ABSTRACT
We have designed a high-throughput system for the identification of novel crystal protein genes (cry) from Bacillus thuringiensis strains. The system was developed with two goals: (i) to acquire the mixed plasmid-enriched genomic sequence of B. thuringiensis using next-generation sequencing biotechnology, and (ii) to identify cry genes with a computational pipeline (using BtToxin_scanner). In our pipeline method, we employed three different kinds of well-developed prediction methods, BLAST, hidden Markov model (HMM), and support vector machine (SVM), to predict the presence of Cry toxin genes. The pipeline proved to be fast (average speed, 1.02 Mb/min for proteins and open reading frames [ORFs] and 1.80 Mb/min for nucleotide sequences), sensitive (it detected 40% more protein toxin genes than a keyword extraction method using genomic sequences downloaded from GenBank), and highly specific. Twenty-one strains from our laboratory's collection were selected based on their plasmid pattern and/or crystal morphology. The plasmid-enriched genomic DNA was extracted from these strains and mixed for Illumina sequencing. The sequencing data were de novo assembled, and a total of 113 candidate cry sequences were identified using the computational pipeline. Twenty-seven candidate sequences were selected on the basis of their low level of sequence identity to known cry genes, and eight full-length genes were obtained with PCR. Finally, three new cry-type genes (primary ranks) and five cry holotypes, which were designated cry8Ac1, cry7Ha1, cry21Ca1, cry32Fa1, and cry21Da1 by the B. thuringiensis Toxin Nomenclature Committee, were identified. The system described here is both efficient and cost-effective and can greatly accelerate the discovery of novel cry genes.
INTRODUCTION
Bacillus thuringiensis is a ubiquitous Gram-positive, spore-forming bacterium that produces parasporal crystals during the stationary phase of its growth cycle (42). The crystals comprise one or more crystal proteins (encoded by cry or cyt genes) that show specific toxicity against several orders of insects, including Lepidoptera, Diptera, Coleoptera, Hymenoptera, Homoptera, Orthoptera, and Mallophaga, and also against nematodes, mites, and protozoa (19, 20, 42). A portion of B. thuringiensis strains also secrete vegetative insecticidal proteins (VIPs) showing activity against Lepidopteran insect larvae (18). Generally, cry genes are located on plasmids, while few of them are reported to reside on the chromosome (31). Since the cloning of the first cry gene by Schnepf and Whiteley (43), more than 700 B. thuringiensis toxin genes (including 586 cry genes in 70 primary ranks, 96 vip genes in 4 primary ranks, and 34 cyt genes in 3 primary ranks; http://www.lifesci.sussex.ac.uk/home/Neil_Crickmore/Bt/) have been isolated and classified according to the nomenclature system described by Crickmore et al. (14).
The insecticidal activity of Cry proteins has led to the global development of bioinsecticides based upon B. thuringiensis for pest control (42). The bacterium is also a key source of genes for transgenic expression to provide pest resistance in plants (6, 26). However, the continuous use of B. thuringiensis products leads to resistance being evolved by insects (1, 22, 47, 49). Several strategies, such as the use of multiple toxins (9, 42, 50), spatial or temporal refugia (2, 27), and high or ultrahigh doses (27), have been employed to delay insect resistance to transgenic plants. The search for novel toxins with high toxicity is considered one of the major approaches to counter the potential resistance evolved by insects as well as in developing products against a wider spectrum of insect pests.
A series of approaches have been utilized for isolating novel cry genes, such as PCR, initially used by Carozzi et al. (10), for predicting the insecticidal activity of previously uncharacterized B. thuringiensis strains, which was followed by variations such as PCR hybridization (29), PCR-RFLP (restriction fragment length polymorphism) (32), E-PCR (exclusive PCR) (28), and PCR-SSCP (PCR-single-stranded conformation polymorphism profiling) (34). The construction of B. thuringiensis DNA libraries in Escherichia coli, followed by screening by Western blotting (36, 43) or a hybridization-based method (5, 30, 33, 35), or the development of DNA libraries in an acrystaliferous mutant of B. thuringiensis followed by microscopic observation and/or SDS-polyacrylamide gel (SDS-PAGE) detection of expressed genes in our laboratory (23) have also been used to detect novel cry protein genes.
In all of these methods, PCR-based systems are the most widely used for the identification of novel cry genes (7, 28, 37, 48). The design of these systems is based on five conserved blocks originally reported by Hofte and Whiteley (26). There are several limitations to these systems. For instance, all of these systems focus on three-domain cry genes, and most of them were limited to finding cry genes with sufficiently high sequence similarity to the primers used; thus, few of them were reported to be able to identify novel cry genes (with less than 45% amino acid sequence identity to the known cry genes) (37). Also, none of these systems are able to obtain the full-length cry gene sequences. The library-based methods are time-consuming and laborious. A general screening strategy for isolating cry genes therefore is required to identify more diverse sequences.
Recently, next-generation sequencing technology has been employed for the discovery of new cry genes (41). The major advantage of this biotechnology is that it provides a great deal of genomic data efficiently. Two issues remain to be resolved for such a strategy for the identification of new cry genes. One is that the average cost for identifying a cry gene is much higher than that of a PCR-based strategy, and the other is that no bioinformatics tool is publicly available for predicting cry genes from genomic sequences. A large number of protein classification algorithms are available, such as the BLAST method using pairwise local alignments to measure sequence similarity (3), the hidden Markov model (HMM) method based on multiple alignments generated by a statistical profile HMM (16), and the support vector machine (SVM) (15) method, which transforms protein sequences into fixed-length feature vectors. They represent three different kinds of protein prediction algorithms.
In this study, we describe a system for isolating new cry genes by combining mixed plasmid-enriched genome sequencing and a computational pipeline. The system was validated by using 21 B. thuringiensis strains from our laboratory. Finally, we identified three new cry-type genes (primary ranks) and five cry-holotype genes (>45% amino acid sequence identity to the known cry genes).
MATERIALS AND METHODS
Strain selection.Most cry genes are located on large plasmids (31). For the purpose of finding novel cry genes efficiently, strains that harbor abundant plasmids and could produce parasporal inclusions were our preferred strains. We screened more than 100 strains, assessing their plasmid profile and/or parasporal inclusion formation. Finally, 21 candidate B. thuringiensis strains were selected for further analysis.
Plasmid-enriched genomic DNA preparation and Illumina sequencing.These selected strains were subjected to plasmid-enriched genomic DNA extraction by incubating the cells with lysozyme (20 mg/ml) in TE (50 mM Tris base, 10 mM EDTA, 20% sucrose, pH 7.5) at 37°C, with shaking at 75 rpm for 2.5 h. Samples were then subjected to alkaline lysis (39) and further purified through ultracentrifugation in the presence of cesium chloride and ethidium bromide (40). After ultracentrifugation, the plasmid DNA (the closed circular plasmid DNA) was identified and withdrawn slowly from the tube. Ethidium bromide was removed from the solution of DNA by repeated extraction with organic solvents, and cesium chloride was removed by ethanol precipitation. The DNA precipitates were then washed twice with 70% ethanol and evaporated at room temperature. Finally, the DNA was dissolved in 400 μl of Tris-EDTA (pH 7.5) buffer and analyzed using pulsed-field gel electrophoresis to determine the concentration. The total plasmid-enriched genomic DNA from 21 candidate B. thuringiensis strains then were mixed together in equal amounts for sequencing.
A library for Illumina paired-end sequencing was prepared from 5 μg mixed plasmid-enriched DNA using a paired-end DNA sample preparation kit (PE-102-1001; Illumina Inc.). The DNA was fragmented by hydrodynamic shearing to generate <800-bp fragments. For end repair and phosphorylation, sheared DNA was purified using a QIAquick PCR purification kit (28104; Qiagen). The end-repaired DNA was A-tailed, and adaptors were ligated according to the manufacturer's instructions. The products of this ligation reaction were size selected by agarose gel electrophoresis and purified. The 5′ adaptor extension and enrichment of the library were performed using 10 PCR cycles with primers PE1.0 and PE2.0, which were supplied by Illumina. The library was finally purified using a QIAquick PCR purification kit and adjusted to an appropriate concentration. The flow cell was prepared according to the manufacturer's instructions using a paired-end cluster generation kit (PE-103-1001; Illumina Inc.) and a Cluster Station. Sequencing reactions were performed on a 1G GA2 equipped with a paired-end module (Illumina Inc.). De novo assembly was done by ABySS (45) using the following parameters: kmer (a parameter of ABySS software), 25; and n, 10.
Computational pipeline construction.To facilitate the identification of cry gene processes, a computational pipeline named BtToxin_scanner was constructed. The program consists of BioPerl software modules (46) and freely available third-party software, including a stand-alone BLAST application (8), HMMER 3.0 (17), and LIBSVM 2.91 (12). The cry database and the background database were integrated into the pipeline. Figure 1 shows the experimental strategy. The details are below.
Workflow for computational pipeline construction.
Preprocessing of input sequences.Based on different types of sequence input, corresponding modules would be called to convert the input sequences into protein sequences. For nucleotide sequences obtained by next-generation sequencing technology, a six-frame translation module was employed to find all possible protein sequences. We did not use a direct open reading frame prediction module, since it might lose dozens of potential cry sequences.
Length filter.Short sequences did not provide any information for the cloning of cry genes, thus sequences with lengths of less than 115 amino acids were eliminated.
Cry candidate prediction module.cry sequences were predicted with three modules: BLAST, HMM, and SVM. The BLAST database was constructed with the data set manually maintained by Crickmore et al. (http://www.lifesci.sussex.ac.uk/home/Neil_Crickmore/Bt/) (B. thuringiensis toxins are available at http://bcam.hzaubmb.org/BtToxin_scanner/), and the expected cutoff value was set to 1e−25. The HMM model (Cry.hmm, Cyt.hmm, and Vip.hmm; all available at http://bcam.hzaubmb.org/BtToxin_scanner/) of Cry proteins was built with the HMMER 3.0 package, and the expected cutoff value was set to 1e−10. The SVM classifier for Cry proteins was developed with amino acid and dipeptide composition using LIBSVM 2.91 (12). We implemented machine learning with radial basis function (RBF) for the training of various models. The training models were optimized for best performance by adjusting the kernel parameter gamma (G = 100) and the regularization parameter C (C = 100) (the SVM model is available at http://bcam.hzaubmb.org/BtToxin_scanner/). Any sequences that passed either of the prediction modules were considered candidate Cry proteins. Candidate Vip protein sequences and Cyt protein sequences were predicted only with the HMM prediction module, and the expected cutoff value was set to 1e−10.
Background elimination.To reduce the number of false-positive Cry protein sequences that would arise as a result of using all three prediction methods, a further background elimination step was employed by screening candidate Cry proteins against the background database and eliminating those proteins with significant similarity (less than 1e−30) (designated the background data set; available at http://bcam.hzaubmb.org/BtToxin_scanner/). The final results become available as a detailed prediction report along with the corresponding protein sequences.
Gene cloning, expression, and microscopic observation.Three strategies were applied for the cloning of novel cry genes from pipeline outputs. For candidate cry sequences with an intact promoter and terminator, they were directly amplified by PCR and then were ligated to the Escherichia coli-B. thuringiensis shuttle vector pHT304 (4). For those that only contained full-length cry genes, the entire genes were amplified, and then they were ligated to vector pBMBL (a plasmid derived from pHT304 and containing a promoter and terminator from cry1Ac10; unpublished data). For candidate cry sequences with only partial segments, either a manual assembly (the contig was searched against the original reads, and those reads that could extend the contig were then extracted and assembled to the contig) or reverse PCR (Fig. 2) followed by sequencing was performed to obtain the remaining coding sequences. After that, recombinant vectors containing the amplified cry genes were introduced by electroporation (39) into an acrystaliferous strain, BMB171 (25), for expression. Parasporal inclusions were observed by phase-contrast microscopy, and the major protein bands were detected by SDS-PAGE analysis.
Schematic diagram of primer design for reverse PCR. The solid dark line represents the contig that needs to be extended, and the dotted line represents DNA downstream of the cloned cry gene. PF means forward primer, and PR means reverse primer. The plasmid DNA was digested completely by a restriction enzyme (such as EcoRI), and the DNA sample was self ligated overnight. The region downstream of the cloned gene was amplified using the PR and PF primers.
Nucleotide sequence accession numbers.The nucleotide sequences published in this paper have been submitted to GenBank and assigned accession numbers JF521572 (cry8Ac1), JF521575 (cry7Ha1), JF521577 (cry21Ca1), JF521578 (cry21Da1), JF521580 (cry32Fa1), JF521582 (Toxin6), JF521583 (Toxin7), and JF521584 (Toxin8).
RESULTS
Construction and validation of the computational pipeline for predicting cry genes.To identify cry genes from genomic sequences efficiently, a computational pipeline named BtToxin_scanner was constructed as described in Materials and Methods (Fig. 1). The performance of BtToxin_scanner was evaluated using two data sets downloaded from GenBank; one includes genomic sequences of B. thuringiensis, which contains a large number of cry genes, and the other includes genomic sequences of Bacillus cereus, which is supposed to be free of cry genes.
BtToxin_scanner shows high sensitivity using data set A with genomic sequences from B. thuringiensis.We tested whether BtToxin_scanner was able to identify B. thuringiensis toxins from genomic sequences by using data set A, which contained 3 completed genome sequences, 15 whole-genome shotgun (WGS) sequences, and 24 complete plasmid sequences from B. thuringiensis (see Table S1 in the supplemental material). We identified 85 candidates from this data set with an average process speed of 1.03 Mb/min. These sequences were downloaded from GenBank, thus they contained annotation information. We extracted protein sequences whose annotation related to B. thuringiensis toxins and compared it to the result obtained with BtToxin_scanner. It was found that BtToxin_scanner correctly identified all of the B. thuringiensis toxins in the keyword extraction list. It also identified an additional 35 candidates (Fig. 3A), among which 20 were considered true positives, 4 of them were marked as unknown due to a lack of homology to any proteins from GenBank, and the remaining 11 were identified as false positives (see Table S2 in the supplemental material).
Comparison of BtToxin_scanner result and keyword extraction results. (A) Venn diagram of comparison of BtToxin_scanner result and keyword extraction result. Inner circle, protein number obtained with BtToxin_scanner; outer circle, protein number obtained with keyword extraction. (B) Categories for the 35 proteins identified only by BtToxin_scanner. Bt toxins indicates B. thuringiensis toxin proteins correctly identified by BtToxin_scanner but missing from the keyword extraction result. Non-Bt toxins indicates non-B. thuringiensis toxin proteins identified by BtToxin_scanner. Unknown indicates proteins identified by BtToxin_scanner with unknown function.
The WGS sequences in data set A were obtained with a run of 454 sequencing technology, with the average cry gene number per run computed as 2.6. We did not include 3 complete genome sequences of B. thuringiensis strain BMB171 (25), B. thuringiensis strain Al Hakam (11), and B. thuringiensis serovar konkukian strain 97-27 (24), since these strains did not produce a parasporal crystal.
BtToxin_scanner shows high specificity using data set B with genomic sequences from B. cereus.Since BtToxin_scanner integrated the three prediction methods, it was important to assess how many false positives were produced. Thus, we evaluated the performance of BtToxin_scanner on data set B (see Table S3 in the supplemental material), which includes 190,062 protein sequences from B. cereus WGS sequences. We identified 11 candidates from data set B with an average process speed of 1.02 Mb/min (protein sequences). These sequences were analyzed by searching against GenBank, and it was found that they contained 6 false positives, 1 unknown protein, 1 Cry protein, and 3 Vip proteins (see Table S4 in the supplemental material).
Mining cry genes from genome sequencing of mixed plasmid-enriched total DNAs from 21 B. thuringiensis strains.As described above, BtToxin_scanner was able to identify B. thuringiensis toxins from genomic sequence efficiently. Thus, we focused on acquiring large numbers of genomic sequences with abundant B. thuringiensis toxin sequences in an efficient way.
Selection of strains.The selection of the 21 working strains was based on a combination of the plasmid pattern, the formation of parasporal inclusion, and the major protein profile detected by SDS-PAGE.
Acquisition of genomic sequences.Plasmid-enriched genomic DNA was extracted and mixed as described in Materials and Methods and then loaded onto the Illumina GA2 sequencer. The raw sequence data were de novo assembled, thus generating the mixed plasmid-enriched genomic sequence data.
A total of 6,070,863 paired-end reads of 100 bp were generated, with an average length of inserts of paired-end reads at 174 bp. All short paired-end reads generated by the Illumina genome analyzer were inputted into the de novo assembly software ABySS (45). A total of 9,846 contigs were produced with a predefined cutoff rate (N50, 1,365 bp; maximum contig size, 58,460 bp). The total size of the produced contigs was 9,884,426 bp, with 2,365 contigs having a length greater than 1 kbp (Table 1) (mixed plasmid-enriched genomic sequences are available at http://bcam.hzaubmb.org/BtToxin_scanner/).
Summary of mixed plasmid-enriched genomic sequence reads and contigs
Prediction of candidate cry gene sequences.Sequence data generated from the mixed plasmid-enriched genome sequencing project was analyzed with BtToxin_scanner, and 143 candidates were identified with an average process speed of 1.80 Mb/min (nucleotide sequences). (The detail report and the sequence file of mixed plasmid-enriched genomic sequence prediction results are available at http://bcam.hzaubmb.org/BtToxin_scanner/.) Among those 143 candidates, there were 113 candidate cry sequences, 23 candidate vip sequences, and 7 candidate cyt sequences. Among the 113 candidate cry sequences, 48 of them showed highest-hit identities of less than 45% to the known cry genes, and 36 hits showed identities of between 45 and 78%. The BLAST module identified 89 cry hits, and the level of identity of those hits to the known cry genes ranged from 27 to 100%. cry32-type hits were the most frequently found among the 89 sequences analyzed, with a frequency of 19%, followed by the cry1-type hits, with a frequency of 18%. These results indicated the existence of multiple potential novel cry genes.
Sequence analysis and characterization of full-length cry genes.Based on the sequence identity to known cry genes and sequence length, we chose 27 sequences from the BtToxin_scanner prediction results. We used different strategies to clone these novel cry genes as described in Materials and Methods. Finally, 8 of them were obtained with full-length sequences. The recombinant plasmids were resequenced and, as expected, all of them shared 100% amino acid sequence identity with the original sequencing data. Their sequences were submitted to the Bacillus thuringiensis delta-endotoxin nomenclature committee. Five of them have received official names (Cry8Ac1, Cry7Ha1, Cry21Ca1, Cry32Fa1, and Cry21Da1), while the remaining 3 were considered novel, since they shared less than 45% amino acid sequence identity to the known cry proteins. Basic information on these selected cry genes is listed in Table 2, and the complete sequences of these proteins are provided as Data set S1 in the supplemental material.
Description of novel crystal proteins from BtToxin_scanner prediction
Cry8Ac1, Cry7Ha1, Cry21Ca1, Cry32Fa1, Cry21Da1, and Toxin7 sequences were considered to constitute complete toxins. Toxin6 protein and Toxin8 protein sequences were considered to constitute naturally truncated proteins, as both of them contained only 802 amino acids. Interestingly, the C-terminal sequence of Toxin6 is similar to that of Clostridium botulinum hemagglutinin HA-17 (21). A BLAST search against GenBank revealed that the Toxin8 protein shares 67% sequence identity to the 83-kDa crystal protein from B. cereus strain AH603, which is not included in the B. thuringiensis Toxin Nomenclature Committee system.
All of these toxin genes were ligated into pHT304 and used to transform the acrystaliferous strain BMB171 for expression. One of them (cry8Ac1) could produce bipyramidal parasporal inclusion (Fig. 4A), and SDS-PAGE analysis revealed the correct size with a predicted molecular mass of 138 kDa (Fig. 4C). The cry8Ac1 gene was amplified from B. thuringiensis strain Sbt006. Sbt006 forms a spherical parasporal inclusion and a >140-kDa protein band as detected by SDS-PAGE (Fig. 4B and C). The difference in protein size and parasporal inclusion shape between the recombinant strain and Sbt006 implies that cry8Ac1 is cryptic in Sbt006. The expression of the other seven crystal genes did not succeed in BMB171.
Phase-contrast micrographs (magnification, ×1,000) and SDS-PAGE analysis of crystal proteins from parent and recombinant strains of B. thuringiensis. (A) Parent strain Sbt006; (B) strain BMB0659, harboring the 138-kDa protein gene cry8Ac1 from Sbt006. (C) Lane M, molecular mass standard; lane 1, BMB0659; lane 2, parent strain Sbt006. The arrow points to the position of crystal protein Cry8Ac1.
Construction of cry gene recognition web server.To make the pipeline method accessible for the broader biological community, we implemented it as a user-friendly web server accessible at http://bcam.hzaubmb.org/BtToxin_scanner. The common gateway interface (CGI) script for BtToxin_scanner was written using PERL version 5.10. The server was installed on an Apache server (2.2.16) in a Linux (Ubuntu 10.10 server) environment. The server accepts three kinds of sequence (proteins, open reading frames [ORFs], and nucleotide sequences). Users can submit sequences for prediction using file uploading. It might take a few minutes (depending on the sequence file size, the average processing speed was 1.02 Mb/min for protein sequences and ORFs and 1.8 Mb/min for nucleotide sequences) to complete the whole job. The output for each run is displayed in a user-friendly table format as well as two downloadable links, one for the detailed information about the prediction and the other for the protein sequences.
The underlying cry database and the background database are updated on a regular basis. The profile HMMs and the SVM model are also updated manually as the new Cry protein members identified by experimental studies are reported.
DISCUSSION
In this study, we combined three different kinds of well-developed algorithms, BLAST, HMM, and SVM, to increase sensitivity, and we used background elimination to increase specificity. We then used two data sets to evaluate the performance of BtToxin_scanner, a computational pipeline, and we proved that it is able to distinguish true positives from negatives efficiently. A small, manually curated data set was used in our pipeline method, which reduced the error level that exists with public databases, such as GenBank (44), and also reduced the time required for the whole prediction process.
Theoretically, next-generation sequencing technology produces large numbers of sequences indiscriminately and quickly, and the computational pipeline could identify all kinds of B. thuringiensis toxin proteins, including Cry proteins, Vip proteins, and Cyt proteins. Thus, it is expected that our system is more efficient in identifying all kinds of B. thuringiensis toxins than PCR-based systems. In fact, we successfully identified 143 candidate B. thuringiensis toxins from the mixed plasmid-enriched genomic sequences. Eight of them were obtained with full-length gene sequences.
A major issue that remains to be resolved is how many strains can simultaneously be analyzed by this method. Previous reports showed that B. thuringiensis strain YBT-1520 contains a total plasmid genetic content of 988 kb (51). To acquire high-quality sequencing data, we mixed 21 plasmid-enriched genomic DNA sequences from B. thuringiensis strains for Illumina sequencing (>50× coverage), and theoretically the average cost for the identification of a cry gene would be about 5% of that of a whole-genome shotgun sequencing strategy. We successfully identified 113 candidate Cry sequences, thus the average cost for the identification of a cry gene was reduced significantly.
We noticed that more than 80% of the sequences from the mixed plasmid-enriched sequencing project were shorter than 1 kbp. Previously, we accomplished the genome sequencing of several strains with the same sequencing strategy. We found that the sequence quality is poor (with more contigs and shorter contig length) when the strain harbors more plasmids (unpublished data). We suppose that the poor sequence quality of the mixed plasmid-enriched sequence data is due to both the abundant plasmids and the sequencing technology itself. The complexity of the plasmid sequence, such as repeat elements, and genes with high sequence identity increases the difficulty of the assembly process. We also used an Illumina sequencing platform, which produces short sequence reads (100 bp). In practice, the assembly of shorter sequence reads yields poorer quality assemblies than those with longer capillary reads (38). The poor sequence quality inflates the number of hits being identified and also increases the time and workload needed to clone the complete coding sequence. The only way to solve the poor sequence quality would be the improvement of the sequencing technology itself.
B. thuringiensis is distinguished from B. cereus because it can produce parasporal crystals (42). However, we found that B. cereus AH603 contained a full-length gene coding for a Cry protein. This strain was originally isolated from a dairy in Norway. The formation of a parasporal crystal phenotype in this strain either has not been established or has not been reported. Thus, whether this strain should be identified as Bacillus cereus or Bacillus thuringiensis is uncertain based on available information.
In conclusion, we have established a system combining mixed plasmid-enriched genome sequencing and a computational pipeline to mine cry genes from B. thuringiensis. The system was able to evaluate 21 B. thuringiensis strains in a fast and efficient way. A total of 113 candidate Cry sequences were extracted from the 21 strains, and 8 of them were identified. Among them, 3 potentially represent novel cry gene types (primary ranks) and 5 of them became cry holotypes. These results proved the efficiency of this system to mine cry genes. Indeed, the mining of novel sequences must be related to the previous strain selection. Still, it is important to note that the selection of the sequencing strategy affects the final prediction results, thus a choice has to be made between cost and efficiency. Additionally, we have developed a computational pipeline, BtToxin_scanner, which is publicly available at http://bcam.hzaubmb.org/BtToxin_scanner.
ACKNOWLEDGMENTS
This work was supported by the National High Technology Research and Development Program (863) of China (2011AA10A203 and 2006AA02Z174), the National Basic Research Program (973) of China (2009CB118902), the National Natural Science Foundation of China (31170047, 30870066, and 31000020), the Genetically Modified Organisms Breeding Major Projects of China (2009ZX08009-032B), and China 948 Program of the Ministry of Agriculture (2011-G25).
We thank the BBSRC for facilitating a closer collaboration with Neil Crickmore through their China Partnering Award scheme. We thank Zheng Jinshui for his help in website construction.
FOOTNOTES
- Received 5 February 2012.
- Accepted 23 April 2012.
- Accepted manuscript posted online 27 April 2012.
Supplemental material for this article may be found at http://dx.doi.org/10.1128/AEM.00340-12.
- Copyright © 2012, American Society for Microbiology. All Rights Reserved.
