Discrimination of Modes of Action of Antifungal Substances by Use of Metabolic Footprinting

Diploid cells of Saccharomyces cerevisiae were grown under controlledconditions with a Bioscreen instrument, which permitted theessentially continuous registration of their growth via opticaldensity measurements. Some cultures were exposed to concentrationsof a number of antifungal substances with different targetsor modes of action (sterol biosynthesis, respiratory chain,amino acid synthesis, and the uncoupler). Culture supernatantswere taken and analyzed for their "metabolic footprints" byusing direct-injection mass spectrometry. Discriminant functionanalysis and hierarchical cluster analysis allowed these antifungalcompounds to be distinguished and classified according to theirmodes of action. Genetic programming, a rule-evolving machinelearning strategy, allowed respiratory inhibitors to be discriminatedfrom others by using just two masses. Metabolic footprintingthus represents a rapid, convenient, and information-rich methodfor classifying the modes of action of antifungal substances.

INTRODUCTION

Top

Introduction

For scientific reasons and because of the needs of the pharmaceuticaland agrochemical industries, there is much current interestin detecting the site or mode of interaction between an exogenousligand and a cell or organism, especially in identifying newones (6, 44, 45). Such studies are nowadays typically carriedout by using high-throughput methods, but there is often aninverse relation between the speed of an assay (involving, e.g.,a cloned receptor with a fluorescence readout) and the amountof information it contains (which in this case is restrictedto the target of interest). However, array-based methods areshowing promise in this regard (3, 20, 34, 37, 46). Genome-widescreens can also provide such information (9, 10, 19) but requirenumerous strains or cell lines to be studied in parallel. Methodswith high information content would combine the virtues of screeningjust a small number of strains with a high-dimensional readoutsimilar to that provided by the array-based methods.

Metabolic control analysis (see, e.g., references 7, 15, 16,21, and 29) tells us that while changes in the activity of atarget enzyme tend to have only small effects on the flux throughmetabolic pathways, they can and do have substantial effectson the concentrations of metabolic intermediates. Since theuse of metabolomics, especially in the form of "metabolic fingerprinting,"has a much higher throughput and is much cheaper than are, say,transcriptomics and proteomics (8, 14, 26, 42), it makes anattractive candidate for mode-of-action studies (18) and hasbeen used to advantage by Ott and colleagues (2, 38), wherethe metabolic fingerprints of cells or tissues exposed to specificsubstances with known targets were analyzed for their modesof binding or action by using pattern recognition techniques.

Normally, metabolomics measures (or seeks to measure) the concentrationsof all the small molecules within a cell or tissue; and forpurposes of functional genomics, we and others have exploitedsuch strategies in the analysis of gene deletion mutants inSaccharomyces cerevisiae (5, 41). However, because of the needto separate cells from medium and to extract the metabolomeof the former separately, these methods are not as convenientas one would wish. In consequence, we have recently shown thatthe analysis of the metabolic footprint, the cocktail of metabolitesleft behind or secreted into the medium, differs reproduciblyin single-gene mutants and can be measured rapidly and convenientlyvia direct-injection mass spectrometry (1) or Fourier transforminfrared spectroscopy (22).

The question of whether metabolic footprinting of strains treatedwith different pharmaceuticals might also be used in discriminatingthe modes of action of different substances then arises. Thepurpose of the present study was to explore this question, andwe show here, using metabolic footprinting and both unsupervisedand supervised methods of machine learning analysis, that wecan indeed classify a number of fungicides according to theirmodes of action.

MATERIALS AND METHODS

Top

Materials and Methods

Fungicide screening. (i) Yeast strain.
The BY4743 diploid strain (MATa/MAT ${alpha}$ ura3 ${Delta}$ 0/ura3 ${Delta}$ 0 leu2 ${Delta}$ 0/leu2 ${Delta}$ 0his3 ${Delta}$ 1/his3 ${Delta}$ 1 +/lys2 ${Delta}$ 0 met15 ${Delta}$ 0/+) was used throughout.

(ii) Screening methodology.
A library of agrochemical fungicides supplied by Aventis CropSciencewas screened for activity against yeast in a Bioscreen microtiterplate (Honeycomb II plates; Labsystems) format. Stocks (25 mgml^–1) of each compound were prepared in dimethyl sulfoxide(DMSO). From these stocks, three additional stocks (2.5 mg ml^–1,250 µg ml^–1, and 25 µg ml^–1) were preparedby serial dilution, such that each compound could be screenedat log concentrations so that a broad picture of its inhibitionbehavior could be obtained. Stocks were aliquoted into glassvials and stored at –40°C.

From a washed starter culture, a stock of 5 x 10⁵ cells ml^–1was prepared, and 550 µl was aliquoted into a series of1.5-ml sterile microcentrifuge tubes. A total of 5.5 µlof fungicide at each concentration to be tested (250, 25, 2.5,and 0.25 µg ml^–1) was then added to the appropriatemicrocentrifuge tube and vortexed to ensure good mixing. DMSOcontrol tubes (5.5 µl of DMSO added to 550 µl ofcell suspension) were also set up. From each microtube, 100µl was pipetted into each of five adjacent Honeycomb platewells to produce a set of five replicate wells per treatment.Plates were then placed in the Bioscreen, where they were incubatedat 30°C and shaken at medium intensity. Optical density(OD) at 600 nm was measured every 20 min for the duration ofthe screen (typically, 24 to 36 h).

The percent inhibition for each compound was calculated as thedifference in the OD of treated cultures between 0 and 17 hexpressed as a percentage of the change in the OD of the DMSOcontrols over the same time span.

(iii) Metabolic footprinting of fungicide-treated cultures.
All footprinting of fungicide-treated cultures was performedin (Bioscreen) microtiter plate format. For all footprintingstudies of fungicide-treated cultures, each compound (whichhad been demonstrated to have reasonable activity against yeastfollowing screening) was added at four different concentrationswhich had been found to inhibit yeast growth over the approximaterange of 20 to 80% (compared with DMSO controls). The aim wasto obtain an approximate fixed percentage of growth inhibitionsuch that any changes in the footprint could be ascribed tothe mode of action of the inhibitor and not, say, changes ingrowth rate.

Footprints were typically collected at 24 h (and no later than28 h) at a point at which the majority of cultures appearedto have entered stationary phase (1). Plates were then removedfrom the Bioscreen and 90 µl was removed from each well,placed in a sterile 0.5-ml microcentrifuge tube, and spun for5 min. Approximately 80 µl of the resulting supernatantwas then removed and stored at –40°C in polystyrenemicrotiter plate (Nalgene) format until mass spectrometric analysis.Samples were diluted for mass spectrometry and analyzed exactlyas described previously for deletion mutant footprints (1).

(iv) Mass spectrometry.
Direct-injection mass spectrometry was performed exactly asdescribed previously (1).

(iv) Numerical methods.
Principal-components (PC) analysis (PCA), discriminant function(DF) analysis (DFA), and hierarchical cluster analysis (HCA)were carried out by using programs written in MATLAB (Mathworks,Cambridge, United Kingdom) exactly as described previously (1).Genetic programming was performed with the program gmax-bio(Aber Genomic Computing [http://www.abergc.com])

RESULTS AND DISCUSSION

Top

Results and Discussion

A typical set of growth curves of the diploid strain in theBioscreen, at different concentrations of inhibitors, is givenin Fig. 1. It is clear that the Bioscreen provides an accurateand convenient system for registering the growth of these cellsand its inhibition by the inhibitors. It also shows that, withthe possible exception of fenpropimorph, there is no regrowthin the presence of the inhibitors, i.e., a phenotypic (or conceivablygenotypic) emergence of resistance in the cells after they havebeen exposed to the inhibitors (as has been noticed, for example,in bacterial experiments [40]).

View larger version (105K):
[in this window]
[in a new window]
FIG. 1. Growth curves (each an average of results from five replicate plate wells) from Bioscreen microtiter plate wild-type diploid cultures treated with 10 different (respiratory and nonrespiratory) fungicides at four different concentrations, plus DMSO control. (Concentrations are given in the graph key as parts per million.)

The effect of the different concentration categories on theinhibition profiles of the nonrespiratory inhibitors is givenin Fig. 2, and the equivalent concentrations are shown in Table1. The two concentrations of respiratory inhibitors used aregiven in Table 2.

View larger version (22K):
[in this window]
[in a new window]
FIG. 2. Inhibition profiles (when screened against wild-type diploid BY4743, on glucose-containing footprinting medium) induced by the (a) nonrespiratory and (b) respiratory fungicides selected for the metabolic footprinting mode of action comparison when applied at the concentrations given in Tables 1 (nonrespiratory inhibitors) and 2 (respiratory inhibitors). Growth inhibition was calculated as the difference in the OD of cultures (an average of six replicates) between 0 and 24 h expressed as a percentage of the change in the OD of DMSO controls over the same time span.

View this table:
[in this window]
[in a new window]
TABLE 1. Concentrations of nonrespiratory inhibitors used in metabolic footprinting comparisons of fungicide-treated wild-type yeast

View this table:
[in this window]
[in a new window]
TABLE 2. Concentrations of respiratory inhibitors used in metabolic footprinting comparisons of fungicide-treated wild-type yeast

Metabolic footprints were taken from these cells and subjectedto mass spectrometric analysis exactly as described previously(1).

Following PCA (data not shown), DFA was performed on the data,a cross-validation analysis demonstrating that PCs 1 to 20 (99.9%of the variance) were sufficient to give reasonable clusteringaccording to compound class without compromising the overlapof training and test data (Fig. 3). HCA was additionally performedon the DFA scores (score, 1 to 3; averaged according to compound)in order that grouping relationships might be better visualized(Fig. 4). Clearly, the greatest separation is between respiratoryand nonrespiratory inhibitors, which form two separate branchesin the resulting dendrogram. Interestingly, within the respiratoryinhibitor branch, the complex III strobilurin-type inhibitorsform their own subbranch, while the complex II inhibitor carboxingroups with the DMSO control form a separate subbranch. A particularlyinteresting feature is that the classification works over arange of doses. This robustness to the actual dose representsa significant benefit in setting up experiments, since thiscould save a lot of time on preliminary, dose-setting experiments.It is noticeable that the DMSO control appears on the same branchwith inhibitory compounds rather than, as one might possiblyhope, forming a separate branch on its own. However, even ata concentration of 1% (128 mM), the level of DMSO is somewhatabove that (50 mM) which was demonstrated by Panek et al. (39)to affect various cellular processes, and so it cannot be consideredentirely benign.

View larger version (45K):
[in this window]
[in a new window]
FIG. 3. DFA was performed (and simultaneously cross-validated [1]) for PC scores (1 to 20, 99.9% of the variance) obtained from an analysis of footprints collected from wild-type diploid yeast cultures treated with 10 different inhibitors. DFA classes were assigned according to compound, regardless of the concentration at which it was applied. Labels are as follows: AE, AEC605025; AZ, azoxystrobin; CE, chlorimuron ethyl; CX, carboxin; E, epoxiconazole; F, fenpropimorph; FQ, fluquinconazole; T, triadimenol; Z, fluazinam. Concentrations 1 to 4 for each nonrespiratory inhibitor are given in Table 1. Concentrations 1 and 2 for each respiratory inhibitor are given in Table 2.

View larger version (17K):
[in this window]
[in a new window]
FIG. 4. The DFA scores (1 to 3) from the analysis illustrated in Fig. 3 were averaged according to compound (i.e., scores for the members of each class were averaged) and subjected to HCA. A separation of the respiratory and nonrespiratory inhibitors was observed in the resulting dendrogram. Fluazinam (marked with an asterisk) is cited as an uncoupler of oxidative phosphorylation (17), and although it might conceivably be regarded as a respiratory inhibitor, it is not, of course, a respiratory chain inhibitor, and the level of inhibition it induces in cells growing on a fermentable carbon source is too great to arise from the inhibition of respiration-coupled processes alone; and therefore, this compound must inhibit other reactions within the cell, most likely on some proton-coupled uptake process necessary for fermentative growth.

In a similar vein, we trained a genetic program (4, 12, 23,24, 30-33, 35, 36) to evolve rules that would discriminate respiratoryfrom nonrespiratory inhibitors. DMSO samples were removed, astheir status would of course be known. To guard against overtraining,the data were split into three partitions, in accordance withaccepted modelling practice (e.g., see reference 43). The geneticprogramming was trained on the first partition, terminatingevolution after 20 generations. This cycle was repeated 10 times.Each of the resulting 10 rules gave perfect classification onthe training partition. Each rule was then applied to the secondpartition in order to test its generalization capability, andof the 10 rules, 6 also gave perfect classification on thispartition. These six rules were then applied to the third partition(which had not been used in any way in forming or selectingthe rules), and in each case a perfect classification was obtained.In partitioning the data, replicates of individual inhibitorshad been retained within single partitions, and thus, the thirdpartition contained only samples treated with inhibitors thatwere not represented in either of the first two partitions butwhose modes of action were nevertheless correctly classified.

Each of the six successful rules used a combination of justthree m/z ratios to discriminate the classes. All but one ofthe six rules contained an m/z ratio of 144, but no other m/zratio appeared more than once. Plotting m/z 144 in turn againsteach of the other variables used by the rules revealed threecases (m/z 141, 179, and 189) where clusters visually separablewith straight lines were obtained (see the example of m/z 144versus 141 in Fig. 5). Thus, as in previously described relatedstrategies (1, 11, 13, 23, 25, 27), it seems that these classificationscan be performed effectively by using only a very small numberof the variables from these high-dimensional data sets, andthere are also likely to be other combinations of peaks in thesespectra that are equally able to discriminate classes. It ispossible that a chemical interpretation of these peaks may leadto the identification of useful markers for understanding thebiochemical basis of these discriminations.

View larger version (26K):
[in this window]
[in a new window]
FIG. 5. Discrimination of respiratory and nonrespiratory inhibitors by using genetic programming to select discriminatory variables. Data were acquired as described above in the legends to Fig. 3 and 4 for all samples except fluazinam, and a genetic program was trained by using gmax-bio to evolve rules that can discriminate respiratory (symbols inside the top left dotted box) from nonrespiratory (symbols outside of the dotted box) modes of action. Variable (m/z) 144 was a member of all but one of the selected rules. In this example, its values are plotted against those of one of several other variables (m/z 141) that were used by the rules, here resulting in linearly separable clusters (shown as dotted lines). Triangles, training set; closed circles, test set for validation purposes; open circles, independent test set.

Within the nonrespiratory branch, it was pleasing to observethat the sterol biosynthesis inhibitors formed one branch (albeitthe morpholine fenpropimorph forms a subbranch with the azoleepoxiconazole) while the amino acid biosynthesis inhibitor anduncoupler form another. The separation of the respiratory inhibitorsfrom the remaining compounds is scarcely surprising given thatthe former inhibitors give rise to only limited or no inhibitionof growth rate. Consequently, in order to determine whethersimilar grouping relationships might be observed between theremaining (nonrespiratory) compounds in the absence of variancecontributed by the respiratory inhibitors, data for these lattercompounds were removed from the data set, and the remainingcompounds were then resubjected to PCA, DFA (PC scores of 1to 20, employing the same compound-based class system), andHCA (DFs 1 to 3, averaged according to compound). The resultingdendrogram (Fig. 6) illustrates that the grouping relationshipsobserved in Fig. 4 are preserved among these nonrespiratoryinhibitors, in that the sterols form a group away from the restof the inhibitors and the uncoupler is separated from the acetolactatesynthase inhibitor.

View larger version (12K):
[in this window]
[in a new window]
FIG. 6. Data for the respiratory inhibitors were removed from the data set employed in the analyses illustrated in Fig. 3 and 4, and PCA and DFA were performed on the data for the remaining (nonrespiratory) inhibitors. Again, DFA classes were assigned according to compound, regardless of their concentration of application. DFA scores (1 to 3, averaged according to compound) were then subjected to HCA, and a dendrogram was produced.

The attractions of the metabolic footprinting approach are itsrapidity, ease of implementation, and reproducibility, whileas a pattern recognition strategy (8, 14), it is not designed(immediately) to determine the molecules that are most importantto or responsible for any particular discrimination. However,the simplicity of some of the rules determined by the geneticprogramming approach means that, in many cases, just a few rulesare sufficient (e.g., as shown in Fig. 5). This result meansthat if a tandem mass spectrometer is available, only thosemasses need to be studied to achieve identification (and thesecould then be confirmed by other methods such as gas chromatography-massspectrometry or nuclear magnetic resonance. Such an iterativestrategy is entirely appropriate for postgenomic studies (28).

In conclusion, the present work has shown, complementarily toprevious studies in which gene knockouts were used (1), thatthe metabolic footprints of wild-type strains treated with inhibitorswith different modes of action contain sufficient informationto allow such modes of action to be discriminated from the metabolicfootprints alone.

ACKNOWLEDGMENTS

This work was supported by a grant from the Biotechnology andBiological Sciences Research Council, United Kingdom, to D.B.K.and S.G.O. J.A. was the recipient of a BBSRC CASE studentshipwith Bayer CropScience (formerly Aventis CropScience).

We thank John Pillmoor, Steve Dunn, and Jane Dancer for theircareful supervision and John Pillmoor for a careful readingof the manuscript. We thank Bharat Rash and Nicola Burton, ofthe Manchester laboratory, for technical assistance and RoyGoodacre (Aberystwyth/UMIST) for useful discussions.

FOOTNOTES

* Corresponding author. Present address: Department of Chemistry, UMIST, Sackville St., P.O. Box 88, Manchester M60 1QD, United Kingdom. Phone: 44 161 200 4492. Fax: 44 161 200 4556. E-mail: dbk{at}umist.ac.uk.

REFERENCES

Top