DNA Microarrays in Analysis of Quorum Sensing: Strengths and Limitations

Quorum sensing in prokaryotic biology refers to the ability of a bacterium to sense information from other cells in the population when they reach a critical concentration (i.e., a quorum). In the context of this commentary the information sensed by some gram-negative bacteria is extracellular low-molecular-weight compounds called acyl homoserine lactones (AHL), whose levels in the medium are related to the population density of the producing species. AHL-mediated quorum sensing was first discovered in marine vibrios over 30 years ago, but Pseudomonas aeruginosa has arguably been the most thoroughly scrutinized species in this regard. Over 100 publications since 1994 were found in the PubMed database under the query “quorum sensing Pseudomonas aeruginosa.” Over one-half of these publications emanated from the laboratories of E. P. Greenberg and B. H. Iglewski, the corresponding authors of two separate reports (14, 20) in this issue describing the use of DNA-based microarray assays to evaluate the transcription of P. aeruginosa genes that are directly or indirectly regulated by AHL-mediated quorum sensing.

During the last decade a cadre of P. aeruginosa genes and phenotypes were reported to be affected by AHL-mediated quorum sensing, including genes and phenotypes that contribute to biofilm development (5, 15, 16), virulence in diverse hosts, including slime molds, plants, insects, invertebrate animals, and mammals (3, 4, 9, 12, 13, 23, 24), and resistance to antimicrobial agents (1, 6, 18). The bulk of these genes and phenotypes were identified by using genetic approaches, but sometimes different researchers reached contradictory conclusions about whether specific P. aeruginosa genes or phenotypes were responsive to quorum sensing (8, 19, 21). It is not unusual in experimental science to encounter apparently conflicting outcomes even when the same hypotheses are addressed by using nearly identical experimental methods. Fortunately, thoughtful analyses of apparently differing experimental data can provide novel insights into biological processes and provide fodder for further inquiry and conflict resolution. On the other hand, different experimental outcomes could be inconsequential or tangential to the main issue. They can be the residue of trivial technical matters or limitations of the methods employed, or they can be the result of a subtle or immaterial phenotypic characteristic that allows different interpretations of similar data. In some situations different outcomes, such as different conclusions concerning whether the transcription of a particular gene is regulated by quorum sensing, can have a salient impact on understanding a more complex phenotype (e.g., virulence). For example, conflicts about whether a major regulatory gene, such as rpoS (8, 19, 21), is influenced by AHL-mediated quorum sensing can affect the development of a cohesive image of how quorum sensing affects phenotypic characteristics (e.g., in vitro motility), gene expression during stationary-phase growth, and ultimately a complex phenotype (e.g., virulence in an animal model).

A little over 2 years ago the sequence of the genome of P. aeruginosa strain PAO1 was published (17), and 1 year later a commercially produced DNA microarray chip (Affymetrix GeneChip) was made available to investigators through the generous support of the Cystic Fibrosis Foundation. Like a tractable genetic system, the DNA microarray chip, which represents an entire genome, is a very powerful, but not omnipotent, tool to experimentally address myriad questions related to the basic biology of prokaryotic pathogens, including questions about quorum sensing. Accordingly, Wagner et al. (20) and Schuster et al. (14) in this issue provide impressive assemblages of data from microarray experiments in which they examined the transcriptional responses of P. aeruginosa genes to the presence of AHL signaling components. The two data sets by themselves are remarkable enough, especially considering how much new information was obtained in a relatively brief time compared to the time that it took to gather the previous anthology of data (more than 10 years). Data from these independent microarray experiments are also of considerable value because they affirm data generated previously by alternative methods (e.g., microbial genetics). They also illustrate that even though there may be some appreciable differences in the otherwise similar experimental designs used by two groups of investigators (see below), transcriptional analysis of gene expression by using DNA microarray chips can rapidly generate a wealth of new information that can be independently and rapidly corroborated when data are generated by different investigators. However, while there is a great deal of concurrence between the data sets acquired by Wagner et al. and Schuster et al., there are significant areas of disagreement worthy of further analysis.

CONCORDANT DATA—OLD AND NEW

Both Wagner et al. and Schuster et al. designed several excellent experiments to address whether the transcription levels of the genes represented on the P. aeruginosa microarray chip change in response to differences in cell density or growth phase or whether they are specifically responsive to AHL levels. These considerations are not trivial and are analogous to the difference between identifying all genes that are responsive to iron limitation (many genes) and identifying the genes that are responsive only to the novel signaling process triggered by the siderophore pyoverdine (a more limited number) (7, 10). To address these issues, both groups of investigators employed mutants deficient in AHL production (defective in luxI and rhlI) and compared the transcriptional responses in the mutant to those in the parental strain in the presence and absence of AHL. Shuster et al. also compared the transcript levels in a strain lacking the AHL receptors (deficient in lasR and rhlR), while Wagner et al. explored the responses in the presence of a range of growth media and other environmental conditions. Both groups reported that a striking number of genes were induced and repressed by AHL-mediated signaling. Shuster et al. found 315 and 38 such genes, respectively, and Wagner et al. found 394 and 222 such genes, respectively. The quorum-sensing regulon thus comprises about 6% of the genome. Many of the genes that were previously identified as being quorum sensing regulated were verified in the data of the two groups (Table 1). The two groups frequently agreed in the assignment of many genes or operons with known or unknown functions which had not previously been identified as responsive to AHL quorum sensing. Independent corroborations of this kind lead to new perspectives regarding the role of quorum sensing in P. aeruginosa gene expression. For example, both groups reported that AHL activates expression of PA0026, PA0027, and PA0028, all of which are currently listed as genes encoding hypothetical proteins in the latest P. aeruginosa database (http://www.pseudomonas.com). However, microarray data which coworkers and I obtained (not related to an analysis of quorum sensing) led us to assign functions to these proteins. PA026 encodes a novel extracellular phospholipase C, and PA0027 and PA0028 encode proteins required for posttranslational modification and secretion of this enzyme (A. Barker, A. Filloux, A. Vasil, and M. Vasil, unpublished data). Thus, it is not necessary to perform similar experiments to evaluate the expression of these genes, since the data in these reports provide us with a high level of confidence that their expression is influenced by AHL-mediated quorum sensing.

TABLE 1.

Comparison of quorum-sensing data sets from Wagner et al. and Schuster et al.

Open reading frame(s) (gene name and/or proposed function)	Change (fold)
	Wagner et al.a	Schuster et al.b
Quorum-activated genes: concordant data
PA0105 (coxB, cytochrome c oxidase)	6	4
PA0122 (fungal hemolysin)	29	51
PA0158 (RND efflux transporter)c	3	3
PA0852 (cbpD, chitin binding protein)	18	95
PA0026 (phospholipase C)	3	6
PA0027 (rotamase peptidyl-prolyl cis-trans isomerase)	4	6
PA0028 (phospholipase C chaperone)	3	8
PA1130 (hypothetical protein)	7	16
PA1246-PA1250 (aprADEFI, alkaline protease precursor and secretion apparatus proteins)	3-12	5-23
PA1901-PA1903 (phzCDE, phenazine biosynthesis)	16-51	18-208
PA2193-PA2194 (hcnABC, cyanide biosynthesis)	30-208	16-187
PA2580 (transcriptional regulator)	7	46
PA2591 (transcriptional regulator)	6	42
PA3479 (rhlA, rhamnolipid biosynthesis)	104	203
PA4175 (prpL, endoproteinase precursor)	6	23
PA4302 (type II secretion)	9.6	7.4
Large difference in magnitudes of changes
PA2300 (chiC, chitinase)	5	103
PA0996 (coenzyme A ligase)	5	218
PA1217 (2-isopropylmalate synthase)	30	383
PA1431 (rslA, regulatory protein)	8	352
PA3724 (lasB, elastase)	39	242
Quorum-activated genes: discordant data
PA0198 (exbB1, transport of small molecules)	NCd	10
PA0263 (hcpC, hemolysin coregulated protein)	NC	9
PA1000 (unknown function)	NC	44
PA1001 (phnA, anthranilate synthase component I)	NC	286
PA1002 (phnB, anthranilate synthase component II)	NC	28
PA1215 (unknown function)	NC	55
PA1914 (similar to halovibrin-ADP-ribosyltransferase)	NC	704
PA2512-PA2514 (antABC, anthranilate dioxygenase)	NC	67-604
PA1927 (metE, methionine biosynthesis)	20	NC
PA1999 (coenzyme A transferase subunit A)	32	NC
PA2000 (coenzyme A transferase subunit B)	46	NC
PA3622 (rpoS, sigma factor)	3	NC
PA4211 (phenazine biosynthesis protein)	74	NC
Quorum-repressed genes: concordant datae
PA0485 (hypothetical membrane protein)	2	3
PA3038 (porin)	2	4
PA3234 (sodium:solute symporter)	3	7
PA3235 (hypothetical membrane protein)	3	7
Quorum-repressed genes: discordant data
PA1978 (transcriptional regulator)	6	NC
PA1983 (cytochrome c550)	12	NC
PA2261 (2-ketogluconate kinase)	18	NC
PA3871 (rotamase peptidyl-prolyl cis-trans isomerase)	6	NC
PA3877 (narK1, nitrite reductase)	13	NC
PA0165 (outer membrane protein)	NC	5
PA0435-PA0437 (hypothetical proteins)	NC	9-34
PA3281-PA3284 (hypothetical proteins)	NC	21-28
PA4442 (cysN, amino acid biosynthesis)	NC	8
PA5168 (dicarboxylate transporter)	NC	6

^aData from reference 20.

^bData from reference 14.

^cRND, resistance-nodulation-cell division.

^dNC, no change.

^eThe genes in this group are all the genes that both groups of investigators agree belong to this classification. For all other groups the genes listed are only subsets of the genes that fall under the classifications.

The gene encoding an endoproteinase (PrpL) which plays a role in the virulence of P. aeruginosa in eye and pulmonary infections in mouse and rat models was previously identified as being induced by iron starvation and regulated by the alternative sigma factor PvdS (22), but it was not known to be affected by AHL-mediated quorum sensing. Clearly, the concordant data presented in the new reports provide new information that will be extremely valuable to any investigator interested in the role of quorum sensing in the basic biology and pathogenesis of P. aeruginosa. They may also provide useful information to researchers interested in developing antimicrobial agents against this opportunist that are targeted to quorum sensing (6).

DISCORDANT DATA—CAN WE AGREE TO DISAGREE?

Although there is considerable agreement between the conclusions described in the reports of Schuster et al. and Wagner et al., there are also noteworthy discordant results. These include (i) major differences in the magnitudes of the responses of numerous transcripts to AHL supplementation; (ii) some curious disparities with respect to whether some genes are responsive to AHL-mediated signaling; and (iii) a remarkably high level of discordance about the number of genes classified as being repressed by AHL signals. There is more than 50% agreement regarding the genes that both groups identified as being AHL inducible, but there is less than 5% concordance regarding the genes listed as being AHL repressible.

With regard to the first issue, although there is considerable qualitative agreement about whether many genes are responsive to quorum-sensing signals, there are prominent differences regarding the magnitudes of the transcriptional responses (Table 1). These differences manifest crucial issues regarding interpretation of microarray data, reflecting in part differences in growth conditions, strain history, and statistical methods used to compare transcript levels. Little significance should be ascribed to large differences in the magnitudes of the reported transcriptional responses, since the DNA microarray method is not a reliable method to quantitatively examine gene transcription. Certainly there must be a significant difference in the transcript levels of a given gene in the presence and absence of AHL so that investigators can determine whether the gene is responsive to AHL signaling. For example, a sixfold increase in the transcript level in response to the presence of AHL may be defined as the cutoff for identification of a gene as being AHL responsive. However, even if another group reports a 60-fold increase for the same gene, assignment of significance to these differences is probably not worthwhile. For example, Schuster et al. (14) reported a >300-fold increase in transcription of PA1431 in the presence of AHL, while Wagner et al. (20) reported an 8-fold increase. This difference is likely not biologically meaningful, and the real value of the data is that both reports support the conclusion that AHL-mediated quorum sensing regulates the expression of PA1431.

How can the discordant data be explained? In models of quorum sensing, the concentration of AHL in a natural environment, such as during growth of P. aeruginosa in soil or in a compromised host, increases progressively from a limiting value to a higher (i.e., inducing) value during growth of the population. However, it is unlikely that all AHL-responsive genes are off at a given concentration and on at a threshold inducing concentration. It is likely that an elaborate cascade of responses triggered by the varying concentrations of AHL dictates whether a particular gene is transcribed during a particular phase of growth. Even though the experiments described in the reports of Schuster et al. and Wagner et al. were carefully and rationally designed, the fact remains that both groups added a specific, arbitrarily chosen concentration of AHL to an arbitrarily chosen density of cells. This experimental protocol probably distorts the usual relationship between AHL concentration and cell density. It is also pertinent that the two groups of investigators added substantially different concentrations of AHLs to the cultures. Schuster et al. added 2 μM N-(3-oxododecanoyl) homoserine lactone and 10 μM N-butyryl homoserine lactone, while Wagner et al. added only 1 μM N-(3-oxododecanoyl) homoserine lactone and 2 μM N-butyryl homoserine lactone. It is not known whether these levels are limiting or saturating with respect to the ability of specific genes to be induced or repressed. Neither group evaluated the levels of the most significant regulatory proteins involved in this process (i.e., LasI, LasR, RhlI, and RhlR). They did examine transcription of these genes, but it is possible that the levels of these proteins were different in the two studies due to differences in the experimental protocols. The two groups used different growth media. Schuster et al. (14) used an undefined medium (buffered Luria-Bertani broth), which contained tryptone, yeast extract, and a significantly higher level of NaCl (86 mM) than the defined minimal medium used by Wagner et al. (0.5 mM).

Wagner et al. (20) reported that 24 genes were regulated by quorum sensing in their defined media but were not responsive in a complex medium (nutrient broth with yeast extract but no additional NaCl). It is probable that some genes are expressed in one of the media used by one group but are not expressed in a medium used by the other group, even though the gene is responsive to AHL signaling. A similar situation is observed for expression of genes encoding siderophore receptors, which requires both iron limitation and the presence of the cognate siderophore (10). Even a simple variable like NaCl concentration can have a tremendous influence on the results of carefully designed microarray assays. An even more subtle factor that may contribute to the differences in the magnitudes of the responses found for the same genes is the different liquid-to-air ratios used during growth of the cultures. This difference in an environmental factor affects the availability of oxygen, which could affect the expression of AHL-responsive genes. Such subtle differences in experimental design may not by themselves have affected conclusions about whether the transcription of some gene responds to AHL-mediated quorum sensing, but combined minor differences in experimental design may have been magnified and may have generated discordant outcomes in the studies discussed here. Other investigators using the same commercially produced microarray chips reported that some genes which appeared to be barely responsive to a particular stimulus (i.e., iron starvation) with the microarray techniques (3-fold increase, which might not be considered to be very interesting, especially when there are many genes that show a >10-fold increase in response) were highly responsive (>100-fold increase) to the same stimulus under identical growth conditions when other detection methods, such as fusions or RNase protection assays (10), were used. In this case the muted response observed in the microarray experiments was likely due to the particular set of oligonucleotides selected to represent the gene on the printed array. These issues provide a glimpse into some of the limitations associated with the use of DNA microarrays that could explain discordant results. Although DNA microarrays provide an impressive amount of data in a relatively short time, the data have to be cautiously interpreted and verified by alternative means.

Another source of discordance occurred when the signal detected by one group was too low to claim that it was regulated under the growth conditions that were used. Another factor to consider is whether secondary mutations arose during construction of the strains used in the studies. The expression of some genes could be affected if a cryptic mutation arose in a regulatory gene that is responsive to AHL signals. This phenomenon has been documented during the construction of P. aeruginosa mutants (2), and other workers have noted significant phenotypic variations between isolates of P. aeruginosa strain PAO1 from different laboratories (11).

The most difficult issue to reconcile is why there is such a high level of disagreement regarding the genes which are repressed by AHL. Only four genes (PA0485, PA0308, PA3234, and PA3235) of the more than 100 genes classified as being quorum sensing repressed were found by both groups to fit this classification. Because the same confounding factors described above should apply to both induced and repressed genes, the level of discordance for repressed genes should be similar to the level of discordance observed for induced genes. This is clearly not the case, and thus these remarkable differences are without a doubt interesting. They are a compelling impetus to carry out further experiments and hopefully to determine whether they result from different experimental conditions, reflect a significant limitation of DNA microarray assays for evaluating more subtle transcriptional responses, or constitute a significant and novel biological phenomenon that may reveal exciting new intricacies about AHL-mediated quorum circuits.