Fig 1.
Visual representation of RIPTiDe workflow.
(A) Example transcript and resultant reaction weight distributions were calculated by RIPTiDe. During the pruning step, reactions for genes that recruit greater abundances of transcript are assigned smaller linear coefficients which in turn result in higher likelihood of usage in an overall flux minimization objective. Alternatively, during the subsequent the flux sampling step, the remaining reactions associated with higher transcription are assigned larger linear coefficients leading to increased possible flux ranges. (B) Schematic outlining computation performed during each step of RIPTiDe. The platform requires two input data structures: a genome-scale network reconstruction and transcript abundances associated with genes. During initialization transcript abundances for genes are transferred to their corresponding reaction, allowing coefficients to be assigned. Each reversible reaction is now also separated into pairs of irreversible reactions. Applying principles of parsimonious FBA to the constrained model, a minimum sum of fluxes optimization is performed with respect to a predefined minimum flux through the cellular objective. Reactions that no longer carry flux in this state are pruned, and flux sampling is performed on the intermediate model to determine context-specific bounds for the remaining reactions. (C) Pictorial representation of the impact of RIPTiDe on the original reconstruction, converting it to a context-specific model of metabolism with a parsimonious metabolic solution space with respect to the given transcriptomic data.
Fig 2.
Example model of bacterial metabolism supports the utility of RIPTiDe for identifying most likely context-specific strategies.
(A) General topology of simplified GENRE that consists of 16 reactions and 14 metabolites in total (S2 Table). The objective of ATP generation can be achieved through two separate means: either the catabolism of glucose via glycolysis or through the paired fermentation of proline and glycine. Glycolysis is the more parsimonious pathway for generating ATP, which is reflected in the pFBA solution fluxes to the right of the diagram (outlined in red). (B) RIPTiDe is able to correctly identify a likely route of flux when provided a transcriptome. (C) When provided with transcriptomic evidence, RIPTiDe is able to identify a less parsimonious, but more concordant solution (Lower weighted sum of fluxes). Despite very low transcription, RIPTiDe still identifies the necessity for carbon dioxide efflux during fermentation to maintain mass balance.
Fig 3.
RIPTiDe identifies established biological differences in E. coli str. K-12 substr. MG1655 across in vitro growth conditions.
Without increased user supervision, RIPTiDe correctly predicts behavior and context-specific pathways for E. coli str. K-12 substr. MG1655 (iJO1366), while simultaneously producing functional models that can be characterized phenotypically. (A) NMDS ordination of Bray-Curtis dissimilarities between exchange reaction flux samples for each version of iJO1366. Table legend indicates flux constraints placed on iJO1366 prior to flux sampling. Flux samples from RIPTiDe transcriptome contextualization without exchange constraints are not significantly different from those where media condition was set a priori (p-value > 0.05). The gray check denotes that the max growth constraint is inherently integrated into the RIPTiDe workflow. Significant differences evaluated by PERMANOVA. (B) Comparison of metabolic reactions included among RIPTiDe-contextualized transcriptomes with iJO1366. The majority of reactions (71.5%) are conserved across models within a central set of pathways. All exchange reaction bounds set ±1000 prior to contextualization. (C) Comparison of importance (essentiality) for conserved genes across pruned RIPTiDe models. 105 core essential genes were identified across all groups (S3 Table). Hierarchical clustering reveals context-specific pathway essentially, labeled across the bottom axis, based on the environment in which the bacterium is growing. All exchanges set to ±1000 prior to contextualization.
Fig 4.
Cellular sources of NADPH in RIPTiDe contextualized models reflect known biological differences in E. coli across media conditions.
Shown in each panel is reaction pruning of the small transhydrogenase circuit found in E. coli, and flux sampling results of NADPH sources from RIPTiDe using the iJO1366 GENRE with contextualized transcriptomic abundances from E. coli K-12 MG1655 grown in (A) M9+glucose minimal media, and (B) LB rich media. This mechanism for restoring NADPH balance is known to be essential for E. coli growth in M9 minimal media but dispensable in LB, which is correctly selected by the unsupervised network pruning from RIPTiDe. Significant difference in flux levels determined with Wilcoxon signed-rank test (p-value << 0.001); pathway maps generated using Escher [77].
Fig 5.
RIPTiDe reveals differential host-associated, metabolism utilizing in vivo metatranscriptomic data.
Transcriptomic reads attributable to E. coli were extracted from a metatranscriptomic dataset sequenced from the cecum of mice in which E. coli is the most highly abundant community member [8]. (A) In vivo transcript abundance of reads recruited to the E. coli K-12 MG-1655 genome (left y-axis) and the weights assigned by RIPTiDe to the associated metabolic reactions during each step (right y-axis). Metrics listed at the top of the plotting area reflect the resultant context-specific model compared to the complete iJO1366 model. (B) PCoA ordination of Bray-Curtis dissimilarities between flux distributions among reactions of each contextualized iJO1366 model either from in vivo or LB rich media (aerobic, in vitro) conditions. Significance was calculated by PERMANOVA. (C) Metabolite substrates of exchange reactions exclusive to the contextualized models shown. Inverse, normalized flux was calculated by dividing each flux sample by the associated flux through biomass in the same distribution, then multiplied by the overall median flux through biomass across all conditions analyzed. Median and interquartile ranges are displayed.