Using Gene Expression to Study Specialized Metabolism-A Practical Guide

Abstract

Plants produce a vast array of chemical compounds that we use as medicines and flavors, but these compounds' biosynthetic pathways are still poorly understood. This paucity precludes us from modifying, improving, and mass-producing these specialized metabolites in suitable bioreactors. Many of the specialized metabolites are expressed in a narrow range of organs, tissues, and cell types, suggesting a tight regulation of the responsible biosynthetic pathways. Fortunately, with unprecedented ease of generating gene expression data and with >200,000 publicly available RNA sequencing samples, we are now able to study the expression of genes from hundreds of plant species. This review demonstrates how gene expression can elucidate the biosynthetic pathways by mining organ-specific genes, gene expression clusters, and applying various types of co-expression analyses. To empower biologists to perform these analyses, we showcase these analyses using recently published, user-friendly tools. Finally, we analyze the performance of co-expression networks and show that they are a valuable addition to elucidating multiple the biosynthetic pathways of specialized metabolism.

Keywords: clustering; co-expression; enrichment; metabolism; online; transcriptomics.

FIGURE 1

Expression profiles of AT1G69500 (CYP704B1), a cytochrome P450 involved in pollen exine formation. (A) User interface of the CoNekT tool used to identify organ-specific genes. (B) The plot shows the expression of CYP704B1 in Arabidopsis. The various organs and tissues are shown on the x-axis, while the y-axis indicates expression levels as Transcripts Per Million (TPM). The gray points indicate the minimum and maximum expression. (C) Expression profiles of CYP704B1 and its orthologs in seed plants. Green and red color indicate low and high expression, respectively, while black cells indicate missing gene expression data. The figure contains expression profiles of genes from Arabidopsis (AT) and Amborella (AMTR). The CoNekT platform groups various tissues (e.g., petals, anthers, and pistils) from an organ (e.g., flower) into one category. Each cell contains the average expression values of samples from the organ. For brevity, only genes from Amborella and Arabidopsis are shown.

FIGURE 2

Analysis of lignin biosynthesis with expression clustering and co-expression network approaches. (A) Biosynthetic pathway of lignin. (B) Hierarchical clustering of PALs, C4H, and CAD genes. The red and blue colors indicate high and low expression in a given organ, respectively. (C) Co-expression network of PAL1. Nodes represent genes, gray edges connect co-expressed genes, while node colors indicate orthogroups of the gene families. The red square in (B,C) indicate genes known to be involved in lignin biosynthesis.

FIGURE 3

Co-expression network of Arabidopsis laccases and MYB transcription factors. Nodes represent genes, gray edges connect co-expressed genes, while node colors indicate MYBs (yellow) or laccases (purple). The red square indicates MYBs and laccases implicated in lignin biosynthesis. For brevity, only genes that are connected to at least one other gene are shown.

FIGURE 4

Comparison of the duplicated clusters involved in suberin and cutin biosynthesis. (A) Co-expression networks of module 26 (blue, left) and 206 (green, right) from Arabidopsis thaliana. The annotation of the colored shapes of the gene families (given as orthogroups OG) is shown below. (B) Expression profiles of At4g39480 from cluster 29 and At5g58860 from cluster 206. Green and red colors indicate low and high expression, respectively. The expression values are scaled by dividing each row by the maximum expression found in the row. The tool used is available at the “Tools/Generate heatmap/Comparative” on CoNekT’s homepage.

FIGURE 5

Performance of co-expression networks in predicting correct enzymes in specialized metabolism. The rows contain different SM classes, as defined by MapMan, while the columns contain four plants: maize (Zea mays, orange box), tomato (Solanum lycopersicum, red box), rice (Oryza sativa, gray box), and Arabidopsis thaliana (green box). For each species, we calculate the performance for three networks, based on: Higest Reciprocal Rank (HRR), Mutual Rank (MR), and Pearsson Correlation Coefficient (PCC). The shade of the cells and the cell numbers correspond to F1 score (harmonic mean of precision and recall), which ranges from 0 (poor performance of prediction or too few genes associated to a specific pathway to perform a prediction) to 1 (perfect performance).