RAX2: a genome-wide detection method of condition-associated transcription variation

Abstract

Most mammalian genes have mRNA variants due to alternative promoter usage, alternative splicing, and alternative cleavage and polyadenylation. Expression of alternative RNA isoforms has been found to be associated with tumorigenesis, proliferation and differentiation. Detection of condition-associated transcription variation requires association methods. Traditional association methods such as Pearson chi-square test and Fisher Exact test are single test methods and do not work on count data with replicates. Although the Cochran Mantel Haenszel (CMH) approach can handle replicated count data, our simulations showed that multiple CMH tests still had very low power. To identify condition-associated variation of transcription, we here proposed a ranking analysis of chi-squares (RAX2) for large-scale association analysis. RAX2 is a nonparametric method and has accurate and conservative estimation of FDR profile. Simulations demonstrated that RAX2 performs well in finding condition-associated transcription variants. We applied RAX2 to primary T-cell transcriptomic data and identified 1610 (16.3%) tags associated in transcription with immune stimulation at FDR < 0.05. Most of these tags also had differential expression. Analysis of two and three tags within genes revealed that under immune stimulation short RNA isoforms were preferably used.

Figure 1.

Demonstration for association of usage of alternative splice and poly(A) sites with conditions: splicing switch and usage switch of poly(A) sites due to change in condition. Under normal condition, the gene transcription product is isoform 1 due to splicing event between exons a and b, but in stress or stimulation, for example, the gene product is isoform 2 by switching splicing from exon b to exon c (top two figures). This is called a splicing switch. Likewise, under normal condition, cells use distal poly(A) site and the transcription produces mRNA isoform I but when cells are stressed or stimulated, poly(A) site usage switches from distal site to proximal site and produces isoform II (bottom two figures). This is called poly(A) site usage switch.

Figure 2.

Transcription unit structures and different 3′ UTR isoforms. A majority of mammalian transcription units are characterized by alternative cleavage and polyadenylation (ACP). The majority of these contain multiple cleavage and polyadenylation sites in their terminal exon (top), impacting untranslated region identity without changing the coding sequence. Transcription units may also be characterized by mutually exclusive terminal exon structure (middle) or composite terminal exon structure (bottom). In the latter two cases, ACP is coupled to mRNA splicing, and both the coding sequence and the untranslated region are impacted. Black box: exon, white box: untranslated region (UTR), red flag: poly(A) site and -//-: intron.

Figure 3.

A model for multiple poly(A) sites or tags within genes. mRNA of Hsp70.3 is a typical model for multiple poly(A) sites (green boxes) and microRNA sites (red boxes). It has a common start point (promoter ), four poly(A) sites (PA1, PA2, PA3 and PA4) in 3′ UTR coding for several mRNA isoforms that are different in their 3′ end. Alternative polyadenylation changes the length of the 3′ UTR and it also can change microRNA binding sites in 3′ UTR. While microRNAs tend to repress translation and promote degradation of the mRNAs they bind to. Since transcription products are one-to-one corresponding to poly(A) sites, we define the transcription products as tags.

Figure 4.

A demonstration plot of ranked treatment chi-square versus null chi-square. The expected linear plot (red line) is given by null hypotheses that the treatment chi-square is equal to null chi-square at each chi-square point. The observed linear plot is given by ranked observed treatment chi-squares versus ranked estimated null chi-squares. Given a threshold Δ, all treatment chi-squares with are declared to be significant or interested where and s = a +1,…, G. All null chi-squares with are defined as potential false positives with probability given in Equation (9) where and t = b+1,…, G.

Figure 5.

Scatter and linear plots of treatment chi-squares versus estimated null chi-squares. (A) Scatter plot of treatment chi-squares against estimated null chi-squares derived from primary experimental data. Over 80% of treatment chi-squares fall in the null chi-square distribution. (B) Linear plot of ranked treatment chi-squares against ranked estimated null chi-squares. The observed linear dots (blue line) were significantly deviated from the expected linear dots (red line) across the estimated null chi-square distribution.

Figure 6.

Heatmap of tags differentially expressed between two conditions detected by RAX2. The heatmap was made with the data of 1279 tags with fold change (average of tag expressions in stimulated cells/average of tag expressions in normal cells or average of tag expressions in normal cells/average of tag expression in stimulated cells)> 1.4 selected from 1610 associated tags detected by RAX2.

Figure 7.

Two-way scatter plot for four association patterns of transcription of two tags within genes between two cell states. A two-way scatter plot displays distributions of scatter dots in four phases: Phase II for forward switch, phase IV for backward switch, phase I for positive accordance and phase III for negative accordance. (A) Plot of proximal tags versus distal tags where coordinates x and y are differences between ratios of counts of tags in stimulation and those in rest state. The ratio = sum of transcription counts of a tag over all replicates in a cell state / sum of transcription counts of this tag over all replicates and all cell states. (B) Numbers of genes with four transcription patterns of proximal and distal tags. (C) Numbers of genes with proximal tags positively and negatively responses to stimulation and numbers of genes with distal tags positively and negatively response to stimulation

Figure 8.

Examples for association patterns of transcriptional representation of two tags within genes between two cell states. Gene products shown here contain two tags defined by usage of distal and proximal poly(A) sites. Association between relative expression of two tags within genes and the cell states, detected by RAX2, shows backward switch (A), forward switch (B), negative accordant changes (C) and positive accordant changes (D). y axis corresponds to ratio of count sum of a tag over all replicate libraries in a cell state (0 h or 48 h post stimulation) to the sum of the tag over all replicate libraries across all cell states.