Unperturbed expression bias of imprinted genes in schizophrenia

Abstract

How gene expression correlates with schizophrenia across individuals is beginning to be examined through analyses of RNA-seq from postmortem brains of individuals with disease and control brains. Here we focus on variation in allele-specific expression, following up on the CommonMind Consortium (CMC) RNA-seq experiments of nearly 600 human dorsolateral prefrontal cortex (DLPFC) samples. Analyzing the extent of allelic expression bias-a hallmark of imprinting-we find that the number of imprinted human genes is consistent with lower estimates (≈0.5% of all genes), and thus contradicts much higher estimates. Moreover, the handful of putatively imprinted genes are all in close genomic proximity to known imprinted genes. Joint analysis of the imprinted genes across hundreds of individuals allowed us to establish how allelic bias depends on various factors. We find that age and genetic ancestry have gene-specific, differential effect on allelic bias. In contrast, allelic bias appears to be independent of schizophrenia.

Fig. 1

Quantifying allelic bias of expression in human individuals using the RNA-seq read count ratio statistic S_ig. The strength of bias towards the expression of the maternal (red) or paternal (blue) allele of a given gene g in individual i is gauged with the count of RNA-seq reads carrying the reference allele (small closed circles), and the count of reads carrying an alternative allele (open squares) at all SNPs for which the individual is heterozygous. The allele with the higher read count tends to match the allele with the higher expression, but measurement errors may occasionally revert this tendency as seen for SNP s₃ in gene g₁ in individual i₂

Fig. 2

Across-individuals distribution of read count ratio S for each gene indicates substantial variation of allelic bias and that <1% of all genes are imprinted. The vertically arranged four main panels present the empirical distribution of S_g across all individuals given each gene g. The upper two panels are distinct representations (survival plot: 1 – ECDF, and “survival heatmap”) of the same three distributions corresponding to a: PEG10, b: ZNF331, and c: AFAP1. PEG10 and ZNF331, previously found to be imprinted in mice or in other human tissues, and one for AFAP1, a gene without prior evidence. The bottom two survival heatmaps present the distribution of S_g for the top 2% and 100% of the 5307 analyzed genes. These are ranked according to gene score defined as 1 – ECDF (0.9) in the bottom far right panels. The score of PEG10, ZNF331, and AFAP1 is marked by a,b,c, respectively, in green circles. As expected, PEG10 and ZNF331 both score high and rank within the top 30 of all genes suggesting they are also imprinted in the present context, the adult human DLFPC. The bottom panels also indicate that <1% of all genes might be imprinted

Fig. 3

The top 50 genes ranked by the gene score defined, for gene g, as 1 – F_g(0.9), where F_g is the empirical cumulative distribution function (ECDF) for g. Thus 1 – F_g(0.9), is the fraction of individuals i for which S_ig > 0.9. Note that the same ranking and score is shown in the bottom half of Fig. 2. The tan colored bars indicate the fraction of individuals with nearly unbiased expression (Eq. 2). Gene names (y axis) are colored according to prior imprinting status: known imprinted (blue), nearby candidate (green), and distant candidate (red). “X” characters next to gene names indicate mapping bias and/or cis-eQTL effects based on the reference/nonreference allele test (“Methods”), while “0” indicates that total allele count was insufficient for this test

Fig. 4

Schizophrenia does not affect allelic bias of imprinted genes. Distribution of read count ratio for control, schizophrenic (SCZ), and affectic spectrum (AFF) individuals within each gene that has been considered as imprinted in the DLPFC brain area in this study

Fig. 5

Allelic bias depends differentially on age across imprinted genes. The panels and colors are consistent with the imprinted genes and psychiatric diagnoses presented in Fig. 4. The differential dependence on age is apparent when comparing PEG3 or ZNF331 (negative dependence) to KCNK9 or RP13-487P22.1 (positive dependence) or to NDN or NLRP2 (no dependence)