Ordering Genes in a Flagella Pathway by Analysis of Expression Kinetics from Living Bacteria

See allHide authors and affiliations

Science  15 Jun 2001:
Vol. 292, Issue 5524, pp. 2080-2083
DOI: 10.1126/science.1058758


The recent advances in large-scale monitoring of gene expression raise the challenge of mapping systems on the basis of kinetic expression data in living cells. To address this, we measured promoter activity in the flagellar system of Escherichia coli at high accuracy and temporal resolution by means of reporter plasmids. The genes in the pathway were ordered by analysis algorithms without dependence on mutant strains. The observed temporal program of transcription was much more detailed than was previously thought and was associated with multiple steps of flagella assembly.

Under the proper conditions, the bacterium E. colisynthesizes multiple flagella, which allow it to swim rapidly. Classical genetics showed that the 14 flagella operons are arranged in a regulatory cascade of three classes (1–5) (Fig. 1). The class 1 operon encodes the transcriptional activator of class 2 operons. Class 2 genes include structural components of a rotary motor called the basal body–hook structure, as well as the transcriptional activator for class 3 operons. Class 3 includes flagellar filament structural genes and the chemotaxis signal transduction system that directs the cells' motion. A checkpoint mechanism ensures that class 3 genes are not transcribed before functional basal body–hook structures are completed (Fig. 1).

Figure 1

The genetically defined hierarchy of flagellar operons in Escherichia coli (1, 2). The master regulator FlhDC turns on class 2 genes, one of which, FliA, turns on class 3 genes. A checkpoint ensures that class 3 genes are not turned on until basal body–hook structures (BBH) are completed. This is implemented by FlgM, which binds and inhibits FliA. When BBH are completed, they export FlgM out of the cell, leaving FliA free to activate the class 3 operons (9, 27,28). Note that flgM is transcribed from both a class 2 (flgAMN) and a class 3 (flgMN) promoter.

Here, we developed a system for real-time monitoring of the transcriptional activation of the flagellar operons by means of a panel of 14 reporter plasmids in which green fluorescent protein (GFP) (6) is under the control of one of the flagellar promoters (7). This allowed us to extend previous timing studies that depended on lacZ fusions to up to four operons (8, 9). Use of GFP eliminates the need for cell lysis required for lacZ and DNA microarray studies (10–13). Therefore, the present system makes it possible to measure accurately continuous time courses from living cells grown in a multiwell plate fluorimeter (14). Average errors between repeat experiments were less than 10%, compared with errors of at least twofold often associated with expression assays requiring cell lysis and manipulation (10–12).

The flagella system is turned on during the exponential phase of growth. Clustering the fluorescence levels of the operons (Fig. 2A) according to similarity in their expression profiles (10–13) showed that they fall into clusters that correspond to the genetically defined classes 1 and 2 (Fig. 2B). Three of the six class 3 operons are close to the compact class 2 cluster, and the other three are in a separate cluster. This separation is based mainly on different coordinated responses of the operon classes. To determine the timing order, we extended the clustering algorithm with a temporal labeling procedure that hierarchically orders the clusters according to the relative timing of their average expression profiles. Log fluorescence of each reporter strain, normalized by its maximum for each experiment, was set to zero mean and variance one, and clustered by means of a standard single-linkage algorithm with a Euclidean metric (Matlab 5.3, Mathworks) (15). In general, clustering algorithms do not specify an ordering of the clusters. In the resulting dendrograms, as the data are split hierarchically into a tree, pairs of subtrees in each splitting are placed in an arbitrary order. To define the temporal order of expression, we first considered each splitting from the top down and computed the average log fluorescence (normalized by the maximal fluorescence) for the two subtrees, log(f 1) and log(f 2). Next, we computedt i = –∫ log[f i(t)]dt(generally the earlier a sigmoidal curve rises, the smaller itst i. Since log fluorescence is used, the initial rise timing is emphasized.) The subtree with the smallert i was then positioned to the left. The present algorithm was able to correctly order simulated gene cascades. The algorithm is available upon request or

Figure 2

(A) Fluorescence of flagella reporter strains as a function of time, normalized by the maximal fluorescence of each strain. Average of five experiments in growth condition A (see below) are shown (bars, SD). Class 1, 2, and 3 operons are marked in blue, red, and green, respectively. (B) Fluorescence of flagella reporter strains as a function of time for two experimental conditions. Log intensity of each promoter, normalized by its maximal value in each experiment, scales from blue (low) to red (high). Operons are arranged according to the temporal clustering results. The first 630 minutes of each experiment, for two growth conditions, with and without preexisting flagella, are shown. Condition A: Stationary-phase cultures with two to five flagella per cell (29) are diluted 1:600 into fresh medium; induction of new flagella begins after about three to four generations, and thus old flagella are diluted out by cell division to a degree that most cells have no preexisting flagella. Condition B: Overnight cultures are diluted 1:60. The flagellar operons are turned on within one cell generation so that old flagella are present. The presence or absence of preexisting flagella was verified by microscopic observation of cell motility as described (23). Dendrogram shows hierarchical gene clustering and temporal order. The statistical significance (P value) for temporal ordering at each splitting was determined by the fraction of times that a larger |t 1t 2| value was found upon clustering and labeling 1000 randomized data sets generated by randomly permuting the gene coordinates at each time point. Similarly, a P value for clusters was determined by the fraction of times that a larger splitting distance occurred in the randomized data sets. Clusters with significance P < 0.001 are marked with filled triangles; P ≈ 0.01 with an open triangle; and P > 0.01, no triangles. Temporal ordering of all tree splittings is significant (P < 0.01), except the splittings marked with a star.

The algorithm arranged the operons in the order: class 1 followed by class 2 followed by class 3 (8, 9) (Fig. 2B). Within the class 2 cluster, the promoters were turned on sequentially, with significant delays, in the order fliL, fliE,fliF, flgA, flgB, flhB, andfliA (Fig. 2). The observed order corresponds to the spatial position of the gene products during flagellar motor assembly, going from the cytoplasmic to the extracellular sides (1,2) (Fig. 3). The fliL operon genes form the cytoplasmic C ring, and fliE and fliF genes form the MS ring in the inner membrane, thought to be the first assembled structure (1). The flgA,flhB, and flgB genes participate in the export and formation of the periplasmic rod, the distal rings in the outer membrane, and the extracellular hook. The transcription factor responsible for turning on class 3 genes, fliA, is the last class 2 gene to turn on.

Figure 3

Present kinetic classification of the flagellar operons. The three clusters and the operons within each cluster are arranged by their relative timing according to the temporal clustering results. Positions of the corresponding gene products in the flagellum (1) are indicated in green.

A separation of class 3 genes into two kinetic groups was seen, with the filament structural operons flgK, fliD, andfliC activated first, and flgM and the chemotaxis operons meche and mocha going on only after a substantial delay (Fig. 2). Thus, the hardware for the flagellar propeller is expressed before the chemotaxis navigation system (Fig. 3). The genes for motor torque generation, motAB in themocha operon, are in the late class 3 group, and indeed, it has been shown that they can be functionally incorporated long after motors are assembled (16, 17).

When flagella were induced in cells with no preexisting flagella, a temporal separation between most class 2 genes and class 3 genes was observed (Fig. 2B, condition A); whereas in cells with preexisting flagella, the delay between class 2 and the early class 3 genes decreased drastically (Fig. 2B, condition B). This probably reflects the checkpoint in flagella biosynthesis (Fig. 1). When preexisting flagella are present, newly synthesized FlgM is exported from the cells even before new basal bodies are completed. This frees FliA to turn on class 3 genes at an earlier time. Such memory effects may be a general kinetic signature of regulatory checkpoints.

A simple hypothesis for the mechanism underlying the temporal order of promoter activation within classes 2 and 3 is that the DNA regulatory sites in the promoter regions of the operons are ranked in affinity. As the concentration of the relevant transcription factor (FlhDC, FliA) gradually increases in the cell, it first binds and activates the operons with the highest affinity sites, and only later does it bind and activate operons with lower affinity sites.

The standard genetic method of pathway analysis suffers from the limitation that conclusions drawn from mutant cells sometimes apply to a physiological state far from wild-type. The present kinetic analysis can complement genetics by probing cells with an intact regulatory system, rather than mutant cells. For example, class 3 operons were subdivided by mutant analysis into class “3a” and class “3b” (Fig. 1), based on residual expression in a fliA mutant of class 3a but not 3b operons (1). This mutant may exemplify a situation never reached by wild-type cells (high FlhDC but no FliA). The present kinetic subdivision of class 3 operons into early and late temporal groups hints at a functionally reasonable order.

The precise order of transcription of the various operons is probably not essential for assembling functional flagella. This is suggested by complementation experiments in which the motility of flagella mutants was rescued by expression of the wild-type gene from a foreign promoter (1). The detailed transcription order could, however, function to make flagella synthesis more efficient, because parts are not transcribed earlier than needed. From the viewpoint of reverse engineering, this may be exploited to decipher detailed assembly steps from transcription data.

The present experimental method can be readily applied to gene systems in a broad range of sequenced prokaryotes, as well as to eukaryotic genes with well-defined regulatory regions. For example, GFP was used to monitor gene expression on a large scale in yeast (18). Studies on various systems could establish whether temporal clustering and memory effects can be a general method in mapping assembly cascades and detecting regulatory checkpoints. It would be important to discover whether, in analogy to the systems-identification principals of engineering, there are ways of mapping additional system motifs, such as feedback loops, by using detailed expression measurements.

  • * To whom correspondence should be addressed. E-mail: urialon{at}


View Abstract

Navigate This Article