Abstract
A microarray capable of detecting genes for resistance to 75 clinically relevant antibiotics encompassing 19 different antimicrobial classes was tested on 132 Gram-negative bacteria. Microarray-positive results correlated >91% with antimicrobial resistance phenotypes, assessed using British Society for Antimicrobial Chemotherapy clinical breakpoints; the overall test specificity was >83%. Microarray-positive results without a corresponding resistance phenotype matched 94% with PCR results, indicating accurate detection of genes present in the respective bacteria by microarray when expression was low or absent and, hence, undetectable by susceptibility testing. The low sensitivity and negative predictive values of the microarray results for identifying resistance to some antimicrobial resistance classes are likely due to the limited number of resistance genes present on the current microarray for those antimicrobial agents or to mutation-based resistance mechanisms. With regular updates, this microarray can be used for clinical diagnostics to help accurate therapeutic options to be taken following infection with multiple-antibiotic-resistant Gram-negative bacteria and prevent treatment failure.
INTRODUCTION
The extensive use of antimicrobials has selected resistance in many bacterial species, and this has become a major public health issue worldwide (1). The problem of multidrug resistance is currently most acute in Gram-negative bacteria, where treatment options can be severely limited, and there are few promising new antibiotics with activity against Gram-negative bacteria under advanced development (1). Clonal spread of resistant strains and horizontal transfer of resistance genes both contribute to the rising global prevalence of multiresistant Gram-negative bacteria, with horizontal transfer enabling the acquisition of multidrug resistance by previously susceptible bacteria (2).
Most methods currently employed by clinical diagnostic laboratories, e.g., the Vitek (bioMérieux, Marcy l'Etoile, France) and MicroScan (Siemens, Camberley, United Kingdom) (3) systems, provide insight only into the resistance phenotypes, require susceptibility testing following bacterial culture, and can take from approximately 9 to 20 h to obtain results. Molecular methods are faster but present their own challenges. In particular, conventional or real-time PCR, single-gene sequencing, or in situ hybridization usually detects only a few resistance genes simultaneously. Microarrays, however, offer the potential for simultaneous detection of large numbers of resistance genes, allowing prediction of an isolate's repertoire of resistances to multiple antibiotic classes (4–7).
We previously developed a microarray that detected 51 resistance genes in Escherichia coli and Salmonella and facilitated epidemiological studies (4, 8). Its coverage included genes that conferred acquired resistance to a range of antimicrobials of clinical importance as well as two integrase genes associated with class 1 and class 2 integrons and not just those genes encoding β-lactamases (6). This work sought to extend the microarray to make it relevant to the broader range of Gram-negative bacterial genera commonly encountered in clinical diagnostic or reference laboratories and to include in its coverage newly critical gene groups, including genes encoding carbapenemases.
MATERIALS AND METHODS
Bacterial strains used and determination of antibiotic susceptibilities.
A panel of 132 bacterial isolates from our laboratory collections and clinical isolates representing a diverse range of Enterobacteriaceae and nonfermenter genera, including Pseudomonas and Acinetobacter, was assembled. The antimicrobial susceptibilities (MICs) of these isolates to 19 antibiotics (tobramycin, amikacin, gentamicin, streptomycin, ampicillin, amoxicillin-clavulanic acid [co-amoxiclav], aztreonam, cefotaxime, ceftazidime, cefpirome, cefoxitin, piperacillin, imipenem, meropenem, chloramphenicol, ciprofloxacin, sulfonamide, tetracycline, and trimethoprim) were determined using the British Society for Antimicrobial Chemotherapy (BSAC) agar dilution methodology or disk diffusion (9). Susceptibility was defined using BSAC/European Committee on Antimicrobial Susceptibility Testing (EUCAST) clinical breakpoints (legacy breakpoints were used for cefpirome, cefoxitin, streptomycin, sulfonamide, and tetracycline) (9). For several species-antimicrobial combinations, BSAC breakpoints were not available, and in these cases, susceptibility categories were not assigned.
Design and validation of the new probes.
Gene probes and labeling primers for the new genes were designed, and their specificity was tested in silico as previously described (8). Multiple probes and/or primers were designed for certain genes (e.g., blaIMP), so as to enable the detection of diverse alleles. The performance of the new probes was assessed in two replicate hybridizations (see below) using control strains taken from the panel of bacterial isolates, in which the presence of the gene had been verified by PCR and sequencing. The specificity of each new probe was estimated by comparing the microarray result with PCR and sequence data for the relevant control strain. The new genes required 39 probe and 41 primer sequences for the linear amplification and labeling reaction; these are listed in Table S1 in the supplemental material, together with the panel of control strains and PCR primers used to validate each new probe.
Microarray procedure.
For testing on the microarray, bacteria were cultivated overnight at 37°C on Luria-Bertani agar plates, and then a loopful of bacteria was lysed and DNA from the lysed cells was quantified using a nanodrop apparatus and labeled in a linear multiplex reaction using all primers (i.e., primers described previously [4, 8] and the new primers [shown in Table S1 in the supplemental material]). The amplified labeled mix was added to the microarrays for hybridization to all probes present on AMR06 (i.e., probes described previously [4, 8]and the new probes [shown in Table S1 in the supplemental material]), with signals from the hybridization duplex read on an ArrayMate apparatus (Alere Technologies, Jena, Germany) using IconoClust software, as already described (standard version; Inverness Technologies, Jena, Germany) (8). Mean signal intensities of three replicate spots per probe were used for analysis. Intensities of ≥0.4 were considered positive, as established previously (8).
Relationship between genotype and phenotype.
Typically, in clinical diagnostic laboratories, the MIC or the result of the disk diffusion method is used to determine antimicrobial resistance (AMR), and such a determination is especially important for strains carrying transferable resistances. We tested all 132 isolates in the bacterial panel with the updated 75-gene microarray and compared the genotypic data obtained with the antimicrobial susceptibilities of the bacterial isolates to 19 antibiotics. The microarray tests and analyses were performed blindly with no prior knowledge of the antimicrobial susceptibility data of the test isolates.
For all species and antimicrobial combinations for which a BSAC breakpoint was available for the particular species/genus, the agreement between antimicrobial susceptibilities (phenotype) and microarray result (genotype) was assessed (Table 1). Where BSAC breakpoints were not available for a species-antimicrobial combination, the data were not interpreted further, as the correlation between microarray results and antibiotic susceptibilities could not be assessed.
Table 1.
Antibiotic groups considered in this study and the relevant antibiotic resistance genes for each group present on the microarray
| Antibiotic | Relevant antibiotic resistance genesa |
|---|---|
| Amikacin | aac(6′)-Ib, armA, rmtC |
| Gentamicin | aac(6′)-IIc, aadB, aac(6′)-aph(2′), armA, rmtC, aac(3′)-Ia, aac(3′)-IVa, aac(6′)-Ib |
| Streptomycin | aadA1-like, aadA2-like, aadA4-like, strA, strB |
| Tobramycin | aadB, armA, rmtC, aac(3)-IVa, aac(6′)-Ib |
| Ampicillin | blaCMY, blaPER-1, blaCTX-M-1, blaCTX-M-2, blaCTX-M-8, blaCTX-M-9, blaCTX-M-26, blaDHA, blaOXA-1, blaOXA-2, blaOXA-7, blaOXA-9, blaPSE-like, blaSHV, blaTEM, blaIMP, blaOXA-23, blaOXA-40, blaOXA-48, blaOXA-51, blaOXA-58, blaSPM-1, blaGIM-1 |
| Aztreonam | blaCMY, blaPER-1, blaVEB-1, blaDHA, blaSHV, blaFOX, blaACC-1, blaGES-1, blaCTX-M-1, blaCTX-M-2, blaCTX-M-8, blaCTX-M-26, blaCTX-M-9, blaOXA-7, blaTEM |
| Cefotaxime | blaCTX-M-1, blaCTX-M-2, blaCTX-M-8, blaCTX-M-26, blaCTX-M-9, blaDHA, blaACC, blaFOX, blaSHV, blaCMY, blaGES-1, blaPER-1, blaVEB-1, blaIMP, blaGIM-1 |
| Cefoxitin | blaCMY, blaGES-1, blaPER-1, blaDHA, blaIMP, blaFOX, blaMOX, blaKHM, blaGIM-1, blaSPM-1 |
| Cefpirome | blaGES-1, blaPER-1, blaACC, blaCTX-M-1, blaCTX-M-2, blaCTX-M-8, blaCTX-M-26, blaCTX-M-9, blaFOX, blaOXA-1, blaSHV, blaIMP |
| Ceftazidime | blaCMY, blaGES-1, blaPER-1, blaVEB-1, blaACC, blaDHA, blaFOX, blaSHV, blaIMP, blaGIM-1, blaSPM-1 |
| Amoxicillin-clavulanic acid | blaOXA-1, blaTEM, blaIMP |
| Imipenem | blaCMY, blaGES-1, blaIMI3, blaIMP, blaKPC-4, blaOXA-23, blaOXA-40, blaOXA-48, blaOXA-51, blaOXA-58, blaGIM-1 |
| Meropenem | blaGES-1, blaIMI3, blaIMP, blaKPC-4, blaOXA-23, blaOXA-40, blaOXA-48, blaOXA-51, blaOXA-58, blaGIM-1 |
| Piperacillin | blaCMY, blaVEB-1, blaACC, blaDHA, blaOXA-1, blaIMP, blaKPC-4, blaCTX-M-1, blaCTX-M-2, blaCTX-M-8, blaCTX-M-9, blaCTX-M-26, blaSHV, blaPER-1, blaGIM-1, blaSPM |
| Chloramphenicol | catA1, catB3-like, catB8, catA3, cmlA1-like, floR1 |
| Ciprofloxacin | qnrA, qnrB, qnrS |
| Sulfonamide | sul1, sul2, sul3 |
| Tetracycline | tet(A), tet(B), tet(C), tet(D), tet(E), tet(G) |
| Trimethoprim | dfrA01, dfrA05, dfrA07, dfrA12, dfrA14, dfrA15, dfrA17, dfrA19 |
The genes relevant for the 19 antibiotic agents included in this study are given in Table 1 and include all relevant genes for that class of antibiotics present on the microarray. To simplify the data analysis, the presence of a gene was always assumed to confer a corresponding antimicrobial resistance phenotype.
The correlation of the microarray results with phenotypes was evaluated by two-by-two table analysis (10), where test specificity, sensitivity, and the predictive value of a positive and negative test were calculated using the following criteria: microarray-positive and antibiotic-resistant results are true positive (TP), microarray-negative and antibiotic-susceptible results are true negative (TN), microarray-positive but antibiotic-susceptible results are false positive (FP), and microarray-negative but antibiotic-resistant results are false negative (FN). The correlation was evaluated for each antibiotic in this manner. Overall test performance was evaluated using the total of all evaluable results. The bacterial panel contained insufficient numbers of isolates of each species to enable separate evaluation.
RESULTS
Validation of new probes.
Probes for 24 new genes were validated using the control strains (see Table S2 in the supplemental material). These variously encoded genes for β-lactamases (blaGES, blaPER, blaVEB, and the carbapenemase genes blaGIM-1, blaIMI, blaIMP, blaKHM, blaKPC, blaOXA-23-like, blaOXA-40-like, blaOXA-48-like, blaOXA-51-like, blaOXA-58-like, blaSPM-1), aminoglycoside resistance [aac(6′)-IIc, aac(6′)-aph(2′), armA, rmtC], chloramphenicol resistance (catB8), macrolide resistance (ereA, ermB, mphA), rifampin resistance (arr-1), and streptogramin resistance (vatE). Additional probes for three genes already present in existing iterations [blaCMY, blaMOX, aac(6′)-Ib] were added to the revised microarray and validated, to enable better detection of allelic variations relevant to these resistances. The primers and probes for the new genes are listed in Table S1 in the supplemental material, while the sequences for the old probes and primers are detailed in previous publications (4, 8).
Each new resistance gene was identified in both replicates of the relevant control strain, and for most strains, other additional resistance genes were also identified by microarray. For some genes where, due to allelic variation, more than one probe was present, different control strains were required to validate each probe, e.g., the probe for aac(6′)-Ib (see Table S2 in the supplemental material).
Another 31 new gene probes (e.g., rmtA, rmtC, qnrC, qnrD; data not shown) were added in the microarray design, but their specificity could not be validated, as control strains containing the target gene were not available; these probes were not included in subsequent analysis and require further work. For other gene probes (e.g., blaVIM), despite showing in silico specificity, in vitro microarray tests showed poor probe specificity, returning false positives, indicating the importance of validating new probes with control strains (data not shown). These probes were not considered for further analysis and will be modified in the next iteration of the microarray.
Antimicrobial resistance genes detected in the bacterial panel.
Antimicrobial resistance genes were found in 110 of the 132 isolates, belonging to 15 of the 23 species/genera tested, and encompassed resistance to all classes of antimicrobials represented on the microarray (Fig. 1). The panel included 72 clinical isolates, of which the majority (n = 50) possessed multiple (between 2 and 21) genes conferring resistance to antimicrobials of one or more antimicrobial drug class; only 3 isolates harbored a single resistance gene; for the remainder, no genes were detected by the microarray. In total, genes detected by the new probes were microarray positive on 163 occasions, and PCR confirmed the presence of 146 out of 153 (95%) of these (10 microarray-positive genes were not tested by PCR; data not shown).
Fig 1.
All genes detected by microarray and the antimicrobial susceptibilities of the bacterial panel studied. For each antibiotic class, the resistance genes represented on the microarray and the antibiotic tested (indicated in capital letters) by MIC determination or the disc diffusion method are grouped, and results are given for all 132 isolates. Microarray-positive genes are indicated by black-filled cells (a blank cell indicates that no gene was detected). Genes represented by more than one probe on the microarray were considered present if at least one probe had an intensity of ≥0.4. (a) Results from the aminoglycoside and β-lactam antimicrobial groups; (b) results from the remaining antimicrobial groups. b, asterisks denote clinical isolates; c, for each antibiotic where a BSAC criterion was available, the antimicrobial susceptibility is given as susceptible (S) and resistant (R) (a blank cell indicates that no interpretive criterion was available or the test was not performed). Resistant or susceptible results which are true positive and true negative are given in shaded boxes; false-positive and false-negative results are given in black. The results of the two-by-two table analysis for each antibiotic for which data were available and the overall performance values are also given.
Resistance genes were not detected by microarray in any isolate (n = 13) of the following eight species: Achromobacter xylosoxidans, Aeromonas hydrophila, Burkholderia cenocepacia, Pantoea agglomerans, Pseudomonas putida, Ralstonia mannitolilytica, Serratia marcescens, and Stenotrophomonas maltophilia (Fig. 1); this likely reflects the fact that their resistances largely involve chromosomal mutation and not acquired resistance genes. The panel contained one Gram-positive isolate (Enterococcus faecium), employed as a control for the erm(B) and vat(E) probes, for which it was positive, but no other resistance genes represented on the microarray were detected in this isolate. These genes were added to the microarray because erm(B) has been identified in Gram-negative bacteria (11) and in E. faecium has been linked with vat(E) (12).
Resistance genes were detected by microarray in four nonfermenter species tested: Acinetobacter baumannii, Moraxella osloensis, Pseudomonas aeruginosa, and Pseudomonas fluorescens (Fig. 1). All of the nine A. baumannii strains tested were microarray positive for multiple resistance genes (from 5 to 11 genes in each isolate; Fig. 1) spanning resistance to antimicrobials of three or more different antibacterial drug classes (aminoglycosides, β-lactams, chloramphenicol, rifampin, sulfonamides, or tetracyclines). Eight A. baumannii strains were microarray positive for the intrinsic carbapenemase blaOXA-51 gene, and seven possessed a second blaOXA carbapenemase gene (blaOXA-23, blaOXA-40, or blaOXA-58). For P. aeruginosa, microarray-positive results were obtained with 9 of the 10 strains tested. For one strain, the integrase 1 gene but no resistance gene was detected. The remaining eight strains harbored one to six genes for resistance to antimicrobials from one or more of the following antibacterial drug classes: aminoglycosides, β-lactams, chloramphenicol, quinolones, sulfonamides, tetracyclines, or trimethoprim (Fig. 1).
Relationship between genotype and phenotype.
The relationship between genotype and phenotype was evaluated for all 19 antimicrobial groups, for all strains, using the gene categories given in Table 1. We thought that this evaluation would be relevant, as the microarray focuses on acquired resistance mechanisms, which, through horizontal gene transfer, may result in a species/genus acquiring genes that had previously been undetected in that organism. For all evaluable microarray results combined, the test specificity [TN/(TN + FP)] was 83.0% and the predictive value of a positive test [TP/(TP + FP)] was 90.6% (Fig. 1), indicating that microarray-positive results (i.e., gene presence) were highly predictive of resistance in a gene-specific manner, correlating well with phenotypic MIC and disk diffusion data. For 9 of the 19 antibiotics tested, both the specificity and predictive value of a positive test were between 90 and 100%; these included many clinically relevant agents and antimicrobial classes: tobramycin, ampicillin, cefoxitin, piperacillin, imipenem, ciprofloxacin, sulfonamide, tetracycline, and trimethoprim (Fig. 1).
Isolates were occasionally microarray positive for a gene but susceptible to a corresponding antimicrobial. In these cases, the specificity and the predictive value of a positive test were lower due to high numbers of FP results, showing a poor correlation with the resistance phenotype (Fig. 1). Across all evaluable results, 87 microarray FPs (as defined in Materials and Methods) arising from the detection of 102 resistance genes (some isolates were microarray positive for more than one relevant gene) were obtained. These results are summarized in Table 2. PCR confirmed the presence of 92 out of 98 (94%) of these genes, indicating that they were actually true positives (4 were not tested; Table 2). This was particularly true for strains harboring genes from the aminoglycoside group (amikacin, gentamicin, and streptomycin), where >97% of the genes were confirmed to be positive by PCR. Another source of FPs was the β-lactamase genes blaSHV and blaTEM. Although the microarray detects these genes, it cannot distinguish between extended-spectrum β-lactamase (ESBL) and non-ESBL variants of these genes (8). In the analysis undertaken, microarray-positive results for either of these genes were interpreted as conferring an ESBL phenotype, which did not necessarily match the result from the susceptibility testing. In all instances where a microarray FP result was obtained with blaSHV or blaTEM, PCR again confirmed the presence of the gene. The detection of either blaSHV or blaTEM represented the only FP results obtained for amoxicillin-clavulanic acid, cefotaxime, and ceftazidime, 1 of the 5 FP results obtained with cefpirome, and 10 of the 13 FP results obtained with aztreonam (Table 2).
Table 2.
Comparison of microarray-positive and antibiotic-susceptible results with PCR data
| Antibiotic | No. of microarray-positive but susceptible isolatesa | No. of microarray-positive relevant genesb | Microarray-positive gene(s)c | No. of isolates with the indicated PCR result |
||
|---|---|---|---|---|---|---|
| Positivec | Negativec | Not done | ||||
| Amikacin | 18 | 18 | aac(6′)-Ib | 18 | 0 | 0 |
| Gentamicin | 12 | 16 | aac(6′)-Ib | 8 | 0 | 0 |
| aac(3′)-Ia | 1 | 0 | 0 | |||
| aac(6′)-aph(2′) | 4 | 0 | 1 | |||
| aadB | 0 | 0 | 2 | |||
| Streptomycin | 14 | 19 | aadA1-like | 7 | 1 | 0 |
| aadA2-like | 6 | 0 | 0 | |||
| aadA4-like | 2 | 0 | 1 | |||
| strA | 1 | 0 | 0 | |||
| strB | 1 | 0 | 0 | |||
| Amoxicillin-clavulanic acid | 2 | 2 | blaTEM | 2 | 0 | 0 |
| Aztreonam | 13 | 16 | blaACC-1 | 1 | 0 | 0 |
| blaDHA | 1 | 0 | 0 | |||
| blaFOX | 1 | 0 | 0 | |||
| blaSHV | 5 | 0 | 0 | |||
| blaTEM | 8 | 0 | 0 | |||
| Cefotaxime | 2 | 2 | blaSHV | 2 | 0 | 0 |
| Ceftazidime | 4 | 4 | blaSHV | 4 | 0 | 0 |
| Cefpirome | 5 | 5 | blaACC-1 | 1 | 0 | 0 |
| blaFOX | 2 | 0 | 0 | |||
| blaOXA-1 | 1 | 0 | 0 | |||
| blaSHV | 1 | 0 | 0 | |||
| Imipenem | 3 | 4 | blaCMY | 2 | 0 | 0 |
| blaOXA-48 | 2 | 0 | 0 | |||
| Meropenem | 4 | 4 | blaOXA-48 | 4 | 0 | 0 |
| Chloramphenicol | 8 | 10 | catB3 | 3 | 5 | 0 |
| catB8 | 2 | 0 | 0 | |||
| Ciprofloxacin | 1 | 1 | qnrA | 1 | 0 | 0 |
| Trimethoprim | 1 | 1 | dfrA01 | 1 | 0 | 0 |
| Total | 87 | 102 | 92 | 6 | 4 | |
The number of false-positive correlations between genotype and phenotype is given for each antibiotic.
The total number of relevant genes detected by microarray. In instances where more than one gene was present in an antibiotic group, more genes than the number of false-positive correlations were detected, as some isolates were microarray positive for more than one gene.
For each antibiotic, the specific genes detected by microarray are given together with the PCR result.
Microarray-negative results showed poorer agreement with antimicrobial susceptibilities than did positive results; the microarray had an overall sensitivity [TP/(TP + FN)] of 64.5%, and the predictive value of a negative test [TN/(TN + FN)] was 47.8%. It was particularly low (<60%) for sensitivity and the negative predictive value for the following agents: cefoxitin, imipenem, meropenem, ceftazidime, ciprofloxacin, and tetracycline (Fig. 1). These results were due to a myriad of reasons, which were mostly explainable. For example, the microarray could not detect the carbapenemase genes blaVIM (due to the nonspecificity of the probes) and blaNDM (which was absent from the microarray) that had been identified by PCR in 5 and 13 test isolates, respectively (data not shown), contributing to the low predictive value and sensitivity for the carbapenem group. In P. aeruginosa, the reduced susceptibility to a range of antibiotics, including imipenem and meropenem (Fig. 1), that was seen is likely associated with porin (OprD) deficiency and/or (in the case of all β-lactams except imipenem) increased efflux activity (13). Similarly, a number of E. coli isolates and a large number of Klebsiella pneumoniae isolates were resistant to cefoxitin, probably due to the loss or reduced expression of porins (14), but were microarray negative (Fig. 1) because the microarray does not detect the presence or expression of porins. Also, the microarray detects three plasmid-mediated quinolone resistance genes (qnrA, qnrB, and qnrS) which confer low-level resistance phenotypically and not the high-level quinolone resistance conferred by single or double amino acid substitutions in the target enzymes (DNA gyrase and topoisomerase IV) or by alterations affecting drug entry or efflux (15), again leading to lower negative predictive and sensitivity values for these genes. However, it is important to stress that our data showed that for microarray-positive isolates harboring genes for acquired resistance to cefoxitin, imipenem, and ciprofloxacin, there was a 90 to 100% correlation with the relevant resistance phenotype, indicating that the correlation was excellent for genes present on the microarray (Fig. 1). For tetracycline, 34 isolates were resistant in the absence of a microarray-positive result for a relevant gene using the BSAC legacy breakpoint (1 mg/liter; Fig. 1). However, if the 8-mg/liter EUCAST epidemiological cutoff (ECOFF) values used for many Enterobacteriaceae (data from the EUCAST MIC distribution website, http://www.eucast.org) are used instead, then the number of FNs decreases from 34 to 8 (data not shown). Again, the phenotypes of isolates harboring a tetracycline resistance gene showed a 100% correlation with the relevant resistance phenotype, indicating that for genes present on the microarray, the correlation is excellent (Fig. 1).
DISCUSSION
In an era when antimicrobial resistance genes are spreading rapidly on multiresistance plasmids, the revised microarray described here provides a powerful alternative to PCR for the rapid, accurate, and simultaneous detection of 75 clinically important acquired resistance genes in a diverse collection of Gram-negative bacteria, and not only in E. coli and Salmonella, associated with human infections. Our data demonstrated that as with PCR, only genes which are being amplified (and present on the microarray) are detected, and the rate of detection of these genes is highly accurate and specific and correlates >94% with the results of PCR even when they are not expressed. However, other genes or mechanisms which may give rise to resistance but are absent from the microarray will, as expected, not be detected. Thus, for the microarray, just as with PCR and any other genotypic test, the absence of a signal cannot be used to infer the absence of resistance mechanisms. The advantage of using the microarray as opposed to multiplex PCR in clinical diagnostic laboratories is that for the current format there is a microarray chip at the bottom of each well of a 96-well plate, allowing a large number of genes and up to 96 samples to be tested simultaneously, providing genotypic results within 5 h after bacterial growth. Although phenotypic screening may still be required to determine the pathogen's antibiogram, rapid genotypic screening of clinical isolates by the microarray can provide results approximately 24 h earlier and can help reduce the risk of administering antibiotics that are unlikely to be effective for treatment and propagate highly resistant strains. However, phenotypic screening also has pitfalls. As demonstrated by this work, low or no expression of genes of certain antibiotic groups, such as aminoglycosides, can give apparent false-positive microarray results. These represent pseudogenes or genes that are present but not expressed or poorly expressed. Silent (nonexpressing) resistance genes have been reported in numerous species; include several aminoglycoside resistance genes, such as strA (16), aadA1 (17), aadA2 (18), and both aac(6′)-Ib and aac(3′)-Ia (19); and probably account for the low correlation in this class. However, although these genes may currently be poorly expressed or silent, this may change if a selection pressure, e.g., administration of an aminoglycoside antibiotic, is applied, hence posing problems in treatment. In the case of the carbapenemase gene blaOXA-48, its underdetection by phenotypic methods is already beginning to pose a serious problem (20), and in this study, several isolates were microarray positive for this gene but carbapenem susceptible (Table 2), again showing the importance of including genotypic methods in routine diagnostics.
It is worth noting that the data in Figure 1 are a summary of the performance of the microarray with each antibiotic for all bacterial species included in this study. These results may differ if each antibiotic was further divided according to bacterial species, as not all species are currently known to harbor all relevant genes included on the microarray (Table 1), although this may change in the future due to gene acquisition through horizontal gene transfer. However, this was not possible for the current study, as for several species the number of bacteria tested was small and the results would not be statistically significant. Such a study merits further investigation in future, as it may alter the performance of the microarray with each antibiotic, affecting the interpretation of microarray data for individual clinical isolates.
In this study, we used BSAC clinical breakpoints to determine susceptibility, as they have direct clinical application and enable an informative comparison to microarray results. ECOFF values (www.eucast.org) provide an alternate method to determine susceptibility, enabling wild-type isolates to be distinguished from those with reduced susceptibility, which are likely to harbor an acquired resistance mechanism, although they may not show clinical resistance. Comparison of microarray results with ECOFF-derived susceptibilities marginally improved the microarray test performance; the specificity was 87.0%, the predictive value of a positive test was 92.6%, and for 12 of the 19 antibiotics tested, both the specificity and predictive value of a positive test were between 90 and 100% (data not shown). For our data set, the difference between using the BSAC cutoff and the ECOFF value was most clearly demonstrated for tetracycline, where use of the latter decreased the number of false negatives detected.
Therefore, the microarray presented in this study shows a very good correlation with gene presence and with regular updates can provide a useful adjunct to current methods employed by clinical diagnostic laboratories and for epidemiological studies to detect antimicrobial resistance genes. A version of this array is commercially available from Alere Technologies (Jena, Germany). In an era when whole-genome sequencing (WGS) is becoming increasingly common for use in clinical diagnostics, there is still a bottleneck in rapid interpretation of WGS data, unlike for microarrays. The utility of microarrays in future could be further improved by developing methods which allow direct detection from clinical samples, thereby increasing the speed with which results are provided to clinicians.
Supplementary Material
ACKNOWLEDGMENTS
We are grateful to Alere Technologies, particularly to Peter Slickers for his help in designing the new probes and primers and to Ralf Ehricht for his general advice. Thanks also go to David M. Livermore for critical reading of the manuscript.
This work was supported by a Department of Trade and Industry Interact Proof-of-Concept grant (to M.A. and N.W.) and an ANTIRESDEV European Union Seventh Framework Programme grant (grant agreement no. 241446 to M.A.).
Footnotes
Published ahead of print 5 November 2012
Supplemental material for this article may be found at http://dx.doi.org/10.1128/AAC.01223-12.
REFERENCES
- 1. Freire-Moran L, Aronsson B, Manz C, Gyssens IC, So AD, Monnet DL, Cars O. 2011. Critical shortage of new antibiotics in development against multidrug-resistant bacteria—time to react is now. Drug Resist. Updat. 14:118–124 [DOI] [PubMed] [Google Scholar]
- 2. Hawkey PM, Jones AM. 2009. The changing epidemiology of resistance. J. Antimicrob. Chemother. 64(Suppl. 1):i3–i10 [DOI] [PubMed] [Google Scholar]
- 3. Sellenriek P, Holmes J, Ferrett R, Drury R, Storch GA. 2005. Comparison of MicroScan Walk-Away®, Phoenix™ and VITEK-TWO® microbiology systems used in the identification and susceptibility testing of bacteria, abstr. C-203. Abstr. 105th Gen. Meet. Am. Soc. Microbiol. American Society for Microbiology, Washington, DC [Google Scholar]
- 4. Anjum MF, Choudhary S, Morrison V, Snow LC, Mafura M, Slickers P, Ehricht R, Woodward MJ. 2011. Identifying antimicrobial resistance genes of human clinical relevance within Salmonella isolated from food animals in Great Britain. J. Antimicrob. Chemother. 66:550–559 [DOI] [PubMed] [Google Scholar]
- 5. Davis MA, Besser TE, Orfe LH, Baker KN, Lanier AS, Broschat SL, New D, Call DR. 2011. Genotypic-phenotypic discrepancies between antibiotic resistance characteristics of Escherichia coli isolates from calves in management settings with high and low antibiotic use. Appl. Environ. Microbiol. 77:3293–3299 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Naas T, Cuzon G, Bogaerts P, Glupczynski Y, Nordmann P. 2011. Evaluation of a DNA microarray (Check-MDR CT102) for rapid detection of TEM, SHV, and CTX-M extended-spectrum beta-lactamases and of KPC, OXA-48, VIM, IMP, and NDM-1 carbapenemases. J. Clin. Microbiol. 49:1608–1613 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Walsh F, Cooke NM, Smith SG, Moran GP, Cooke FJ, Ivens A, Wain J, Rogers TR. 2010. Comparison of two DNA microarrays for detection of plasmid-mediated antimicrobial resistance and virulence factor genes in clinical isolates of Enterobacteriaceae and non-Enterobacteriaceae. Int. J. Antimicrob. Agents 35:593–598 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Batchelor M, Hopkins KL, Liebana E, Slickers P, Ehricht R, Mafura M, Aarestrup F, Mevius D, Clifton-Hadley FA, Woodward MJ, Davies RH, Threlfall EJ, Anjum MF. 2008. Development of a miniaturised microarray-based assay for the rapid identification of antimicrobial resistance genes in Gram-negative bacteria. Int. J. Antimicrob. Agents 31:440–451 [DOI] [PubMed] [Google Scholar]
- 9. British Society for Antimicrobial Chemotherapy 2012. BSAC methods for antimicrobial susceptibility testing, version 11.1, May 2012. British Society for Antimicrobial Chemotherapy, Birmingham, United Kingdom: http://bsac.org.uk/ [Google Scholar]
- 10. Mackinnon A. 2000. A spreadsheet for the calculation of comprehensive statistics for the assessment of diagnostic tests and inter-rater agreement. Comput. Biol. Med. 30:127–134 [DOI] [PubMed] [Google Scholar]
- 11. Courvalin P. 1994. Transfer of antibiotic resistance genes between gram-positive and gram-negative bacteria. Antimicrob. Agents Chemother. 38:1447–1451 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Jensen LB, Hammerum AM, Aarestrup FM. 2000. Linkage of vat(E) and erm(B) in streptogramin-resistant Enterococcus faecium isolates from Europe. Antimicrob. Agents Chemother. 44:2231–2232 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Strateva T, Yordanov D. 2009. Pseudomonas aeruginosa—a phenomenon of bacterial resistance. J. Med. Microbiol. 58:1133–1148 [DOI] [PubMed] [Google Scholar]
- 14. Delcour AH. 2009. Outer membrane permeability and antibiotic resistance. Biochim. Biophys. Acta 1794:808–816 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Jacoby GA. 2005. Mechanisms of resistance to quinolones. Clin. Infect. Dis. 41(Suppl. 2):S120–S126 [DOI] [PubMed] [Google Scholar]
- 16. Randall LP, Cooles SW, Osborn MK, Piddock LJ, Woodward MJ. 2004. Antibiotic resistance genes, integrons and multiple antibiotic resistance in thirty-five serotypes of Salmonella enterica isolated from humans and animals in the UK. J. Antimicrob. Chemother. 53:208–216 [DOI] [PubMed] [Google Scholar]
- 17. Enne VI, Delsol AA, Roe JM, Bennett PM. 2006. Evidence of antibiotic resistance gene silencing in Escherichia coli. Antimicrob. Agents Chemother. 50:3003–3010 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. White DG, Zhao S, Sudler R, Ayers S, Friedman S, Chen S, McDermott PF, McDermott S, Wagner DD, Meng J. 2001. The isolation of antibiotic-resistant Salmonella from retail ground meats. N. Engl. J. Med. 345:1147–1154 [DOI] [PubMed] [Google Scholar]
- 19. Dai N, Li DZ, Chen JC, Chen YS, Geng R, Hu YH, Yang JP, Du J, Hu CP, Zhang W, Li JS, Yu Q, Wan HY, Mu L, Zhong XN, Wei LP, Ma JJ, Wang QY, Hu K, Tian GZ, Cai SX, Wang RQ, He B, Wang SQ, Wang ZW, Zhao SR, Gao ZC. 2010. Drug-resistant genes carried by Acinetobacter baumannii isolated from patients with lower respiratory tract infection. Chin. Med. J. (Engl.) 123:2571–2575 [PubMed] [Google Scholar]
- 20. Nordmann P, Gniadkowski M, Giske CG, Poirel L, Woodford N, Miriagou V, the European Network on Carbapenemases 2012. Identification and screening of carbapenemase-producing Enterobacteriaceae. Clin. Microbiol. Infect. 18:432–438 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.