Identifying Missing Biosynthesis Enzymes of Plant Natural Products

Abstract

Elucidating plant-specialized biosynthetic pathways has always constituted a laborious task, notably for natural products with high pharmaceutical values. Here, we discuss emerging omics-based strategies that facilitate the identification of genes from these complex metabolic pathways, paving the way to engineered supplies of these compounds through synthetic biology approaches. Elucidating plant-specialized biosynthetic pathways has always constituted a laborious task, notably for natural products with high pharmaceutical values. Here, we discuss emerging omics-based strategies that facilitate the identification of genes from these complex metabolic pathways, paving the way to engineered supplies of these compounds through synthetic biology approaches. Plants represent a remarkable source of natural metabolites, many of which remain the basis of the pharmacopoeia used by humans to treat various disorders [1.Newman D.J. Cragg G.M. Natural products as sources of new drugs from 1981 to 2014.J. Nat. Prod. 2016; 79: 629-661Crossref PubMed Scopus (3055) Google Scholar]. Highly potent compounds are found within monoterpene indole alkaloids of the Apocynaceae plant family, yew taxane-type terpenoids, mayapple lignans, poppy isoquinoline alkaloids, and hemp cannabinoids. Due to their pharmaceutical importance, the biosynthetic pathways responsible for the production of these compounds in planta have attracted the attention of many research groups for decades. This area remains in focus as it is essential to be able to propose cheaper and high-throughput alternatives of production of these valuable pharmaceutical compounds, in a short period. While most of the enzymes catalyzing biosynthesis steps have been progressively identified using sequence homology (see Glossary)-based cloning or protein purification, these discovery efforts have always been laborious, particularly until the arrival of massive omics resources such as next generation sequencing (NGS). This is likely because downstream branches of these pathways: (i) differ among plant species, even within the same genus; and (ii) involve enzymes from superfamilies containing several tens to hundreds of members [2.Noda-Garcia L. et al.Metabolite-enzyme coevolution: from single enzymes to metabolic pathways and networks.Annu. Rev. Biochem. 2018; 87: 187-216Crossref PubMed Scopus (40) Google Scholar]. Such superfamilies include reductases, oxidases, methyltransferases, and cytochrome P450s [3.Dastmalchi M. et al.Family portraits: the enzymes behind benzylisoquinoline alkaloid diversity.Phytochem. Rev. 2017; 17: 1-29PubMed Google Scholar]. Cytochrome P450s in particular are key components of plant biosynthetic pathways and may catalyze multiple types of reactions, including hydroxylation, epoxidation, demethylation, dealkylation, decarboxylation, and C–C bond cleavage [4.Wei Y. et al.Recent developments in the application of P450 based biocatalysts.Curr. Opin. Chem. Biol. 2018; 43: 1-7Crossref PubMed Scopus (50) Google Scholar]. Fortunately, recent advances in omics-based strategies have accelerated the identification of missing biosynthetic enzymes. Below, we briefly highlight these advances, also summarized in Figure 1. Identifying enzymes catalyzing steps of given pathways requires a prior knowledge on potential reaction scenarios leading to the formation of the expected compound, as illustrated on the left part of Figure 1. Metabolomics are typically used to identify preferential accumulation sites of specialized metabolites and to detect plausible reaction intermediates. Molecular networking also represents a considerable advance in the processing of the resulting metabolomics data acquired with either liquid chromatography or gas chromatography coupled to mass spectrometry by connecting related separated compounds based on their mass spectra [5.Fox Ramos A.E. et al.Natural products targeting strategies involving molecular networking: different manners, one goal.Nat. Prod. Rep. 2019; 36: 960-980Crossref PubMed Google Scholar]. Ideally, connected compounds may reflect an entire biosynthetic branch. Once a reaction sequence has been postulated, a type of enzymatic activity required for each step may be hypothesized based on literature data or bioinspired organic chemistry. Activity of enzymes relies on specific functional conserved domains that can drive their identification. For instance, a hydroxylation step identified as described above may be catalyzed by a cytochrome P450 or a dioxygenase, each bearing their own functional domain. Defining a first set of candidate genes requires access to genomes or transcriptome assemblies, providing gene or transcript sequences, respectively, to look for such specific domains. Although challenging in terms of computational requirements, whole genomes may be sequenced at lower cost from short DNA reads generated through NGS. RNA-seq-based transcriptome analysis is simpler to conduct than whole genome assembly because the transcriptome only covers a small part of the genome. Based on genome or transcriptome annotation, functional domains can be systematically attributed to define first sets of candidate genes that will require a further prioritization through the methodological aspects described below. One of the most direct and obvious ways to unravel missing steps is to take inspiration from previous works reported in the literature and screen for homology to orthologous gene sequences of already characterized biosynthesis steps in other plant species (Figure 1). This can be done by searching homologies [e.g., with basic local alignment search tool (BLAST)] or constructing phylogenies statistically fitted on evolutionary models encompassing protein sequences from multiple species [6.Salim V. et al.Camptotheca acuminata 10-hydroxycamptothecin O-methyltransferase: an alkaloid biosynthetic enzyme co-opted from flavonoid metabolism.Plant J. 2018; 95: 112-125Crossref PubMed Scopus (8) Google Scholar]. For example, Farrow et al. [7.Farrow S.C. et al.Cytochrome P450 and O-methyltransferase catalyze the final steps in the biosynthesis of the anti-addictive alkaloid ibogaine from Tabernanthe iboga.J. Biol. Chem. 2018; 293: 13821-13833Abstract Full Text Full Text PDF PubMed Scopus (14) Google Scholar] recently identified the ibogamine 10-hydroxylase from the African shrub (Tabernanthe iboga) by searching for genes with homology to tabersonine 16-hydroxylase from Madagascar periwinkle (Catharanthus roseus), which catalyzes indole ring hydroxylation at the same position. However, while often successful, this approach must be considered with caution since it has been demonstrated in the above example that the closest ortholog does not necessarily catalyze a similar reaction [7.Farrow S.C. et al.Cytochrome P450 and O-methyltransferase catalyze the final steps in the biosynthesis of the anti-addictive alkaloid ibogaine from Tabernanthe iboga.J. Biol. Chem. 2018; 293: 13821-13833Abstract Full Text Full Text PDF PubMed Scopus (14) Google Scholar]. Unlike the above described targeted approach, wide-scale untargeted studies can be deployed using ethyl methanesulfonate-based random mutagenesis (Figure 1) [8.Edge A. et al.A tabersonine 3-reductase Catharanthus roseus mutant accumulates vindoline pathway intermediates.Planta. 2018; 247: 155-169Crossref PubMed Scopus (8) Google Scholar]. In this approach, mutagenized plants are screened for changes in the production of the desired metabolite. Plants with altered profiles are then sequenced to identify the source of the mutation by NGS that is responsible for changes in the metabolism. This is a labor-intensive but powerful strategy already applied to C. roseus to isolate mutants with altered alkaloid contents. This approach led to the identification of an O-acetylstemmadenine oxidase catalyzing a major oxidation step in catharanthine and tabersonine biosynthesis [9.Qu Y. et al.Completion of the canonical pathway for assembly of anticancer drugs vincristine/vinblastine in Catharanthus roseus.Plant J. 2019; 97: 257-266Crossref PubMed Scopus (31) Google Scholar]. Since plant biosynthetic pathways sometimes involve enzymes encoded by genes physically clustered in the same genomic region, genome sequence may improve homology searches by the analysis of genomic neighbors. Candidate genes can thus be refined through the in silico processing of large sequencing data using previously characterized biosynthesis genes as baits (Figure 1). It is particularly efficient with more complete genome sequences composed of very long DNA sequence assemblies, as recently exemplified with the identification of a gene cluster encoding five enzymes involved in thebaine biosynthesis in opium poppy (Papaver somniferum) [10.Chen et al.A pathogenesis-related 10 protein catalyzes the final step in thebaine biosynthesis.Nat. Chem. Biol. 2018; 14: 738-743Crossref PubMed Scopus (36) Google Scholar]. With the explosion in the availability of plant genome sequences in the coming years, new gene clusters are likely to be discovered. However, one of the disadvantages of this method is that biosynthetic pathways show only few, if no, physically clustered genes, such as recently observed in happy tree (Camptotheca acuminata) [11.Zhao D. et al.De novo genome assembly of Camptotheca acuminata, a natural source of the anti-cancer compound camptothecin.Gigascience. 2017; 6: 1-7Crossref Scopus (36) Google Scholar]. Candidate gene selection may also rely on expression pattern similarities among tissues or experimental conditions displayed by genes related to a common biosynthetic branch (Figure 1). The underlying hypothesis is that metabolites accumulated in a tissue-specific manner should be produced through a chain of enzymes displaying a similar spatio-temporal expression pattern, which should be visible at the transcript level. Such coexpression can guide identification of missing steps by generating gene lists corresponding to groups of genes that have similar expression across tissues. Groups of similarly expressed transcripts that contain genes already identified as part of the investigated metabolic pathway are used to identify candidate genes for the missing steps. It typically requires the measurement of genome-wide gene expression in different tissues or experimental conditions using RNA-seq [12.Liesecke F. et al.Ranking genome-wide correlation measurements improves microarray and RNA-seq based global and targeted co-expression networks.Sci. Rep. 2018; 8: 10885Crossref PubMed Scopus (24) Google Scholar]. Three major approaches can be used on the generated data (expression matrices containing genes in rows and sample types in columns) for coexpression analyses and eventual identification of missing biosynthetic enzymes. A first approach entails the statistical comparison of contrasted samples (e.g., roots versus leaves or control versus treatment) to identify differentially expressed genes. In this way, geissoschizine oxidase was found to be upregulated in stressed leaves of C. roseus, together with strictosidine glucosidase that commonly act in the monoterpene indole alkaloid biosynthetic pathway [13.Tatsis E.C. et al.A three enzyme system to generate the Strychnos alkaloid scaffold from a central biosynthetic intermediate.Nat. Commun. 2017; 8: 316Crossref PubMed Scopus (48) Google Scholar]. More extensively, gene expression profiles can be compared by a hierarchical clustering approach in multiple sample comparisons, such as performed to identify enzymes converting matairesinol into etoposide aglycone in the mayapple (Podophyllum hexandrum) lignan biosynthetic pathway [14.Lau W. Sattely E.S. Six enzymes from mayapple that complete the biosynthetic pathway to the etoposide aglycone.Science. 2015; 349: 1224-1228Crossref PubMed Scopus (183) Google Scholar]. A second method is based on the construction of gene coexpression networks (GCNs) to provide a more exhaustive view of coexpression relationships among genes [12.Liesecke F. et al.Ranking genome-wide correlation measurements improves microarray and RNA-seq based global and targeted co-expression networks.Sci. Rep. 2018; 8: 10885Crossref PubMed Scopus (24) Google Scholar]. GCNs visualize similarities in gene expression profiles where distances between each possible gene pairs are calculated using specific metrics (e.g., Pearson correlation coefficient) and only the best coexpressed gene pairs are retained to construct a GCN. In GCN type representations, genes (also referred to as ‘nodes’) are connected by edges representing these distances [12.Liesecke F. et al.Ranking genome-wide correlation measurements improves microarray and RNA-seq based global and targeted co-expression networks.Sci. Rep. 2018; 8: 10885Crossref PubMed Scopus (24) Google Scholar]. This approach has been successfully used to streamline the identification of precondylocarpine acetate synthase and tabersonine synthase in C. roseus [15.Caputi L. et al.Missing enzymes in the biosynthesis of the anticancer drug vinblastine in Madagascar periwinkle.Science. 2018; 360: 1235-1239Crossref PubMed Scopus (111) Google Scholar]. Another method for analyzing gene coexpression is based on unsupervised machine learning (UML). UML methods group similarly expressed genes into a specific expression cluster. These methods are unsupervised because they cluster genes according to their expression levels without prior knowledge of their function. UML has been successfully used several times, in particular with self-organizing maps. As an example, this algorithm has been used to identify the sarpagan-bridge enzyme involved in ajmaline biosynthesis in the devil pepper (Rauvolfia serpentina), an Apocynaceae closely related to C. roseus [16.Dang T.-T.T. et al.Sarpagan bridge enzyme has substrate-controlled cyclization and aromatization modes.Nat. Chem. Biol. 2018; 14: 760-763Crossref PubMed Scopus (15) Google Scholar]. Elucidating candidate gene functions is a mandatory step but can be laborious and time consuming especially in nonmodel plant species for which no mutant libraries or no quantitative genetics data are available, albeit they are the main source of specialized metabolites. While biochemical characterization relying on recombinant protein expression can be envisaged for a small number of candidates, it always requires access to potential enzyme substrates. Over the last few years, wider gene function analyses have been enabled through the development of efficient and straightforward reverse genetic approaches based on transient transcript degradation through virus-induced gene silencing (VIGS), notably [17.Wijekoon C.P. Facchini P.J. Systematic knockdown of morphine pathway enzymes in opium poppy using virus-induced gene silencing.Plant J. 2012; 69: 1052-1063Crossref PubMed Scopus (48) Google Scholar]. Besides confirming involvement in biosynthetic pathways, such an approach can also provide evidence of gene function through the identification of accumulated biosynthetic intermediates resulting from silencing [13.Tatsis E.C. et al.A three enzyme system to generate the Strychnos alkaloid scaffold from a central biosynthetic intermediate.Nat. Commun. 2017; 8: 316Crossref PubMed Scopus (48) Google Scholar]. However, final biochemical characterization is still required to confirm VIGS results. Finally, for validation of multiples genes from the same pathway, heterologous reconstitution of partial pathways can be performed by simultaneous gene co-overexpression combined to biosynthetic precursor feeding, as performed for podophyllotoxin and tabersonine pathways [14.Lau W. Sattely E.S. Six enzymes from mayapple that complete the biosynthetic pathway to the etoposide aglycone.Science. 2015; 349: 1224-1228Crossref PubMed Scopus (183) Google Scholar,15.Caputi L. et al.Missing enzymes in the biosynthesis of the anticancer drug vinblastine in Madagascar periwinkle.Science. 2018; 360: 1235-1239Crossref PubMed Scopus (111) Google Scholar]. The present forum article briefly describes current procedures used to characterize missing steps from plant metabolic pathways. The advent of NGS technologies have largely fueled these procedures, allowing completion of pathway knowledge in several medicinal plants. According to the different strategies depicted in Figure 1 and described above, candidate genes are selected using their sequence and/or expression properties. However, it is likely the case that, in the near future, integrative approaches combining several of these features will streamline the candidate gene prioritization process. For example, Carqueijeiro et al. [18.Carqueijeiro I. et al.A BAHD acyltransferase catalyzing 19-O-acetylation of tabersonine derivatives in roots of Catharanthus roseus enables combinatorial synthesis of monoterpene indole alkaloids.Plant J. 2018; 94: 469-484Crossref PubMed Scopus (25) Google Scholar] combined gene expression and physical clustering analysis to identify an acetyl transferase involved in the biosynthesis of root alkaloids in C. roseus. Further automated integrative approaches will undoubtedly facilitate the identification of missing enzymes, as knowledge on plant biosynthetic machineries is rapidly progressing. Because many biosynthetic genes are now fully characterized, this knowledge may indeed drive supervised machine learning (SML)-based approaches to predict gene function [19.Kim G.B. et al.Machine learning applications in systems metabolic engineering.Curr. Opin. Biotechnol. 2019; 64: 1-9Crossref PubMed Scopus (39) Google Scholar]. A training set representing 80% of a transcriptome labeled as following: characterized genes or orthologs as ‘alkaloid-related’ and the remaining ones as ‘nonalkaloid’. SML tries to construct the best model from input data to correctly classify genes with the correct label and predicts possible labels for the remaining 20% of genes. Although promising, fine optimization is required to correctly deploy SML approaches for missing gene identification. It is important to note that although natural product biosynthesis in plants displays a certain form of homogeneity (coexpression, physical clustering, or sequence homology), some steps are catalyzed by enzymes encoded by genes without obvious distinguishing features, making them difficult to detect through these genomic approaches. In this case, many candidate genes identified in silico must be tested before the desired catalytic activity is discovered. High-throughput techniques to clone candidate genes or more conventional procedures such as native protein purification through cell fractionation must thus be considered as part of the elucidation process. Based on this whole set of approaches, completion of biosynthetic pathways will accelerate in the coming years and, in turn, will facilitate the development of microbial cell factories synthetizing plant-derived drugs [20.Luo X. et al.Complete biosynthesis of cannabinoids and their unnatural analogues in yeast.Nature. 2019; 567: 123-126Crossref PubMed Scopus (193) Google Scholar,21.Li Y. et al.Complete biosynthesis of noscapine and halogenated alkaloids in yeast.Proc. Natl. Acad. Sci. U. S. A. 2018; 115: E3922-E3931Crossref PubMed Scopus (113) Google Scholar]. This remains an essential prerequisite to be able to propose, in the short term, cheaper and high-throughput alternatives of production of these valuable pharmaceutical compounds. We acknowledge funding from the EU Horizon 2020 research and innovation program (MIAMi project-Grant agreement N°814645), ARD2020 Biopharmaceutical program of the Région Centre Val de Loire (BioPROPHARM and CatharSIS projects), La ligue Contre le Cancer (Yeast4LiFE), and from le Studium (Consortium fellowship) and ERC 788301. The authors want to apologize for other excellent studies which have not been cited here due to length restriction. second generation technologies refers to the high-throughput sequencing of DNA fragment ends, usually outputting millions of short reads (up to 250 nt long). The third generation technologies sequence single DNA molecules, generating long reads (at least >1000 nt long). high-throughput sequencing of reverse-transcribed mRNA. Short read-based RNA-seq is typically a two-step workflow, including: (i) transcript reconstruction (genome-based or de novo); and (ii) gene expression inference from read attribution to reconstructed transcript sequences. an unsupervised machine learning algorithm used to cluster genes into a given number of nodes forming a 2D map. In this map, regions contain nodes reflecting genes with similar expression levels. homology defines similarity between two nucleic or protein sequences. Similarities can be searched against specific databases with the well-known BLAST algorithm. Phylogenies a set of sequences to highlight evolutionary relationships through statistical approaches. a reverse genetics approach allowing transient plant gene by the of a a short sequence targeting a specific a complex due to genome and requires a of short reads and long reads to the reconstruction of long

Topics & Concepts