Analysis of long non-coding RNAs highlights tissue-specific expression patterns and epigenetic profiles in normal and psoriatic skin

Abstract

Background: Although analysis pipelines have been developed to use RNA-seq to identify long non-coding RNAs (lncRNAs), inference of their biological and pathological relevance remains a challenge. As a result, most transcriptome studies of autoimmune disease have only assessed protein-coding transcripts.

Results: We used RNA-seq data from 99 lesional psoriatic, 27 uninvolved psoriatic, and 90 normal skin biopsies, and applied computational approaches to identify and characterize expressed lncRNAs. We detect 2,942 previously annotated and 1,080 novel lncRNAs which are expected to be skin specific. Notably, over 40% of the novel lncRNAs are differentially expressed and the proportions of differentially expressed transcripts among protein-coding mRNAs and previously-annotated lncRNAs are lower in psoriasis lesions versus uninvolved or normal skin. We find that many lncRNAs, in particular those that are differentially expressed, are co-expressed with genes involved in immune related functions, and that novel lncRNAs are enriched for localization in the epidermal differentiation complex. We also identify distinct tissue-specific expression patterns and epigenetic profiles for novel lncRNAs, some of which are shown to be regulated by cytokine treatment in cultured human keratinocytes.

Conclusions: Together, our results implicate many lncRNAs in the immunopathogenesis of psoriasis, and our results provide a resource for lncRNA studies in other autoimmune diseases.

Figure 1

Overview of the analysis pipeline. We first performed Tophat alignment and identified uniquely mapped reads for each RNA-seq sample, we then assembled the transcripts using Cufflinks for each sample. We used a computational approach to nominate potential novel transcripts (Prensner JR et al., [13]) by comparing with Ensembl gene set. We removed those potential novel transcripts which are close (that is, <2 kb) to any exons from any annotated transcripts, inhabited in regions with lower mappability/alignability, or less than 200 bp in length. We quantified the gene expressions using read counts. We then normalized the values across the samples and performed differential expression analysis using DESeq. We inferred the properties and biological functions of the lncRNAs by comparing results with other RNA-seq experiments and using co-expression analysis.

Figure 2

Genomic map of lncRNAs expressed in skin tissues across the genome. The number of lncRNAs identified in this study (y-axis) per megabase across the genome (x-axis).

Figure 3

Expression behaviors for different gene categories. The mean gene expression in RPKM is shown in (a) and coefficient of variation in the normal skin samples for different transcript categories is shown in (b).

Figure 4

Tissue specificity analysis for different gene categories. (a) Heatmap showing the proportion of genes from each category expressed in different tissue types. (b) Tissue specificity (T _s) for different gene categories in skin when comparing with 16 other tissue types.

Figure 5

Relative distance to enhancers (a) and promoters (b) for different transcript classes. The means and error bars depict the relative distance (D _ecto /D _average) to the enhancer (a) and promoter (b) elements for genes in each category in these two ectodermally derived cell types (HMEC and NHEK). D _ecto is the closest distance to the enhancer (or promoter) in NHEK (or HMEC), and D _average is the average closest distance to the enhancer (or promoter) to the other cell types.