INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

The VITEK 2 automated antimicrobial susceptibility test system is a new integrated system which automatically performs rapid identification and antimicrobial susceptibility testing after an inoculum has been prepared manually (J.-P. Gayral, R. Robinson, and D. Stamstedt, Abstr. Eur. Cong. Clin. Microbiol. Infect. Dis., abstr. P254, p. 53, 1997). Its improved performance over those of earlier rapid systems is due to the larger number of wells in its card, enhanced optics, and new algorithms based on kinetic analysis of data (S. Dib, J. Nguyen, V. Jarlier, and A. Philippon, Abstr. 37th Intersci. Conf. Antimicrob. Agents Chemother., abstr. D50, p. 92, 1997; Gayral et al., Abstr. Eur. Cong. Clin. Microbiol. Infect. Dis.; M. Ghanem, C. Bradford, D. Freiner, M. Ullery, and J. Gerst, Abstr. 98th Annu. Meet. Am. Soc. Microbiol. 1998, abstr. C-416, p. 200, 1998; J. H. Jorgensen, A. L. Barry, M. M. Traczewski, D. F. Sahm, M. L. McElmeel, and S. A. Crawford, Abstr. 98th Annu. Meet. Am. Soc. Microbiol. 1998, abstr. C-422, p. 201, 1998; L. A. Meeh, C. Shubert, S. Weber, P. Kim, and M. Peyret, Abstr. 98th Annu. Meet. Am. Soc. Microbiol. 1998, abstr. V-66, p. 523, 1998; E. S. Moland, K. S. Thomson, and C. C. Sanders, Abstr. 37th Intersci. Conf. Antimicrob. Agents Chemother., abstr. D48, p. 92, 1997; C. Shubert, R. Griffith, W. McLaughlin, M. Ullery, and M. Peyret, Abstr. 98th Annu. Meet. Am. Soc. Microbiol. 1998, abstr. C-478, p. 211, 1998). The Advanced Expert System (AES) is an expert system designed to analyze results generated by the VITEK 2 system for biologic validity and then provide comments on results (J. M. Boeufgras, A. Lazzarini, M. Peyret, and J. Zindel, Abstr. Eur. Cong. Clin. Microbiol. Infect. Dis., abstr. P299, p. 64, 1997; J. M. Boeufgras, R. Vachon, J. L. Balzer, C. Davenas, A. Rongier, M. Tarpin, and M. Peyret, Abstr. Eur. Cong. Clin. Microbiol. Infect. Dis., abstr. P300, p. 64, 1997). Unlike previous expert systems, the AES is based upon an extensive knowledge base that comprises over 2,000 phenotypes and 20,000 MIC distributions (Boeufgras et al., Abstr. Eur. Cong. Clin. Microbiol. Infect. Dis., abstr. P299). This allows it to recognize certain susceptibility patterns as indicative of specific phenotypes and interpret the results accordingly.

A phenotype is defined as the expression of a specific mechanism of susceptibility or resistance to a given drug class within a particular species. The word phenotype is used in this context in preference to the word genotype since categorization is often based upon tests for gene products or phenotypes that result from the expression of the gene product rather than upon tests for the actual gene or gene sequence involved. The wild-type phenotype is defined as the phenotype for that species in the "wild," i.e., prior to any mutation of chromosomal genes or acquisition of new DNA that alters susceptibility to the drug class in question. Thus, if one considers phenotypes for Escherichia coli and -lactam drugs, there are a number of possible phenotypes (Table 1). The wild type is devoid of any significant levels of -lactamase and is thus susceptible to ampicillin and most other -lactam antibiotics. Once the strain has acquired a plasmid-mediated penicillinase like TEM-1 or SHV-1, it is resistant to penicillins and perhaps cephalothin and now possesses a penicillinase phenotype (Table 1). If the strain acquires an extended-spectrum -lactamase (ESBL) or if the gene encoding its resident TEM-1 or SHV-1 mutates to produce an ESBL derivative, the strain will have an ESBL phenotype of reduced susceptibility or resistance to expanded-spectrum cephalosporins and aztreonam, in addition to penicillins and cephalothin (Table 1). If the strain acquires a plasmid-mediated AmpC -lactamase from an organism such as Enterobacter spp. or Citrobacter freundii or if mutations in the promoter region of its own chromosomal ampC gene occur, giving rise to elevated levels of AmpC -lactamase, the strain will have a cephalosporinase phenotype and will display reduced susceptibility or resistance to virtually every -lactam drug except the carbapenems (Table 1). Thus, for each species, one can prepare a list of phenotypes based upon possible mechanisms of susceptibility and resistance for each drug class, and these phenotypes can be associated with certain susceptibility patterns (3-5, 7, 12, 15, 17).

View this table: [in this window] [in a new window]   TABLE 1.   Examples of -lactam phenotypes for E. colia

From this association between phenotype and susceptibility patterns, the AES was developed by using a large knowledge base of MIC distributions for each of the known phenotypes. This knowledge base was obtained from published reports, human experts with their own databases on phenotypes, and in-house data at bioMérieux (Boeufgras et al., Abstr. Eur. Cong. Clin. Microbiol. Infect. Dis., abstr. P299 and P300). For each of the recognized phenotypes, the range of MICs obtained from tests with each specific drug was determined and was defined as the MIC distribution for that phenotype. For example, from data available worldwide, the MIC distribution for wild-type E. coli and ampicillin was found to be 0.5 to 16.0 µg/ml, while that for the acquired penicillinase phenotype for E. coli was found to be 256 µg/ml. Thus, with a large knowledge base of over 20,000 MIC distributions for over 2,000 phenotypes, it should be possible for the AES to predict a phenotype by using the susceptibility pattern obtained from tests with the VITEK 2 system.

Therefore, a study was designed to ascertain the ability of the AES to correctly identify the -lactam phenotypes of 196 isolates of the family Enterobacteriaceae and the species Pseudomonas aeruginosa. These isolates had been collected from laboratories worldwide and had been characterized for their -lactam phenotypes by biochemical and molecular methods.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Strains. A total of 196 isolates of the family Enterobacteriaceae and the species P. aeruginosa were collected from laboratories worldwide and their -lactam phenotypes were characterized by biochemical and molecular techniques (2, 6, 10, 11, 13, 16, 18). Many of these have been described previously (8, 13, 14, 17). -Lactamases were identified biochemically as to their isoelectric points and inhibitor-substrate profiles. For species that produce chromosomally encoded inducible AmpC -lactamases, basal enzyme levels and levels following induction with cefoxitin were ascertained by using cephalothin as a substrate to determine if the strain was a wild type or had a mutant phenotype. For Klebsiella oxytoca, levels of enzyme were measured by using nitrocefin as a substrate to ascertain if the strain was a wild type or a hyperproducer of the chromosomal K1 -lactamase (8, 13, 17). Permeability mutants were determined by analysis of outer membrane porins (2). The -lactam phenotypes and species included among the 196 isolates are shown in Table 2. These included both common and rare or atypical resistance phenotypes. For strains with more than one -lactamase, the -lactam phenotype produced by the dominant enzyme was used for analysis. For example, among C. freundii strains with the high-level cephalosporinase phenotype there were strains that also produced an acquired penicillinase. However, the broader substrate profile of the AmpC -lactamase masked the presence of the acquired penicillinase. Thus, the phenotypes were the same for all high-level cephalosporinase producers regardless of the presence or absence of an acquired penicillinase, and the presence of the second enzyme was ignored for the purposes of data evaluation.

View this table: [in this window] [in a new window]   TABLE 2.   -Lactam phenotypes included in the test panel of 196 isolates

Susceptibility tests. Antibiotic susceptibilities were determined according to the manufacturer's recommendations by using the VITEK 2 instrument. The cards used for the test were standard European cards and contained the following antibiotics and concentration ranges: (i) AST-N009 for members of the family Enterobacteriaceae (ampicillin, 2 to 32 µg/ml; amoxicillin-clavulanate [2:1 ratio], 2-1 to 32-16 µg/ml; cephalothin, 2 to 64 µg/ml; cefoxitin, 4 to 64 µg/ml; cefotaxime, 1 to 64 µg/ml; ceftazidime, 1 to 64 µg/ml; ticarcillin, 8 to 128 µg/ml; ticarcillin-clavulanate [clavulanate at 2 µg/ml with ticarcillin at a twofold dilution], 8-2 to 128-2 µg/ml; piperacillin-tazobactam [tazobactam at 4 µg/ml with piperacillin at a twofold dilution], 4-4 to 128-4 µg/ml; and imipenem, 0.5 to 16 µg/ml) and (ii) AST-N008 for P. aeruginosa (cefepime, 1 to 64 µg/ml; ceftazidime, 1 to 64 µg/ml; piperacillin, 4 to 128 µg/ml; pipercillin-tazobactam [tazobactam at 4 µg/ml with piperacillin at a twofold dilution], 4-4 to 128-4 µg/ml; ticarcillin, 8 to 128 µg/ml; ticarcillin-clavulanate [clavulanate at 2 µg/ml with ticarcillin at a twofold dilution], 8-2 to 128-2 µg/ml; imipenem, 0.5 to 16 µg/ml; meropenem, 0.25 to 16 µg/ml; and aztreonam, 1 to 64 µg/ml). Quality control was performed with each run by using E. coli ATCC 25922 and P. aeruginosa ATCC 27853.

Data analysis. Since this study was not designed to assess the ability of the VITEK 2 system to identify these gram-negative species, the species name of the strain was manually entered into the instrument and only susceptibility tests were performed. (It is acknowledged that the interpretive abilities of the AES are dependent on the accuracy of the organism identification and that the fact that different laboratories use different identification systems is a relevant issue.) The results were then analyzed by the AES with a test version of software, and a hard-copy report of that analysis was obtained. The -lactam phenotype identified by the AES was then compared to the phenotype that had been identified by biochemical and molecular methods. If the MIC distributions of a given phenotype were unique for the drugs tested, then a single phenotype was identified by the AES. However, if the MIC distributions for the drugs tested overlapped for several phenotypes, the AES would list all of the possible phenotypes. The AES was considered to have correctly identified the -lactam phenotype of a strain if it listed the same -lactam phenotype identified by biochemical and molecular methods (i) in a single choice or (ii) in one of several possibilities.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Overall, the AES was able to identify a -lactam phenotype for 183 of the 196 (93.4%) isolates tested. The correct phenotype was identified by the AES in one or more choices for 157 of the 183 (85.8%) isolates, and for 111 of the 183 (60.7%), the correct phenotype was identified in a single choice. The 13 isolates for which a -lactam phenotype could not be identified are described below.

Species. The performance of the AES by species is shown in Table 3. The percentage of strains tested for which the AES correctly identified the -lactam phenotype in one or more choices varied from a high of 92% for Enterobacter spp. to a low of 74% for E. coli and C. freundii.

View this table: [in this window] [in a new window]   TABLE 3.   Identification of -lactam phenotypes by the AES with the data analyzed by species

Phenotypes. Certain phenotypes were more difficult than others for the AES to identify (Table 4). Among the wild-type phenotypes, 10 were incorrectly identified by the AES. Many of these were incorrectly identified as having an acquired penicillinase phenotype, and none of these errors was due to incorrect MICs obtained with the VITEK 2 instrument (Table 5). One wild-type Enterobacter aerogenes strain was identified by the AES as an ESBL or acquired penicillinase producer due to a falsely elevated ceftazidime MIC. One of the three problems with the wild-type-acquired penicillinase and acquired penicillinase phenotypes was due to a falsely elevated cefotaxime MIC (Table 5). The elevated ceftazidime and cefotaxime MICs reverted to the correct result of susceptible on repeat testing. The phenotypes of all five of the K. pneumoniae strains with a plasmid-mediated cephalosporinase (AmpC) were incorrectly identified as ESBL plus impermeability by the AES because the correct phenotype was not in the database (Table 5). Although for the purposes of this study this was considered an incorrect identification, the MIC distributions for the two phenotypes are similar. Had the cephalosporinase phenotype been in the database for this species, the AES would have listed cephalosporinase or ESBL plus impermeability as the two phenotypes possible. Two E. coli strains that produced ESBLs were incorrectly identified as acquired penicillinase producers by the AES (Table 5). These two strains were unusual in that the MICs of cefotaxime and ceftazidime, the two expanded-spectrum cephalosporins on the test card, for the strains were below 1.0 µg/ml. For the current ESBL-producing indicator strains (9), only cefpodoxime MICs are 2.0 µg/ml in tests with these strains. Thus, the inability of the AES to ascertain the ESBL phenotypes for these strains was due to the absence of cefpodoxime on the test card. Two high-level cephalosporinase-producing Serratia marcescens strains were incorrectly identified by the AES as belonging to more susceptible phenotypes (Table 5). These errors were due to the fact that the MICs of a variety of drugs for these strains are uncharacteristically low. Thus, any MIC-based system would incorrectly categorize these strains. The failure of the AES to identify an impermeability phenotype for a single strain of P. mirabilis was due to the absence of the phenotype in the database (Table 5). However, when the AES identified the strain as having an acquired penicillinase phenotype, it did note that the MIC pattern was highly unusual.

View this table: [in this window] [in a new window]   TABLE 4.   Identification of -lactam phenotypes by the AES with the data analyzed by phenotype

Unidentifiable phenotypes. The AES was unable to identify a phenotype for 13 of the 196 strains tested (Table 6). In most instances, there were either real errors in the MICs that made it impossible for the AES to match the susceptibility pattern to a phenotype for the species or the AES indicated that there were errors in MICs that made it impossible for it to match the pattern to a phenotype. For example, the AES could not identify a phenotype for three cephalosporinase-producing E. coli strains because the susceptibility pattern (which was in fact accurate) looked similar to that expected of an organism like Enterobacter or Citrobacter with an inducible AmpC -lactamase (Table 6). True errors in the MICs of ampicillin, amoxicillin-clavulanate, cephalothin, and/or cefoxitin led the AES to suggest that three wild-type C. freundii strains were in fact Escherichia, Citrobacter youngae, or Citrobacter braakii. Repeat testing of most of these isolates did not resolve these problems.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

Overall, the AES performed well in this validation study, identifying correctly the -lactam phenotypes of 157 of the 196 isolates tested in one or more choices. It should be noted that for the purposes of this study, the isolate panel selected included many strains with phenotypes rarely or infrequently encountered in the clinical laboratory (e.g., ESBL- and cephalosporinase-producing E. coli) as well as strains with rarely encountered phenotypes that gave atypical results in susceptibility tests (i.e., ESBL-producing E. coli and K. pneumoniae for which ceftazidime MICs were <1 µg/ml). Thus, the overall performance of the AES reflects the strains and phenotypes tested and is not a reflection of overall performance in the average clinical laboratory.

Examination of the causes for the incorrect phenotypes identified by the AES revealed the need for several improvements. First, the AES, like any database system, needs to be updated often to ensure that all known phenotypes and MIC distributions are in the database. The occurrence of plasmid-mediated AmpC -lactamases in clinical isolates of E. coli and K. pneumoniae was extremely rare during the time when the database for the AES was being developed. Thus, these phenotypes were not in the database and strains with these phenotypes were either unidentifiable or incorrectly identified by the AES.

Certain errors in identification of the cephalosporinase and ESBL phenotypes highlighted the need for a specific ESBL test on the card. At this time, ESBL producers are recognized by MIC distributions which cannot always distinguish between the cephalosporinase and ESBL phenotypes. Cefoxitin resistance in E. coli and K. pneumoniae may be due to impermeability rather than to the cephalosporinase. Thus, an isolate of these species with a cephalosporinase will have a susceptibility pattern similar to that of a porin mutant with an ESBL. The use of a specific test for ESBL production that compares the MICs of certain drugs in the presence and absence of clavulanic acid may improve discrimination between these phenotypes (13, 17, 18); M. M. Traczewski, A. L. Barry, S. D. Brown, J. A. Hindler, D. A. Bruckner, and D. F. Sham, Abstr. 97th Gen. Meet. Am. Soc. Microbiol. 1998, abstr. C-37, p. 137, 1998.

A final problem concerns the difficulty with identifying the impact of permeability changes on -lactam susceptibility. For most species, the impact of permeability changes even on the wild-type phenotype has not been studied adequately to provide a database for phenotype recognition. Furthermore, certain susceptibility patterns can arise from the presence of -lactamase, altered permeability to a drug, or a combination of the two factors. In only a few instances, e.g., resistance to imipenem in P. aeruginosa, can the role of altered permeability be clearly defined and predicted by use of MIC-based tests. In most instances, e.g., cefoxitin resistance in E. coli and K. pneumoniae or ceftazidime resistance in S. marcescens, MICs alone are inadequate for identification of the actual mechanism responsible for the resistance. Thus, it is likely that the problem of recognition of most impermeability phenotypes will not be resolved in the near future. However, tests for recognition of phenotypes such as ESBL plus impermeability in E. coli and acquired cephalosporinase in K. pneumoniae will be added to the commercial version of the software.

In summary, the AES was able to provide correct the -lactam phenotypes of 157 of the 196 gram-negative isolates tested, including strains with phenotypes rarely encountered in the routine clinical laboratory, by using the card configurations in this study. Certain remediable problems with the system were identified, and remediation of these problems should lead to improved performance in the future. These results suggest that the AES should be very useful for the identification of the -lactam phenotypes of gram-negative isolates and that further study of its utility for the clinical laboratory is warranted.