
Violin plots and scatterplots comparing the distribution of the absolute MFE difference (|ΔMFE|) between ASD genes and controls using a 21-, 61-, 81-, and 101-nt region surrounding SNPs. Compared with controls, ASD genes always exhibited larger allelic differences.
- Figure S10. Violin plots and scatterplots comparing the distribution of the absolute MFE difference (|ΔMFE|) between ASD genes and controls using MFE calculated with RNAstructure.
To validate our findings for MFE using RNAfold (Fig 3C), a similar analysis was performed using a different MFE calculating algorithm RNAstructure (version 6.0.1, http://rna.urmc.rochester.edu/RNAstructure.html) with the following parameters Fold 41nt-window.fa output -MFE. Specifically, for each sequence variant, we calculated the MFE of a 41-nt RNA segments (20-nt flanking each variant) along the whole transcript for the two alleles separately, and then calculated their absolute difference. For each gene, we used the maximum |ΔMFE| among all the variants to represent the allelic difference in mRNA secondary structure. The horizontal lines indicate the median. Compared with controls, ASD genes exhibited larger allelic differences (P-value = 0.042, two-sided Mann–Whitney U test).
- Figure S11. RNA secondary structure comparison in different regions.
Violin plots and scatterplots comparing the distribution of the absolute MFE difference (|ΔMFE|) between ASD genes and controls using SNPs in 5′ UTR, CDS, and 3′ UTR. Compared with controls, ASD genes exhibited larger allelic differences using SNPs in CDS or 3′ UTR but not in 5′ UTR.
- Figure S12. AREScore comparison.
Box plots and scatterplots showing the distribution of the value (A) and the absolute value (B) of AREScore difference between the stable and unstable alleles for genes with significant ASD and controls. No significant difference was observed between the control and ASD gene group for both value and the absolute value of AREScore allelic difference.
- Figure S13. Codon adaption index comparison.
Box plots and scatterplots showing the distribution of codon adaptation index difference between the stable and unstable alleles for genes with significant ASD and controls. No significant difference was observed between ASD gene and control group.
- Figure 4. The role of ASD in the allelic difference of RNA abundances.
Scatterplot comparing each gene's allele-specific expression (log2-transformed fold change at y-axis) and decay (Δλ at x-axis). Dashed gray lines indicate twofold change for gene expression and 0.06 for decay rate difference, respectively (FDR < 0.05). Genes with significant allelic bias at only RNA abundance level, only decay level, and both levels were depicted in green, orange, and purple, respectively.
- Figure S14. Identification of genes with significant ASA.
(A) FDR (y-axis) was plotted against different Δλ threshold (x-axis) in identifying genes with significant ASA. See the Materials and Methods section for details. (B) Scatterplot showing the bootstrap means (x-axis) and standard deviations (y-axis) of estimated ASA. Dashed blue lines indicate the Benjamini–Hochberg–adjusted P-value of 0.05 and dashed black lines indicate twofold divergence of gene expression. Out of 8,815 genes (black), 1,241 (red) exhibited significant ASE.
- Figure S15. The role of ASD in the allelic difference of RNA abundances under different combinations of FDR thresholds.
Bar plots showing the percentage of ASD genes in those with significant ASA (blue bars) and the percentage of ASA genes in those with significant ASD (red bars) at different combinations of FDR thresholds (0.005, 0.0075, 0.01, 0.025, 0.05, 0.075, and 0.1 for ASD and ASA).
- Figure S16. RNA secondary structure comparison.
Violin plots and scatterplots comparing the distribution of the MFE difference (ΔMFE) between the stable and unstable alleles for genes with significant ASD and controls. No significant correlation (or anti-correlation) between the stability of RNA secondary structure and the rate of RNA decay was observed.
Supplementary Materials
