Targeted RNA-seq improves efficiency, resolution, and accuracy of allele specific expression for human term placentas

Abstract

Genomic imprinting is an epigenetic mechanism that results in allele-specific expression (ASE) based on the parent of origin. It is known to play a role in the prenatal and postnatal allocation of maternal resources in mammals. ASE detected by whole transcriptome RNA-seq (wht-RNAseq) has been widely used to analyze imprinted genes using reciprocal crosses in mice to generate large numbers of informative SNPs. Studies in humans are more challenging due to the paucity of SNPs and the poor preservation of RNA in term placentas and other tissues. Targeted RNA-seq (tar-RNAseq) can potentially mitigate these challenges by focusing sequencing resources on the regions of interest in the transcriptome. Here, we compared tar-RNAseq and wht-RNAseq in a study of ASE in known imprinted genes in placental tissue collected from a healthy human cohort in Mali, West Africa. As expected, tar-RNAseq substantially improved the coverage of SNPs. Compared to wht-RNAseq, tar-RNAseq produced on average four times more SNPs in twice as many genes per sample and read depth at the SNPs increased fourfold. In previous research on humans, discordant ASE values for SNPs of the same gene have limited the ability to accurately quantify ASE. We show that tar-RNAseq reduces this limitation as it unexpectedly increased the concordance of ASE between SNPs of the same gene, even in cases of degraded RNA. Studies aimed at discovering associations between individual variation in ASE and phenotypes in mammals and flowering plants will benefit from the improved power and accuracy of tar-RNAseq.

Keywords: allele-specific expression; genomic imprinting; human placenta; quantification of ASE; targeted RNA-seq.

Figure 1

Distribution of parent of origin expression and ASE in PEGs, MEGs, and CEGs. The box plots show the distributions of (A) Pat-Freq and (B) ASE at the SNPs in PEGs, MEGs, or CEGs. The mean of each distribution is indicated by a diamond. Orange and green colors denote wht-RNAseq and tar-RNAseq, respectively.

Figure 2

Correlation of gene-level average Pat-Freq between wht-RNAseq and tar-RNAseq. The scatter plot shows the Pat-Freq in each gene that had data in a sample sequenced with both wht-RNAseq and tar-RNAseq. PEGs, MEGs, and CEGs are denoted by blue, pink, and gray, respectively. The linear regression is shown by a dashed line.

Figure 3

Comparison of SNP coverage between wht-RNAseq and tar-RNAseq. The violin plots show the distributions of (A) the counts of SNPs per sample, (B) the counts of SNPs per gene, (C) the counts of genes with at least one SNP, and (D) the read counts at SNPs. Orange and green colors denote wht-RNAseq and tar-RNAseq, respectively.

Figure 4

Concordance of ASE for SNPs in the same gene. The Pearson correlation coefficients were calculated from pairwise combinations of the SNPs from the same genes and their distributions are shown in box plots, stratified by escalated depth filtering thresholds. The mean of each distribution is indicated by a diamond. Data from tar-RNAseq (with deduping), wht-RNAseq (with deduping), and wht-RNAseq (without deduping) are denoted by green, orange, and gray colors, respectively.

Figure 5

Combination of tar-RNAseq and RiboErase rescued most low RIN samples. The scatter plots show the fraction of SNPs covered by no fewer than 10 reads versus RIN. The ribodepleted tar-RNAseq, nonribodepleted tar-RNAseq, and ribodepleted wht-RNAseq samples are denoted by green circles, green triangles, and orange circles, respectively. The linear regression for tar-RNAseq and wht-RNAseq data points is shown by a green and an orange dashed line, respectively.