RGAugury: a pipeline for genome-wide prediction of resistance gene analogs (RGAs) in plants

Abstract

Background: Resistance gene analogs (RGAs), such as NBS-encoding proteins, receptor-like protein kinases (RLKs) and receptor-like proteins (RLPs), are potential R-genes that contain specific conserved domains and motifs. Thus, RGAs can be predicted based on their conserved structural features using bioinformatics tools. Computer programs have been developed for the identification of individual domains and motifs from the protein sequences of RGAs but none offer a systematic assessment of the different types of RGAs. A user-friendly and efficient pipeline is needed for large-scale genome-wide RGA predictions of the growing number of sequenced plant genomes.

Results: An integrative pipeline, named RGAugury, was developed to automate RGA prediction. The pipeline first identifies RGA-related protein domains and motifs, namely nucleotide binding site (NB-ARC), leucine rich repeat (LRR), transmembrane (TM), serine/threonine and tyrosine kinase (STTK), lysin motif (LysM), coiled-coil (CC) and Toll/Interleukin-1 receptor (TIR). RGA candidates are identified and classified into four major families based on the presence of combinations of these RGA domains and motifs: NBS-encoding, TM-CC, and membrane associated RLP and RLK. All time-consuming analyses of the pipeline are paralleled to improve performance. The pipeline was evaluated using the well-annotated Arabidopsis genome. A total of 98.5, 85.2, and 100 % of the reported NBS-encoding genes, membrane associated RLPs and RLKs were validated, respectively. The pipeline was also successfully applied to predict RGAs for 50 sequenced plant genomes. A user-friendly web interface was implemented to ease command line operations, facilitate visualization and simplify result management for multiple datasets.

Conclusions: RGAugury is an efficiently integrative bioinformatics tool for large scale genome-wide identification of RGAs. It is freely available at Bitbucket: https://bitbucket.org/yaanlpc/rgaugury .

Keywords: Genome-wide prediction; Nucleotide binding site (NBS); Pipeline; Receptor like kinase (RLK); Receptor like protein (RLP); Resistance gene analog (RGA).

Fig. 1

Workflow of RGAugury. The pipeline was designed to use protein sequences to detect conserved domains and motifs found in genes involved in plant resistance and identify RGAs by integrating results generated from five programs: BLAST, InterProScan, pfam_scan, nCoil and Phobius. The annotated candidates for four different RGA types are exported as plain files. Analyses performed in parallel mode are labelled in blue. Intermediate results are indicated by a dashed-line box. GFF3: Generic Feature Format version 3; CC: coiled-coil; LRR: leucine-rich repeat; NB-ARC: nucleotide binding adapter shared by APAF-1, R gene products and CED-4; STTK: serine/threonine and tyrosine kinase; LysM: lysin motif; TM: transmembrane

Fig. 2

RGA identification based on domain structures of genes. CC: coiled-coil; CN: CC-NBS; CNL: CC-NBS-LRR; LRR: leucine rich repeat; LysM: lysin motif; NB-ARC: nucleotide binding site-activity regulated cytoskeleton; NBS: nucleotide-binding site; NL: NBS-LRR; RGA: resistance gene analog; RLK: receptor like kinase; RLP: receptor like protein; STTK: serine/threonine and tyrosine kinase; TIR: Toll/Interleukin-1 receptor; TM: transmembrane; TN: TIR-NBS; TNL: TIR-NBS-LRR; TX: TIR-unknown domain

Fig. 3

Web user interface pages of RGAugury. a The main page of RGAuguary for data input. All parameter values required in the command line version are specified directly on this page. Only protein sequences in FASTA format are required. A GFF3 file corresponding to the input protein sequences is optional but recommended. Databases for InterProscan can be selected by choosing either a predesigned ‘Quick’ mode or a ‘Deep’ mode. The default E-value cut-off for the initial RGA filtering with BLASTP is 1e-5. b The RGA prediction result summary page

Fig. 4

Performance of RGAugury. Forty-nine sequenced plant genomes (Zea mays was excluded, see text) with varying numbers of protein coding genes were used for RGA identification on a server embedded with 40 CPUs. Time to complete the processing of the entire pipeline for each dataset was recorded as a performance measurement. Performance for the ‘Quick’ mode (Pfam + Gene3D databases) and ‘Deep’ mode (Pfam + Gene3D + SMART + Superfamily) were compared. The dots and R ² value in red represent results for the ‘Quick’ mode and those in red represent the results for the ‘Deep’ mode