Systems biology approaches to understanding the Arabidopsis hypersensitive response

Vikas Agrawal¹, Chu Zhang¹, Prasad S. Dhurjati² and Allan D. Shapiro¹

¹Department of Plant and Soil Sciences, ²Department of Chemical Engineering, University of Delaware, Newark, DE

Motivation Cross-talk and feedback regulation in signal transduction networks are often quite complex. The signaling network governing Arabidopsis hypersensitive response to avirulent Pseudomonas syringae was analyzed using a systems biology approach. A systems approach uses quantitative data collection and mathematical modeling to understand higher order relationships governing network behavior. The goal was to create a framework for efficient design of experiments by simulating the time course progression of the most important components of the response in silico. These predictions were then compared to experimental data for validation (Agrawal, et al., Biotechnology Progress 2003).

Strategy The most important components of the response were taken as model variables. These included death of individual cells (PCD), salicylic acid (SA) and reactive oxygen (ROS) accumulation, and level of apoplastic superoxide dismutase (SOD) (Zhang, et al., Molecular Plant-Microbe Interactions, in press). Initial estimates of kinetic parameters and time delays were made from experimental data and subsequently refined by global fitting of simulated to experimental data. This process resulted in a system of ten delay differential equations governed by expert-system type rules. This system was solved numerically using engineering software (MATLAB). A one-to-one correspondence was maintained between model variables and specific signaling components. The mathematical forms for relationships between model variables also corresponded one-to-one with experimentally observed relationships between signaling components. As such, the assumptions questioned by new data can be readily identified.

Results 1) In silico simulations of the time course of changes in levels of salicylic acid, PCD and hydrogen peroxide match experimental data. 2) We have used the model to prove that direct negative autoregulation of salicylic acid biosynthesis does not exist in this system. Including terms for this extra negative feedback loop made it impossible for simulated data to match experimental results. 3) Simulations also determined that NPR1-dependent negative feedback on PCD cannot affect the fraction of total PCD seen late in HR progression as a direct consequence of high levels of superoxide. 4) The dynamic profiles of apoplastic superoxide dismutase (SOD) activity and two putative gene induction events have been predicted. 5) “Sensitivity” analysis has been used to predict which model components have the most significant influence on overall system dynamics. These predictions will aid in design of further experiments to test our knowledge of control of the HR.

Our current strategy to improve the explanatory power of the model via explicit modeling of cell-to-cell signaling events using population balance modeling will be described.

Comparison of large scale gene expression measurement technologies

Vikas Agrawal, Hassan Ghazal and Blake C. Meyers

Department of Plant and Soil Sciences, University of Delaware. Newark, DE

We are in the process of comparing four large scale gene expression technologies, based on expression data from more than 10 tissues/conditions in the model plant Arabidopsis thaliana. Data comprising 40 Agilent arrays, 40 Affymetrix arrays and 63 MPSS sequencing runs was obtained using the same mRNA samples for each library. Qiagen/Operon data is forthcoming.

Agilent technology is known to be more reproducible in technical replicates than MPSS than Affymetrix in turn. We have shown that MPSS technology has a dynamic range about 5 fold larger than that of Agilent and 10 times that of Affymetrix. Via global analysis of the data across the libraries we found that MPSS often underestimates the expression levels (relative to the other two) while Agilent overestimates them or vice versa.

We identified 45 genes that were within a 2-fold range in all technologies and all libraries. There were more than 2000 genes that were measured within a factor of 2 across all technologies in each specific library. The overall correlations across technologies for specific libraries were in the range of 0.65-0.75. The measurements of differential expression had a poorer correlation across technologies. There were 6 genes that were expressed above a concentration of 500 transcripts per million (TPM), 592 genes expressed below 10 TPM and 85 genes expressed below 4 TPM in all libraries as measured by all technologies.

Thus we have examined issues of reproducibility, correlation, dynamic range and estimation of the absolute level of transcripts. We are conducting QRT-PCR measurements that will place these comparisons on a stronger foundation. There are issues of the cost to benefit ratios that we have not presented in this study.