Coordinate Changes In RNA Expression, Transcript Splicing, and DNA Methylation In Pediatric AML

Given reported associations between loss of PASG function and abnormalities of genomic methylation, we tested 48 AML specimens at the extremes of the PASGΔ75 distribution for total 5-methylcytosine (5-mC) content by liquid chromatography/tandem mass spectrometry. The mean total methylation was significantly lower in the high compared to the low PASGΔ75 groups (mean 5-mC 3.95% vs. 4.22% of cytosine, p=0.015 Mann-Whitney). To identify specific regions of altered methylation, we used high-throughput sequencing of DNA enriched by pull-down with the methyl binding domain fragment of MBD2 (MBD-Seq). Comparison of summed methylation signal across regions flanking RefSeq transcriptional start sites (TSS) showed the expected decrease in methylation just upstream of the TSS in both groups. However, methylation more distal to the TSS was proportionally lower in PASGΔ75 high than PASGΔ75 low samples (Fig 1a). To evaluate methylation at CpG islands (CGI), UCSC CGI were scaled to 500 bp and MBD-Seq data were summed across 20Kb flanking CGI. While both groups showed the anticipated increase in methylation signal on CGI, methylation in the shore regions immediately flanking CGI was proportionally lower in high PASGΔ75 compared to low PASGΔ75 samples (Fig 1b), further suggesting epigenetic differences between these sample groups.

Because we were unable to identify sequence variants in PASG intron 18 or flanking exons that might explain alternative splicing, we asked whether expression of PASGΔ75 was associated with global changes in transcript splicing. We evaluated gene expression patterns on the Affymetrix HuGene array, with assessment of alternative splicing using Partek software alternative splicing (altsplice) algorithm for quartile-grouped samples. In contrast to comparison of adjacent quartile groups, which showed modest changes in expression and relatively few transcripts with significant altsplice scores, comparison of the highest and lowest quartile samples showed marked changes in gene expression and a large number of alternatively spiced transcripts as assessed by significance of the altsplice score (Fig 2). In addition to splicing changes, these analyses suggested marked differences in gene expression patterns of AML specimens grouped by PASGΔ75 quartile, with clear separation of Q1 and Q4 samples by principal components analysis. Using a conservative Wilcoxon gene sets test and limiting ourselves to small, curated Biocarta pathways, we found expression patterns associated with high deletion variant expression strongly linked to overlapping pathways involving DNA repair, replication, and cell cycle progression.