Section: New Results

Data: High-precision measurements of gene expression in bacteria

During this past year we spent much effort in improving the quality and extent of the experimental observations by (i) modifying the reporter gene system and validating key steps of the reporter system, (ii) constructing expression vectors that allow subtle perturbations of the regulatory system, and (iii) developing a new method for identifying the pertinent genes to be included in a model.

Improvements of gene expression measurements

We have improved the gene expression measurements by exploring new variants of the green fluorescent protein (GFP), by transferring some of the constructions to the chromosome and by measuring and validating intermediate steps of gene expression. This work involves Corinne Pinel and Caroline Ranquet-Brazzolotto, post-doctoral researcher in the framework of the EC-MOAN project.

We are using two types of reporter genes: bacterial luciferase and variants of the GFP. The luciferase is much more sensitive than GFP, but since it is an enzyme, its activity depends not only on gene expression, but also on the metabolic state of the cell. We therefore systematically construct an identical reporter vector using GFP. An important consideration for the measurements is the half-life of the reporter protein. Rapid changes in gene expression are difficult to measure with very stable reporter proteins. We have therefore engineered GFP by adding (or not) a degradation tag in order to control its half-life.

Our reporter constructs are located on a low copy-number plasmid (about 20 copies per cell). However, we can not formally exclude that some of our experimental manipulations change the plasmid copy number. In this case, we would misinterpret signal changes as changes in gene expression when really they only reflect changes in plasmid copy number. We have therefore constructed two "platforms" on the chromosome of E. coli into which we can transfer our reporter constructs, assuring that there is exactly one copy per cell. We have adapted this new system to the constraint, particularly important for the gfp constructs, that a single copy decreases the signal intensity by a factor of 20.

The relevant variable in our models, described by most equations, are the promoter activity of a gene under investigation and the concentration of its protein. However, reporter genes measure the accumulation of the reporter protein after transcription of the gene into mRNA and translation of this mRNA into protein. In order to relate the signal (reporter gene activity) to the relevant variables (promoter activity, protein concentration) we have to make assumptions about the kinetics of the intermediate steps. In order to validate these assumptions we have directly measured the relevant quantities (mRNA accumulation and stability) for one model gene.

An article describing the validation of the models used for the interpretation of the reporter gene data is currently under revision. The computer tool WellReader (Section  5.3 ) assists the user in transforming the primary data in biologically relevant variables. Version 3 of this tool has been deposited at the APP and was released in the fall of this year. An application note on WellReader has been accepted for publication [2] ."

System perturbations

In order to probe the system properties, we have to perturb the system and observe the dynamics of the system response by means of reporter genes. Until now, most of our perturbations consisted in changing the nutrient source of the bacteria. This affects the activity of a small number of genes. In order to perturb the system more systematically, Caroline Ranquet-Brazzolotto and Corine Pinel have constructed a series of expression vectors that allow the controlled induction of any particular gene at a precise moment during the time course of the experiment. We have characterized the induction profile of our expression vector and have constructed a second vector, using a different external signal, which will allow us to externally control the expression of two different genes in the bacterium.

We have also modified some of the proteins that are expressed from these vectors by adding a degradation tag to the proteins. The second vector can then be used to express a specific protease that will destroy the target protein when the external signal is given. We are currently calibrating this system, developed by Guillaume Baptist in the framework of his PhD thesis. These constructs will allow us to perturb the system in many different and very controlled ways, in exact analogy to modifications that can be introduced into the dynamical model of the network.

Somewhat 'cruder' perturbations consist in eliminating an entire node of the network. This corresponds biologically to the deletion of the corresponding gene. We have explored many of these single mutants (see below), but we have also begun to construct multiple mutants, where two or more genes of the network are removed. This serves to further probe the system behavior and test model predictions. This serves to further probe the system behavior and test model predictions. An even more radical approach consists in the redesign of part of this network, as is for instance undertaken in the PhD thesis of Jérôme Izard, funded by INRIA in the framework of the Action d'Envergure ColAge (Section  8.1 )."

Detecting the input function of a gene

Another important task for the modeling and the biological understanding of a process is to well delimit the system boundaries. While high-throughput methods for detecting the target of a particular regulator are now classical (typically, DNA microarrays are used to study the effect of a particular mutation on the expression of all genes of a genome), no efficient method exists to determine the regulators that affect, directly or indirectly, the expression of a gene under investigation (the 'input function').

We have developed such a technique by making use of the Keio mutant collection of E. coli and devising a method for efficiently transforming our reporter plasmid into more than 4000 different clones. We have optimized the different steps of the procedure: transformation, detection of different signals (luminescence or coloration), image analysis, mutant selection and dynamical measurements of gene expression in the selected mutants. We have applied the method for identifying regulators that control the expression of genes that are critical for growth on acetate by E. coli , and regulators that modulate the expression of genes responsible for extracellular structures, so-called curli. This work has involved all members of the experimental side of IBIS and is currently being prepared for publication.

In both cases we have confirmed known regulatory mechanisms, but we have also discovered novel inputs to the regulation of these genes. We have established collaborations with other laboratories in France and Europe, specialized in the measurement of metabolites or metabolic activities, that will complement our measurements of gene expression. This direction is paralleled on the modeling side by the inclusion of metabolic reactions in the regulatory scheme and the resulting development of coarse-grained models and methods for model reduction (Section  3.1 ).