Selecting targets from eukaryotic parasites for structural genomics and drug discovery

Abstract

The selection of targets is the first step for any structural genomics project. The application of structural genomics approaches to drug discovery also starts with the selection of targets. Here, three protocols are described that were developed to select targets from eukaryotic pathogens. These protocols could also be applied to other drug discovery projects.

1 Introduction

The success of the methods developed in early structural genomics projects led to the application of the approach to specific scientific problems. Early projects focused on the exploration of protein structure and selected targets for that purpose [1, 2]. More recently, there has been a transition to applying structural genomics methods to structure-aided drug discovery. The Seattle Structural Genomics Center for Infectious Disease (SSGCID) has focused on providing experimentally determined structures that serve as an initial step in the development process of structure-based drugs, vaccines, and diagnostics for infectious diseases [3, 4].

The SSGCID has focused its structure determination efforts on human pathogens, including bacterial, eukaryotic, and viral organisms from the National Institute of Allergy and Infectious Diseases (NIAID) list of biodefense organisms and those causing emerging and re-emerging diseases. Here, we describe three protocols for selecting targets as applied to seven eukaryotic pathogens: Babesia bovis, Coccidioides immitis, Cryptosporidium parvum, Encephalitozoon cuniculi, Entamoeba histolytica, Giardia lamblia, and Toxoplasma gondii.

The first protocol involves identification of potential drug targets in the seven eukaryotic species listed above. The approach involves searching for proteins with sequence similarity (>50 % over >75 % of their length) with protein targets in the DrugBank database [5]. DrugBank represents a comprehensive and publicly available resource that combines detailed drug (i.e., chemical) data with comprehensive target (i.e., protein) information [5]. The database contains over 6,700 drug entries and over 4,000 non-redundant protein sequences that are linked to these drug entries. Selecting proteins with sequence similarity to known drug targets substantially increases the likelihood that selected proteins are “druggable.” In addition, knowledge of chemical ligands (e.g., the drugs that act against their DrugBank homologues) that are likely to bind these proteins should increase their success in traversing the structure determination pipeline and provide ligands for co-crystallization. Determination of their three-dimensional structures will facilitate basic biomedical research by significantly shortening the time needed for development of novel chemotherapeutic agents. The application of this protocol yielded a total of 679 targets in the seven chosen eukaryotic pathogens.

The second protocol focuses on identifying representatives of hand-selected drug candidates. In this case, drug targets were obtained through a literature survey, discussions, and communications with pharmaceutical and academic researchers. A total of 93 targets representing 42 protein families from 32 organisms were collected. Orthologs were identified in the above seven eukaryotic pathogens using OrthoMCL clustering. OrthoMCL is a genome-scale algorithm for grouping orthologous protein sequences [6]. It provides not only groups shared by two or more species/genomes but also groups representing species-specific gene expansion families. This protocol yielded a total of 65 targets in the seven chosen eukaryotic pathogens.

The third protocol uses the TDRtargets public repository [7]. The TDRtargets project has collected diverse information relevant to drug target identification for a variety of important human pathogens [7] and provides a website where researchers can look for information on targets of interest. In addition, by using the TDRtargets database tools, researchers can quickly prioritize genes of interest by running simple queries (such as looking for small enzymes or proteins with high-quality structural models), assigning numerical weights to each query (in the history page), and combining these results to produce a ranked list of candidate targets. This protocol yielded a total of 614 targets in the eukaryotic pathogens of the genera Babesia, Brugia, Cryptosporidium, Leishmania, and Trypanosoma.

The methods in this chapter describe the following steps of the target selection strategy: (1) creating the reference genome sequence dataset, (2) selecting candidate targets, and (3) filtering out sequences not conducive to structural genomics approaches. Data management is a major component of target selection in structural genomics; however, it is outside the scope of this chapter.

2 Materials

1. Computer running the UNIX operating system, Internet connection (see Note 1).

2. Installed bioinformatics software: NCBI blast [8], OrthoMCL [6], Phobius [9], or TMHMM [10] (see Note 2).

3. Proficiency in a scripting language such as Python or Perl for parsing and combining the results of the target selection steps. A workable alternative that has been tested is the cloud-based database service SQLShare [11].

3 Methods

3.1 Create the Reference Genome Sequence Dataset

1. Select a representative strain for each genus (see Note 3).

2. Download CDS (i.e., DNA) and protein sequences in fasta format from EupathDB [12], which stores all organisms described in the introduction, except for Coccidioides, for which the sequences were downloaded from the Broad Institute (see Note 4).

3.2 Select Candidate Targets

3.2.1 DrugBank Homologues

1. Download target protein sequences from DrugBank in fasta format.

2. Perform a sequence similarity search using BlastP of these reference sequences against DrugBank, and keep the hits with at least 50 % similarity over 75 % of their length (see Note 5).

3. The remaining protein sequences were Jaccard clustered [13] to remove paralogs that shared >75 % similarity over 75 % of their length.

3.2.2 Representatives of Known Drug Targets

1. Obtain protein sequences of nominated drug candidates from the relevant repository if a database identifier is provided or search the UniProt database (see Note 6).

2. Combine sequences from the first step with reference protein sequences obtained in Subheading 3.1, and perform all-against-all BlastP, followed by OrthoMCL clustering using a Markov inflation index of 1.2 (see Note 7).

3. Select sequences from the reference genomes that cluster with the original nominated sequences from step 1.

3.2.3 TDR Targets

1. Query database using the TDRtargets “search for targets” web form http://tdrtargets.org/search (see Note 8).

2. Download results. In the horizontal menu on top of the page, click on “my queries.” The results are listed under “My target queries,” click on the “Export” link, and export using the default format.

3. Download sequences from the Source Database (see Note 9).

3.3 Remove Targets Not Conducive to Structural Genomics

1. Remove targets containing introns unless cDNA is available (see Note 10).

2. Screen proteins with known structure or those selected by other structural genomics centers to remove targets showing greater than 95 % conservation and 95 % coverage to targets in the Structural Biology Target Registration Database (TargetTrack, formerly TargetDB [14]) and sequences in the Protein Data Bank [15] by performing a BlastP search against these two databases (see Note 11).

3. Remove targets that contain transmembrane domains predicted by TMHMM or Phobius (see Note 12), except for N-terminal signal sequences, which are removed before PCR amplification.

4. Remove targets that are longer than 750 amino acid residues in length and have a cysteine content greater than 10 (see Note 13). Those criteria are “rules of thumb,” but it is known that limiting the number of cysteine residues decreases the likelihood of protein aggregation.

Established high-throughput centers will run these steps routinely using customized automated software pipelines. The SSGCID uses the Ergatis workflow management system that executes jobs in parallel on a computer cluster [16]. Another popular tool is the Galaxy platform [17].