Metabolic systems analysis of LPS induced endothelial dysfunction applied to sepsis patient stratification

Endothelial dysfunction contributes to sepsis outcome. Metabolic phenotypes associated with endothelial dysfunction are not well characterised in part due to difficulties in assessing endothelial metabolism in situ. Here, we describe the construction of iEC2812, a genome scale metabolic reconstruction of endothelial cells and its application to describe metabolic changes that occur following endothelial dysfunction. Metabolic gene expression analysis of three endothelial subtypes using iEC2812 suggested their similar metabolism in culture. To mimic endothelial dysfunction, an in vitro sepsis endothelial cell culture model was established and the metabotypes associated with increased endothelial permeability and glycocalyx loss after inflammatory stimuli were quantitatively defined through metabolomics. These data and transcriptomic data were then used to parametrize iEC2812 and investigate the metabotypes of endothelial dysfunction. Glycan production and increased fatty acid metabolism accompany increased glycocalyx shedding and endothelial permeability after inflammatory stimulation. iEC2812 was then used to analyse sepsis patient plasma metabolome profiles and predict changes to endothelial derived biomarkers. These analyses revealed increased changes in glycan metabolism in sepsis non-survivors corresponding to metabolism of endothelial dysfunction in culture. The results show concordance between endothelial health and sepsis survival in particular between endothelial cell metabolism and the plasma metabolome in patients with sepsis.


Supplement IIa -Assessment of iHUVEC2812
We built a metabolic reconstruction of EC metabolism by curating RECON1 then using Fastcore 1 to apply a transcriptomic data-set. Removal of inconsistent reactions by Fastcore, those that are dead ends or otherwise unconnected in the model, removed 34 % (1,270/3,743) of reactions in RECON1 but less than 14% (301 reactions) in the HUVEC model based on iEC2812 before application of metabolic constraints and provides a measure of the influence of literature curation.
Roughly 80 % of nucleotide and amino acid metabolism reactions are retained. There are extra reactions added in fatty acid and signalling metabolism. Around 70 % of reactions of sugar and central carbon reactions are retained. Whereas only around 60 % the reactions classified as peripheral metabolism (other), vitamin metabolism and exchange reactions were retained (Figure 1).
In comparison to RECON1 2 , iEC2812 had a higher proportion of reactions linked to genes, around 70 % compared to around 60 % in RECON1 2 . Also of the genes present in the model only around 50 % in RECON1 2 were used whereas in iEC2812 around 70 % of genes were used (Table1).

Supplement IIb -Comparing HUVEC, HPAEC and HMVEC models
To determine if and how metabolism differed between three endothelial subtypes, commonly used in laboratory research, we built three GEMs based upon iEC2812.
These are slightly more extensive than a previously published model of HUVEC metabolism containing ~2,550 reactions (Table1) compared to 2088 active reactions 3 . This is partly due to using transcriptomic rather than proteomic data to constrain reactions and partly due to the improved base model.
Analysis of reactions and genes essential for survival showed roughly 90 % of essential reactions were common to at least two reconstructions, this is partly a reflection of the composition of the biomass function. Glycerophosphlipid metabolism, nucleotide metabolism and sphingolipid metabolism were strongly represented in all sets of essential reactions. Essential reactions found only in HUVECs were in nucleotide and nucleotide sugar reactions. Essential reactions only in HPAECs four are reactions of nucleotide metabolism or transport, one is a ceramide reaction, and the other three are amino acid metabolism or transporters. Essential reactions only in HMVECs include nucleotide metabolism and transport reactions, and an amino acid transport reaction. On the other hand common essential reactions include cardiolipin synthase 1, phosphatidylglycerophosphate synthase 1 and sphingomyelin synthase 1 all of which have previously been associated with cardiovascular disorders 4,5 (Figure1).
The ability of iEC2812 and sub-type models to describe known features of endothelial metabolism were queried by their ability to secrete known biomarkers of human endothelium. Nitric oxide, meth-oxy-tryptophan, kynurenine, spingosine-1phosphate, prostaglandin D2 and prostaglandin E2 and gamma amino butyric acid have all been previously shown to be secreted by ECs and function in signalling both by and to ECs [6][7][8][9][10][11] .

IFN LPS(ng/mL) Mean SD Change v blank control
Change

IFN LPS (ng/mL) Mean SD Change v blank control
Change The mass distribution of extracellular lactate was determined using UPLC-MS followed by Iso-core processing after growth with 50% 1,2 13 C labelled glucose (A). It was confirmed that the overall glucose consumption/lactate secretion rate was unaffected by the extra glucose. It was also shown that there was no significant difference in mass distribution values between 0,1,2 or 3 labelled lactate with and without LPS after 24 hours. On t-tests of the proportion of each mass of lactate mass+0 p = 0.568, mass+1 p = 0.878, mass+2 p = 0.487, mass+3 p = 0.680 and the overall mean enrichment of distribution value was 0.143 in control cells and 0.146 in LPS treated cells (p = 0.530).
This distribution was compared, using linear least squares fitting in Matlab, to the theoretical mass distribution of lactate in cells if all lactate was produced either via glycolysis or via the pentose phosphate pathway (B). This suggests a mean of 68% glycolysis in control cells and 70% in LPS treated cells, this small difference is in line with the non-significant difference seen in the mass distributions, and is reflected in the constraints placed on the models shown in Supplement Ib.