MicroScope—an integrated microbial resource for the curation and comparative analysis of genomic and metabolic data

New website design

To make all the PkGDB data and the comparative analysis tools widely accessible to the scientific community, we have developed the new web site structure shown in Supplementary Figure S1. The MicroScope analysis tools and graphical interfaces are organized as several categories (main tool bar in the black rectangle): The ‘MaGe’ and ‘Genomic tools’ menus contain functionalities needed for the curation process and for the summary of the automatic and expert annotation (Genome overview, COG functional classification, etc.). The ‘Comparative genomics’ menu gathers tools, which compare the query genome to others microbial genomes (Supplementary Figure S1): Gene phyloprofile, Genomic island prediction, Line plot using synteny results, computation of fusion/fission events and statistics about synteny results using the PkGDB and the RefSeq genome resources (3). The ‘Metabolism’ menu initially contained tools allowing the user to browse KEGG or MicroCyc metabolic data and to compare metabolic pathways across several organisms (4). This menu has recently been extended to several new tools dedicated to the curation of metabolic data (see section ‘Metabolic data curation tools’). The ‘Searches’ menu allows the user to perform Blast searches in PkGDB and to use the ‘Search by keywords’ interface, which allows for the identification of genes and functions of interest using a variety of selection filters (queries can be made on the annotations and on the pre-computed results). Sequence and annotation data available in PkGDB can be downloaded in standard file formats (Genbank, EMBL, tabulated, etc.; ‘Export’ menu); this now includes the PGDBs computed with the Pathologic software (4,10), which can be then loaded directly into any local installation of BioCyc (11). We also added some functionalities to extract non-coding DNA sequences and/or DNA fragments. Several new tools, presented in the ‘Analysis of experimental data’ section, have been developed and organized in the ‘Experimental Data’ menu (see later in the text). Finally, using the ‘User Panel’ menu, various MicroScope settings can be modified (i.e. the selected genomes in the synteny maps, the set of favourite organisms to work with, etc.). In addition, the main coordinator of a project is now able to manage the personal accounts related to his/her project (i.e. remove/add accounts, setup access rights on sequences).

Gene carts in MicroScope handle lists of genes resulting from analysis tools and/or queries (4). The ‘Gene carts’ interface allows one to perform and combine various gene list operations: merge, intersect, differentiate of two gene carts, extract gene/protein sequences and run multiple alignments via the plugged Jalview software (22). MicroScope users, which are involved in the same project(s), can now exchange results from their analyses/queries through the upload of gene carts (in an XML file format). In addition, the ‘Search by keywords’ interface has been extended to use data from one or several gene carts in the context of a query.

Comparative genomics tools

Analysis of experimental data

High-Throughput Sequencing technologies increased not only the flow of genome sequencing projects but also paved the way to new applications of sequencing in both population genetics and functional genomics. To deal with these new types of experimental data, we set up two MicroScope extensions following the three-way architecture of the platform: background analysis pipelines, a database for storing results produced by the pipeline and Graphical User Interfaces (GUI) for the exploration and analysis of data. The pipelines of the data from evolution projects (i.e. clones of the same species at different generation time) and from transcriptomic projects (i.e. RNAseq experiments) rely on a common base: the two first steps consist in the pre-processing of reads (i.e. discarding low quality reads, trimming low quality bases at 3′-ends, etc.) and in their mapping onto a reference genome.

Automatic genome-scale metabolic network reconstructions

Each genome stored in PkGDB is processed by an in-house workflow mainly based on Pathologic (version 16.0), one component of the Pathway Tools software suite (30), to generate a genome-scale metabolic network reconstruction. Pathologic uses a two-step process to build PGDBs from genome annotations. The ‘Reactome’ projection step associates genes with metabolic reactions from MetaCyc [i.e. the BioCyc reference pathways, (11)]. Owing to wrongly formatted or unspecific Enzyme Commission (EC) numbers, or to insufficiently explicit textual annotations, Pathologic may in some cases either over-predict or miss enzymatic reactions. We thus enhanced the matching procedure by directly using the official MetaCyc reaction frame identifier as the functional annotation. Our export procedure from the MicroScope PkGDB to Pathologic input format directly associates the genes to MetaCyc reaction identifiers if any; alternatively, the procedure also exports the partial EC numbers and/or the common gene name product. The set of predicted catalysed reactions in the studied genome is used in the pathway projection step to infer metabolic pathways stored in the MetaCyc resource. The precise rules applied for inferring pathways can be found in (30). This tailor-made reconstruction strategy has been used to build and analyse the genome-scale metabolic networks of 29 Escherichia coli strains (8 commensal and 21 pathogenic strains, including six Shigella strains) (31). Evaluation of the benefits of our reconstruction strategy showed that (i) the genome annotation quality directly impacted the number of matched reactions and improved the completion of predicted pathways and (ii) the curation done on a reference metabolic network can be efficiently adapted to closely related organisms (31). Indeed, using a unified source of genome annotations and a common reconstruction process for all metabolic networks limits the biases originating from the reconstruction process, thus making the networks reliably comparable.

The reconstructed metabolic networks are accessible via the MicroScope web interfaces and are also stored in the MicroCyc repository (http://www.genoscope.cns.fr/agc/microcyc), which contains to date >1600 complete metabolic networks (>800 are publicly available). The reconstruction strategy is re-run every day taking into account new expert annotations saved in PkGDB. It provides a list of pathways that need to be asserted as existing or not, and a list of pathway holes (i.e. enzymatic reactions which are not linked to any gene). Two dedicated user interfaces have been developed to ease the expert curation process, and our recently published CanOE strategy (32) has also been integrated in the MicroScope workflow to guide the user in the process of pathway hole filling.

Finding gene candidates for orphan enzymes: the CanOE strategy

In the last update of MicroScope, we described the ‘Pathway Synteny’ functionality that allowed annotators to retrieve groups of genes in a given organism sharing conserved syntenies and encoding enzymes involved in a same metabolic pathway (4). The listed results could be used to quickly check for reaction-hole candidate coding genes among the conserved miss-annotated genes of a given group. We have now developed a new tool named Candidate genes for Orphan Enzymes (CanOE) that allows one to automatically analyse the conservation of both genomic and metabolic contexts to provide candidate genes for orphan enzymatic reactions. Details on the CanOE algorithm can be found in (32). This strategy has been applied to all prokaryotic genomes stored in the MicroScope platform, and the results (i.e. genomic metabolons and ranked candidate genes for orphan enzymatic activities) were stored in PkGDB. MicroScope users can query these results in various ways (see the online tutorial for CanOE). An illustration of the use of CanOE is given in Figure 4A and corresponds to the search for S. typhimurium LT2 orphan reactions. Four candidate genes have been found for local orphan reactions (i.e. orphan reactions in the target organism), and one candidate gene is associated to a global orphan reaction (i.e. orphan reaction in all known organisms). As explained in the section ‘Using the new MicroScope functionalities: a case study’ (see later in the text), this last candidate gene is the starting point of the manual curation of the ‘CDP-3,6-dideoxyhexose biosynthesis’ pathway in S. typhimurium.

Curation of metabolic pathways

To help the user in the curation of false-positive predicted pathways and incomplete pathways (i.e. gene-reaction associations are missing), a new ‘Pathway curation’ interface, available in the ‘Metabolism’ menu of the MicroScope platform (Supplementary Figure S1), has been developed.

For each predicted pathway x in the reference organism, a pathway completion is computed that corresponds to the number of reactions from pathway x that are found in the genome divided by the total number of reactions in pathway x, as described in the MetaCyc resource (this value ranges between 0 = absence of the pathway, and 1 = complete pathway). The developed interface displays the list of predicted pathways, using the MetaCyc pathway hierarchy, together with the completion value and the number of reactions in the corresponding pathway (see Figure 4B). The users can curate each pathway by assigning different pathway statuses, which reflect the degree of functionality of any predicted pathway: (i) predicted (default status); (ii) validated (the pathway is known to be functional in the organism under study and thus, all the reactions of the pathway should be present); (iii) variant needed (the organism under study metabolizes the compound in a way different from that of the corresponding pathway described in MetaCyc), (iv) unknown (not enough evidence to select another status), (v) non-functional (inactive pathway, i.e. ‘pseudogenization’ events have affected some enzymatic steps); and (vi) deleted (false-positive predictions of the pathway projection algorithm). Figure 4B illustrates the use of this curation interface for S. typhimurium LT2: among the four listed predicted pathways (default status shown in orange), one is missing a reaction (PWY-5833, completion value = 0.67). The proposed candidate gene for this reaction (i.e. STM2090) is found both by CanOE and the ‘co-evolved genes’ methods. The complete curation process is described in section ‘Using the new MicroScope functionalities: a case study’, and leads to the ‘validated’ status for the ‘CPD-3,6 dideoxyhexose biosynthesis’ in S. typhimurium LT2 (Figure 5B).

Curation of Gene-Protein-Reaction associations in the Gene Editor

An essential step in the curation of metabolic data is the curation of associations between genes coding for enzymatic activities and the biochemical reactions catalysed by these enzymes [isozymes, multifunctional enzymes and protein complexes (33)]. The MicroScope Gene Editor has recently evolved to allow association of the gene being annotated to two main enzymatic reactions resources: MetaCyc (11) and RhEA (34). RhEA is the primary resource data for reactions in the Microme project. It defines reactions, which are unique and chemically balanced at the level of mass and charge, using chemical compounds stored in the ChEBI database (35).

Two new fields ‘MetaCyc reactions’ and ‘RhEA reactions’ have been added in the Gene Editor (see Figure 5A). The annotator can use the search modules (in grey background; Figure 5A) to retrieve MetaCyc and/or RhEA reactions using the EC number(s) available in the corresponding field of the Gene Editor, or using keywords entered in the ‘Search reaction by keyword’ area. The result of this query is a multiple selection list of reaction(s) organized as follows: the identifier of the reaction, the name of the reaction and, in parenthesis, the gene(s) in the target genome already associated to the corresponding reaction, if any. These Gene-Reaction (G-R) associations in the genome may have three statuses: (i) ‘validated’ for associations that have been manually validated; (ii) ‘annotated’ for reactions that have been transferred from an homologous gene of a closely related genome; and (iii) ‘predicted’ for G-R associations that have been matched by Pathologic using EC numbers or product names. Such information is very helpful to assert if the enzyme acts as an isoenzyme or as a protein complex. In the example shown in Figure 5A, curation of the STM2090 gene in S. typhimurium led to the association of this gene with the 1.17.1.1-RXN MetaCyc reaction (using generic NAD(P) coenzymes) and with two RhEA reactions: RHEA: 19653 (using the NAD coenzyme) and RHEA: 19657 (using the NADP coenzyme).

The PkGDB resource already stores 2800 validated G-R associations such as those of E. coli K-12 and Bacillus subtilis 168. It also gathers 32 224 transferred and 1 482 039 predicted GR associations.

Using the new MicroScope functionalities: a case study

An example of the use of these different curation tools and interfaces is illustrated in Figure 4 and 5. In the context of the curation of S. typhimurium LT2 metabolic data, results of the CanOE strategy were explored to search for predicted gene associations to orphan reactions. Figure 4A shows that only one global orphan reaction (i.e. a reaction without any known coding genes in all MicroScope organisms) corresponding to the EC number 1.17.1.1 is proposed to be associated with a S. typhimurium candidate gene, namely STM2090. Indeed, the corresponding genomic metabolon shows that this potential association (dotted purple line in the graph) is found in the genomic region STM2089-STM2092 of the S. typhimurium chromosome, the co-localized genes of which have known associations (shown in green) with three others reactions (2.7.7.33-RNX, CDP-GLUCOSE-46-DEHYDRATASE-RNX and RNX9158) that all belong to a same metabolic pathway. The CDP-3,6-dideoxyhexose is a basic component of the O-antigen part of lipopolysaccharides in enterobacteria, and a dominant antigenic determinant in the membranes of these organisms (36). This pathway appears incomplete in the pathway curation interface of MicrosScope, because it has a pathway completion below 1 (Figure 4B). Using the link PWY-5833, the list of reactions involved in the CDP-3,6-dideoxyhexose biosynthesis pathway is displayed (Figure 4C): the automatic projection procedure failed to associate the EC:1.17.1.1 to a S. typhimurium gene, leading to a pathway hole for the CDP-4-dehydro-6-deoxyglucose reductase activity (i.e. the global orphan enzymatic activity for which CanOE is able to propose a candidate gene, STM2090).

Looking at the co-evolution scores of S. typhimurium LT2 genes with the other genes of this pathway (‘co-evolved genes’ functionality, Figure 4C), it appears that the STM2090 gene has the third best score just after STM2092 and STM2091, pointing out a common evolutionary profile of presence–absence with other genes of the CDP-3,6-dideoxyhexose biosynthesis pathway. The STM2090 candidate gene was initially annotated as a ‘lipopolysaccharide biosynthesis protein rfbH’ (Figure 4C), and no enzymatic activity was associated to this gene. However, looking at the blast results available in the Gene Editor, it appears that STM2090 shares significant similarity with Yersinia pseudotuberculosis TrEMBL entries (>80% amino-acid identity over the whole protein length); the protein encodes the CDP-6-deoxy-L-threo-D-glycero-4-hexulose-3-dehydrase subunit of the CDP-4-dehydro-6-deoxyglucose reductase complex, the activity of which has been experimentally demonstrated in this organism (36). The analysis of the flanking genes in the metabolon (STM2089-STM2092) led us to identify STM2093 as the gene coding for the CDP-6-deoxy-L-threo-D-glycero-4-hexulose-3-dehydrase reductase subunit of the same complex. Thus, we can assume, with a high confidence, that the EC:1.17.1.1 is catalysed by a CDP-4-dehydro-6-deoxyglucose reductase enzyme complex encoded by the STM2090 and STM2093 genes in S. typhimurium. Based on these evidences, the functional annotations of STM2090 and STM02093 have been updated accordingly and associated to the EC number 1.17.1.1. The MetaCyc and RhEA reactions linked to this EC number have been validated (Figure 5A). Following this curation step, the pathway projection tool has been re-executed: the CDP-3,6-dideoxyhexose biosynthesis pathway is now complete (Figure 5C) and has been validated in the pathway curation interface of MicroScope selecting the ‘validated’ status (Figure 5B).