NGS (Next-generation sequencing) platforms have brought about a revolution in the growth of biological knowledge [1]. These devices feature high data throughput at a reduced cost compared to platforms using the Sanger method [2,3].

The benchtop sequencing platforms have been launched in the past three years, for instance, Miseq by Illumina (www.illumina.com) and the Ion Torrent PGM (Personal genome machine) by Life Technologies [4].

The Miseq system performs sequencing by synthesis, similar to other Illumina sequencers [4]. However, its technology has dramatically reduced the time required per round compared to the Illumina HiSeq [5,6]. In 2011, Life Technologies released the Ion Torrent PGM, which uses sequencing based on semiconductor technology. This system detects hydrogen ions released during DNA sequencing, thereby inaugurating the post-light era [4,5].

Despite the increased accuracy of data produced by such sequencers, many errors remain (substitutions and indels), which are inherent to the chemistry used in these platforms, the method of library preparation and the type of sequencing used. These errors increase the complexity of data processing, and each of these devices introduces specific errors, such as the substitution errors that are prevalent on the Illumina and SOLiD platforms [7].

An error common to the Ion Torrent PGM and Proton platforms, which was observed in Roche 454, is related to the recognition of homopolymers due to the sequencing of these regions in a single cycle. Additionally, there is the issue of detecting the byproducts of the nucleotide incorporation reactions, i.e., pyrophosphates (454) and hydrogen ions (Ion Torrent and Proton). These are synthesis-based methods that measure reagent flow because the intensity of the flow of reactants is directly proportional to the amount of nucleotides incorporated. However, the relationship between the measured flow intensity and the number of nucleotides incorporated is nonlinear in homopolymeric regions, causing frequent errors in determining the length of such regions, which results in insertions and deletions [8].

The read mapping errors affect the de novo and reference assembly processes due to wrong insertions and deletions, which can cause frame-shifts that are identified during genome annotation and described, mainly, to Ion Torrent PGM [3,9,10].

To improve the quality of data produced on the Ion Torrent PGM sequencing platform, Life Technologies has developed a sequencing enzyme called Hi-Q with site-directed mutations in its molecular structure that reduce insertion and deletion error rates compared to the traditional chemistry.

This study aimed to demonstrate the performance during assembly and sequencing of a fragment library from the Corynebacterium pseudotuberculosis Cp31 genome using the traditionally marketed enzyme and Hi-Q, and compare the occurrence of pseudogenes between the genome available at NCBI and that assembled with Hi-Q enzyme.

Origin of the isolate, organism growth and DNA extraction

The model organism used in this study was C. pseudotuberculosis 31, which was isolated from a buffalo in Egypt [11]. The bacterium was grown and the DNA was extracted according to Ramos et al. [12].

Construction of libraries and sequencing

The steps to prepare the library included enzymatic DNA fragmentation into 400 bp fragments (Ion plus fragment library kit, PN#4471252 and Ion Shear Kit, PN#4471248), the binding of specific adapters, size selection performed on 2% E-Gel, PN#G661002, the amplification and dilution of the library, and, finally, emulsion in Ion Onetouch 2 with the Ion PGM Template OT2 400 kit, PN#4479882. The Ion OneTouch ES system was used in the enrichment step.

The Ion PGM Hi-Q Sequencing Solutions kit PN#A24569 was used for sequencing, along with the reagents available in the Ion PGM Hi-Q Sequencing Reagents Kit, PN#A24568. The sequencing was performed on an Ion 318 v2 chip, PN#4484354, following the manufacturer's recommended protocol for producing 400 bp reads.

Data analysis

The quality assessment for the raw data generated by sequencing using the two kits was performed using the FastQC tool (http://www. bioinformatics.babraham.ac.uk/), followed by data processing using the Fastx-toolkit (http://hannonlab.cshl.edu/).

De novo assembly

The assembly process for both datasets was performed using two assembly tools, MIRA version 4.0.2 (http://www.chevreux.org/) and SPADES version 3.1.1 (http://bioinf.spbau.ru/).

The parameters used in the SPADES assembler were as follows: -iontorrent that enable a pipeline specific to data of the Ion Torrent PGM platform; additionally and -careful the error correcting step was performed before the assembly process. For Mira, the default parameters were used.

Assembly quality assessment

The assembly quality was assessed using the QUAST computational tool (http://quast.bioinf.spbau.ru/), which uses an annotated reference genome as the template to validate gene completeness, misassemblies, GC content, and other statistical metrics.

The software also allows multiple assemblies to be analyzed together, thus facilitating comparisons between different assemblies. Therefore, it is possible to evaluate the performance of the Hi-Q enzyme for genome assembly.

Genome annotation

The genome annotation was performed using RAST platform [13].

Bacterial growth, the extraction of chromosomal DNA and library construction for sequencing were performed using the same protocol for both samples. The raw data obtained from the two datasets, Cp31fragments and Cp31Hi-Q, were evaluated using FastQC (Figure 1). This analysis showed the higher quality (above Phred 20) of the Cp31Hi-Q data throughout the reads, so no quality filter was applied to the data sets.

Figure 1: Quality assessment for the Cp31fragments (A) and Cp31Hi-Q (B) libraries.

The results for the assemblies obtained using the computational tools MIRA and SPADES are listed in Table 1. It was observed that even when using different assembly software, the Hi-Q library yielded better results than the library without Hi-Q. Furthermore, the SPADES tool yielded better values for N50, total contigs and total bases.

Table 1: Assembly results using the MIRA and SPADES assemblers.

The assemblies were evaluated using the QUAST software [14], and the Hi-Q data were of higher quality, as shown in Figure 2. Figure 2A shows the gene completion of all four assemblies, and it is evident that the two assemblies using Hi-Q (blue and purple) featured better values for completeness relative to the reference (black dashes), which had fewer contigs than the assemblies without the enzyme (red and green); those without the enzyme had more contigs and lower gene completeness. These results demonstrate that the Hi-Q enzyme yields better performances in the assembly process.

Figure 2: Assembly quality assessment. (A) Analysis of gene completeness and (B) Plot of contig length.

Furthermore, the use of the Hi-Q enzyme was associated with the generation of longer contigs (Figure 2B), which is reflected by the blue and purple lines representing contigs >200 kb and >500 kb, respectively. In contrast, the data without Hi-Q (red and green) had contigs well below 100 kb. These results demonstrate once again the higher quality of the data produced with the Hi-Q enzyme.

These results demonstrate a significant improvement in the chemistry used in the sequencing kit, despite the improvements made for the generation of 400 bp reads. According to Loman et al. [4], there was lower accuracy (with a rate below 60% for homopolymers with a length equal to or greater than six bases) in the assembly of data from the Ion Torrent PGM platform compared to the results obtained on the 454 GS Junior and MiSeq platforms. This result is directly linked to difficulty in recognizing homopolymers.

After compare the genome annotation of C. pseudotuberculosis 31 present in the NCBI database and that sequenced with Hi-Q enzyme, 12 pseudogenes were shared between them and a reduction of 3-fold in amount of pseudogenes in the sequence produced by Hi-Q enzyme was observed (Figure 3). The list used to produce the Venn graph (Figure 3) is available in supplementary material. As an example, Figure 4 shows how the Hi-Q data improved the assembly in homopolymeric regions and fixed a frame-shift.

Figure 3: Venn graph of annotated pseudogenes for the C. pseudotubeculosis 31 (Cp31) sequences. Cp31_Hi-Q: the sequence produced with Hi-Q enzyme; Cp31_NCBI: the sequence present in NCBI database; Shared: pseudogenes shared by both genome sequences.

Figure 4: Frame-shift correction by Ion Torrent PGM, sequenced by Hi-Q enzyme. The Artemis interface present a frame-shift generated by 3 deletions events which are highlighted (A). After performed the Blast against Corynebacterium pseudotuberculosis 258, we confirmed the deletions were wrong (C). The same region was represented by Ion Torrent PGM (B), but the deletions were fixed, which shows an example of the efficiency of Hi-Q enzyme to address the problems to represent homopolymeric regions.

To evaluate how the Hi-Q improved the results, another analysis was performed using the data produced by the same sequencer (Ion Torrent PGM) using mate-pair and the traditional enzyme, and then we compared it to Hi-Q genome assembly. Despite using a mate-paired library (Cp31_mate), the final results show an inferior performance when compared to the assembly statistics of the Hi-Q enzyme data (Table 2), which presented the highest N50 and base pairs.

Table 2: Comparasion between mate-pair and Hi-Q libraries.

The efficiency of the Hi-Q enzyme was initially evident in the quality assessment of the raw data: better quality data were obtained from the Cp31Hi-Q library than data from the Cp31fragments library that was sequenced without the Hi-Q enzyme.

After the process of assembly an assessment of the completeness of the genes combined with the statistical parameters of the assembly is essential for assessing the accuracy of the assembly process. These analyses revealed better results for data produced with the Hi-Q enzyme on all metrics evaluated.

This result directly implies a reduction of frameshift errors, given that the number of broken genes is directly proportional to the number of such errors in the annotation.

Thus, the use of the Hi-Q enzyme is expected to reduce problems associated with sequencing errors and other experiments that rely on the Ion Torrent PGM sequencing platform.

This work was part of the Rede Paraense de Genômica e Proteômica supported by Fundação de Amparo a Pesquisa do Estado do Pará. AAOV, PHGS, VA, AS, RTJR were supported by Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq). KCP was supported by Fundação Amazônia Paraense de Amparo à pesquisa (FAPESPA).