The discovery of penicillin in 1928 by Alexander Fleming has revolutionized the treatment of bacterial infections. The large-scale use of antibiotics, however, has also led to an increase in the number of microbes that can resist treatment. Drug resistant bacteria are an increasing threat to public health, as highlighted by a recent estimate that in the US methicillin-resistant Staphylococcus aureus (MRSA) may contribute to more deaths than HIV (1). Methicillin-resistant strains of S. aureus were initially documented in the 1960s (2) and have been associated with higher mortality rates (3,4) than their drug-sensitive counterparts. Similar challenges are posed by the emergence of multidrug- and extensively-drug resistant tuberculosis (MDR-TB and XDR-TB, respectively) (5,6). Antibiotic resistance can result from large genomic changes, such as the acquisition of entire plasmids or mobile elements encoding resistance factors. Recent studies are, however, revealing the important role small mutations play in the evolution of resistance. For example, only 35 point-mutations distinguish a vancomycin-resistant strain of S. aureus from its sensitive counterpart, and these mutations evolved in just 3 months within an infected patient (7). Furthermore, antibiotic resistance genes have the potential to be used for bioterrorism purposes through genetically modified organisms. These factors emphasize the urgent need for a better understanding of the mechanisms through which bacteria develop resistance, as well as for the development of new techniques for the rapid identification of resistance factors. The database presented in this article provides a first component of an informatics infrastructure aimed at enabling such studies.

Several mechanisms have been characterized through which bacteria become resistant to antibiotics (8): (i) the production of enzymes that digest/metabolize the antibiotic; (ii) efflux pumps that eliminate the drug from the cell; (iii) modifications to the cellular target of the antibiotic that prevent binding; (iv) activation of an alternate pathway that bypasses drug action; and (v) particularly for gram-negative bacteria, down-regulation or elimination of transmembrane porins through which drugs enter the cell (9). The annotation information commonly associated with genes deposited in public databases is insufficiently detailed for representing this variety of resistance mechanisms and the additional meta-information relevant in this context. Specifically, each resistance gene is associated with a resistance profile (set of antibiotics or classes of antibiotics targeted by the gene), yet this information is usually not available. Second, resistance often requires the cooperation of multiple genes, usually within a same operon [e.g. vancomycin resistance VanA operon requires seven genes (10)], while most annotation information is targeted at individual genes. Finally, resistance frequently results from modifications to, or the disruption of an individual gene (e.g. modifications of the drug target), information incompatible with standard annotation procedures. Consequently, specialized resources are necessary for annotating and cataloging information related to antibiotic resistance.

Several recent efforts have been made to partially unify this information, such as Antibiotic Resistance Genes Online (ARGO) (11), MvirDB (12) and a compendium of TEM β-lactamase genes at the Lahey Clinic (http://www.lahey.org/Studies/). All, however, have limited functionality. ARGO only contains part of β-lactamase, vancomycin and tetracycline resistance genes. In addition, it does not include rich annotation information such as resistance profile, mechanism of action, operon information or gene sequence. Furthermore, many of the links between ARGO and GenBank target incorrect records (e.g. links to a genome instead of the relevant gene record). MvirDB is a broad repository of virulence-associated genes, including toxins, virulence factors and antibiotic resistance. The latter information is simply a replicate of the ARGO database. The Lahey Clinic website is a comprehensive collection of TEM type β-lactamases, which attempts to standardize the nomenclature for these genes. In addition to these specialized resources, antibiotic resistance information can be extracted in a restricted manner from GenBank (13) and SwissProt (14), databases that lack many important types of information relevant in this domain.

To address the limitations of currently available public resources, and to facilitate the identification and characterization of antibiotic resistance genes, we have created a manually curated database [Antibiotic Resistance Genes Database (ARDB)] unifying most of the publicly available genes and related information. Our motivations in creating ARDB are (i) to provide a centralized compendium of information on antibiotic resistance; (ii) to facilitate the consistent annotation of resistance information in newly sequenced organisms; and (iii) to facilitate the identification and characterization of new genes. We believe this resource will be found useful by a broad range of scientists, including microbiologists, clinicians and the bio-defense research community.

The diversity of antibiotic resistance genes, types and mechanisms, combined with the fact that related information, such as resistance profile, is mostly ‘paper-bound’ made the construction of ARDB both difficult and time-consuming. To compile, confirm and validate this collection of data, several textbooks and several hundred journal articles were searched and summarized.

The majority of protein and nucleic acid sequences of known antibiotic resistance genes were retrieved from the NCBI nucleotide and protein databases and additional sequences were retrieved from the Swiss-Prot database. Genes were grouped into resistance types based on their protein sequence similarity using the following approach. First, the sequence of an experimentally confirmed representative was identified for every type of resistance, based on literature searches and meta-information provided by the NCBI protein database. These representative resistance genes were then used to ‘fish out’ additional homologues using similarity searches against the NCBI nr database. The similarity cutoff was set at 80% unless a different value was recommended in the literature for a specific resistance type. Using this approach we identified 13 254 protein sequences putatively involved in antibiotic resistance. We filtered this set by removing vector sequences, synthetic constructs and redundant genes, resulting in a non-redundant set of 6206 proteins. This set was further refined by removing incomplete sequences, thereby yielding a core set of 4554 antibiotic resistance proteins. Each sequence was associated with corresponding CDD, COG, ontology and source organism information. Furthermore, the genes were grouped into resistance types, corresponding to clusters of genes with similar resistance profiles, operon membership and mechanism of action. In addition, basic information about known antibiotics was extracted from KEGG DRUG (15), PubChem, PubMed MeSH database and the Chemical Entities of Biological Interest (ChEBI) ontology. Although ARDB is mainly targeted at antibiotic resistance genes, 12 additional drug targets have also been included into ARDB with relevant information [16S rRNA (16), 23S rRNA, gyrA (17), gyrB, parC, parE, rpoB, katG, pncA, embB, folP, dfr], whose modification has been shown to confer resistance.

The data flow for the curation process is highlighted in Figure 1. ARDB is implemented as a MySQL relational database, and the corresponding schema is available on our website. Access to this database is provided through a CGI-based web interface.

Open in a separate window

Figure 1.

ARDB curation data flow.

No comprehensive ontology is currently available for annotating antibiotic resistance information. To facilitate the computational analysis of antibiotic resistance information we have created a set of ontology terms aimed at characterizing both the resistance profile conferred by a specific gene and its specific mechanism of action. Specifically, for every antibiotic X, we have created a set of ‘X resistance’ terms. Furthermore, we classify several mechanisms of action, including drug target modification, replacement or protection, drug enzymatic destruction and drug transport. Drug transport is further subclassified into ATP-binding cassette (ABC) drug efflux, major facilitator superfamily (MFS) drug efflux, small multidrug resistance (SMR) drug efflux and resistance-nodulation-cell division (RND) drug efflux, following the terminology used in (18). These terms are defined within an Antibiotic Resistance (AR) ontology and are associated with each record present in our database. We are currently working with the broader ontology community to further refine this information and integrate it within existing ontology development efforts.

Users can access our database through a web interface at http://ardb.cbcb.umd.edu. This interface provides several modes of interaction as highlighted below.

Keyword searches

Simple keyword search is available at the top of each page of ARDB website (Figure 2a), providing a quick means for searching a specific object in our database (gene, type, antibiotic, genome and genus) (Figure 2b and f). Users can search all of the data, or narrow down the search to a specific type of information. For example, users interested in the molecular mechanisms of resistance to tetracycline can search for the keyword ‘tetracycline’ within the ‘Resistance Type’ database. An advanced search function is also available, allowing users to select from among the available keywords associated with each database field.

Open in a separate window

Figure 2.

Sample web pages from ARDB. (a) Front page, (b) resistance type, (c) blast result, (d) mutation annotation, (e) browse and (f) genome information.

The database described in this article, ARDB, unifies most of the publicly available antibiotic resistance genes and provides a reliable annotation service to researchers investigating the molecular basis for resistance in bacteria. Because of the large diversity and the rapid identification of new resistance genes, the current version of ARDB is just a first catalog of currently available information, and will continue to be updated over the coming months and years. We plan to coordinate our development efforts with researchers actively involved in antibiotic resistance research as well as with the developers of biological ontologies and of databases storing related information (such as virulence factors or toxins). As part of these efforts we aim to refine the structure of our database, better determine the types of information stored and identify additional requirements for the user interface. Future efforts will also target the development of new approaches for cataloguing and characterizing polymorphisms correlated with resistance, as well as for annotating changes to cellular regulatory networks that underlie the mechanisms of drug tolerance.