Plants produce diverse low-molecular-weight compounds via specialized metabolism. Discovery of the pathways underlying production of these metabolites is an important challenge for harnessing the huge chemical diversity and catalytic potential in the plant kingdom for human uses, but this effort is often encumbered by the necessity to initially identify compounds of interest or purify a catalyst involved in their synthesis. As an alternative approach, we have performed untargeted metabolite profiling and genome-wide association analysis on 440 natural accessions of Arabidopsis thaliana. This approach allowed us to establish genetic linkages between metabolites and genes. Investigation of one of the metabolitegene associations led to the identification of N-malonyl-D-alloisoleucine, and the discovery of a novel amino acid racemase involved in its biosynthesis. This finding provides, to our knowledge, the first functional characterization of a eukaryotic member of a large and widely conserved phenazine biosynthesis protein PhzF-like protein family. Unlike most of known eukaryotic amino acid racemases, the newly discovered enzyme does not require pyridoxal 5′-phosphate for its activity. This study thus identifies a new D-amino acid racemase gene family and advances our knowledge of plant Damino acid metabolism that is currently largely unexplored. It also demonstrates that exploitation of natural metabolic variation by integrating metabolomics with genome-wide association is a powerful approach for functional genomics study of specialized metabolism.D-amino acid | racemase | genome-wide association | secondary metabolism | natural variation P lants have the ability to create over 200,000 small compounds known as secondary or specialized metabolites (1). These chemically diverse compounds help mediate plant adaptation to their environment and play important roles in plant defense mechanisms, pigmentation, and development. In addition, many of these metabolites are desirable to humans as medicinal and nutritional compounds. Therefore, furthering our understanding of plant specialized metabolism will have profound impacts on various applications from crop improvement to human health.To date, only a small fraction of the chemical and catalytic space in plant specialized metabolism has been explored. Even in the best-studied model plant Arabidopsis thaliana, there are still many uncharacterized metabolites, and the vast majority of genes encoding enzymes implied to be involved in specialized metabolism do not have known associations with any metabolites. Several studies of Arabidopsis natural accessions (individuals collected from wild populations) revealed considerable qualitative and quantitative variation in the accumulation of various compounds such as glucosinolates, terpenoids, and phenylpropanoids (2-4). This extensive metabolite variation can be attributed to genetic variation in genes encoding enzymes and regulatory factors of the pathways involved; quantitative trait locus (QTL) mapping has successfully uncove...