Novel Sequence Discovery by Subtractive Genomics

Podsumowanie

The purpose of this protocol is to use a combination of computational and bench research to find novel sequences that cannot be easily separated from a co-purifying sequence, which may be only partially known.

Streszczenie

Subtractive genomics can be used in any research where the goal is to identify the sequence of a gene, protein, or general region that is embedded in a larger genomic context. Subtractive genomics enables a researcher to isolate a target sequence of interest (T) by comprehensive sequencing and subtracting out known genetic elements (reference, R). The method can be used to identify novel sequences such as mitochondria, chloroplasts, viruses, or germline restricted chromosomes, and is particularly useful when T cannot be easily isolated from R. Beginning with the comprehensive genomic data (R + T), the method uses Basic Local Alignment Search Tool (BLAST) against a reference sequence, or sequences, to remove the matching known sequences (R), leaving behind the target (T). For subtraction to work best, R should be a relatively complete draft that is missing T. Since sequences remaining after subtraction are tested through quantitative Polymerase Chain Reaction (qPCR), R does not need to be complete for the method to work. Here we link computational steps with experimental steps into a cycle that can be iterated as needed, sequentially removing multiple reference sequences and refining the search for T. The advantage of subtractive genomics is that a completely novel target sequence can be identified even in cases in which physical purification is difficult, impossible, or expensive. A drawback of the method is finding a suitable reference for subtraction and obtaining T-positive and negative samples for qPCR testing. We describe our implementation of the method in the identification of the first gene from the germline-restricted chromosome of zebra finch. In that case computational filtering involved three references (R), sequentially removed over three cycles: an incomplete genomic assembly, raw genomic data, and transcriptomic data.

Wprowadzenie

The purpose of this method is to identify a novel target (T) genomic sequence, either DNA or RNA, from a genomic context, or reference (R) (Figure 1). The method is most useful if the target cannot be physically separated, or it would be expensive to do so. Only a few organisms have perfectly finished genomes for subtraction, so a key innovation of our method is the combination of computational and bench methods into a cycle enabling researchers to isolate target sequences when the reference is imperfect, or a draft genome from a non-model organism. At the end of a cycle, qPCR testing is used to determine whether more subtraction is needed. A validated candidate T sequence will show statistically greater detection in known T-positive samples by qPCR.

Incarnations of the method have been implemented in discovery of new bacterial drug targets that do not have host homologs^1,2,3,4 and identification of novel viruses from infected hosts^5,6. In addition to identification of T, the method can improve R: we recently used the method to identify 936 missing genes from the zebra finch reference genome and a new gene from a germline-only chromosome (T)⁷. Subtractive genomics is particularly valuable when T is likely to be extremely divergent from known sequences, or when the identity of T is broadly undefined, as in the zebra finch germline-restricted chromosome⁷.

By not requiring positive identification of T beforehand, a key advantage of subtractive genomics is that it is unbiased. In a recent study, Readhead et al. examined the relationship between Alzheimer's disease and viral abundance in four brain regions. For viral identification, Readhead et al. created a database of 515 viruses⁸, severely limiting the viral agents that their study could identify. Subtractive genomics could have been used to compare the healthy and Alzheimer's genomes in order to isolate possible novel viruses associated with the disease, regardless of their similarity to known infectious agents. While there are 263 known human-targeting viruses, it has been estimated that approximately 1.67 million undiscovered viral species exist, with 631,000-827,000 of them having a potential to infect humans⁹.

Isolation of novel viruses is an area in which subtractive genomics is particularly effective, but some studies may not need such a stringent method. For example, studies identifying novel viruses have used unbiased high-throughput sequencing followed by reverse transcription and BLASTx for viral sequences⁵ or enriching of viral nucleic acids to extract and reverse transcribe viral sequences⁶. While these studies employed de novo sequencing and assembly, subtraction was not used because the target sequences were positively identified through BLAST. If the viruses were completely novel and not related (or distantly related) to other viruses, subtractive genomics would have been a useful technique. The benefit of subtractive genomics is that sequences that are completely new can be obtained. If the organism's genome is known, it can be subtracted out to leave any viral sequences. For example, in our published study we isolated a novel viral sequence from zebra finch through subtractive genomics, though it was not our original intent⁷.

Subtractive genomics has also proved useful in the identification of bacterial vaccine targets, motivated by the dramatic rise in antibiotic resistance^1,2,3,4. To minimize the risk of autoimmune reaction, researchers narrowed down the potential vaccine targets by subtracting any proteins that have homologs in the human host. One particular study, looking at Corynebacterium pseudotuberculosis, performed subtraction of vertebrate host genomes from several bacterial genomes to ensure that possible drug targets would not affect proteins in the hosts leading to side effects¹. The basic work flow of these studies is to download the bacterial proteome, determine vital proteins, remove redundant proteins, use BLASTp to isolate the essential proteins, and BLASTp against host proteome to remove any proteins with host homologs^1,2,3,4. In this case, subtractive genomics ensure that the vaccines developed will not have any off-target effects in the host^1,2,3,4.

We used subtractive genomics to identify the first protein-coding gene on a germline-restricted chromosome (GRC) (in this case, T), which is found in germlines but not somatic tissue of both sexes¹⁰. Before this study, the only genomic information that was known about the GRC was a repetitive region¹¹. De novo assembly was performed on RNA sequenced from ovary and teste tissues (R+T) from adult zebra finches. The computational elimination of sequences was performed using published somatic (muscle) genome sequence (R₁)¹², its raw (Sanger) read data (R₂), and a somatic (brain) transcriptome (R₃)¹³. The sequential use of three references was driven by the qPCR testing at step 5 of each cycle (Figure 2A), showing that additional filtering was required. The discovered α-SNAP gene was confirmed through qPCR from DNA and RNA, and cloning and sequencing. We show in our example that this method is flexible: it is not dependent on matching nucleic acids (DNA vs RNA) and that subtraction can be performed with references (R) that are comprised of assemblies or raw reads.

Protokół

1. De novo Assemble Starting Sequence

NOTE: Any Next-Generation Sequence (NGS) data can be used, as long as an assembly can be produced from those data. Suitable input data includes Illumina, PacBio, or Oxford Nanopore reads assembled into a fasta file. For concreteness, this section describes an Illumina-based transcriptomic assembly specific to the zebra finch study we performed⁷; however be aware that the specifics will vary by project. For our example project, raw data were derived from a MiSeq and approximately 10 million paired reads were obtained from each sample.

1. Use Trimmomatic 0.32¹⁴ to remove Illumina adaptors and low-quality bases. On the command line, enter:
   java -jar trimmomatic-0.32.jar PE -phred33 forward.fq.gz reverse.fq.gz -baseout quality_and_adaptor_trimmed ILLUMINACLIP:TruSeq3-PE.fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:20 MINLEN:40
2. Use PEAR¹⁵ v. 0.9.6 to create high-quality merged reads from trimmomatic output paired reads, using default parameters. On the command line, enter:
   pear -f <quality_and_adaptor_trimmed_1P.fastq> -r <quality_and_adaptor_trimmed_2P.fastq>
3. Use Reptile v. 1.1¹⁶ to error-correct the reads produced through PEAR. Follow the step-by-step protocol described in¹⁷.
4. Use Trinity v. 2.4.0¹⁸ in default mode to assemble the corrected sequences. For strand-specific libraries, use the -SS_lib_type parameter. The output is a fasta file (your_assembly.fasta). On the command line, enter:
   Trinity --seqType fq --SS_lib_type FR –max_memory 10G –output Trinity_output --left quality_and_adaptor_trimmed_forward_paired_reads.fq –right quality_and_adaptor_trimmed_reverse_paired_reads.fq –CPU 10
   NOTE: The output will be placed in a new directory, Trinity_output, and the assembly will be named 'Trinity.fasta' which can be renamed as Your_assembly.fasta if desired. See the Trinity website for more details: https://github.com/trinityrnaseq/trinityrnaseq/wiki/Running-Trinity.

2. BLAST the Assembly against the Reference Sequence

NOTE: Use this step when the reference is an assembly or long reads like Sanger; if it is composed of raw Illumina reads, see step 3 below for mapping reads to the query. All BLAST steps were completed with version 2.2.29+ though the commands should work on any recent BLAST version.

1. Make a BLAST database of the reference sequence (nucleotide_reference.fasta) at the command line. Enter into the command line the following:
   makeblastdb -dbtype nucl -in nucleotide_reference.fasta -out nucleotide_reference.db
2. BLAST-match the query assembly (generated in step 1) to the reference database. To obtain an output file, use [-out BLAST_results.txt] and to generate tabular output (required for subsequent processing steps with Python scripts), use [-outfmt 6]. These options can be combined in any order, so an example complete command is [blastn -query your_assembly.fasta -db nucleotide_reference.db -out BLAST_results.txt -outfmt 6]. If an e-value setting is desired, use the -evalue option with an appropriate number, for example [-evalue 1e-6]. Be aware however that the subtractive cycle effectively inverts the evalue setting in as described in the discussion.
3. For increased stringency, use protein sequences from the assembly as the BLAST query with translated nucleotide BLAST (tBLASTn), which performs 6-way translation of the (nucleotide) database. This method is recommended for most non-model systems, avoiding the problem of incomplete protein annotations.
   1. Ensure the correct genetic code is selected for the organism being studied, using the -db_gencode option. To obtain protein sequences for the query, run the TransDecoder.LongOrfs command (from TransDecoder package v. 3.0.1) to identify the longest open reading frames from assembled query sequences. The command is [TransDecoder.LongOrfs -t your_assembly.fasta]; the output will be placed in directory called 'transcripts.transdecoder_dir' and will contain a file called longest_orfs.pep containing the longest predicted protein sequences from each sequence in your_assembly.fasta.
   2. To use tBLASTn, run the command [tblastn -query longest_orfs.pep -db nucleotide_reference.db -out BLAST_results.txt -outfmt 6]. If a high-quality protein reference is available, use protein-protein matching with BLASTp rather than tBLASTn.
   3. Make a BLAST database of the protein reference [makeblastdb -dbtype prot -in protein_reference.fasta -out protein_reference.db] and then [blastp -query longest_orfs.pep -db protein_reference.db -out BLAST_results.txt -outfmt 6]. Make sure to save the results as a file for downstream processing, and use tabular (outfmt 6) to ensure the Python scripts can parse them correctly.

3. Map Reads onto the Assembly

NOTE: This method can be used if the reference dataset consists of raw genomic reads, rather than assembled sequences or Sanger sequences, in which case use BLAST (step 2.1).

1. Using BWA -MEM v. 0.7.12¹⁹ or bowtie2²⁰, map the downloaded raw reads (raw_reads.fastq) onto the query assembly. The output will be .sam format. Commands are as follows: first index the assembly: [bwa index your_assembly.fasta], and then map the reads [bwa mem your_assembly.fasta raw_reads.fastq >mapped.sam]. (Note the '>' symbol here is not a greater-than sign; instead it instructs the output to go into the file mapped.sam).

4. Use Python Script to Remove any Matching Sequences

NOTE: Provided scripts work with Python 2.7.

1. Following Step 2, use subtractive Python script by using the command [./Non-matching_sequences.py your_assembly.fasta BLAST_results.txt]. Before running the script, ensure that the BLAST output file is in format 6 (tabular). The script will output a file with non-matching sequences in fasta format named your_assembly.fasta_non-matching_sequences_BLAST_results.txt.fasta and also the matching sequences for records, as your_assembly.fasta_matching_sequences_BLAST_results.txt.fasta. The non-matching file will be the most important, as a source of potential T sequences for testing and further cycles of subtractive genomics.
2. Following Step 3, run the Python script removeUnmapped.py to take as input the .sam from step 3.1, and identifies the names of query sequences without any matching reads and saves them to a new text file. Use the command [./removeUnmapped.py mapped.sam] and the output will be mapped.sam_contigs_with_no_reads.txt. (The program will generate a slimmed-down sam file with all unmapped reads removed; this file can be ignored for purposes of this protocol but may be useful for other analyses).
3. As the output of the previous step is a list of sequence names in a text file called mapped.sam_contigs_with_no_reads.txt, extract a fasta file with these sequences: [./getContig.py your_assembly.fasta mapped.sam_contigs_with_no_reads.txt]. The output will be a file called mapped.sam_contigs_with_no_reads.txt.fasta.

5. Design Primers for the Sequence that Remains

NOTE: At this point there is a fasta file containing candidate T sequences. This section describes qPCR to experimentally test whether they come from T or from previously unknown regions of R. If the subtraction in step 4 removed all sequences, then either the initial assembly failed to include T, or the subtraction may have been too stringent.

1. Use Geneious²¹ to determine optimal primer sequences manually.
   1. Highlight a candidate sequence of 21-28 bp for the Forward primer. Avoid runs of 4 or more of any base. Try to target a region with a fairly uniform combination of all basepairs. A single G or C at the 3' end is beneficial, helping to anchor the primer.
   2. Click on the Statistics tab on the right-hand side of the screen to view that sequence's estimated melting temperature (Tm) as the candidate region is highlighted. Look to obtain a melting temperature between 55-60 °C, while avoiding repeats and long runs of G/C.
   3. Follow steps 5.1.1. and 5.1.2 to choose a reverse primer, situated 150-250 base pairs 3' of the forward primer. While the primer lengths do not need to match, the predicted Tm should be as close as possible to the Tm of the forward primer. Be sure to reverse complement the sequence (if right-clicking in Geneious while the sequence is highlighted it is a menu option).
2. Use the Primer Design function, which is found in the top tool bar in the sequence window.
   1. Click on the Primer Design button. Insert the region to amplify under Target Region.
   2. Under the Characteristics tab, insert desired size, melting temperature (Tm), and %GC (see step 5.1.1.).
   3. Click OK to have primers generated. Order the primers through a custom oligo service.
3. Validate primers with control DNA (encoding both T and R) to optimize Tm and extension time. Use regular Taq and gel electrophoresis to see the band size, but optimization can also be performed with qPCR following the methods in step 6.
   1. Make 10X dilutions of both forward and reverse primers so that the primers have a concentration of 10 μM.
   2. Use a PCR mix of 0.5 μL of dNTP, 0.5 μL of forward primer, 0.5 μL of reverse primer, 0.1 μL of Taq polymerase, 2 μL of template, 0.75 μL of magnesium, 2.5 μL of buffer, and 18.15 μL of water so that there is 25 μL per template with a concentration of 5 ng/μL.
   3. Test the primers at different melting temperatures in the PCR program. Usually optimal performance is observed melt temperatures slightly below the predicted Tm of the primers, but not usually above 60 °C. Also test for optimal extension times using this guide: 1 min per 1000 bp (thus, usually 10-30 seconds depending on amplicon length).
   4. Perform end-point gel electrophoresis to confirm that the primers amplify the expected sequence. Run 25 μL of the qPCR product mixed with 5 μL of 6X glycerol dye on a 2% TAE agarose gel at 200 V for 20 min.

6. qPCR Validation of the Remaining Sequence

NOTE: This step requires primers validated and PCR conditions established in step 5.

1. Run each template in triplicate with the following mix; 12.5 μL of PowerSYBR Green master mix, 0.5 μL of forward primer with a concentration of 10 μM, 0.5 μL of reverse primer with a concentration of 10 μM, 10.5 μL of water, and 1 μL of template DNA (at a concentration of 2 ng/μL), so that each well contains 25 μL of total volume.
2. Run a qPCR program informed by the validated temperature and extension time from step 4. We designed and validated all primers to be compatible with a two-stage cycle, 95 °C for 10 min initial melt, then 40 cycles of 95 °C for 30 s and 60 °C for 1 min. However, a three-stage (melt-anneal-extend) program may be more optimal for the primers and should be adapted if necessary. We recommend that final denaturing curves be generated at least the first time the primers are employed in qPCR to validate the amplification of a single DNA product.
3. Measure qPCR/SYBR Green signals relative to actin (or any other suitable 'R' control) by Ct. For all cases calculate the average and standard deviation of 2^{-(gene Ct - β-actin Ct)}.
4. (Optional) Perform end-point gel electrophoresis to confirm correct product size detection by qPCR. Here, run 25 μL of the qPCR product mixed with 5 μL of 6x glycerol dye on a 2% TAE agarose gel at 200 V for 20 min.

7. Repeat with a New Reference to Pare Down the Data.

NOTE: If step 6 validated the identified sequences from T, end the cycle here (Figure 2A). However, a variety of considerations may motivate a continuation of the cycle, for example if many R sequences remain in the file or if none of the candidate T sequences were validated by qPCR in step 6.

1. Obtain a new reference. This step enables a new iteration of the cycle and may include raw genomic data, raw RNA-seq data, or other assembled datasets. Valuable resources for reference data include the Genome database at the National Center for Biotechnology Information (https://www.ncbi.nlm.nih.gov/genome) which stores assembled genomes accessible through FTP (ftp://ftp.ncbi.nlm.nih.gov/genomes/), and the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) where raw next-generation sequence reads are stored. Genome projects may provide their raw sequence data through other project-associated websites and databases.

Wyniki

After running BLAST, the output file will have a list of sequences from the query that match the database. After Python subtraction, a number of nonmatching sequences will be obtained, and tested by qPCR. The results of this, and next steps, are discussed below.

Negative result. There are two possible negative results that can be seen after BLAST to the reference sequence. There may be no BLAST results, meaning ...

Dyskusje

While subtractive genomics is powerful, it is not a cookie-cutter approach, requiring customization at several key steps, and careful selection of reference sequences and test samples. If the query assembly is of poor quality, filtering steps might only isolate assembly artifacts. Therefore, it is important to thoroughly validate the de novo assembly using an appropriate validation protocol to the specific project. For RNA-seq, guidelines are provided on the Trinity website¹⁸ and for DNA,...

Materiały

Name	Company	Catalog Number	Comments
Accustart II Taq DNA Polymerase	Quanta Bio	95141
Blasic Local Alignment Search Tool (BLAST)			https://github.com/trinityrnaseq/trinityrnaseq/wiki/Transcriptome-Assembly-Quality-Assessment
Bowtie 2			https://www.python.org/download/releases/2.7/
BWA-MEM v. 0.7.12			https://github.com/BenLangmead/bowtie2
Geneious			https://blast.ncbi.nlm.nih.gov/Blast.cgi
PEAR v. 0.9.6			http://www.mybiosoftware.com/reptile-1-1-short-read-error-correction.html
Personal Computer	Biomatters		http://www.geneious.com/
PowerSYBR qPCR mix	ThermoFisher	4367659
Python v. 2.7			https://sco.h-its.org/exelixis/web/software/pear/
Reptile v.1.1			https://alurulab.cc.gatech.edu/reptile
Stratagene Mx3005P	Agilent Technologies	401456
TransDecoder v. 3.0.1			https://sourceforge.net/projects/bio-bwa/files/
Trinity v. 2.4.0			https://github.com/TransDecoder/TransDecoder/wiki