Fig 1.
Mammalian central energy metabolism.
Glucose metabolism, including the pentose phosphate pathway, glycolysis, and the TCA cycle, was modeled as a network using a stoichiometric matrix. Each colored square represents a biochemical reaction, and each line represents a metabolite. Reversible reactions shown in blue, and irreversible reactions shown in red. Arrows indicate the direction of irreversible reactions. Anaplerotic reactions that replenish TCA reactants are represented by hashed squares. The curved arrow indicates the direction of normal flux in the TCA cycle. Selected metabolites (i.e.glucose, lactate, and glutamine) shown explicitly.
Fig 2.
Median metabolite LC-MS intensities over time and experimental group.
Colors indicate sample time, from left to right in the group 0,15, and 30 minutes. Superscript numbers indicate the probability of no difference between population means is less than or equal to p = 0.15, as calculated by ANOVA. P-values for all metabolites are found in the supplementary material (S2). All intensities scaled by 0.001 for ease of display.
Fig 3.
Hypothesis-based predicted flux profiles, v*.
Flux profiles were predicted in silico using optimization to maximize production of (A) type II collagen and (B) lipid synthesis. Reactions are represented by labeled boxes connected to substrates by incoming dark gray arrows and to products by outgoing dark gray arrows. Reaction fluxes are represented by arrows colored to match either positive (blue) or negative (yellow) flux. Larger arrows represent larger flux with linear scaling. Negative flux indicates a reversible reaction running in the direction opposite to its description in the stoichiometric matrix. Electron transport chain and synthesis reactions not shown for figure simplicity.
Fig 4.
Intensities of metabolites that changed significantly over time as indicated by ANOVA were used to cluster the samples. The distance between the samples is the correlation of their intensities. This distance was then used to cluster rows (intensity of a single metabolite over all samples) and columns (intensity of all metabolites belonging to a single sample). While the 15 and 0 minute samples have similarities, all but one of the 30 minute samples cluster together, indicating more similarity within 30 minute samples than to any other group. The intensities have been standarized by column.
Fig 5.
Experimentally-derived flux profiles for compressed chondrocytes display both similarities and differences to hypothetical predictions.
Fluxes calculated for compressed cells for time intervals (A) 0 to 15 min. and (B) 15 to 30 min using the collagen synthesis reaction as described in the text. Similarities (in comparison to predictions in Fig 3A) include large relative fluxes in G1-3 and TCA1-3. Differences include small negative fluxes as discussed in the manuscript. Reactions are represented by labeled boxes connected to substrates by incoming dark gray arrows and to products by outgoing dark gray arrows. Reaction fluxes are represented by arrows colored to match either positive (blue) or negative (yellow) flux. Larger arrows represent larger flux with linear scaling. Negative flux indicates a reversible reaction running in the direction opposite to the direction specified in the stoichiometric matrix. Electron transport chain and synthesis reactions not shown for figure simplicity.
Fig 6.
Overall flux trends highlight differences between compressed cells for time interval 0 to 30 min.
Reactions are represented by labeled boxes connected to substrates by incoming dark gray arrows and to products by outgoing dark gray arrows. Fluxes are represented by arrows colored to match either positive (blue) or negative (yellow) flux. Larger arrows represent larger flux with linear scaling. Negative flux indicates a reversible reaction running in the direction opposite to the direction specified in the stoichiometric matrix. Electron transport chain and synthesis reactions not shown for figure simplicity.
Fig 7.
The majority of perturbed fluxes correlate with fluxes calculated from experimental data in a sensitivity analysis.
Histogram of correlation values for 10000 randomized data sets compared with the empirical flux calculated for the unperturbed dataset. Each randomized data set yielded three randomized flux vectors which were then compared to the corresponding empirical flux vector calculated from the raw data.
Fig 8.
Pairwise correlation between calculated reaction flux and experimental metabolite accumulation.
Positive correlation indicates that high flux is paired with high accumulation, and vice versa. Zero correlation indicates the flux cannot be predicted by the accumulation. Hierarchical clustering based on the Euclidean distance between fluxes and metabolite accumulation.