A major goal of systems biology is to elucidate the molecular networks that underlie cellular decision-making and predict emergent properties of the system. Knowledge of molecular networks provides novel insight into the mechanisms underlying both physiological and pathological cellular processes. Such networks were constructed in yeast 
, Escherichia coli 
, Saccharomyces cerevisiae 
and human 
, mostly using large-scale genetic manipulation in order to identify gene to gene interactions, non-coding RNA interactions, and gene to phenotype interactions. These networks were analyzed, and the function of several network components was elucidated 
High-throughput gene expression assays, such as microarrays and quantitative real-time PCR, provide insights into mechanisms mediating normal physiology and disease states. Gene assays have been used to identify novel genes associated with specific cellular events or phenotypes, and to unravel interaction networks between the genes. Still, for some of the important questions facing cell biologists, the statistical and mathematical approaches used to analyze these data are not applicable. Specifically, the activity state of many signaling components mediating the cellular response (e.g. some scaffold proteins or transcription factors) cannot be measured in systematic high throughput assays, and therefore the interactions between them are not directly decipherable by these approaches.
Several methods have been developed to reconstruct signaling networks from experimental data 
. However, most of these methods rely on measuring the activity levels of the signaling components in question under several conditions, and therefore require a large number of experiments for each signaling component. This applies for bottom-up approaches that use experimental determination of individual biochemical interactions to reconstruct the network 
, as well as for many top-down approaches such as partial least squares (PLS) 
, modular response analysis (MRA) 
, many methods using Bayesian inference 
, and methods based on dynamic properties 
. The few methods that do not require measuring the activity of the signaling components rely on creating large interaction databases by performing many experiments 
, or integration of large databases from several sources 
. Although the latter methods can be useful for finalizing well-studied networks, they are not appropriate when little data is available about the network.
Gene profiling has been previously used to find interactions between various molecules 
, but the focus has been on late time points, when gene activity reaches a quasi-steady state. At these time-points the initial signal from the signaling component is partly degraded due to feedback and cross-talk between the genes. At later time points, the changes in expression of many genes may fail to show a simple function that correlates with the activity of the upstream signaling components responsible for regulating that gene 
. Thus the use of gene expression measured more than ~one hour after modulation of the system can provide a non-mechanistic pattern-matching representation of cellular state. However such approaches may not allow a quantitative reconstruction of the biochemical network in a way that is analogous to construction of a network using measurement of the protein activity states themselves.
A potential technique to measure indirectly the activity levels of signaling components is presented by measuring the activity of early genes, which are defined as genes that do not require any de
-novo synthesis in order to start their transcription. Specifically, their regulatory transcription factors are pre-formed and the activation states of these factors are altered by modulations in cellular signaling. As a result, their promoters act as direct, quantitative sensors of the cellular signaling state 
. Such genes are thus the first genes to be induced following a change in the cell's condition, and are usually activated within minutes. To illustrate the linear function correlating signaling components and early genes measured at early time points, we exposed gonadotrope cells to the hormone GnRH at varying concentration and measured, the resulting levels of one active signaling molecule, phosphoERK, as well as the levels of transcripts for several early genes and non-early genes at 0.75 and 5 hours (). The results show that all of the early genes are linearly correlated with the levels of phosphoERK and the correlation is much higher, (R2
ranging from 0.92 to 0.99) when measured at 0.75 hours than when measured at 5 hours. Therefore, if additional experiments were performed, such as ERK inhibition, to determine which of these genes are most dependent on activation of ERK, the measurement of such genes would provide an indirect, yet sensitive and accurate measurement of the levels of phosphoERK. The relationship is most linear and most direct for early genes measured at early time points. In contrast, in secondary and tertiary genes, which require newly synthesized transcription factors or enhanceosome components to regulate their activity, their activity levels normally do not have a simple function relating it to the activity levels of upstream signaling components at any time point. Notably, the linear amplification between signals and early genes is the basis of the widespread use of synthetic gene reporter constructs to provide quantitative measurements which accurately reflect changes in cell signaling (e.g. using the activity of a cAMP response element reporter to reflect changes in adenylate cyclase activity). Because of these considerations, the utilization of early gene profiling provides an experimentally and computationally tractable approach to reverse engineer the interaction network of signaling components.
Response of early and late genes and their correlation to signaling activity.
Here we present a robust and efficient algorithm named PLACA that uses high throughput assays of early gene expression at early time points combines with perturbation of cellular components in order to uncover experimentally verifiable functional interactions between the components upstream of these early genes. Notably, in addition to the reverse engineered network, PLACA also identifies the specific genes that manifest the functional interaction. Thus PLACA facilitates experiments to validate the inferred interactions.
We tested the performance of PLACA by reconstructing a synthetic network, and found that when using several independent experiments it is robust to experimental noise. Additionally, we studied the early gene responses to signaling component perturbations in the pituitary gonadotrope and used PLACA to reverse engineer the network of this crucial component of the reproductive system. Many of the inferred functional interaction have been previously observed, and novel functional interaction predictions were then successfully tested experimentally.