Pentose Phosphate Pathway

We reconstructed the metabolic network of lysine producing C. glutamicum based on published information [39 (link)] and additional modelling details kindly provided by the authors [see Additional file 2]. The input substrate used was [1-¹³C]-glucose (with 99% enrichment purity), and all fluxes were normalized with respect to the glucose uptake rate (i.e., fluxes are expressed in percentage of glucose uptake rate). As the published MIDs are uncorrected, all the simulated EMU variables were modified for mass interference from non-carbon backbone isotopes using the molecular formula of the amino acids fragments (i.e., parent ion cluster). The first n+1 signal elements were normalized (n indicates number of backbone carbon), and then truncated to the correct vector length (equivalent to the measured MIDs) before performing weighted least-square analysis. The inferred metabolic model consisted of a total of 71 reactions and 42 balanceable metabolites. The metabolite model yielded a total of 26 degrees-of-freedom and 18 fluxes were determined experimentally: anabolic precursor yields (11), biomass yield (1), secreted product yields (5), and glucose uptake rate (1). To reduce the number of unknown parameters, these 18 fluxes were chosen as free fluxes, and the associated flux values were used deterministically as no redundant data exist in the measurement set. Note that if one suspects gross measurement errors in the flux measurement set, then these fluxes should be set free and the flux values subjected to the least-square analysis together with the MIDs. Five (5) of the remaining 8 free fluxes are associated with the reversibility of non-oxidative pentose-phosphate pathway enzymes (3), glucose-6P isomerase (1) and intercellular CO₂ exchange (1). The other 3 free fluxes were assigned (by the software) to the irreversible fluxes of glucose-6P dehydrogenase, pyruvate carboxylase, and glycine synthesis via the serine route.
The MID of 9 amino acids and trehalose were reported. Three "S" type reactions were included in the metabolite network to directly map label distribution of alanine, aspartate and glutamate to pyruvate, oxaloacetate and α-ketoglutarate, respectively. This was not necessary for all other amino acids and trehalose because these metabolites were already described in the isotopomer balances.

After defining the high-confidence core and ranking all non-core reactions, our algorithm attempts to sequentially remove each non-core reaction, starting from those ranked at the bottom (lowest evidence). The selected reaction will be removed only if (i) the core set of reaction remains consistent; and (ii) removal does not prevent model from producing any key metabolites. Reactions in high-confidence core set can only be removed when (i) reactions in the negative reaction set (reactions with E_x(r) =0) are needed to enable flux through the high confidence core reactions; (ii) by removing the high confidence core reactions, more non-core reactions (including those in the negative reaction set) will be removed. Consistency of the core reaction set is confirmed by calculating the maximum and minimum flux for each reaction, and ensuring that at least one is non-zero. As the naïve implementation of flux variability analysis (FVA) is extremely slow, we adapted the checkModelConsistency module described by Jerby et al. in [14 (link)] for optimal performance in Matlab—in particular, we included the option to use the efficient fastFVA algorithm [27 (link)].
The list of key metabolites that must be produced from glucose is compiled based on the universal metabolic model validation test in [18 (link)]. This includes metabolites in glycolysis, TCA cycle, pentose phosphate pathway, as well as non-essential amino acids, nucleotides, palmital-CoA, cholesterol, and several membrane lipids. A full list of these key metabolites is in Additional file 3: Table S1. Instead of testing the production of all non-essential fatty acids, as in [18 (link)], we only tested the production of palmital-CoA, which is derived from palmitate, the first fatty acid produced in fatty acid synthesis, and the precursor of longer chain fatty acids. Similarly, we only tested those membrane lipids that can be derived from glucose and non-essential amino acids. With the addition of essential nutrients like choline, these membrane lipids can be transformed to other membrane lipids such as phosphatidylcholine and sphingomyelin that cannot be directly synthesized from glucose. We only check the production of pyrimidine nucleotides from glucose, as de novo pyrimidine synthesis can occur in a variety of tissues [22 ]. As de novo purine synthesis occurs primarily in the liver and other tissues use the salvage pathway [22 ], we test the ability of all tissues to synthesize purine nucleotides from purines bases and 5-phosphoribosyl 1-pyrophophate (PRPP).

For cDNA synthesis, 1000 ng of intact RNA was reversed to cDNA with a reverse transcription kit (YEASEN, Shanghai, China) with the concentration detected. The synthesized cDNA was diluted to 200 ng/μL as a template for RT-qPCR. To detect carbohydrate metabolism, the expression of genes related to glycogen synthesis (ugp2b, gys2), glycogen degradation (pygl), gluconeogenesis (pck1, pcxb), glycolysis (gck), TCA cycle pathway (idh), and pentose phosphate pathway (g6pd) were evaluated. The expression of genes related to lipid synthesis (fasn, acaca, aclyb) and decomposition (acadl, acaa1, lpl) were determined to illustrate the influence on lipid metabolism, and the expression of genes related to the urea cycle (gs, cps3, otc, ass, asl, and arg1) was also detected. The qPCR was performed in Jena qTOWER3G system using the real-time quantitative PCR detection kit EvaGreen 2 × qPCR Master mix (YEASEN, Shanghai, China): pre-denaturation at 95 °C for 5 min; denaturation at 95 °C for 10 s; annealing and extension at 60 °C for 30 s; and PCR reaction step running 40 cycles. After RT-qPCR, melting curves were analyzed to ensure the specificity of the reaction. Using 18s as the internal reference gene, the relative quantitative data analysis was performed by the 2^−ΔΔCt method. RT-qPCR data were analyzed using GraphPad Prism 6. The primers used in the present study are shown in Table 1.

We next assessed the 12 metabolic gene sets in MSigDB, including reactome pyrimidine catabolism, reactome pentose phosphate pathway, reactome purine catabolism, reactome metabolism of amino acids and derivatives, reactome citric acid/TCA cycle and respiratory electron transport, reactome glycogen metabolism, reactome metabolism of lipids, reactome fatty acid metabolism, reactome glutamate and glutamine metabolism, reactome pyruvate metabolism, reactome glucose metabolism, and reactome metabolism of nucleotides. After merging of the genes and removing duplicates, a total of 1,489 metabolism-related genes were selected. We conducted consensus clustering using NMF to identify different metabolic preference patterns, based on the expression of 1,489 regulators. The expression of 1,489 metabolic regulators [Matrix V, Gene (F) × Patient (N): 1,489 × 350] was factorized into 2 non-negative matrices W [Gene (F) × Patient program (K): 1,489 × 3] and H [Expression program (K) × Patient (N): 3 × 350] (i.e., V ≈ WH). Specifically, we decompose the matrix V into a basis matrix W and a coefficient matrix H. On the one hand, the basis matrix W is characterized by patient programs: each one of patient programs is a vector in W. On the other hand, each column vector of the coefficient matrix H can be regarded as the coordinates obtained by projecting the corresponding column vector of the matrix V onto W: each expression program is a row in H. Repeated factorization of matrix V was performed, and its outputs were aggregated to obtain consensus clustering of GC samples. The optimal number of clusters was then selected, according to the cophenetic, dispersion, and silhouette coefficients. The R package ‘NMF’ (version 0.23.0) with the brunet algorithm and 100 n runs were used to perform the consensus clustering.

Metabolism-related genes were identified from the Molecular Signatures Database (MSigDB) (https://www.gsea-msigdb.org/gsea/msigdb/index.jsp) (Subramanian et al., 2005 (link)). Overall, 12 metabolic gene sets were obtained, which included the reactome pyrimidine catabolism, reactome pentose phosphate pathway, reactome purine catabolism, reactome metabolism of amino acids and derivatives, reactome citric acid/TCA cycle and respiratory electron transport, reactome glycogen metabolism, reactome metabolism of lipids, reactome fatty acid metabolism, reactome glutamate and glutamine metabolism, reactome pyruvate metabolism, reactome glucose metabolism, and reactome metabolism of nucleotides.