Comprehensive exon array data processing method for quantitative analysis of alternative spliced variants

Abstract

Alternative splicing of pre-mRNA generates protein diversity. Dysfunction of splicing machinery and expression of specific transcripts has been linked to cancer progression and drug response. Exon microarray technology enables genome-wide quantification of expression levels of the majority of exons and facilitates the discovery of alternative splicing events. Analysis of exon array data is more challenging than the analysis of gene expression data and there is a need for reliable quantification of exons and alternatively spliced variants. We introduce a novel, computationally efficient methodology, Multiple Exon Array Preprocessing (MEAP), for exon array data pre-processing, analysis and visualization. We compared MEAP with existing pre-processing methods, and validation of six exons and two alternatively spliced variants with qPCR corroborated MEAP expression estimates. Analysis of exon array data from head and neck squamous cell carcinoma (HNSCC) cell lines revealed several transcripts associated with 11q13 amplification, which is related with decreased survival and metastasis in HNSCC patients. Our results demonstrate that MEAP produces reliable expression values at exon, alternatively spliced variant and gene levels, which allows generating novel experimentally testable predictions.

Figure 1.

MEAP workflow. The workflow contains modules for data pre-processing, differential expression analysis, and visualization. Data pre-processing uses a novel background estimation model (PM-BayesBG) and is followed by collective quantile normalization and multi-dimensional expression summarization based on user defined data type (probeset, exon, spliced variant or gene). Differential analysis enables finding biologically interesting targets. MEAP also includes web-based visualization for alternatively spliced events. See http://csbi.ltdk.helsinki.fi/meap/example/MEAPvisual/MEAP_visual_homepage.html for more information.

Figure 2.

Boxplot for the ratios of raw intensity of each probe versus estimated background signal. We used human colon cancer exon array data from Affymetrix public resource and calculated signals of antigenomic and genomic background probes on global, PM-GCBG, MAT and PM-BayesBG background correction models. Ideally, the background intensities of these probes from a background model should be the same as their detected signals. MEAP has the lowest mean ratio according to the mean ratios labeled in the figure.

Figure 3.

Boxplot of the normalized −C_t values for randomly selected exons in HNSCC samples. Exons ENSE00000833461, ENSE00000855493, ENSE00001131505 and ENSE00001382781 are statistically significantly differentially expressed between 11q13+ and 11q13− groups (**P < 0.01; *P < 0.05). ○ represents an outlier.

Figure 4.

MEAP splice visualization plot for gene ORAOV1. Top ‘P-value’, ‘Fold Change’ and ‘Expression Curve’ sections display the t-test P-values (−), fold change (), expression intensities of each exon in a gene between two sample groups (Group1: 11q13+; Group2: 11q13−). Gaps may occur when there are no probes designed against a specific exon region. Sections `Expression' and `Transcript' give the exon expression profile and the expression of each splice variant of ORAOV1 in 11q13+/11q13−. G1/G2 in the exon expression profile corresponds to the fold change between 11q13+ and 11q13− samples, in which green represents downregulation and red represents upregulation. Expression values are on -scale.

Figure 5.

Relative expression level of spliced variants from qRT-PCR. NEO1-201 and NEO1-202 both have high relative expression values in the 11q13+ group compared to the 11q13− group. Meanwhile, the expression level of NEO1-201 is lower than NEO1-202 in both the 11q13+ and 11q13− groups. For ORAOV1, transcript ORAOV1-201 is overexpressed in the 11q13+ group. The SD of ΔC_t values in the 11q13+ group for transcript ORAOV1-202 is high and such a variant does not have significant expression changes between two groups. (**P < 0.01; *P < 0.05).