Systems biology provides a new paradigm to understand complex traits, such as carbon metabolism 
, homeostasis 
, development 
, response to environmental change 
, longevity 
, the clock 
, and life itself 
. This new approach has a number of common elements 
, including viewing living systems as biochemical and regulatory networks, measuring a system-wide response with genomic approaches as in RNA and protein profiling 
, and cycling through a rationalized discovery process to identify the true underlying network explaining a complex trait of interest 
. One application of this approach is being exploited to identify how DNA sequence variation elucidates molecular networks that cause disease 
. The challenge is that systems biology approaches are still in their infancy and require careful evaluation of their utility on particularly well studied systems.
Ideker et al
chose one early paradigm for eukaryotic gene regulation, the GAL
genes in Saccharomyces cerevisiae 
to develop and test new approaches in systems biology. We have chosen another early paradigm for eukaryotic gene regulation, the qa
gene cluster in Neurospora crassa 
to develop and test these new systems biology approaches. Both the GAL
genes and qa
cluster have been biochemically and genetically studied for more than 40 years. A wealth of molecular biology experiments are then available to specify detailed biochemical and regulatory network models or genetic networks
, for short, for galactose and quinic acid (QA) metabolism 
. The GAL
genes provided a testing ground for new genomic scale methodologies for measuring relative mRNA and protein abundances and an iterative process of genetic network identification 
. The qa
gene cluster provided a testing ground for ensemble methods of network identification containing many parameters and limited data and an iterative model-guided discovery process in genomics experiments called Computing Life 
. Here we examine some of the strengths and limitations of these approaches on the qa
In prior work we have identified a working genetic network model to describe how the qa
gene cluster functions in the cell to metabolize quinic acid (QA) as a sole carbon source 
. We followed this work with exploration of how widespread the response to shift to QA is on the transcriptome with microaray experiments and used additional microarray data to refine the ensemble of genetic network models describing the qa
cluster behavior 
. As in a previous study of the GAL
, we found the effects of the qa
cluster are widespread, involving a QA response by more than 100 known genes in varied functional categories including carbohydrate metabolism, protein degradation and modification, ribosome biogenesis, and amino acid metabolism 
. The challenge in understanding both galactose (GAL) and QA metabolism is that other processes could elicit a similar response to shift to these carbon sources. How do we distinguish the effects of these other secondary processes from the main effects of shift to galactose or quinic acid? For example, neither GAL nor QA are preferred carbon sources and could elicit a starvation response on the part of the cell 
, thereby unleashing a whole cascade of stress responses unrelated to the direct effect of the non-preferred carbon source. There is also a need to reconcile the 997 and 895 genes responding to GAL and QA with the average number of targets per transcription factor of ~38 found in S. cerevisiae 
In this work our goals are several fold. First we wish to identify all QA-responsive genes. Second, we wish to distinguish a QA-response from other ancillary responses, such as to starvation. Three, we wish to begin the more or less complete description of the qa
genetic network as is beginning to be achieved in Escherichia coli
for carbon metabolism 
. Finally, we wish to evaluate the performance of some of the new systems biology tools for rationalized discovery about genetic networks in the cell used here to discover the role of the qa
gene cluster in metabolism 
. In particular, we wish to extend the ensemble method to operate on a genomic scale transcriptional network. To address this problem we have developed a parallelized version of the ensemble method for network identification as described under Materials and Methods
Genetic network model for the qa gene cluster
gene cluster consists of 7 genes on linkage group VII of N. crassa 
. Four of the genes are structural genes (qa-2
); one has an unknown function (qa-x
); and two are regulatory genes (qa-1F
). The genes qa-1F
encode a transcriptional activator and repressor, respectively, 
that turn on and turn off the qa
gene cluster. The gene qa-1F
gene product QA-1F activates all genes in the qa
cluster, allowing the use of QA as a sole carbon and energy source. The cluster is also known to be linked to a parallel biosynthetic pathway in aromatic amino acid metabolism as well 
This information enabled formulation of an initial detailed genetic network shown in (minus the QA-responsive genes (qag
) on the left hand side) that explains how QA metabolism functions 
. Following the notation of 
, circles denote reactions, and boxes, molecular species (i.e
., genes, mRNAs, proteins, and metabolites) appearing in this chemical reaction network. Arrows entering a circle denote reactants, and arrows leaving a circle denote products. Double arrows indicate that a molecular species appears on both the left and right side of the reaction and is a catalyst. As an example, enzymes enter reactions with double arrows. The overall structure of this genetic network for carbon metabolism has the Central Dogma at the top and the biochemical pathway for QA metabolism, at the bottom. On the left side of the network is the transport process for QA involving the permease, QA-Y, and on the right side is the genetic regulatory mechanism involving the regulators, QA-1F as QA-1S, as well as metabolites hypothesized to affect these regulators.
Figure 1 A genetic network for the qa gene cluster derived from , .
In this network model, the QA-1F protein activates all of the qa
cluster genes, including a gene (qa-1S
) leading to its own inactivation, by means of the QA-1S repressor protein 
. The active qa-1F1
gene is transcribed into its cognate mRNA qa-1Fr
, which in turn is translated into its cognate protein QA-1F. The QA-1F protein, in turn, activates all of the qa
cluster genes in the A-reactions in in a positive feedback loop. One of these is the qa-1S
gene, encoding a repressor. The encoded QA-1S protein thereby counteracts QA-1F in the I1 reaction and shuts down QA-1F as QA-1S accumulates by sequestering QA-1F. The action of the QA-1S protein is facilitated in reaction I3 by a preferred carbon source, such as sucrose, binding to QA-1F, thereby providing a mechanism for catabolite repression. In addition, QA-1F may also activate a number of QA-responsive genes (qag)
that serve as yet to be identified outputs of this circuit. The number of these QA-responsive genes
in the genome, and hence the extent of QA control over metabolism is largely unknown 
then specifies fully the null hypothesis for this paper with 204 rate constants and 147 initial molecular species concentrations as parameters. A number of alternative hypotheses to will be considered. For example, one alternative hypothesis is catabolite repression by inducer exclusion 
Predictions about QA-responsive genes from the genetic network
Six predictions about the behavior of QA-responsive genes can be made from the genetic network in and prior work.
- When WT cells are shifted from sucrose to quinic acid as described in Materials and Methods, the mRNA levels are predicted to respond . This experiment is referred to as experiment 1-QA response to identify a QA-response in WT. In contrast as a control, we can also predict that if WT cells were shifted from sucrose to sucrose, under the genetic network model there should be no response in mRNA levels by QA-responsive genes . We will refer to this as the control experiment.
- Since QA is a poor non-preferred carbon source, it is possible that cells will not differentiate a shift to QA from a shift to no carbon source, i.e., a starvation response. So, in experiment 2-starvation we shift the cells from sucrose to Fries Medium without a carbon source  to differentiate a response to QA in experiment 1 from a response to starvation in experiment 2. The prediction is that genes under qa cluster control should not have a starvation response because there is no such mechanism in .
- The first two perturbation experiments represent environmental perturbations. The last experiment 3-QA response by qa-1F is a genetic perturbation. If qa-1F mutant cells were shifted from sucrose to QA, the genetic network would predict that there should be no QA-response  because some qa-1F mutants, such as the one selected in Materials and Methods, do not make functional transcriptional activator protein QA-1F. It is still possible that qa-1F mutant cells might perceive that the shift to a poor carbon source, such as QA, as a starvation signal, eliciting other “starvation” genes to respond in a qa-1F mutant. Alternatively, if there were alternative pathways for the metabolism of QA  or if there were other transcription factors that can substitute for QA-1F in function, we may see a response by genes in the absence of a functional qa gene cluster. The QA-1F protein is a member of the largest family of transcription factors in N. crassa, and such transcription factors are well known to act redundantly .
- Another prediction that can be made about the dynamics of the mRNA levels of QA-responsive genes from the genetic network in . If we were to compare such a gene's mRNA level at time 0 in the shift from sucrose to QA to the average mRNA level at later times, we should see a dramatic rise in mRNA abundance with time. We search for this kind of change. This does not preclude the qa gene cluster acting to decrease expression of other genes, but if such a decrease were observed, it would have to be an indirect effect in the genetic network. For example, the QA-1F protein would need to target a repressor, for example, which in turn would down-regulate other genes under its control. This last prediction places an additional constraint on what are considered genes directly regulated by the qa cluster.
These four microarray experiments and the expected dynamics of mRNA levels of genes under QA-1F control provide a means to sift through the 895 genes that respond to the initial shift experiment under experiment 1 
. A QA-responsive gene
will be defined as one whose mRNA level: (1) in WT responds to shift from sucrose to QA; (2) in WT does not respond to shift from sucrose to Fries (i.e
., no starvation response); (3) in a qa-1F
mutant does not respond to shift from sucrose to QA; (4) in WT does not respond to shift from sucrose to sucrose (i.e
., in the control experiment); (5) increases significantly from time 0 to later time points in WT when shifted from sucrose to QA.
- There are two more predictions from prior work about QA-responsive genes. Case and colleagues  presented both genetic and biochemical evidence for the interrelationships of the QA and shikimate (aro) pathways. A mutation in aro-1 (which converts dehydroshikimic acid (DHS)→shikimic acid (SA)) was suppressed by qa-3+ in an aro-1/qa-4 double mutant background. The double mutant leads to the accumulation of DHS, which by mass action allows the QA-3 protein to convert DHS to SA. The aro-1 mutation was thus sidestepped. Also a block in the aro pathway via an aro-9 mutation was demonstrated to lead to the internal induction of the qa cluster. The qa and aro pathways are coupled by redundancy of function and mass action. We thus predict that aromatic amino acid metabolism will have encoding genes that are QA-responsive.
- It has been long known that sorbose-resistant mutants constitutively activated QA-metabolism on sucrose (M. E. CASE, unpublished results). We then expect that some of the QA-responsive genes will be sorbose-resistance mutants (sor-1-sor-4 in  as well. We thus expect that starch metabolism should be represented among QA-responsive genes. (While fungi generally use glycogen instead of starch, we continue to refer to the “sucrose and starch metabolism” because that is the label of this metabolic module used in the Kyoto Encyclopedia of Genes and Genomes.