Prediction of pupylation sites using the composition of k-spaced amino acid pairs

Highlights

• The composition of k-spaced amino acid pairs is utilized to predict and analyze pupylation sites.

• The proposed method is more accurate than existing methods.

• Top ranked k-spaced amino acid pairs are analyzed to provide insights into the substrate specificity of pupylation.

• Both web server and standalone software are available for prediction.

• Probability scores are available for prioritizing pupylation sites.

Abstract

Pupylation is an important post-translational modification in prokaryotes. A prokaryotic ubiquitin-like protein (Pup) is attached to proteins as a signal for selective degradation by proteasome. Several proteomics methods have been developed for the identification of pupylated proteins and pupylation sites. However, pupylation sites of many experimentally identified pupylated proteins are still unknown. The development of sequence-based prediction methods can help to accelerate the identification of pupylation sites and gain insights into the substrate specificity and regulatory functions of pupylation. A novel tool iPUP is developed for the computational identification of pupylation sites. A composition of k-spaced amino acid pairs is utilized to represent a peptide sequence. Top ranked k-spaced amino acid pairs are subsequently selected by using a sequential backward feature elimination algorithm. The 10-fold cross-validation performance of iPUP trained by using the composition of 150 top ranked k-spaced amino acid pairs and support vector machines is 0.83 for the area under receiver operating characteristic curve. The importance analysis of k-spaced amino acid pairs shows that terminal space-containing pairs are useful for discriminating pupylation sites from non-pupylation sites. A sequence analysis confirms that lysines close to C-terminus tend to be pupylated. In contrast, lysines close to N-terminus are less likely to be pupylated. The iPUP tool can predict pupylation sites with probability scores for prioritizing promising pupylation sites. Both the online server and the standalone software of iPUP are freely available for academic use at http://cwtung.kmu.edu.tw/ipup.

Introduction

Selective protein degradation is essential for regulating protein functions in all living organisms. In eukaryotes, ubiquitylation plays important roles in the regulation of DNA repair and transcription, control of signal transduction, and implication of endocytosis and sorting (Herrmann et al., 2007, Welchman et al., 2005). The conjugation of ubiquitins to substrate proteins leads to selective degradation by proteasome (Herrmann et al., 2007, Welchman et al., 2005). Pupylation is recently identified as a functional analogue of ubiquitylation in prokaryotes. A prokaryotic ubiquitin-like protein (Pup) will be linked to specific lysine residues of target proteins by forming isopeptide bonds for proteasomal degradation (Burns et al., 2009, Pearce et al., 2008). Pup is an intrinsically disordered protein (Chen et al., 2009, Liao et al., 2009) with 64 amino acids. Although Pup and ubiquitin are functional analogues, their sequences share only a diglycine motif (GGQ) at C-terminus (Knipfer and Shrader, 1997, Pearce et al., 2008). In contrast to ubiquitylation requiring three enzymes, only two enzymes are involved in pupylation. First, the C-terminal glutamine of Pup is deamidated to glutamate by deamidase of Pup (Dop) (Striebel et al., 2009). Subsequently, proteasome accessory factor A (PafA) attaches the deamidated Pup to specific lysine residues of substrate proteins (Guth et al., 2011).

To better understand the functions of pupylation, several large-scale proteomics methods were developed to identify pupylated proteins and corresponding pupylation sites (Cerda-Maira et al., 2011, Festa et al., 2010, Poulsen et al., 2010, Watrous et al., 2010). As the number of identified pupylated proteins and sites grows, a structured and searchable database PupDB has been developed for management and analysis of pupylation sites by integrating information of pupylated proteins and sites, protein structures and functional annotations (Tung, 2012). However, there are still many pupylation sites to be discovered. It is desirable to develop computational methods to accelerate the identification of pupylated proteins and pupylation sites.

Unlike sumoylation sites with a well-defined motif of [ILVMF]KxE (Melchior, 2000), most post-translational modifications have no clearly defined motif that might be due to the vast number of enzymes for attaching modifiers to substrates such as phosphorylation and ubiquitylation (Li et al., 2008). Similar to ubiquitylation sites, no clear motif has been found in pupylation sites using a two-sample logo analysis based on the dataset of PupDB (Tung, 2012). The development of composition-based prediction methods for pupylation sites.

To develop a prediction method for identifying pupylation sties, pupylation sites are firstly extracted from PupDB and divided into a training dataset and an independent test dataset. Subsequently, a composition of k-spaced amino acid pairs (CKSAAP) is utilized to encode pupylation sites. The original CKSAAP has been successfully applied to the predictions of protein flexible/rigid regions (Chen et al., 2007a), protein crystallization (Chen et al., 2007b), mucin-type O-glycosylation sites (Chen et al., 2008), palmitoylation sites (Wang et al., 2009), ubiquitylation sites (Chen et al., 2011), phosphorylation sites (Zhao et al., 2012) and identification of weakly conserved sequence motifs (Dong et al., 2013). By introducing an additional symbol for a terminal space, CKSAAP can be utilized to analyze k-spaced amino acid pairs containing a terminal space.

Feature selection is a critical step for eliminating unrelated features and improving prediction performances. A sequential backward feature elimination algorithm is proposed to iteratively remove k-spaced amino acid pairs with lowest ranks obtained from chi-square test. Finally, the top 150 k-spaced amino acid pairs are selected to construct the prediction method iPUP that is based on a support vector machine (SVM) classifier. Compared to the only existing predictor GPS-PUP (Liu et al., 2011), iPUP shows better prediction performances of 10-fold cross-validation (10-CV) on the training dataset and independent test on the test dataset with 8% and 5% improvements of area under receiver operating characteristic curve (AUC) values, respectively. An analysis of the top 150 k-spaced amino acid pairs shows that lysines near the protein C-terminus tend to be pupylated. In contrast, lysines near the protein N-terminus is less likely to be pupylated.

In the previous proteomics studies, there are many identified pupylated proteins whose pupylation sites are still unknown. To facilitate the identification of pupylation sites of the pupylated proteins, a total of 1116 experimentally identified pupylated proteins without known pupylation sites are firstly extracted from PupDB. The iPUP is subsequently applied to predict pupylation sites for the pupylated proteins. Based on the threshold of high specificity, iPUP successfully identified 828 out of 1116 (74%) proteins. The average number of predicted pupylation sites for each pupylated protein is 2.59. The prediction method iPUP has been implemented as a web server. It can predict pupylation sites with probability scores for prioritizing promising pupylation sites. A standalone version of iPUP software has also been implemented for the large-scale prediction of pupylation sites for a large number of proteins. The iPUP is a cross-platform program running on the Java Virtual Machine (JVM). Both the web server and software are freely accessible and downloadable at http://cwtung.kmu.edu.tw/ipup.