br Materials and methods br Results
Materials and methods
Discussion We present an updated version of Recon 2.2 that was curated and extended to correctly represent the flavoprotein-catalysed reactions. Furthermore, we introduced a new method to study the role of enzyme-bound cofactors, such as FAD. Curating the representation of FAD in Recon 2.2 allowed to correctly simulate aberrant metabolic behaviour upon single enzyme deficiencies. Since the predecessor of Recon 2, the metabolic reconstruction Recon 1 , was published, many groups have extended and improved model versions. They used it as a basis for tissue specific models [5,23,, , , ], studied the effects of diet , and predicted biomarkers for enzymopathies [17,23,24,37]. Work by Smallbone  and Swainston et al.  focused on a full mass and charge balance and on simulations of energy metabolism. However, despite their crucial role in metabolism, none of the curative efforts in human reconstructions, including the most recent Recon 3D , focused on cofactors, not even organic cofactors that are (in part) synthesized in the cell. Metabolism related to other apoenzymes requiring other bound cofactors for their activity (metals, iron‑sulfur clusters, or heme) would potentially profit from the same solution to further enhance biomarker research in genome-scale models. Flavoprotein-linked diseases can lead to very strong metabolic responses in patients, such as episodes of severe metabolic derangement, hypoglycaemia, metabolic acidosis, sarcosinemia and cardiovascular failure in MADD patients. Acylcarnitines, as well as sarcosine are known to be changed in the plasma and urine of MADD patients [, , ]. The original Recon 2.2 model could not predict any of the known biomarkers and no systemic effects of MADD were seen in the simulations, because the model incorrectly comprised alternative routes to reoxidize FADH2. In our new model, in which the electrons are transferred to the final electron acceptor of each flavoprotein-catalysed reaction rather than to a soluble FAD pool, and in which flavoprotein-dependent reactions are dependent on flavin synthesis, both systemic effects and metabolic changes linked to biomarkers were predicted correctly. This is seen clearly by a full block of mFAO capacity while peroxisomal FAO remained functional in MADD. In contrast, when simulating deficiency of the peroxisomal enzyme ACOX1, only the peroxisomal FAO was impaired, leading to reduced metabolism of long-chain fatty ML-210 substrates (Fig. 5). This extension is relevant for a correct description of mitochondrial fatty-acid oxidation defects, which can be partly rescued by peroxisomes . In total, we tested metabolic changes for 45 diseases, out of which 31 are associated with biochemical biomarkers. A caveat of the existing methods for biomarker predictions is that they only include the metabolites that are known as biomarkers. The models, however predict many more metabolites with altered production or consumption rates. These are potential novel biomarkers. Since they have most often not been explored experimentally, however, we do not know if the predictions are correct. If these would be tested, we would get a more complete insight into the accuracy of our predictions. Therefore, we propose usage of true positive rates for more correct description of model performance. Using Recon 2, Recon 2.2, and Recon 2.2_flavo, we predicted biomarkers for diseases included in the compendium of inborn errors of metabolism published by Sahoo et al.  with True Positive Rates of 26%, 31% and 33% respectively, while accuracies, as calculated in Thiele et al. , remained similar to previously published 77% (77%, 75% and 76% respectively). A lower (17% and 24%) TPR was reported with Recon2.2 and Recon2.2_flavo respectively for biomarkers of the flavoprotein-related diseases subset, with accuracies of 78% and 59% respectively. However more detailed studies of metabolism, using ATP production yield estimation performed for 16 flavoprotein-related diseases linked to the core metabolism, showed promising results for both our models. This method allowed us to test if alternative metabolic pathways exists that allow ATP production from the single carbon sources in various IEMs. The metabolic changes identified with this method were in line with clinical data, including impaired FAO and sarcosine degradation in all MADD cases, no proline degradation in PRODH deficiency and blocked very-long chain FAO in ACOX1 deficiency . Interestingly, ATP-generating breakdown of amino acids has been predicted to be affected in several diseases analysed. Valine breakdown has been predicted by our model to be significantly impaired in isobutyryl-CoA dehydrogenase deficiency (ACAD8) which is in line with the literature knowledge about this disease . Furthermore, our models predict a decreased ATP yield from breakdown of several amino acids, and a general impairment of energy metabolism in SDHA deficiency. This extremely rare disease is indeed known to affect energy metabolism. However, due to its low prevalence, no specific biomarkers are known . Our models predict valine, leucine, threonine, and methionine degradation pathways to be most severely affected in this disease. FAD-containing enzymes are crucial in both fatty acid oxidation and amino acid metabolism as was highlighted in flavoproteome mapping (Figs. 2B and S1B). Consistently, their impact on biomarkers became more pronounced after our curation (Fig. 5). For all 16 tested flavoprotein–related diseases we predicted new metabolic changes that may lead to new biomarker patterns. Our data suggests that these diseases might have multiple identifiable biomarkers. Using a multimarker approach or specific biomarkers ratios, which is common in cardiovascular risk assessment , instead of only single compounds, we could better differentiate between different diseases and potentially also between patients with different severity of the defects which has been proven recently for Zellweger syndrome patients differentiation . The latter was not pursued here, since we only studied complete enzyme deficiencies.