A general computational model for predicting ribosomal frameshifts in genome sequences

Abstract

Programmed ribosomal frameshifts are frequently used by RNA viruses to synthesize a single fusion protein from two or more overlapping open reading frames. We previously developed a program called FSFinder for predicting $-1$ and $+1$ frameshifts for Windows systems. With our new web application and web service called FSFinder2, users can predict frameshifts of any type from any web browser and operating system. We tested it on Shewanella oneidensis MR-1, for which exact frameshift sites are not known. FSFinder2 is the first program capable of finding frameshifts of general type and it is a powerful tool for predicting and analyzing genes that utilize frameshifts for their expression. FSFinder2 is available at http://wilab.inha.ac.kr/FSFinder.

Introduction

During translation, two kinds of error can occur: missense errors and processivity errors. In missense errors, an aminoacyl-tRNA synthetase mischarges a tRNA with a wrong amino acid due to substitution of one of the three bases. Processivity errors, including frameshift errors, occur when a sense codon is recognized incorrectly by a release factor [1]. The frequency of programmed translational frameshifts has been estimated to be more than 10,000-fold lower than missense errors [1], [2], [3]. However, in some cases processivity errors are more sequence-dependent than in others [1]. This is mainly because of the existence of stimulatory sequences and structures in the mRNA [4], [5], [6]. On frameshift sites the ribosome shifts to an alternative reading frame by one or more nucleotides [7]. Frameshifts are classified into different types depending on the number of nucleotides shifted and the shifting direction. The most common type is a $-1$ frameshift, in which the ribosome slips a single nucleotide in the upstream direction. The $-1$ frameshifting requires a frameshift cassette that consists of a slippery site, a stimulatory RNA structure and a spacer. The slippery site generally consists of a heptameric tandem sequence of the form X XXY YYZ [6], but there are other slippery sequences. For example, Escherichia coli dnaX mRNA uses a Shine–Dalgarno (SD) sequence upstream of its frameshift sequence A AAA AAG, as well as a stimulatory RNA secondary structure such as a pseudoknot or stem-loop. A 5–9 nucleotide sequence called a spacer is located between the slippery sequence and the stimulatory RNA secondary structure.

The $+1$ frameshifts are much less common than $-1$ frameshifts, but have been observed in diverse organisms [5]. The prfB gene encoding release factor 2 (RF2) of E. coli is a well-known example of genes caused by $+1$ frameshifting [8], [9]. In this gene a SD sequence is observed upstream of the slippery sequence CUU URA C, where R is either adenine or guanine. The ornithine decarboxylase antizyme (oaz) gene encoding antizyme 1 is caused by $+1$ frameshifting, and its frameshift signal consists of a slippery sequence and a downstream RNA secondary structure [10].

Most frameshift elements are located near the end of an open reading frame. Since these elements are located between two overlapping genes, focusing on the overlapping region can be very efficient. However, Barry et al. [11] consider that frameshifting on Barley yellow dwarf virus RNA requires a viral sequence located four kilobases downstream.

No program exists to predict general types of frameshift. In addition, existing computational models predict too many false positives. In previous work we developed a program called FSFinder (frameshift signal finder) for predicting $-1$ and $+1$ frameshift sites [12]. Trials of FSFinder on $\sim 190$ genomic and partial DNA sequences showed that it predicted frameshift sites efficiently and with greater sensitivity and specificity than other programs because it focused on the overlapping regions of open reading frames and prioritized candidate signals (for $-1$ frameshifts, sensitivity was 0.88 and specificity 0.97; for $+1$ frameshifts, sensitivity was 0.91 and specificity 0.94) [13], [14], [15].

FSFinder is written in Microsoft C# and is executable on Windows systems only. To remove this limitation and to handle frameshifts of general type, we developed a new web-based application program called FSFinder2. Users can predict frameshifts of any type online from any web browser and operating system. We tested FSFinder2 on the facultative gram-negative bacterium, Shewanella oneidensis MR-1 to search for six patterns of frameshifts defined by our collaborators. We believe this is the first program capable of predicting frameshift signals of general type.