Assessment of Phenotype Microarray plates for rapid and high-throughput analysis of collateral sensitivity networks

The crisis of antimicrobial resistance is driving research into the phenomenon of collateral sensitivity. Sometimes, when a bacterium evolves resistance to one antimicrobial, it becomes sensitive to others. In this study, we have investigated the utility of Phenotype Microarray (PM) plates for identifying collateral sensitivities with unprecedented throughput. We assessed the relative resistance/sensitivity phenotypes of nine strains of Staphylococcus aureus (two laboratory strains and seven clinical isolates) towards the 72 antimicrobials contained in three PM plates. In general, the PM plates reported on resistance and sensitivity with a high degree of reproducibility. However, a rigorous comparison of PM growth phenotypes with minimum inhibitory concentration (MIC) measurements revealed a trade-off between throughput and accuracy. Small differences in PM growth phenotype did not necessarily correlate with changes in MIC. Thus, we conclude that PM plates are useful for the rapid and high-throughput assessment of large changes in collateral sensitivity phenotypes during the evolution of antimicrobial resistance, but more subtle examples of cross-resistance or collateral sensitivity cannot be identified reliably using this approach.


32
Collateral sensitivity is when bacteria develop resistance to one antibiotic, and in doing so 33 increase their susceptibility to one or more others. The phenomenon was first observed in  subjected to evolution in vitro. In each case, a strain was exposed to one antibiotic, resistant 45 mutants were isolated, and their sensitivities towards up to 25 other antibiotics were tested. 46 The clinical relevance of this in vitro approach has been questioned because of the 47 stochasticity of evolution [12] and because it emphasises mutationally acquired resistance, 48 rather than the more common scenario of horizontally transferred resistance [13]. In turn, this 49 has led to studies seeking to identify collateral sensitivities that are more directly relevant in 50 clinical settings. For example, 10 genetically-diverse clinical urinary tract isolates of E. coli 51 showed broadly conserved patterns of collateral sensitivity to a panel of 16 antimicrobials 52 [14]. With evidence accumulating that collateral sensitivities may indeed be predictable [15, 53 16] -but with outstanding questions about how to apply this knowledge -the field is primed 54 to advance rapidly.

56
In this work we aimed to develop, implement and validate a high-throughput screen for 57 collateral sensitivities. Agar dilution and broth dilution methods [17] to determine differences 58 in minimum inhibitory concentration (MIC) are highly accurate but labour intensive. As a 59 result, it is unusual for any given study to test for collateral sensitivities towards >20 60 antimicrobials. We hypothesised that higher throughput experiments might reveal previously-61 overlooked sensitivities, which in turn could help to accelerate the field. suggesting that data collection by eye or by OmniLog is equally valid. We set out to assess 73 whether PM plates could provide sensitive and reproducible enough data to be useful in 74 building collateral sensitivity networks. We concentrated on scoring growth data by eye, in 75 order to develop a protocol with the broadest possible applicability.

77
In our proof-of-principle experiments, we have focused on the Gram-positive bacterium, S. 78 aureus. We have compared resistance and sensitivity to 72 antimicrobials (three PM plates) 79 between seven clinical isolates of methicillin-resistant S. aureus (MRSA), one laboratory 80 strain, and one descendent of this strain that was evolved in vitro towards oxacillin resistance.

81
Overall we found PM plates to provide reproducible data, although their correlation with 82 broth microdilution was more variable.  Cell lines to be assayed were cultured overnight in 3 mL of MH broth. Each saturated culture 112 was diluted to an OD 600 of 0.5 ± 0.05, for use as the PM plate inoculum as described below.   After incubation, growth in each well was scored by eye. Each well was given a score out of The first well which showed ≥95% inhibition was deemed to be the MIC cut-off for that 156 isolate.

159
We assessed the accuracy of PM plates for the discovery of collateral sensitivity networks by 160 assaying seven clinical isolates and two laboratory strains of S. aureus. We set out to test two 161 aspects of the accuracy of these assays: reproducibility and reliability. Reproducibility was 162 tested by comparing independent biological replicates in duplicate microarray assays.

175
The full list of compounds is provided in the Supporting Information (S1 Table). As 176 described in the Materials and Methods, relative resistance to each antimicrobial was 177 assessed using a growth score that ranged from 0 (completely sensitive) to 8 (maximally   Comparing the replicated sets of PM data revealed that the growth scores for each 208 antimicrobial were identical between duplicates 84% of the time (Fig 2). When the growth      On the other hand, differences in PM growth score of less than 2 units could not reliably 285 predict differences in MIC. For example, growth scores of 4, 5 or 6 could all correspond to a 286 demeclocycline MIC of 0.0156 µg/ml (Fig 4). Similarly, growth scores of 5, 7 or even 7.5 287 could correspond to a nickel chloride MIC of 64 µg/ml. PM wells; that is, they had growth scores of 8. However, the cefazolin MICs of these strains 295 ranged from 1 µg/ml to 16 µg/ml (Fig 4). The most likely explanation is that the four PM 296 wells containing cefazolin range from a very low concentration up to, perhaps, 0.5 µg/ml. If 297 this is the case, any strain with an MIC > 0.5 µg/ml will show full growth in all four wells; 298 however, it becomes impossible to assess the relative resistance or sensitivity of any strain 299 that fulfils this criterion.

302
This study has emphasized the power -and potential pitfalls -of PM plates for the large-303 scale assessment of cross resistance and collateral sensitivity. We were able to rapidly obtain 304 resistance and sensitivity data for 72 antimicrobials, which is many more than have been 305 tested in previous studies [8,12,14,15]. It would be straightforward to expand our approach 306 to more of the antimicrobial-containing PM plates. In total, plates 11-20 contain 237 307 antimicrobials; testing the entire set would represent an order of magnitude increase in 308 screening breadth compared to current approaches. Moreover, scoring PM growth by eye (Fig   309   1) proved to be a fast, technically straightforward, cost effective and reproducible way to 310 collect resistance and sensitivity data. In independent duplicates, carried out several weeks 311 apart, we obtained identical growth scores in 544 of 648 antimicrobial/strain combinations. When PM growth scores were carefully compared with MIC data obtained by broth 318 microdilution, a trade-off between throughput and accuracy became apparent. Small 319 differences in PM growth score did not reliably correlate with MIC. Our data suggest that a 320 difference in growth score of at least 2 points is required to indicate a genuine difference in 321 MIC between two S. aureus strains. For example, the PM assays suggested that many of the 322 MRSA strains were more sensitive to demeclocyline, nickel chloride or 2,2-dipyridyl than 323 the laboratory strain S. aureus ATCC 25923 (Fig 3). However, the differences in PM growth 324 score were small and the evidence for increased sensitivity was not borne out by MIC testing 325 (Fig 4). The level of agreement (or disagreement) between PM scores and MICs was