TissueEnrich: Tissue-specific gene enrichment analysis

Abstract

Summary

RNA-Seq data analysis results in lists of genes that may have a similar function, based on differential gene expression analysis or co-expression network analysis. While tools have been developed to identify biological processes that are enriched in the genes sets, there remains a need for tools that identify enrichment of tissue-specific genes. Therefore, we developed TissueEnrich, a tool that calculates tissue-specific gene enrichment in an input gene set. We demonstrated that TissueEnrich can assign tissue identities to single cell clusters and differentiated embryonic stem cells.

Availability and implementation

The TissueEnrich web application is freely available at http://tissueenrich.gdcb.iastate.edu/. The R package is available through Bioconductor at https://bioconductor.org/packages/TissueEnrich. Both the web application and R package are for non-profit academic use under the MIT license.

Supplementary information

Supplementary data are available at Bioinformatics online.

1 Introduction

The development of RNA-Seq technology has enabled large-scale comparison of gene expression in a multitude of developmental stages, cell-types and conditions. RNA-Seq data analysis identifies genes that likely have a shared function, either through differential gene expression analysis or co-expression network analysis. Gene ontology (GO) enrichment analysis is widely used to assign function to the gene sets and to gain insights into the biological processes they are involved in. While GO analysis identifies enriched processes in sets of genes, it does not determine enrichment of tissue-specific genes. Understanding which groups of genes are tissue-specific is valuable, as tissue-specific genes are more likely to be associated with human disease (Winter et al., 2004). To this end, tools have been developed that include tissue enrichment or tissue-specific enrichment calculations (http://genetics.wustl.edu/jdlab/tsea/; Komljenovic et al., 2016; Watanabe et al., 2017; Xu et al., 2014). However, these tools are limited in the flexibility of parameters for defining tissue-specificity, or the ability to allow users to define their own tissue-specific gene sets to use for enrichment analysis (see Supplementary Material for more details). Therefore, we developed ‘TissueEnrich’, a tool to carry out tissue-specific gene enrichment. TissueEnrich is available as an interactive web application, allowing the user to visualize tissue-specific gene enrichment and visualize expression of genes from the input set that were determined to be tissue-specific. We also developed an R package that further allows users to define tissue-specific genes in custom datasets, which they can then use for tissue-specific enrichment analysis.

2 Materials and methods

2.1 RNA-Seq datasets

To calculate tissue-specificity of genes, we used RNA-Seq data from GTEx (Ardlie et al., 2015), HPA (Uhlén et al., 2015) (https://v18.proteinatlas.org/about/download) and the mouse ENCODE project (Shen et al., 2012). We used data from tissues with >1 biological replicate. Processed GTEx data were downloaded from the Expression Atlas (E-MTAB-5214; Petryszak et al., 2016), and the samples from sub-tissues were grouped and considered to be biological replicates. The one-to-one orthologous genes between human and mouse were downloaded from Ensemble (Version 91; Zerbino et al., 2018). TissueEnrich only uses protein-coding genes for analysis. Tissue-specific metrics and group definitions are provided in the Supplementary Material.

2.2 Tissue-specific gene enrichment

We used the hypergeometric test to calculate the enrichment of tissue-specific genes in the input gene set (see Supplementary Material).

3 Results

3.1 Web application

We used R Shiny (Winston et al., 2018) to develop a user-friendly web application to calculate tissue-specific gene enrichment in a user-provided gene set. From the ‘Tissue Enrichment’ tab, the user can select the organism their data is from and the RNA-Seq dataset to use for tissue-specificity information. The output is an interactive bar chart, depicting the significance of tissue-specific gene enrichment, plotted as –Log₁₀(P-adjusted) on the y-axis, across each tissue. The user can view the expression values of the genes that were part of a tissue-specific group in an interactive heatmap by clicking the corresponding bar for that tissue in the bar chart. Furthermore, users can search for the expression values of individual genes along with their tissue-specific groups under the ‘Tissue Specific Genes’ tab. The ‘Help’ tab provides detailed usage instructions.

3.2 R package

The TissueEnrich R package contains similar functions to those in the web application. The package includes the ‘teGeneRetrieval’ function, which can be used to define tissue-specific genes in any given dataset. In this function, users can adjust thresholds for calculating tissue-specificity. The resulting tissue-specific genes can be used to carry out tissue-specific gene enrichment, using the ‘teEnrichmentCustom’ function.

3.3 Case study 1: Defining tissue-specificity of genes expressed in differentiated embryonic stem cells

We used two RNA-Seq datasets generated from differentiated embryonic stem cells (ESCs) to test TissueEnrich. The first dataset is from ESCs differentiated into cardiomyocytes (Szabo et al., 2015). We ran TissueEnrich on the 1000 most highly expressed genes, processed as previously described in Jain et al. (2017) and found enrichment for heart-specific genes using both the HPA and mouse ENCODE datasets (Supplementary Fig. S1 and Supplementary Data S1). These results validate the robustness of TissueEnrich and highlight heart-specific genes that may be interesting for follow up experiments. The second RNA-Seq dataset is from ESCs differentiated into trophoblast-like cells (Yabe et al., 2016). While Yabe et al. concluded that the differentiated cells are of trophoblast origin, the origin has been debated (Roberts et al., 2014). We previously used an approach similar to what we have now packaged into TissueEnrich to determine that the 1000 most highly expressed genes in these cells have strong enrichment for placenta-specific genes defined through the HPA (Jain et al., 2017) (Supplementary Data S2). Here, we used TissueEnrich for the same genes and found that they also show enrichment for placenta-specific genes defined through the mouse ENCODE project (Supplementary Fig. S2), indicating trophoblast-like cells may have a conserved function in mouse.

3.4 Case study 2: Annotation of cell clusters from single-cell RNA-Seq data

Next, we used single-cell RNA-Seq data from mouse embryos at e6.5 (Scialdone et al., 2016). Scialdone et al. clustered the cells using gene expression data and assigned tissue identities to the cell clusters based on marker genes that were upregulated in each cluster compared to all other clusters. Here, we ran TissueEnrich for each set of genes that were specifically up-regulated in cell clusters associated with e6.5 to determine if more information could be obtained about the cell identities. We found that cluster 1 showed strongest tissue-specific enrichment for intestine, liver and kidney (tissues derived from visceral endoderm); cluster 2 for placenta (tissue derived from extraembryonic ectoderm); and cluster 3 for brain and olfactory bulb tissue (Supplementary Fig. S3 and Supplementary Data S3). These results are in agreement with the tissue identities assigned in Scialdone et al., and provide additional detail on tissue-specific gene expression that may further help understand cell specification during early development.

4 Conclusions

We developed the TissueEnrich web application and R package, and demonstrated that it provides an unbiased way to carry out tissue-specific gene enrichment in mouse and human RNA-Seq data.