Figure 1.
Workflow for integrating multiple data types with TEAM.
TEAM integrates three types of experimental data: starting media composition, expression data, and biomass data. Pre-calculations include normalization of the gene expression data, interpolation of all data sets, and calculation of gene penalties based on the gene expression data. For a given time interval, TEAM calculates the metabolic flux distribution most consistent with gene expression and biomass data. It applies this result to update media conditions for the subsequent time interval.
Figure 2.
A comparison of results across different methods for a representative penalty threshold.
The media contained 36 mM L-Lactate, 13 mM D-lactate, 9 mM ammonium, and other minimally required nutrients. The oxygen concentration was set to 10 mM at each time point, mimicking the controlled 100% dissolved oxygen (DO) concentration from the experiment. The resulting usage dynamics of several metabolites of interest (including combined DL-lactate, ammonium, pyruvate, acetate, formate and glycolate) as predicted by dFBA are compared to experimental data. (A) HPLC Data, (B) dFBA, (C) TEAM with a global penalty threshold (Type 1), (D) TEAM with a gene-specific penalty threshold (Type 2), (E) TEAM with a gene-specific penalty threshold normalized by standard deviation (Type 3). Black dots represent hours when microarray measurements were taken.
Figure 3.
Overall distribution of S. oneidensis gene expression measurements with two individual genes highlighted.
Distributions calculated using a pooled set of data from M3D [17] and time-course experimental data [18]. For both example genes, the distribution of gene expression measurements and the corresponding cumulative distribution function (CDF) are shown. For each CDF, individual gene penalty thresholds are found for common percentiles θ = 25%, 50% and 75%. D-lactate dehydrogenase expression measurements (A) have a higher mean expression and a more pronounced peak than acetate kinase (C), which is more uniformly distributed. The corresponding CDFs capture this variation in distribution. D-lactate dehydrogenase expression penalties (B) are higher and less distributed than those of acetate kinase (D). Mean gene expression (E) and standard deviation (F) over all genes for a single time point are also shown. All microarrays contain 4230 gene products, and each individual distribution contains 310 data points; 19 come from the experimental time-course and 281 from M3D.
Figure 4.
Sensitivity analysis for Type 2 gene-specific threshold.
(A) Total carbon concentration in media for each penalty threshold θ, summed over all time points. Penalty thresholds between 40% and 75% exhibit enrichment for intermediate carbon sources formate, glycolate, pyruvate. (B) Extinction time of lactate and ammonium in the media. Lactate runs out significantly earlier for intermediate penalty thresholds. Heatmap indicates the total media concentration of secreted carbon sources (acetate, pyruvate, glycolate, formate).
Figure 5.
A measure of predictive accuracy between pyruvate and acetate excretion behavior.
For all percentage thresholds θ between 1% and 99%, the quality of predictions for (A) pyruvate and (B) acetate secretion and utilization behavior was calculated using the residual sum of squares between the experimental HPLC measurements and the model predictions for all three gene penalty calculation types. Only the Type 2 and 3 penalty thresholds predict the secretion of pyruvate, occurring between θ = 55% and θ = 82%.
Figure 6.
A comparison of internal flux profiles for two different penalty thresholds.
Superposition of metabolic flux onto central carbon metabolism of S. oneidensis. Top panel corresponds to a Type 2 penalty threshold of 65%, and bottom panel to a Type 2 penalty threshold of 85%. Large nodes and edges on the networks represent reactions and small nodes correspond to metabolites. The colors of the large nodes correspond to the penalty associated with that reaction. Colored squares on the network plots identify the transport reactions for each exchange metabolite. A network key and reaction and metabolite details can be found in Figure S3, Table S1 and Table S2 respectively. Detailed time-course flux predictions are provided in Supplementary Dataset S1.