Transcriptome classification reveals molecular subtypes in psoriasis

Abstract

Background: Psoriasis is an immune-mediated disease characterised by chronically elevated pro-inflammatory cytokine levels, leading to aberrant keratinocyte proliferation and differentiation. Although certain clinical phenotypes, such as plaque psoriasis, are well defined, it is currently unclear whether there are molecular subtypes that might impact on prognosis or treatment outcomes.

Results: We present a pipeline for patient stratification through a comprehensive analysis of gene expression in paired lesional and non-lesional psoriatic tissue samples, compared with controls, to establish differences in RNA expression patterns across all tissue types. Ensembles of decision tree predictors were employed to cluster psoriatic samples on the basis of gene expression patterns and reveal gene expression signatures that best discriminate molecular disease subtypes. This multi-stage procedure was applied to several published psoriasis studies and a comparison of gene expression patterns across datasets was performed.

Conclusion: Overall, classification of psoriasis gene expression patterns revealed distinct molecular sub-groups within the clinical phenotype of plaque psoriasis. Enrichment for TGFb and ErbB signaling pathways, noted in one of the two psoriasis subgroups, suggested that this group may be more amenable to therapies targeting these pathways. Our study highlights the potential biological relevance of using ensemble decision tree predictors to determine molecular disease subtypes, in what may initially appear to be a homogenous clinical group. The R code used in this paper is available upon request.

Figure 1

Pipeline for patient stratification. For further details on related methodology, please see main text.

Figure 2

Differential gene expression in lesional (PP), non-lesional (PN) and normal (NN) skin tissue. Gene expression was analysed to reveal probe sets that were differentially expressed between pairwise comparisons of PP, PN and NN tissue groups. (A) The Venn diagram shows the number of probe sets identified in each of the three differential analyses performed. Probe sets common to all three pairwise comparisons were 228 (206 genes). (B) Microarray analysis of 108 skin tissue samples (in columns) for 206 genes (in rows) common to all tissue types, identified through differential expression analysis. Tissues have been grouped according to disease phenotype (normal NN, non-lesional PN and lesional PP) and heatmap colours indicate z-score of each gene expression value against the mean of corresponding normal values, (green: decreased expression, red: increased expression, inset). Similarity of gene expression vectors across all samples is represented by the dendrogram on the left. (C) Principal Component Analysis to suggest sample clustering across skin types according to gene expression patterns. Good separation of inflamed (PP) and non-inflamed (PN, NN) tissues was observed, indicating a progression from normal (red) to lesional skin (blue) through the non-lesional cases (green).

Figure 3

Informative genes for the classification of skin samples in lesional and non-lesional classes (PP and PN, respectively). Gini Index (GI) was used to generate a variable importance measure and provide an estimate of feature (gene) relevance to disease state. The five most important genes in determining disease classes were IL1F7, C7orf59, AQP9, BTC and TUFT1.

Figure 4

A multidimensional scaling (MDS) plot to illustrate the molecular grouping of samples. A dissimilarity matrix in random forest is constructed through the use of synthetic data drawn from the distribution of psoriatic samples (see Methods). Patients are clustered according to these dissimilarities and two distinct psoriatic groups are identified, PP01 (green) and PP02 (purple). All lesional samples (PP) cluster away from both normal (NN) and non-lesional (PN) tissue samples, in accordance to observations in figure 2c.

Figure 5

Genes identified as most informative through RF classification of skin tissues in four molecular groups (NN, PN, PP01 and PP02). Gini Index (GI) was used as variable importance measure for estimating the discriminative power of relevant features (genes) and, consequently, their relevance to disease state. The five most important genes in determining disease classes were BTC, CNFN, C20orf11, BUB3 and IL1F7. Genes with related annotation are listed in Table 1.

Figure 6

Graphical representation to illustrate the relation between 43 highly discriminative genes and disease sub-groups. Contributions shown according to Gini Index, calculated from random forest classification. The four skin-types (PP01: light blue, NN: green, PP02: blue, PN: light green) followed by relevant genes are arranged clockwise. Skin groups and genes are ordered according to shared pairing links.

Figure 7

Text mining results for validation process according to the literature. Co-occurrence of gene names with disease-related terms, such as”psoriasis”, “NK cells”, “T cells”, “immune response”, “Wnt signaling pathway”, “Notch signalling pathway”, “TGF – beta signaling pathway” and “ErbB signaling pathway” was searched in Pubmed abstracts through PubMatrix.