Colistin susceptibility testing: evaluation of reliability for cystic fibrosis isolates of Pseudomonas aeruginosa and Stenotrophomonas maltophilia

Abstract

Objectives

Antibiotic susceptibility methods that are commonly used to test bacterial isolates from patients with cystic fibrosis are of uncertain reliability for the polymyxins. To assess the reliability of four standard testing methods, this pilot study used a challenge set that included polymyxin-resistant isolates of Pseudomonas aeruginosa and Stenotrophomonas maltophilia.

Methods

Twenty-five P. aeruginosa and 12 S. maltophilia isolates were tested for susceptibility to colistin (polymyxin E). Repeatability (concordance of replicates performed concurrently), reproducibility (concordance of replicates performed over time) and comparability (concordance of different methods) of agar dilution, broth microdilution, Etest and disc diffusion were assessed through the use of descriptive statistics and scatterplot analyses.

Results

All four methods displayed excellent repeatability (overall concordance rate of 99%). However, analysis of reproducibility revealed substantially lower rates of concordance (74% for agar dilution, 84% for broth microdilution and Etest, and 91% for disc diffusion). In addition, comparability to agar dilution of the three other methods was generally poor, with overall rates of very major error ranging from 12% for broth microdilution to 18% for Etest and disc diffusion.

Conclusions

Compared with agar dilution, other susceptibility testing methods give high rates of apparent false polymyxin susceptibility for cystic fibrosis isolates of P. aeruginosa and S. maltophilia. Prospective study of the correlation between in vitro susceptibility and clinical response is needed to clarify whether these discrepancies reflect oversensitivity of the agar dilution method or insensitivity of the other methods.

Introduction

Cystic fibrosis (CF) is a lethal genetic disorder that primarily affects the respiratory and digestive tracts.¹ The most common cause of morbidity and mortality in CF is chronic airway infection, primarily with Pseudomonas aeruginosa, although other intrinsically resistant Gram-negative organisms including Burkholderia cepacia complex and Stenotrophomonas maltophilia are also seen.² Episodes of acute exacerbation punctuate the course of chronic airway infection and hasten disease progression. To slow this progression, CF patients are treated with multiple courses of oral and intravenous antibiotics, and, increasingly, chronic administration of inhaled antibiotics.³ As a consequence, multidrug-resistant (MDR) strains emerge, thereby limiting therapeutic options. Clinicians are increasingly using colistin (polymyxin E), administered as the prodrug colistimethate either intravenously or via nebulization.^3,4 Colistin and other polymyxins (a family of acylated cyclic peptides produced by the Gram-positive organism Bacillus polymyxa) bind to lipopolysaccharide (LPS), thereby disrupting the Gram-negative cell wall.⁵

The first polymyxin-resistant laboratory strain of P. aeruginosa was reported in 1954, only 7 years after the discovery of polymyxins.⁶ However, for four decades after clinical introduction of colistin in 1961, isolates of P. aeruginosa from both CF and non-CF patients generally remained susceptible to colistin, with an MIC₅₀ (50th percentile of the MIC) of 1–2 mg/L, an MIC₉₀ (90th percentile of the MIC) of 4–8 mg/L and maximum MIC values of 16–32 mg/L.^7,8 As colistin use has become increasingly common in European and Australian CF centres,^4,9 the emergence of highly resistant isolates of P. aeruginosa has been reported (colistin MIC > 128 mg/L).^10,11 One study reported that 48% of P. aeruginosa isolates associated with acute exacerbations were resistant to polymyxin.⁹ Epidemic spread of polymyxin-resistant P. aeruginosa strains has also been documented or suspected at several CF centres.^10–12 Similarly, as polymyxin B use in non-CF MDR Gram-negative infections has become more common at some hospitals in the USA, the prevalence of polymyxin-resistant P. aeruginosa isolates has increased.¹³

Detection of polymyxin-resistant P. aeruginosa isolates in the studies cited above has largely depended on the use of agar dilution methodology. In contrast, the commonly used disc diffusion methodology is prone to inconsistent detection of resistance.¹⁴ Previous work has sought to define optimal methods for in vitro colistin susceptibility testing of various CF and non-CF Gram-negative pathogens.^14–16 However, highly colistin-resistant isolates of P. aeruginosa and S. maltophilia were quite rare at the time of those studies. The CLSI has recently revised the colistin interpretive criteria for P. aeruginosa and other non-Enterobacteriaceae (susceptible MIC, ≤2 mg/L; intermediate MIC, 4 mg/L; resistant MIC, ≥8 mg/L).¹⁷ However, the CLSI has not yet formulated recommendations delineating optimal methods for colistin susceptibility testing, particularly in light of the emergence of highly colistin-resistant P. aeruginosa and S. maltophilia isolates.

The specific objective of this pilot study was to evaluate the reliability of four standard susceptibility methods using a challenge set of P. aeruginosa and S. maltophilia CF isolates that included highly colistin-resistant isolates.

Materials and methods

Bacterial isolates and strains

The CF Referral Center for Susceptibility and Synergy Studies at Columbia University received the 25 P. aeruginosa isolates and 12 S. maltophilia isolates tested in this study from 28 CF clinical centres located in the USA during the period 2003–05 for the clinical purpose of performing chequerboard synergy testing.¹⁸ Each isolate came from a different patient. This research was conducted in accordance with the Declaration of Helsinki and the US Code of Federal Regulations Title 45 Part 46 (Protection of Human Subjects). The Institutional Review Boards of Columbia University (project #AAAA5796) and Seattle Children's Hospital (project #E-06-234-01) reviewed and approved the use of these isolates for research under a waiver of written informed consent, owing to the authors' lack of access to the patients from whom the isolates had been referred for clinical testing. The P. aeruginosa isolates were chosen to represent a wide range of colistin MICs (≤2 to >64 mg/L) with an MIC₅₀ of 16 mg/L and an MIC₉₀ of >64 mg/L based on initial broth microdilution testing at the time of receipt at Columbia University; the S. maltophilia isolates were similarly chosen to represent a range of colistin MICs based on initial testing at concentrations of 100 and 200 mg/L. In all, 84% of the P. aeruginosa isolates and 93% of the S. maltophilia isolates satisfied a published MDR definition for CF isolates.¹⁸

Prior to use in this study, the isolates had been stored in glycerol stocks at −70°C. To ensure accurate speciation of these isolates, all were re-identified by API 20NE (bioMérieux, Hazelwood, MO, USA), and speciation of non-mucoid and non-pigmented P. aeruginosa isolates was also confirmed using molecular techniques similar to those used for the B. cepacia complex.¹⁹ The S. maltophilia isolates were plated on DNase agar with Toluidine Blue medium (Remel, Lenexa, KS, USA) as previously described.²⁰ Quality control strains used in this study included the colistin-susceptible ATCC strains Escherichia coli 25922 and P. aeruginosa 27853, the colistin-intermediate ATCC strain S. maltophilia 13637 and three clinical isolates of P. aeruginosa previously found to be highly resistant to colistin by all testing methods (data not shown).

Susceptibility testing

All susceptibility testing was performed according to CLSI recommendations,^17,21 in duplicate on three separate occasions (four separate occasions for agar dilution), with independent review of the results by two members of the study team. A single batch of colistin sulphate salt (Sigma–Aldrich, St Louis, MO, USA) was used for all dilution testing. Agar dilution testing was performed as described previously;²¹ a Nunc 96-pin replicator with 1 mm pins and an OmniTray Copier (Nunc International, Rochester, NY, USA) were used to inoculate OmniTray plates (Nunc) containing 2-fold dilutions of colistin (0.125–512 mg/L) in Difco Mueller–Hinton agar (Becton Dickinson Diagnostic Systems, Sparks, MD, USA). Broth microdilution testing was performed with commercially prepared microtitre plates (Microtech Medical Systems, Aurora, CO, USA) containing 2-fold dilutions of colistin (0.25–512 mg/L) in cation-adjusted Mueller–Hinton broth. Etest strips (AB BIODISK, Piscataway, NJ, USA) calibrated to a colistin concentration range of 0.064–1024 mg/L (first round of testing) or 0.016–256 mg/L (second and third rounds of testing) and Kirby Bauer (KB) antimicrobial discs (Becton Dickinson, Franklin Lakes, NJ, USA) containing 10 µg of colistin were used in accordance with recommendations of the manufacturers. All assay plates were inspected after incubation for 16–20 h at 35 ± 2°C as per CLSI recommendations, and re-inspected after an additional 24 h at 35 ± 2°C. Because prolonged incubation did not substantially alter the MIC results for P. aeruginosa and S. maltophilia isolates from CF patients, only MIC results obtained after initial incubation were analysed.

For agar dilution testing, if an isolate displayed a skipped-concentration phenomenon wherein no growth was observed at a given concentration of colistin (e.g. 0.25 mg/L) but growth was observed at higher concentrations (e.g. 0.5 and 1 mg/L), the MIC was read as one dilution above the highest concentration with growth (e.g. 2 mg/L); more complex skip patterns were considered not interpretable as per CLSI recommendations.²¹ Etest results between 2-fold dilutions were rounded to the next highest 2-fold dilution (e.g. 3 mg/L was rounded to 4 mg/L). Because the lower end of the colistin test range for agar dilution testing and Etest (0.125 and 0.016 mg/L, respectively) was less than for broth microdilution testing (0.25 mg/L), agar dilution or Etest results ≤0.125 mg/L were reclassified as ≤0.25 mg/L. Similarly, because the upper end of the colistin test range for agar dilution and broth microdilution testing (512 mg/L) was greater than for Etest (256 mg/L), agar dilution or broth microdilution results ≥512 mg/L were reclassified as >256 mg/L. For KB disc diffusion testing, if isolates displayed growth of individual colonies up to the edge of the disc within a zone of inhibition, the inhibitory diameters were read as 0 mm as per CLSI recommendations.²² Results for control strains were as expected for all four methods on all rounds of testing.

Data analysis

To assess intra-assay discrepancies at a single point in time (repeatability), duplicate MIC or disc diameter values and their interpretations were compared with each other for the second and third round of testing (third and fourth round of testing for agar dilution).

To assess intra-assay discrepancies at multiple points in time (reproducibility), each of the duplicate MIC or disc diameter values and interpretations from the second round of testing (third round for agar dilution) was compared with each of the duplicate values and interpretations from the third round of testing (fourth round for agar dilution), resulting in four pairwise comparisons for each isolate.

To assess inter-assay discrepancies (comparability), interpretations of each of the duplicate MICs or disc diameters for one method were compared with interpretations of each of the duplicate values for a second method, for the second and third rounds of testing (third and fourth round for agar dilution). Thus each analysis of comparability included eight pairwise comparisons for each isolate.

Two types of discrepancies were considered; discordance of numerical values and differences in interpretation of the values. Discordance of numerical values was defined as a difference between MICs of ≥2 log₂ dilutions for agar dilution, broth microdilution and Etest, or a difference between inhibitory zone diameters of ≥2 mm for KB disc diffusion testing. CLSI-recommended colistin susceptibility breakpoints for P. aeruginosa were used to assess differences in interpretation (for dilution and Etest assay MIC values: susceptible, ≤2 mg/L; intermediate, 4 mg/L; and resistant, ≥8 mg/L; for disc diffusion inhibitory zone diameters: susceptible, ≥11 mm; and resistant, ≤10 mm).¹⁷ The same breakpoints were used for S. maltophilia because CLSI has not defined colistin susceptibility breakpoints for this organism. A serious difference in interpretation was defined as one value in a pair corresponding to resistant and the other corresponding to susceptible. A minor difference in interpretation was defined as one value corresponding to intermediate and the other corresponding to resistant or susceptible.

Scatterplot analyses were also used to assess comparability between methods. For each method and isolate, median MICs or disc diameters were calculated from duplicate tests performed on three rounds of testing (four rounds for agar dilution). Median values that fell between 2-fold dilutions (for MICs) or whole number measurements (for disc diameters) were rounded to the next highest dilution or next lowest whole number measurement, respectively. For each pair of methods, a scatterplot of median values was created. To analyse each scatterplot according to current interpretative criteria, lines representing the CLSI-recommended colistin susceptibility breakpoints for P. aeruginosa were drawn at right angles to each axis.

Because previous studies indicated that agar dilution gives higher colistin MICs for P. aeruginosa isolates than does broth microdilution,¹⁵ and might provide greater sensitivity than other susceptibility test methods in detecting polymyxin-resistant isolates,^16,23,24 agar dilution was used as the provisional reference method to determine the relative proportions of very major errors (false susceptibility), major errors (false resistance) and minor errors (susceptible/intermediate and resistant/intermediate discrepancies). For scatterplot analysis, these criteria defined regions of essential agreement {lower left and upper right quadrants and centre square in Figures S1 and S2 [available as Supplementary data at JAC Online (http://jac.oxfordjournals.org/)], and lower right and upper left quadrants in Figure S3 [available as Supplementary data at JAC Online (http://jac.oxfordjournals.org/)]}, very major error (upper left quadrant in Figures S1 and S2, and upper right quadrant in Figure S3), major error (lower right quadrant in Figures S1 and S2, and lower left quadrant in Figure S3) and minor error (central vertical and horizontal strips in Figures S1 and S2, and central horizontal strip in Figure S3). Very major errors and major errors together were defined as serious errors.

To assess whether the colistin disc diffusion error rates reflected an acceptable or unacceptable degree of discrepancy, the CLSI (formerly NCCLS) guidelines were applied for evaluating disc diffusion susceptibility tests using challenge sets of organisms.²⁵ Thus, within one log₂ dilution of the intermediate MIC value (for colistin agar dilution MICs, from 2 to 8 mg/L), discrepancy rates of <40% for minor errors, <10% for major errors and <10% for very major errors were considered acceptable. At higher MIC values (colistin agar dilution MICs ≥ 16 mg/L), lower discrepancy rates of <5% for minor errors and <2% for very major errors were considered acceptable, since by definition such very major errors would represent between-method differences of at least three log₂ dilutions (i.e. comparison method MIC ≤ 2 mg/L). Conversely, at MICs ≤ 1 mg/L, discrepancy rates <5% for minor errors and <2% for major errors were considered acceptable.

Results

Repeatability of colistin susceptibility testing methods

Repeatability (concordance of replicates performed concurrently) of colistin susceptibility testing methods was excellent for both P. aeruginosa and S. maltophilia isolates (Table 1). Concordant results were noted for 294 of 296 duplicate tests (99%). Seven of 294 concordant duplicates (2%) were associated with a difference in interpretation, whilst neither of two discordant duplicates was associated with such a difference (Table 1). Discordant duplicates were observed only for agar dilution (3%).

Table 1.

Repeatability of colistin susceptibility testing methods for CF isolates of P. aeruginosa and S. maltophilia

		Concordant duplicates,bn (%)		Discordant duplicates,cn (%)
Susceptibility testing method	Number of duplicates,aN	no interpretation difference	interpretation differenced	no interpretation difference	interpretation differenced
Agar dilution	74	71 (96)	1 (1)	2 (3)	0 (0)
Broth microdilution	74	71 (96)	3 (4)	0 (0)	0 (0)
Etest	74	71 (96)	3 (4)	0 (0)	0 (0)
Disc diffusion	74	74 (100)	0 (0)	0 (0)	0 (0)
Total	296	287 (97)	7 (2)	2 (1)	0 (0)

^aBased on testing of 37 isolates in duplicate on two separate occasions.

^bDefined as <2 log₂ dilution or <2 mm zone of inhibition difference between duplicate results.

^cDefined as ≥2 log₂ dilution or ≥2 mm zone of inhibition difference between duplicate results.

^dAll interpretation differences were minor (susceptible/intermediate or resistant/intermediate).

Reproducibility of colistin susceptibility testing methods

Reproducibility (concordance of replicates performed over time) of colistin susceptibility testing methods was assessed by testing P. aeruginosa and S. maltophilia isolates in duplicate on two separate occasions (Table 2). Of 148 pairwise MIC comparisons for agar dilution, 38 (26%) were discordant, of which 6 (4%) were associated with a difference in interpretation. Of 148 pairwise MIC comparisons for broth microdilution, 24 (16%) were discordant, of which 18 (12%) were associated with a difference in interpretation. Of 148 pairwise MIC comparisons for Etest, 24 (16%) were discordant, of which 12 (8%) were associated with a difference in interpretation. Of 148 pairwise zone of inhibition comparisons for disc diffusion, 14 (9%) were discordant, of which 4 (3%) were associated with a difference in interpretation. The overall rates of serious and minor differences in interpretation for these reproducibility comparisons (including the 22 concordant pairs that were associated with interpretation differences) were 4% and 6%, respectively.

Table 2.

Reproducibility of colistin susceptibility testing methods for CF isolates of P. aeruginosa and S. maltophilia

		Concordant pairs,bn (%)			Discordant pairs,cn (%)
			interpretation differences			interpretation differences
Susceptibility testing method	Number of pairwise comparisons,aN	no interpretation differences	serious	minor	no interpretation differences	serious	minor
Agar dilution	148	106 (72)	0 (0)	4 (3)	32 (22)	4 (3)	2 (1)
Broth microdilution	148	122 (82)	0 (0)	2 (1)	6 (4)	6 (4)	12 (8)
Etest	148	112 (76)	0 (0)	12 (8)	12 (8)	6 (4)	6 (4)
Disc diffusion	148	130 (88)	4 (3)	—	10 (7)	4 (3)	—
Total	592	470 (79)	4 (1)	18 (3)	60 (10)	20 (3)	20 (3)

^aBased on testing of 37 isolates in duplicate on two separate occasions.

^bDefined as <2 log₂ dilution or <2 mm zone of inhibition difference.

^cDefined as ≥2 log₂ dilution or ≥2 mm zone of inhibition difference.

Comparability of colistin susceptibility testing methods

Serious differences in interpretation between agar dilution and the other three methods ranged from 12% to 18% (Table 3). Scatterplots representing comparisons of agar dilution with each of the other methods (Figures S1, S2 and S3) showed that serious differences in interpretation were partly attributable to three P. aeruginosa isolates and an S. maltophilia isolate for which agar dilution results corresponded to resistance (colistin MICs 32–64 mg/L) whilst the other methods gave MICs or zones of inhibition that corresponded to susceptibility.

Table 3.

Comparability of colistin susceptibility testing methods for CF isolates of P. aeruginosa and S. maltophilia

	Rates of differences in interpretation (%)a
	agar dilution		broth microdilution		Etest
Assay	serious	minor	serious	minor	serious	minor
Broth microdilution	12	8	—	—	—	—
Etest	18	8	3	10	—	—
Disc diffusion	18	2	8	6	5	6

^aRates of serious and minor differences in interpretation were derived from eight pairwise comparisons for each of 37 isolates tested in duplicate on two separate occasions.

For the P. aeruginosa isolates, agar dilution testing indicated a colistin MIC₅₀ that corresponded to susceptible but a colistin MIC₉₀ that corresponded to resistant (Table 4). Broth microdilution also gave a colistin MIC₉₀ that corresponded to resistant but was 16-fold lower that that of agar dilution, and Etest gave a colistin MIC₉₀ that corresponded to intermediate. For the S. maltophilia isolates, the colistin MIC₅₀ and MIC₉₀ given by agar dilution were more similar to those given by broth microdilution and Etest (Table 4). Consistent with these results, broth microdilution, Etest and disc diffusion gave high rates of very major error for P. aeruginosa isolates, and more moderate rates of very major error for S. maltophilia when compared with agar dilution as the provisional reference method (Table 4).

Table 4.

Comparability studies of polymyxin susceptibility methods for P. aeruginosa and S. maltophilia published since 2001

									Interpretation,d %				Error rates,d,e %
Study	Speciesa	Isolates, n	Source	Reference methodb	Test range,c mg/L	MIC range,c mg/L	MIC50,c mg/L	MIC90,c mg/L	S	Ic	R	Comparison methodb	vmj	mj	mnc
									This study	PA	25	CF	CST AD	0.125–512	≤0.5 to >256	2	>256	60	4	36	BMiD	16	0	4
												Etest-256	20	0	12
												DD-10	24	0	4
	SM	12	CF	CST AD	0.125–512	≤0.5 to >256	>256	>256	8	0	92	BMiD	8	0	8
												Etest-256	17	0	8
												DD-10	8	0	0
Gales et al.14	PA	80	non-CF	CST BMiD	1–128	≤1–2	≤1	≤1	100	NA	0	DD-10	0	0	0
				PMB BMiD	1–128	≤1–2	≤1	2	100	NA	0	DD-300	0	0	0
	SM	23	non-CF	CST BMiD	1–128	≤1–64	≤1	32	70	NA	30	DD-10	30	0	0
				PMB BMiD	1–128	≤1–64	2	8	70	NA	30	DD-300	30	0	0
Nicodemo et al.23	SM	66	non-CF	CST AD	NA	0.125–32	2	4	76	NA	24	Etest-1024	9	5	NA
												DD-10	23	0	6
				PMB AD	NA	0.25–16	2	4	77	NA	23	Etest-1024	12	11	NA
												DD-300	18	0	9
Hogardt et al.15	PA	401	CF	CST AD	0.5–16	≤0.5 to >16	2	4	92	NA	8	BMiD	5	1	NA
												BMiD-48	3	3	NA
				PMB AD	0.5–16	≤0.5 to >16	2	2	95	NA	5	BMiDf	3	2	NA
												BMiD-48f	2	3	NA
Tan and Ng16,24	PA	47g	non-CF	CST AD	0.25–128	2–16	2	8	89	NA	11	DD-25	11	0	NA
									68	NA	32	DD-10	32	0	0
												DD-50	30	0	NA
												BMiD-V	30	0	NA
												Etest-1024	11	30	NA
	SM	9	non-CF	CST AD	0.25–128	NA	NA	NA	0	NA	100	BMiD-V	22	0	NA
												Etest-1024	11	0	NA
Van der Heijden et al.31	PA	109h	non-CF	CST BMiD	NA	≤0.25–2	1	1	100	0	0	Etest-1024	0	0	7
												DD-10	0	0	NA
				PMB BMiD	NA	≤0.25–8	0.5	1	99	0	1	Etest-1024	1	0	49
												DD-300	1	13	NA
Galani et al.30	PA	124	non-CF	CST Etest	0.0625–1024	≤0.5 to >128	1	2	94	1	6	DD-10	1	0	1
	SM	36	non-CF	CST Etest	0.0625–1024	≤0.5 to >128	8	256	42	8	50	DD-10	0	0	39

^aPA, P. aeruginosa; SM, S. maltophilia.

^bCST, colistin sulphate; PMB, polymyxin B sulphate; BMiD, broth microdilution; BMiD-48, 48 h incubation; BMiD-V, Vitek2 (1–8 mg/L CST); AD, agar dilution; DD, disc diffusion; DD-10, 10 µg of CST; DD-25, 25 µg of CST (sanctioned by the BSAC); DD-50, 50 µg of CST (sanctioned by the French Society for Microbiology); DD-300, 300 U of PMB; Etest-256, 0.016–256 mg/L CST; Etest-1024, 0.0625–1024 mg/L CST or PMB.

^cNA, not available or not applicable.

^dS, susceptible; I, intermediate; R, resistant. Breakpoints were those sanctioned at the time and locale of each study, as follows: dilution methods and Etest, R ≥ 8 mg/L and S ≤ 2 mg/L (this study, Van der Heijden, Galani), R ≥ 4 mg/L (Gales, Nicodemo, Tan except for calculation of DD-25 error rates) or R ≥ 8 mg/L (Hogardt, Tan calculation of DD-25 error rates only); DD-10, R ≤ 10 mm (this study, Van der Heijden, Galani) or R ≤ 8 mm and S ≥ 11 mm (Gales, Nicodemo, Tan); DD-25, R ≤ 13 mm; DD-50, R ≤ 14 mm; DD-300, R ≤ 8 mm and S ≥ 12 mm (Gales, Nicodemo) or R ≤ 11 mm (Van der Heijden).

^evmj, very major error; mj, major error; mn, minor error.

^fThe BMiD and BMiD-48 test methods in Hogardt et al.¹⁵ used CST instead of PMB.

^gPA in the Tan and Ng¹⁶ 2007 study (n = 47) were a subset of those in the Tan and Ng²⁴ 2006 study (n = 56) and included all colistin-resistant PA in the latter (T. Y. Tan, Changi General Hospital, Singapore, personal communication); interpretation and error rates for the 2006 results were recalculated based on n = 47. Two interpretations are given for CST AD results due to use of different breakpoints for calculation of DD-25 error rates (see footnote d).

^hOf these strains, a non-clonal subset (n = 78) was tested by disc diffusion and Etest.

Discrepancy rates for colistin disc diffusion testing

Many clinical laboratories rely on disc diffusion testing of CF isolates of P. aeruginosa, S. maltophilia and other pathogens due to its relative ease and good comparability for antibiotics other than colistin.²⁶ Of the 20 isolates with colistin MIC ≥ 16 mg/L by agar dilution, colistin disc diffusion testing deemed 7 (35%) to be susceptible (Figure S3), representing an unacceptable rate of very major error.

Discussion

Colistin is increasingly important for the treatment of infections caused by MDR Gram-negative bacilli, yet the reliability of colistin susceptibility testing for these organisms remains problematic. We found that current colistin susceptibility methods have excellent repeatability for P. aeruginosa and S. maltophilia CF isolates representing a range of colistin MICs, but that their reproducibility varies and their comparability is poor, as evidenced by a high rate of serious differences in interpretation between agar dilution and the other methods.

Studies of MDR CF isolates have shown that for antibiotics other than colistin, broth microdilution is comparable to agar dilution,²⁷ and Etest and disc diffusion are comparable to broth microdilution.²⁶ However, studies of non-CF Gram-negative clinical isolates in the late 1960s and early 1970s indicated that colistin disc diffusion correlated poorly with agar dilution, owing to very major errors associated with non-pseudomonal isolates.^28,29 Colistin-resistant isolates of P. aeruginosa remained exceedingly rare in CF and other clinical populations through the early 1990s,⁷ but more recently epidemic spread of highly resistant strains has been reported among CF patients treated chronically with inhaled colistin.^10,11 Recent studies demonstrate that broth microdilution, Etest and disc diffusion may substantially underestimate the prevalence of P. aeruginosa and S. maltophilia colistin resistance in at-risk clinical populations (Table 4).^{14–16,23,24,30,31} We found that a substantial number of P. aeruginosa isolates with colistin disc diffusion inhibitory diameters in the range of 13–17 mm (defined as susceptible according to CLSI guidelines) are resistant to colistin by agar dilution (Figure S3). Because some clinical laboratories strictly rely on disc diffusion testing to determine the colistin susceptibility of these pathogens, such very major errors could result in ineffective clinical use of colistin and failure to detect epidemic spread of colistin-resistant clinical strains.

Whether disc diffusion testing reliably detects polymyxin resistance may depend on the specific combination of resistance mechanisms expressed in a given isolate. In vitro exposure of P. aeruginosa to polymyxin B sulphate selects for moderately resistant mutants that are also cross-resistant to other cationic antimicrobial peptides.³² In such laboratory mutants, activating mutations in the pmrAB locus promote attachment of aminoarabinose to phosphate groups of lipid A and core oligosaccharide, hindering the interaction of cationic antimicrobial peptides with LPS.³² It is plausible that in clinical strains, mutations in pmrAB or other polymyxin resistance loci could alter resistance phenotypes in response to polymyxin concentration gradients³³ or to growth on a surface,³⁴ and thus might account for discrepancies between dilution and diffusion methods or between broth- and agar-based methods, respectively. We are currently studying such strains to examine this hypothesis. More generally, improved understanding of colistin resistance mechanisms could facilitate the development of molecular susceptibility methods.

Factors that cause variations in drug–target stoichiometry such as bacterial inoculum size and growth rate might also contribute to discrepancies in colistin susceptibility testing. CF isolates of P. aeruginosa sometimes require extended incubation as long as 24 h beyond the initial 16–20 h incubation period.^17,27 Most of the highly polymyxin-resistant P. aeruginosa isolates that we studied were slow growing, and some reached a stable agar dilution endpoint only after such extended incubation; however, this had a minimal effect on comparability (data not shown), as others have also observed (Table 4, Hogardt et al.¹⁵).

In summary, our findings indicate that agar dilution is the most sensitive CLSI-sanctioned method for detection of P. aeruginosa polymyxin resistance. In our study and others, disc diffusion has an unacceptably high rate of very major errors; however, other comparison methods have similarly poor comparability such that we are unable to recommend a preferred method for these organisms. Microbiological and clinical outcome data are not available for the patients from whom the bacterial isolates in this study were obtained, thus we could not assess the clinical relevance of the discrepant colistin susceptibility results. Indeed, no study has adequately assessed the predictive value of colistin susceptibility testing. Considering that chronic use of inhaled colistin is relatively common among CF patients in several European countries and is becoming more common among CF patients in the USA,^3,11 prospective correlation of in vitro susceptibility, colistin pharmacokinetics^35,36 and clinical outcomes is needed to enhance the predictive value of colistin susceptibility testing, and to determine whether discrepancies reflect oversensitivity of the agar dilution method or insensitivity of the other methods.

Funding

This work was supported by a grant from the NIH to S. M. M. (R01 AI067653).

Transparency declarations

S. M. M. and L. S. have received lecture fees from the France Foundation, a third-party sponsor of continuing medical education activities. L. S. has received grant support from Chiesi Pharmaceuticals and Bayer, and has served on advisory boards to Chiesi, Novartis, Gilead, Axio and Transave. All other authors: none to declare.

Author contributions

S. M. M., E. G. and L. S. conceived and designed the study, E. G., Y. C., S. A. C., S. T., A. K. M. and M. D. performed laboratory work and acquired data, S. M. M., E. G., S. C. and L. S. analysed and interpreted data, S. M. M. and E. G. drafted the manuscript and S. M. M., E. G., S. A. C. and L. S. critically revised its content.

Supplementary data

Figures S1, S2 and S3 are available as Supplementary data at JAC Online (http://jac.oxfordjournals.org/).