University of Groningen In silico patient Wegrzyn, Agnieszka

In silico patient

Wegrzyn, Agnieszka

10.33612/diss.126805978

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.

Document Version

Publisher's PDF, also known as Version of record

Publication date: 2020

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):

Wegrzyn, A. (2020). In silico patient: systems medicine approach to inborn errors of metabolism. University of Groningen. https://doi.org/10.33612/diss.126805978

Copyright

Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policy

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.

1

S

M

The term metabolism includes a multitude of intricate biochemical processes that are vital to sustaining life. As constant advancements in healthcare and life quality are made, people are living longer than ever before. However, they are often burdened with one or more chronical diseases. This provides a serious challenge for doctors and patients to decide on the right treat-ment, taking into account various comorbidities. Majority of human diseases arise from the imbalance in biochemical processes affecting the metabolism of proteins, fats, or carbohy-drates, or resulting in impaired organelle function presenting as complex medical conditions including several human organ systems. The imbalance in metabolic processes can be caused either by environmental factors (including lifestyle choices, or exposure to toxins) or by genetic mutations (both hereditary, and non-hereditary). Inborn errors of metabolism (IEM) are a sub-group of genetic disorders which results from a single enzyme deficiency. While individually rare, their birth prevalence ranges from 0.4 in 1000 to 1.3 in 1000 live births [1–5]. Most treat-ment options are focused on handling the symptoms of the disorders to improve the pa-tient’s quality of life, since successful gene therapy is not yet available, despite the tremendous progress made in the recent years [6–8]. However, as patients suffering from an identical mu-tation may display a spectrum of disease phenotypes it can be challenging to decide on the right treatment.

Modern healthcare gradually moves away from a reactive, ‘one-size-fits-all’ approach to pro-active, personalised care, tailored to people’s individual health needs [9]. This paradigm shift requires a thorough knowledge of the patient’s genetic background, medical history, lifestyle, and environment. Only then a clinician will be able to determine health risks, early diagnostic markers, and most effective interventions. Fortunately, over the past decades, exponential ad-vances in the technologies and informatics have allowed researchers to generate and process large biological data sets (‘omics’ data). With various ‘omics’ technologies becoming cheaper and readily available, we see a rise in the interdisciplinary, integrative approaches that support the development of systems medicine. This systems approach requires physicians and experts in biochemistry, statistics, informatics, mathematics, and computational modelling to collabo-rate to transcend the boundaries of each of their respective fields for the benefit of the patient [10,11].

The cycle of systems medicine (Fig. 1.1) starts with a particular question at hand and requires identification of relevant parameters/variables to observe. Next, knowledge about the

biological interactions between the chosen variables is gathered to represent the system in the form of a mathematical model. The type of models depends largely on the question to be an-swered and the type of data available. They range from machine learning models [12–14] to mechanistic models based on rate equations [15–19], logical models [20] or agent-based sim-ulations [21]. In the final steps, the system is investigated for emergent patterns and validated by data obtained from the patient. Each cycle brings more understanding of the patient-specific state of the system [22].

)LJXUH1.16FKHPDWLFUHSUHVHQWDWLRQRIWKHV\VWHPPHGLFLQHF\FOH

C

Currently computational approaches used within systems medicine can be divided into two distinct branches: 1) a knowledge discovery branch, based mostly on data mining, machine learning, and neural networks that are able to extract emerging patterns from vast amounts of available data; 2) simulation-based analysis which utilizes mechanistic models based on pre-existing information to test hypotheses LQVLOLFR, and provide predictions [23].

ARTIFICIAL INTELLIGENCE

The knowledge discovery branch is used with great success in assisting clinicians at the diag-nosis stage. Several recent studies showed that deep learning algorithms were able to match

1

the performance of trained experts or even outperform clinicians in accuracy and sensitivity of diagnosis in breast cancer [24], Alzheimer disease [25], cardiac diseases [14] and lung pathol-ogies [26]. Furthermore, such algorithms can be used for LQsilico prediction of opportunities for therapy optimisation [12,13] or drug discovery [27,28]. However, the ‘black-box’ nature of artificial intelligence approaches limits their usefulness in areas where causality and an under-standing of the mechanisms underlying the disease are required.

PREDICTIVE MECHANISTIC MODELS

In contrast to artificial intelligence, mechanistic models rely on a detailed understanding of the system studied: its architecture, components, and their interactions, to build a mathematical model representing the system. Hypothesis-driven simulations using these models are com-pared with experimental observations for validation, and if inconsistent, they reveal that our knowledge about the system is incomplete. Once the model predictions are in agreement with the experimental observations, they can be used to generate novel hypotheses, as well as to explore questions that cannot be easily addressed experimentally. For example, genome-scale models have helped predict novel biomarkers for various diseases [18,29–32], while kinetic models have been used to predict novel treatment targets [19] or response to novel therapies [33,34]. There are many different types of models utilised within the systems medicine frame-work, including gene regulatory networks, signal transduction pathways, metabolic networks, and pharmacokinetics and pharmacodynamic models. Their usage depends on the type of data available and the scale at which we want to study the system of interest [35]. For example, in studies of metabolism genome-scale constraint-based models and detailed, dynamic models are ubiquitously used. Both model types can be personalised by integration with patient-specific data available, typically transcript, protein, and metabolite levels.

GENOME6&$/(&21675$,17%$6('02'(/6

In recent years, following the growth of ‘omics’ technologies, genome-scale metabolic models (GEMs) have been gaining popularity as they provide a snapshot of metabolism based on ge-netic and environmental constraints [36]. A GEM reconstruction describes a whole set of stoi-chiometry-based, mass-balanced metabolic reactions of an organism. It uses gene-protein-re-action (GPR) associations to link the regene-protein-re-actions with associated enzymes based on genome an-notation and experimental data [37]. Furthermore, genetic and environmental constraints such as transcriptomic, metabolomic, or fluxomic data are incorporated in the model to represent the modelled organism, tissue, or cell better. Using optimisation techniques based on linear pro-gramming, GEM models can predict the metabolic flux distribution within a network [38]. By

studying the changes in the flux distribution caused by a specific perturbation (either genetical or environmental), GEM models can reveal which pathways play a role in the observed phe-notype, or predict novel biomarkers. In studies of human metabolism, two major metabolic reconstructions are used: the human metabolic reaction (HMR) series with its latest version known as human-GEM [39] and the ReconX series with the newest update named Recon3D [40]. Human GEMs have contributed to a better understanding of the mechanisms underlying various diseases, and the prediction of appropriate disease biomarkers or treatments [41]. For example, the iHepatocytes2322 model (HMR-based) was used to identify serine deficiency in NAFLD [18], and recently, the multi-organ metabolic models based on Recon3D were coupled together into two gender-specific whole-body models (Harvey and Harvetta) which allowed to correctly represent inter-organ metabolic interactions, such as the Cori and Cahill cycles, and organ essentiality [42].

Despite the continuous improvements in human reconstruction models, they are still subject to several limitations. Firstly, the reconstructions are only as complete as our current knowledge and therefore, are susceptible to errors due to missing or inaccurate data gathered in various databases. Furthermore, metabolism of biopolymers such as glycogen and phospholipids car-ries a unique combinatorial challenge due to the enzyme promiscuity that has yet remained unsolved in the context of GEMs, and therefore, remains misrepresented. Also, the interaction between cofactors and apoenzymes, which are crucial for their activity, is often simplified and can lead to artefacts. Lastly, GEMs work under the assumption of a steady-state of a system and lack kinetic information and regulation, whereas, in living organisms, time-dependent dy-namics play an essential role.

DETAILED,.,1(7,&02'(/6

In contrast to GEMs, dynamic models comprise only of a subset of metabolic reactions, usually within a specific pathway, but at much greater detail, including enzyme kinetics and metabolic regulation. They are commonly represented as a set of coupled, ordinary differential equations (ODEs) based on rate equations for the enzymes in the system. ODEs describe all the reactions and transport processes in the model that either consume or produce the variable metabolites, one ODE per each metabolite pool. Dynamic models allow predictions to be made of metabo-lite concentrations and fluxes, both in a steady-state and kinetically over time. Moreover, sen-sitivity analysis or Metabolic Control Analysis, which analyses the response of variable me-tabolites to the changes in model parameters, allows identification of the metabolic steps that significantly influence the system’s behaviour under specific conditions [43]. The importance

1

of this approach in systems medicine has been beautifully illustrated by the work by Haanstra et al. They showed that the response to an enzyme inhibition in the same metabolic pathway might differ between a host and pathogen, allowing for targeted treatment. The same study also is an excellent example of how one can experimentally validate phenomenon discovered LQ

silico [19] as a model’s accuracy can vary significantly based on the parameters availability

and the level of detail. A similar approach was applied successfully to specifically target cancer cells without harming other cells or tissues in the body [34].

Just like GEMs, dynamic models also have their limitations. As mentioned before, they are prone to error due to insufficient quality and/or quantity of the available input data, e.g. missing parameter values, unknown enzyme kinetics, or unknown metabolic regulation. Furthermore, since the studied pathways are not connected with the systemic metabolic network in the model, but are instead studied in isolation, their simulated behaviour might not accurately represent the intricate interplay between various pathways in the organism.

I

Cancer [12,13,20,24] and other multifactorial, acquired metabolic diseases such as cardiac dis-eases [14,44–47], insulin resistance [48,49], and obesity [50] are amongst leading causes of morbidity in our society and therefore gaining much attention in the systems medicine field. However, their complexity poses a significant limitation for detailed mechanistic modelling. On the contrary, inborn errors of metabolism (IEM), inherited diseases caused primarily by a single gene mutation rendering its protein product inactive or blocking its expression, allow for a more tangible starting point for detailed modelling [15,16,29,32,51–56]. Despite their per-ceived scarcity (individual IEM affect less than 1 in 10 000 people), collectively, the IEM’s global prevalence is estimated to be between 0.4 and 1.3 per 1000 live births [1–5] with 33% mortality rate especially in the mid- and low-income countries [4]. Roughly two-thirds of all the IEM patients are children, often suffering from severe and life-threatening symptoms, stressing the urgency of diagnosis and early treatment to prevent the irreversible damage to the patient. The low numbers of patients per individual disease pose a problem in applying tradi-tional statistical. However, this makes IEMs the perfect subject for the personalised systems medicine approach, where each patient is treated as their own control [57].

The long-held approach to IEM as simple “one gene, one disease” problem has seen a paradigm shift in the last two decades and has forced clinicians and researchers to recognise IEM as complex diseases [58] requiring a systems approach to achieve a better mechanistic

understanding. As patients suffering from an identical mutation usually display a spectrum of disease phenotypes, the environment in which the deficient enzyme acts should not be ignored. Multiple modifying factors could explain the observed differences, such as environment, epi-genetics, and unique genetic background of individual patients. Together they cause either a decrease or increase of metabolic robustness leading to more or less severe phenotypes, respec-tively [59]. Systems medicine offers unique opportunities to study the system-wide effects of various IEM within their metabolic network and cellular environment, to dissect the underlying mechanisms in disease development and progression in each patient. However, to date, only a few studies integrating ‘omics’ data with modelling approaches have been published for IEM [15,16,29,32,51–56]. Among them, genome-scale approaches have shown up to 74% accuracy in predicting known biomarkers of various IEM [18,32,52], providing the first step for model-based biomarker predictions for other IEM. These genome-scale approaches are particularly suited to study the mechanism and predict biomarkers of diseases that impact multiple path-ways, such as fatty-acid oxidation (FAO) disorders, or disorders related to the cofactor metab-olism.

On the other hand, when a smaller (sub)system is studied, the dynamic models have yielded valuable insights in identifying and studying novel treatments or suggesting a new metabolic mechanism of disease. For example, publications analysing defects in the mitochondrial fatty-acid metabolism (mFAO) [15,16,55] and the role of blood-brain-barrier (BBB) transport in the phenylketonuria (PKU) [56] have been able to prove the usefulness of this approach in identi-fying and studying novel treatments or elucidating the metabolic mechanism of disease.

F

-

Flavin adenine dinucleotide (FAD) and its precursor flavin mononucleotide (FMN) are redox cofactors that are required for the activity of more than a hundred human enzymes. Out of those, 52 are known to cause human diseases if inactive, including several fatty acid oxidation (FAO) disorders [60]. In human cells, flavins are synthesised from their precursor riboflavin, also known as vitamin B2. Unlike nicotinamide adenine dinucleotide (NAD), which diffuses freely between enzymes, flavins are bound to enzymes called flavoproteins [61].

1

)LJXUH1.26FKHPDWLFUHSUHVHQWDWLRQRIWKH(7)F\FOH Flavoproteins are depicted in light grey, CoenzymeQ

(CoQ) and the lumped oxidative phosphorylation pathway (OxPhos) are indicated in dark grey ellipses in the mitochondrial inner membrane. Abbreviations: ACADs – acyl-CoA dehydrogenases, SARDH – sarcosine dehy-drogenase, DMG – dimethylglycine, DMGDH -dimethylglycine dehydehy-drogenase, ETF(ox) – oxidised electron-transfer-flavoprotein, ETF(red) – reduced electron-transfer-flavoprotein

A classical flavoprotein-dependent FAO disease is multiple-acyl-CoA-dehydrogenase defi-ciency (MADD), also known as glutaric aciduria type II. It is caused by a defect in one of the ETF cycle flavoproteins (ETFA, ETFB or ETFDHgenes) [62]. These FAD-containing en-zymes are crucial to link not only mitochondrial FAO but also amino acid metabolism of sar-cosine and dimethylglycine to oxidative phosphorylation (Fig. 1.2). Depending on the residual ETF cycle activity, MADD may lead to a life-threatening lack of energy supply to the body, with episodes of severe metabolic decompensation, hypoglycaemia, metabolic acidosis, sar-cosinemia and cardiovascular failure. The available treatment consists of low-fat, low-protein and high-carbohydrate diet with riboflavin, glycine and L-carnitine supplementation.

How-ever, this treatment is not sufficient for neonatal patients [63] for whom experimental treat-ment with sodium-D, L-3-hydroxybutyrate showed promising results [64]. This example shows clearly the crucial and complex role of the FAD in the metabolism.

Cofactors, are among the most central nodes in the metabolic network, being the most con-nected molecules [65]. However, current genome-scale models do not distinguish between the cofactors that are bound to their co-enzymes (e.g. FAD) and those freely diffusing (e.g. NAD), which introduces artificial degrees of freedom in the studied network. This problem should be urgently addressed to allow more accurate model predictions with regard to various cofactor-related IEMs, many of which play a central role in fatty acid metabolism.

F

Fatty acid oxidation (FAO) defects are among the most common IEM identified worldwide with a prevalence between 0.0139 and 0.0789 in 1000 live births [1,2,4]. However, the preva-lence if IEMs is regionally-dependent with Portugal having the highest incidence of FAO re-ported of 0.1575 in 1000 live births [66]. Traditionally FAO defects include only diseases

affecting mitochondrial transport and FAO, whereas peroxisomal FAO defects are part of the peroxisomal disorders (see Fig. 1.3). If counted together, their prevalence would increase to 0.0792-0.1337 per 1000 live births [4]. Lipids are an essential source of energy during periods of fasting, decreased food intake, or increased energy demands due to physical activity or ill-ness. Depending on the length of the fatty acid (FA) and its structure (straight or branched-chain; saturated or unsaturated), different metabolic degradation pathways are predominant. These pathways are localised in mitochondria, peroxisomes, or endoplasmic reticulum. In total, around 35 different diseases affecting either mitochondrial or peroxisomal FAO have been found to date. They have variable presentations, varying in time of disease onset and disease severity. Presentations range from neonatal onset with malformation syndrome, metabolic ac-idosis, cardiomyopathy, sudden death, retinopathy, developmental delay, myelopathy and pe-ripheral neuropathy, to late-onset with neuropathy, myopathy, and retinopathy, whereas some patients even remain asymptomatic [67–69].

FATTY-ACID OXIDATION

Mitochondria can utilise both saturated and unsaturated FAs with up to 22-carbon chains in a process called β-oxidation. It generates acetyl-CoA, a substrate for the tricarboxylic acid (TCA) cycle, and reducing agents flavin adenine dinucleotide (FADH2) and nicotinamide adenine

di-nucleotide (NADH). FADH2 and NADH donate electrons to the respiratory chain (OxPhos)

for ATP generation (Fig. 1.3). In total, mitochondrial β-oxidation of palmitate (16 carbons) yields 106 molecules of ATP [70]. However, very-long-chain FAs (above 22 carbons), branched-chain FAs, and long-chain dicarboxylic FAs cannot be broken down in mitochondria, but require peroxisomal α- and β-oxidation to be processed [71]. Peroxisomal β-oxidation fol-lows the same chemical steps as its mitochondrial counterpart of stepwise shortening of acyl-CoAs, producing acetyl-CoA from straight-chain FAs and propionyl-CoA from branched-chain FAs. Despite the similarities, peroxisomal and mitochondrial enzymes are encoded by distinct genes and show different substrate specificity. Moreover, the enzymes catalysing the first step of mitochondrial FAO are FAD-dependent dehydrogenases, donating their electrons to the respiratory chain via the Electron-Transfer-Flavoprotein (ETF) cycle, while their perox-isomal counterparts, FAD-dependent acyl-CoA oxidases, donate their electrons directly to mo-lecular oxygen. Additionally, mitochondrial FAO can fully break down FAs to CO2 and water,

with the active participation of TCA cycle oxidative phosphorylation, while peroxisomal FAO can only shorten the FAs, and its products need to be transported to the mitochondria for com-plete oxidation. For this, the CoA esters produced in the peroxisomal β-oxidation need to be

1

converted to carnitine esters before they can be exported from peroxisomes. However, the exact transport mechanism of peroxisomal FAO products is currently unknown [71].

)LJXUH1.36FKHPDWLFUHSUHVHQWDWLRQRIWKHPDPPDOLDQIDWW\DFLGR[LGDWLRQ)$2SDWKZD\V Transporter

proteins and complexes I-V of the respiratory chain or oxidative phosphorylation are indicated as grey boxes in the mitochondrial inner membrane. General substrates and products of the FAO pathways are shown. Abbrevia-tions: CoA – coenzyme A, Car – carnitine, FA – fatty acid, SCFA – short-chain fatty acids, MCFA – medium-chain fatty acids, LCFA- long-medium-chain fatty acids, VLCFA – very-long-medium-chain fatty acids, SDCFAs – short-dicar-boxylic fatty acids, TCA – tricarshort-dicar-boxylic acid cycle, OxPhos – oxidative phosphorylation

Despite a broad enzyme specificity of peroxisomal β-oxidation, some branched-chain FAs con-tain a functional group at the 3-carbon that cannot be β-oxidised. These FAs require the perox-isomal α-oxidation pathway to be broken down. This unique pathway shortens the FAs by only one carbon, leading to a formation of formyl-CoA and a shortened FA that can be a substrate for the β-oxidation. A classical substrate of α-oxidation is phytanoyl-CoA [72–74]. Alterna-tively, ω-oxidation in the endoplasmic reticulum has been found to play a supportive role in the breakdown of fatty acids. In this pathway, FAs are first converted to ω-hydroxy-FA and then oxidised to dicarboxylic fatty acids (DCFAs) [75–77]. DCFAs are a substrate for peroxi-somal β-oxidation and can be broken down to medium- and short-chain dicarboxylic acids, which are hydrophilic enough to be removed from the body via urine. Together, all FAO path-ways form a complex system that does not only provide the body with energy but is also

required to break down potentially toxic compounds such as very-long-chain fatty acids and phytanic acid [78,79].

REFSUM DISEASE

As mentioned earlier, peroxisomes are organelles that, among other functions, are crucial for cellular lipid metabolism. Refsum disease is an inborn error of metabolism (IEM) that is caused by biallelic mutations in the gene encoding phytanoyl-CoA 2-hydroxylase (PHYH), resulting in defective α-oxidation of the branched-chain fatty acid phytanic acid (3,7,11,15-tetra-methylhexadecanoic acid) [74]. Since phytanic acid contains a 3-methyl group, it needs to un-dergo α-oxidation first, thereby producing pristanic acid, which then can be further degraded by β-oxidation (see Fig. 1.2) [80]. Alternatively, phytanic acid can be ω-oxidised in the endo-plasmic reticulum [76]. The end product of ω-oxidation and subsequent β-oxidation of phytanic acid is 3-methyladipic acid (3-MAA), which has been described to be upregulated in patients with Refsum disease [81]. Refsum disease was first described in 1945, with symptoms includ-ing progressive retinitis pigmentosa, polyneuropathy, cerebellar ataxia, and deafness [81]. Phytanic acid is a xenobiotic compound for humans, solely derived from the diet, and patients with Refsum disease are mostly diagnosed in late childhood by elevated levels of phytanic acid in plasma and tissues [74,81]. Currently, plasmapheresis and a strict diet to reduce the intake of dairy-derived fat are the only available treatments for patients. However, a study by Baldwin et al. shows that 33% of patients had problems with compliance to the diet. Furthermore, epi-sodes of illness, weight loss, or pregnancy may lead to an increase in plasma phytanic acid and cause rapid progression of the disease [82]. Therefore, further studies focusing on novel bi-omarkers of disease progression as well as diet and treatment optimisation will be essential for better patient care.

A

Amino acids are essential building blocks of proteins and serve as neurotransmitters, precursors of hormones, coenzymes, pigments, neurotransmitters, and nucleic acids. Many of them can be synthesised by humans, except for nine essential amino acids which must be obtained from the diet. All amino acids have their unique degradative pathways in which nitrogen and carbon components are recycled to build other amino acids, carbohydrates, or lipids. Disorders of amino acid metabolism and transport are individually rare, with a prevalence from 0.0655 in 1000 live births for phenylketonuria to 0.0041 in 1000 live births for homocystinuria [4]. Tra-ditionally, they are named after the compound that accumulates in the blood (i.e. tyrosinemia) or urine (i.e. alkaptonuria). However, both biochemical and genetic heterogeneity are common,

1

leading to a range of distinct forms of each of the diseases. Similarly, symptoms vary greatly depending on the disease. While sarcosinemia has no clinical phenotype, ornithine transcar-bamylase is lethal if left untreated in the neonate. In more than half of the disorders central nervous system dysfunction, developmental retardation, or behavioural disturbances are pre-sent [83].

PHENYLKETONURIA

Phenylketonuria (PKU) is a classic example of an inborn error of amino-acid metabolism. It is caused by a deficiency of the hepatic phenylalanine hydroxylase (PAH) enzyme, which con-verts phenylalanine into tyrosine [84]. If left untreated, phenylalanine levels in plasma and brain build-up, which leads to PKU symptomatology including severe intellectual disability, seizures, and psychiatric problems. High brain phenylalanine levels are neurotoxic and affect brain metabolism [85–89]. Moreover, phenylalanine inhibits tyrosine and tryptophan hydrox-ylases, which are critical enzymes in the synthesis of cerebral monoaminergic neurotransmit-ters, dopamine and serotonin. [90–93]. Together, a combination of high phenylalanine and low tyrosine and tryptophan may lead to low concentrations of monoaminergic neurotransmitters. This has been suggested to play an essential role in the mood and psychosocial problems of PKU patients [94–96]. Today, neonatal screening allows PKU diagnosis and initiation of treat-ment shortly after birth. Treattreat-ment consists of a strict phenylalanine-restricted diet: low in nat-ural protein and a tyrosine-enriched amino acid supplement including all amino acids but phe-nylalanine. Similarly, to riboflavin supplementation in MADD, some PKU patients respond well to tetrahydrobiopterin supplementation, which is a cofactor of PAH. While a strict diet can prevent severe intellectual disability, the outcome remains suboptimal and warrants addi-tional/alternative pathophysiology-based treatment strategies aiming to decrease the brain phe-nylalanine concentrations and limit the abnormalities in cerebral neurotransmitter, protein and myelin metabolism [97]. Furthermore, in only 65% of patients above 16 years old phenylala-nine levels were within guideline levels, suggesting a decrease in compliance to the died by adult patients [98]. As a possible novel dietary treatment strategy, supplementation of non-phenylalanine large neutral amino acids (LNAA) instead of restricting dietary non-phenylalanine intake has been suggested. It has shown some promises in studies with mice. However, further studies and diet optimisation will be essential to include this treatment in patients [90,99,100].

O

With the rapid growth of the systems medicine field, we are able to acquire an unprecedented amount of patient-specific data. This should lead to better diagnosis and understanding of dis-eases, and identification of the most efficient treatment for each individual patient. However, this aim cannot be achieved without robust and validated mathematical models that integrate the current knowledge and experimental data to analyse the system behaviour. In this thesis, I addressed this problem using different modelling approaches as tools for multi-level data inte-gration, hypothesis generation, and studies of disease mechanism in the set of inborn errors of metabolism. Through collaborative work with clinicians and experimentalists, I was able to study and provide new insights into the biochemical consequences of primary genetic defects by applying both genome-scale, constraint-based, and detailed, dynamic modelling coupled with data analysis and integration.

In the first two chapters, I aimed to identify novel biomarkers and better understand the meta-bolic consequences of genetic defects using genome-scale models of human metabolism. As explained abovecofactors play a central role in human metabolism, yet genome-scale meta-bolic models do not distinguish between the cofactors that are bound to their co-enzymes (e.g. FAD) and those that diffuse freely (e.g. NAD). Therefore, in &KDSWHU, I addressed this problem by curating the FAD-related metabolism in the human genome-scale model Recon3D. I validated the new model by simulating the metabolic consequences of the MADD disease. Finally, I predicted biomarkers for 16 flavoproteome-related IEMs.

In &KDSWHU, I used the model generated in the previous chapter and integrated it with multi-omics data (transcriptmulti-omics, protemulti-omics, and metabolmulti-omics) to generate a fibroblast-specific model. This model was used to study the metabolic consequences of Refsum disease, a defect in peroxisomal FAO. To accurately represent the complexity of the human FAO system ex-plained above, pathways related to the oxidation of phytanic acid were curated and extended. Using this model, I investigated the metabolic phenotype of Refsum disease at the genome-scale. The curated Refsum model provides a basis for the identification of novel biomarkers and targets for treatment optimisation. In &KDSWHU, my aim was to apply systems medicine approach to optimise a treatment strategy in an IEM. Because of the availability of relevant data, I decided to focus on phenylketonuria (PKU) which is the most frequent disease among all the IEMs. As described above, diet is currently the only treatment available for PKU pa-tients. Nevertheless, their adherence to the diet tends to decrease with age. Therefore, in

1

&KDSWHU, I studied the relationship between the plasma large neutral amino acids (LNAA)

levels and brain LNAA and neurotransmitter levels. Focusing on LNAA transport via the blood-brain-barrier and the subsequent brain neurotransmitter synthesis, I constructed a de-tailed, kinetic model. The model was based on the available enzyme kinetics data and validated using experimental data. Furthermore, I used this model to test alternative dietary treatments based on the LNAA supplementation in PKU.

The final part of this thesis includes a general discussion of all chapters and provides conclud-ing remarks and future perspectives and challenges for systems medicine &KDSWHU.

R

1. Sanderson S, Green A, Preece MA, Burton H. The incidence of inherited metabolic disorders in the West Midlands, UK. Arch Dis Child. 2006;91: 896–899. doi:10.1136/adc.2005.091637 2. Dionisi-Vici C, Rizzo C, Burlina AB, Caruso U, Sabetta G, Uziel G, et al. Inborn errors of

metabolism in the Italian pediatric population: A national retrospective survey. J Pediatr. 2002;140: 321–329. doi:10.1067/mpd.2002.122394

3. Applegarth DA, Toone JR, Lowry R and RB. Incidence of Inborn Errors of Metabolism in British Columbia, 1969–1996. Pediatrics. 2000;105: e10–e10. doi:10.1542/peds.105.1.e10 4. Waters D, Adeloye D, Woolham D, Wastnedge E, Patel S, Rudan I. Global birth prevalence and

mortality from inborn errors of metabolism: a systematic analysis of the evidence. J Glob Health. 2018;8: 021102. doi:10.7189/jogh.08.021102

5. Mak CM, Lee H-CH, Chan AY-W, Lam C-W. Inborn errors of metabolism and expanded newborn screening: review and update. Crit Rev Clin Lab Sci. 2013;50: 142–162. doi:10.3109/10408363.2013.847896

6. Dunbar CE, High KA, Joung JK, Kohn DB, Ozawa K, Sadelain M. Gene therapy comes of age. Science (80- ). 2018;359: eaan4672. doi:10.1126/science.aan4672

7. Kishnani PS, Sun B, Koeberl DD. Gene therapy for glycogen storage diseases. Hum Mol Genet. 2019;28: R31–R41. doi:10.1093/hmg/ddz133

8. Ultragenyx Announces Positive Data from Phase 1/2 Study of DTX401 Gene Therapy in Glycogen Storage Disease Type Ia | Ultragenyx Pharmaceutical Inc. [Internet]. [cited 30 Oct 2019]. Available: http://ir.ultragenyx.com/news-releases/news-release-details/ultragenyx-announces-positive-data-phase-12-study-dtx401-gene

9. Hood L, Friend SH. Predictive, personalized, preventive, participatory (P4) cancer medicine. Nat Rev Clin Oncol. 2011;8: 184–187. doi:10.1038/nrclinonc.2010.227

10. Wolkenhauer O, Auffray C, Jaster R, Steinhoff G, Dammann O. The road from systems biology to systems medicine. Pediatr Res. 2013;73: 502–507. doi:10.1038/pr.2013.4

11. Apweiler R, Beissbarth T, Berthold MR, Blüthgen N, Burmeister Y, Dammann O, et al. Whither systems medicine? Exp Mol Med. 2018;50: e453–e453. doi:10.1038/emm.2017.290

12. Bibault JE, Giraud P, Durdux C, Taieb J, Berger A, Coriat R, et al. Deep Learning and Radiomics predict complete response after neo-adjuvant chemoradiation for locally advanced rectal cancer. Sci Rep. 2018;8: 12611. doi:10.1038/s41598-018-30657-6

13. Yousefi S, Amrollahi F, Amgad M, Dong C, Lewis JE, Song C, et al. Predicting clinical outcomes from large scale cancer genomic profiles with deep survival models. Sci Rep. 2017;7: 11707. doi:10.1038/s41598-017-11817-6

14. Madani A, Ong JR, Tibrewal A, Mofrad MRK. Deep echocardiography: data-efficient supervised and semi-supervised deep learning towards automated diagnosis of cardiac disease. npj Digit Med. 2018;1: 59. doi:10.1038/s41746-018-0065-x

15. Martines A-CMF, van Eunen K, Reijngoud D-J, Bakker BM. The promiscuous enzyme medium-chain 3-keto-acyl-CoA thiolase triggers a vicious cycle in fatty-acid beta-oxidation. Nielsen J, editor. PLOS Comput Biol. 2017;13: e1005461. doi:10.1371/journal.pcbi.1005461 16. van Eunen K, Simons SMJ, Gerding A, Bleeker A, den Besten G, Touw CML, et al. Biochemical

1

editor. PLoS Comput Biol. 2013;9: e1003186. doi:10.1371/journal.pcbi.1003186

17. Stavrum A-K, Heiland I, Schuster S, Puntervoll P, Ziegler M. Model of tryptophan metabolism, readily scalable using tissue-specific gene expression data. J Biol Chem. 2013;288: 34555–66. doi:10.1074/jbc.M113.474908

18. Mardinoglu A, Agren R, Kampf C, Asplund A, Uhlen M, Nielsen J. Genome-scale metabolic modelling of hepatocytes reveals serine deficiency in patients with non-alcoholic fatty liver disease. Nat Commun. 2014;5: 3083. doi:10.1038/ncomms4083

19. Haanstra JR, Gerding A, Dolga AM, Sorgdrager FJH, Buist-Homan M, du Toit F, et al. Targeting pathogen metabolism without collateral damage to the host. Sci Rep. 2017;7: 40406. doi:10.1038/srep40406

20. Eduati F, Doldàn-Martelli V, Klinger B, Cokelaer T, Sieber A, Kogera F, et al. Drug Resistance Mechanisms in Colorectal Cancer Dissected with Cell Type–Specific Dynamic Logic Models. Cancer Res. 2017;77: 3364–3375. doi:10.1158/0008-5472.CAN-17-0078

21. Cosgrove J, Butler J, Alden K, Read M, Kumar V, Cucurull-Sanchez L, et al. Agent-Based Modeling in Systems Pharmacology. CPT Pharmacometrics Syst Pharmacol. 2015;4: 615–629. doi:10.1002/psp4.12018

22. Westerhoff H V, Winder C, Messiha H, Simeonidis E, Adamczyk M, Verma M, et al. Systems Biology: The elements and principles of Life. FEBS Lett. 2009;583: 3882–3890. doi:10.1016/j.febslet.2009.11.018

23. Kitano H. Computational systems biology. Nature. 2002;420: 206–210. doi:10.1038/nature01254

24. Ehteshami Bejnordi B, Veta M, Johannes van Diest P, van Ginneken B, Karssemeijer N, Litjens G, et al. Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer. JAMA. 2017;318: 2199. doi:10.1001/jama.2017.14585

25. Ding Y, Sohn JH, Kawczynski MG, Trivedi H, Harnish R, Jenkins NW, et al. A deep learning model to predict a diagnosis of Alzheimer disease by using 18 F-FDG PET of the brain. Radiology. 2019;290: 456–464. doi:10.1148/radiol.2018180958

26. Rajpurkar P, Irvin J, Ball RL, Zhu K, Yang B, Mehta H, et al. Deep learning for chest radiograph diagnosis: A retrospective comparison of the CheXNeXt algorithm to practicing radiologists. Sheikh A, editor. PLOS Med. 2018;15: e1002686. doi:10.1371/journal.pmed.1002686 27. Chen H, Engkvist O, Wang Y, Olivecrona M, Blaschke T. The rise of deep learning in drug

discovery [Internet]. Drug Discovery Today. Elsevier Current Trends; 2018. pp. 1241–1250. doi:10.1016/j.drudis.2018.01.039

28. Mak KK, Pichika MR. Artificial intelligence in drug development: present status and future prospects [Internet]. Drug Discovery Today. Elsevier Current Trends; 2019. pp. 773–780. doi:10.1016/j.drudis.2018.11.014

29. Sahoo S, Thiele I. Predicting the impact of diet and enzymopathies on human small intestinal epithelial cells. Hum Mol Genet. 2013;22: 2705–2722. doi:10.1093/hmg/ddt119

30. Yizhak K, Chaneton B, Gottlieb E, Ruppin E. Modeling cancer metabolism on a genome scale. Mol Syst Biol. 2015;11. doi:10.15252/msb.20145307

31. Asgari Y, Khosravi P, Zabihinpour Z, Habibi M. Exploring candidate biomarkers for lung and prostate cancers using gene expression and flux variability analysis. Integr Biol. 2018;10: 113–

120. doi:10.1039/C7IB00135E

32. Wegrzyn AB, Stolle S, Rienksma RA, Martins dos Santos VAP, Bakker BM, Suarez-Diez M. Cofactors revisited – Predicting the impact of flavoprotein-related diseases on a genome scale. Biochim Biophys Acta - Mol Basis Dis. 2019;1865: 360–370. doi:10.1016/j.bbadis.2018.10.021 33. da Veiga Moreira J, Hamraz M, Abolhassani M, Schwartz L, Jolicœur M, Peres S. Metabolic therapies inhibit tumor growth in vivo and in silico. Sci Rep. 2019;9: 3153. doi:10.1038/s41598-019-39109-1

34. Liberti M V., Dai Z, Wardell SE, Baccile JA, Liu X, Gao X, et al. A Predictive Model for Selective Targeting of the Warburg Effect through GAPDH Inhibition with a Natural Product. Cell Metab. 2017;26: 648-659.e8. doi:10.1016/j.cmet.2017.08.017

35. Wolkenhauer O, Auffray C, Brass O, Clairambault J, Deutsch A, Drasdo D, et al. Enabling multiscale modeling in systems medicine. Genome Med. 2014;6: 21. doi:10.1186/gm538 36. Ryu JY, Kim HU, Lee SY. Reconstruction of genome-scale human metabolic models using

omics data. Integr Biol. 2015;7: 859–868. doi:10.1039/c5ib00002e

37. Thiele I, Palsson BØB. A protocol for generating a high-quality genome-scale metabolic reconstruction. Nat Protoc. 2010;5: 93–121. doi:10.1038/nprot.2009.203.

38. Orth JDJJD, Thiele I, Palsson BØOBBØO. What is flux balance analysis? Nat Biotechnol. 2010;28: 245–8. doi:10.1038/nbt.1614

39. Metabolic atlas [Internet]. [cited 3 Sep 2019]. Available: https://www.metabolicatlas.org/ 40. Brunk E, Sahoo S, Zielinski DC, Altunkaya A, Dräger A, Mih N, et al. Recon3D enables a

three-dimensional view of gene variation in human metabolism. Nat Biotechnol. 2018;36: 272–281. doi:10.1038/nbt.4072

41. Gu C, Kim GB, Kim WJ, Kim HU, Lee SY. Current status and applications of genome-scale metabolic models. Genome Biol. 2019;20: 121. doi:10.1186/s13059-019-1730-3

42. Thiele I, Sahoo S, Heinken A, Heirendt L, Aurich MK, Noronha A, et al. When metabolism meets physiology: Harvey and Harvetta. bioRxiv. 2018; 255885. doi:10.1101/255885

43. Bruggeman FJ, Westerhoff H V. The nature of systems biology. Trends Microbiol. 2007;15: 45–50. doi:10.1016/j.tim.2006.11.003

44. Penn MS, Klemes AB. Multimarker approach for identifying and documenting mitigation of cardiovascular risk. Intergovernmental Panel on Climate Change, editor. Future Cardiol. 2013;9: 497–506. doi:10.2217/fca.13.27

45. Edwards LM, Sigurdsson MI, Robbins PA, Weale ME, Cavalleri GL, Montgomery HE, et al. Genome-Scale Methods Converge on Key Mitochondrial Genes for the Survival of Human Cardiomyocytes in Hypoxia. Circ Cardiovasc Genet. 2014;7: 407–415. doi:10.1161/CIRCGENETICS.113.000269

46. Kosta S, Negroni J, Lascano E, Dauby PC. Multiscale model of the human cardiovascular system: description of heart failure and comparison of contractility indices. Math Biosci. 2016;0: 1–9. doi:10.1016/j.mbs.2016.05.007

47. Karlstädt A, Fliegner D, Kararigas G, Ruderisch HS, Regitz-Zagrosek V, Holzhütter H-G. CardioNet: a human metabolic network suited for the study of cardiomyocyte metabolism. BMC Syst Biol. 2012;6: 114. doi:10.1186/1752-0509-6-114

1

48. Lee S, Zhang C, Kilicarslan M, Piening BD, Bjornson E, Hallström BM, et al. Integrated Network Analysis Reveals an Association between Plasma Mannose Levels and Insulin Resistance. Cell Metab. 2016;24: 172–184. doi:10.1016/j.cmet.2016.05.026

49. Kumar A, Harrelson T, Lewis NE, Gallagher EJ, LeRoith D, Shiloach J, et al. Multi-Tissue Computational Modeling Analyzes Pathophysiology of Type 2 Diabetes in MKR Mice. Alquier T, editor. PLoS One. 2014;9: e102319. doi:10.1371/journal.pone.0102319

50. Mardinoglu A, Kampf C, Asplund A, Fagerberg L, Hallström BM, Edlund K, et al. Defining the Human Adipose Tissue Proteome To Reveal Metabolic Alterations in Obesity. J Proteome Res. 2014;13: 5106–5119. doi:10.1021/pr500586e

51. Pagliarini R, di Bernardo D. A Genome-Scale Modeling Approach to Study Inborn Errors of Liver Metabolism: Toward an In Silico Patient. J Comput Biol. 2013;20: 383–397. doi:10.1089/cmb.2012.0276

52. Sahoo S, Franzson L, Jonsson JJ, Thiele I, Sahoo S, Franzson L, et al. A compendium of inborn errors of metabolism mapped onto the human metabolic network. Mol Biosyst. 2012;8: 2545. doi:10.1039/c2mb25075f

53. Sahoo S. Systems biology of inborn errors of metabolism. University of Iceland. 2013. 54. Tebani A, Afonso C, Marret S, Bekri S. Omics-Based Strategies in Precision Medicine: Toward

a Paradigm Shift in Inborn Errors of Metabolism Investigations. Int J Mol Sci. 2016;17. doi:10.3390/ijms17091555

55. van Eunen K, Volker-Touw CML, Gerding A, Bleeker A, Wolters JC, van Rijt WJ, et al. Living on the edge: substrate competition explains loss of robustness in mitochondrial fatty-acid oxidation disorders. BMC Biol. 2016;14: 107. doi:10.1186/s12915-016-0327-5

56. Taslimifar M, Buoso S, Verrey F, Kurtcuoglu V. Propagation of Plasma L-Phenylalanine Concentration Fluctuations to the Neurovascular Unit in Phenylketonuria: An in silico Study. Front Physiol. 2019;10: 360. doi:10.3389/fphys.2019.00360

57. Schee genannt Halfmann S, Mählmann L, Leyens L, Reumann M, Brand A. Personalized Medicine: What’s in it for Rare Diseases? Adv Exp Med Biol. 2017;1031: 387–404. doi:10.1007/978-3-319-67144-4_22

58. Dipple KM, McCabe ERB. Phenotypes of Patients with “Simple” Mendelian Disorders Are Complex Traits: Thresholds, Modifiers, and Systems Dynamics. Am J Hum Genet. 2000;66: 1729–1735. doi:10.1086/302938

59. Dipple KM, Phelan JK, McCabe ERB. Consequences of complexity within biological networks: Robustness and health, or vulnerability and disease. Mol Genet Metab. 2001;74: 45–50. doi:10.1006/mgme.2001.3227

60. Lienhart WD, Gudipati V, MacHeroux P. The human flavoproteome. Arch Biochem Biophys. 2013;535: 150–162. doi:10.1016/j.abb.2013.02.015

61. Heuts DPHM, Scrutton NS, McIntire WS, Fraaije MW. What’s in a covalent bond? FEBS J. 2009;276: 3405–3427. doi:10.1111/j.1742-4658.2009.07053.x

62. Frerman FE, Goodman SI. Deficiency of electron transfer flavoprotein or electron transfer flavoprotein:ubiquinone oxidoreductase in glutaric acidemia type II fibroblasts. Proc Natl Acad Sci U S A. 1985;82: 4517–20. doi:10.1073/pnas.82.13.4517

63. Frerman FE, Goodman SI, Frerman FE GS, Frerman, Goodman F, Frerman FE, Goodman SI. Defects of Electron Transfer Flavoprotein and Electron Transfer Flavoprotein-Ubiquinone

Oxidoreductase: Glutaric Acidemia Type II. In: Valle D, Beaudet AL, Vogelstein B, Kinzler KW, Antonarakis SE, Ballabio A, Gibson K MG eds., editor. The Metabolic and Molecular Basis of Inherited Disease. New York, NY: McGraw-Hill; 2014. pp. 2357–2365. doi:10.1036/ommbid.131

64. Van Rijt WJ, Heiner-Fokkema MR, du Marchie Sarvaas GJ, Waterham HR, Blokpoel RGT, van Spronsen FJ, et al. Favorable Outcome After Physiologic Dose of Sodium-D,L-3-Hydroxybutyrate in Severe MADD. Pediatrics. 2014;134: e1224–e1228. doi:10.1542/peds.2013-4254

65. Wagner A, Fell DA. The small world inside large metabolic networks. Proc R Soc London Ser B Biol Sci. 2001;268: 1803–1810. doi:10.1098/rspb.2001.1711

66. Rocha H, Castiñeiras D, Delgado C, Egea J, Yahyaoui R, González Y, et al. Birth Prevalence of Fatty Acid β-Oxidation Disorders in Iberia. JIMD Rep. 2014;16: 89–94. doi:10.1007/8904_2014_324

67. Argyriou C, D’Agostino MD, Braverman N. Peroxisome biogenesis disorders. Transl Sci rare Dis. 2016;1: 111–144. doi:10.3233/TRD-160003

68. Waterham HR, Ferdinandusse S, Wanders RJA. Human disorders of peroxisome metabolism and biogenesis. Biochim Biophys Acta - Mol Cell Res. 2016;1863: 922–933. doi:10.1016/j.bbamcr.2015.11.015

69. Merritt II JL, Norris M, Kanungo S. Fatty acid oxidation disorders. Ann Transl Med. 2018;6: 473–473. doi:10.21037/atm.2018.10.57

70. Stryer L, Berg JM (Jeremy M, Tymoczko JL, Gatto Jr. GJ. Biochemistry [Internet]. 9th ed. W.H. Freeman; 2019. Available: https://www.ncbi.nlm.nih.gov/books/NBK21154/

71. Wanders RJA, Waterham HR, Ferdinandusse S. Metabolic Interplay between Peroxisomes and Other Subcellular Organelles Including Mitochondria and the Endoplasmic Reticulum. Front Cell Dev Biol. 2016;3: 83. doi:10.3389/fcell.2015.00083

72. Croes K, Van Veldhoven PP, Mannaerts GP, Casteels M. Production of formyl-CoA during peroxisomal alpha-oxidation of 3-methyl-branched fatty acids. FEBS Lett. 1997;407: 197–200. doi:10.1016/S0014-5793(97)00343-8

73. Foulon V, Sniekers M, Huysmans E, Asselberghs S, Mahieu V, Mannaerts GP, et al. Breakdown of 2-hydroxylated straight chain fatty acids via peroxisomal 2-hydroxyphytanoyl-CoA lyase: A revised pathway for the ??-oxidation of straight chain fatty acids. J Biol Chem. 2005;280: 9802– 9812. doi:10.1074/jbc.M413362200

74. Wanders RJA, Komen J, Ferdinandusse S. Phytanic acid metabolism in health and disease. Biochim Biophys Acta. 2011;1811: 498–507. doi:10.1016/j.bbalip.2011.06.006

75. Kramar R. The contribution of peroxisomes to lipid metabolism. J Clin Chem Clin Biochem. 1986;24: 109–118.

76. Wanders RJA, Komen J, Kemp S. Fatty acid omega-oxidation as a rescue pathway for fatty acid oxidation disorders in humans. FEBS J. 2011;278: 182–194. doi:10.1111/j.1742-4658.2010.07947.x

77. Sanders R-J, Ofman R, Dacremont G, Wanders RJA, Kemp S. Characterization of the human omega-oxidation pathway for omega-hydroxy-very-long-chain fatty acids. FASEB J. 2008;22: 2064–2071. doi:10.1096/fj.07-099150

1

phytanic acid, a marker of Refsum disease, in astrocytes involves mitochondrial impairment. International journal of developmental …. 2006. pp. 113–22. doi:10.1016/j.ijdevneu.2005.11.002

79. Khan M, Singh J, Gilg AG, Uto T, Singh I. Very long-chain fatty acid accumulation causes lipotoxic response via 5-lipoxygenase in cerebral adrenoleukodystrophy. J Lipid Res. 2010;51: 1685–95. doi:10.1194/jlr.M002329

80. Van Veldhoven PP. Biochemistry and genetics of inherited disorders of peroxisomal fatty acid metabolism. J Lipid Res. 2010;51: 2863–2895. doi:10.1194/jlr.R005959

81. Wierzbicki ASA, Lloyd MDM, Schofield CJ, Feher MD, Gibberd FB. Refsum’s disease: a peroxisomal disorder affecting phytanic acid alpha-oxidation. J Neurochem. 2002;80: 727–735. doi:10.1046/j.0022-3042.2002.00766.x

82. Baldwin EJ, Gibberd FB, Harley C, Sidey MC, Feher MD, Wierzbicki AS. The effectiveness of long-term dietary therapy in the treatment of adult Refsum disease. J Neurol Neurosurg Psychiatry. 2010;81: 954–957. doi:10.1136/jnnp.2008.161059

83. Aliu E, Kanungo S, Arnold GL. Amino acid disorders. Ann Transl Med. 2018;6: 471–471. doi:10.21037/atm.2018.12.12

84. Scriver CR, Clow CL. Phenylketonuria: Epitome of Human Biochemical Genetics. N Engl J Med. 1980;303: 1336–1342. doi:10.1056/NEJM198012043032305

85. de Groot MJ, Sijens PE, Reijngoud D-J, Paans AM, van Spronsen FJ. Phenylketonuria: brain phenylalanine concentrations relate inversely to cerebral protein synthesis. J Cereb blood flow Metab. 2015;35: 200–205. doi:10.1038/jcbfm.2014.183

86. Martynyuk AE, Glushakov AV, Sumners C, Laipis PJ, Dennis DM, Seubert CN. Impaired glutamatergic synaptic transmission in the PKU brain. Mol Genet Metab. 2005;86: 34–42. doi:10.1016/j.ymgme.2005.06.014

87. Glushakov A V, Glushakova O, Varshney M, Bajpai LK, Sumners C, Laipis PJ, et al. Long-term changes in glutamatergic synaptic transmission in phenylketonuria. Brain A J Neurol. 2005;128: 300–307. doi:10.1093/brain/awh354

88. Hörster F, Schwab MA, Sauer SW, Pietz J, Hoffmann GF, Okun JG, et al. Phenylalanine reduces synaptic density in mixed cortical cultures from mice. Pediatr Res. 2006;59: 544–548. doi:10.1203/01.pdr.0000203091.45988.8d

89. Shefer S, Tint GS, Jean-Guillaume D, Daikhin E, Kendler A, Nguyen LB, et al. Is there a relationship between 3-hydroxy-3-methylglutaryl coenzyme a reductase activity and forebrain pathology in the PKU mouse? J Neurosci Res. 2000;61: 549–563. doi:10.1002/1097-4547(20000901)61:5<549::AID-JNR10>3.0.CO;2-0

90. Van Vliet D, Bruinenberg VM, Mazzola PN, Van Faassen MHJR, De Blaauw P, Pascucci T, et al. Therapeutic brain modulation with targeted large neutral amino acid supplements in the Pah-enu2 phenylketonuria mouse model. Am J Clin Nutr. 2016;104: 1292–1300. doi:10.3945/ajcn.116.135996

91. Andolina D, Conversi D, Cabib S, Trabalza A, Ventura R, Puglisi-Allegra S, et al. 5-Hydroxytryptophan during critical postnatal period improves cognitive performances and promotes dendritic spine maturation in genetic mouse model of phenylketonuria. Int J Neuropsychopharmacol. 2011;14: 479–489. doi:10.1017/S1461145710001288

92. Hommes FA, Lee JS. The control of 5-hydroxytryptamine and dopamine synthesis in the brain: a theoretical approach. J Inherit Metab Dis. 1990;13: 37–57.

93. Pascucci T, Giacovazzo G, Andolina D, Conversi D, Cruciani F, Cabib S, et al. In vivo catecholaminergic metabolism in the medial prefrontal cortex of ENU2 mice: an investigation of the cortical dopamine deficit in phenylketonuria. J Inherit Metab Dis. 2012;35: 1001–1009. doi:10.1007/s10545-012-9473-2

94. Lou H. Large doses of tryptophan and tyrosine as potential therapeutic alternative to dietary phenylalanine restriction in phenylketonuria. Lancet (London, England). 1985;2: 150–1. 95. Lykkelund C, Nielsen JB, Lou HC, Rasmussen V, Gerdes AM, Christensen E, et al. Increased

neurotransmitter biosynthesis in phenylketonuria induced by phenylalanine restriction or by supplementation of unrestricted diet with large amounts of tyrosine. Eur J Pediatr. 1988;148: 238–245. doi:10.1007/BF00441411

96. Boot E, Hollak CEM, Huijbregts SCJ, Jahja R, van Vliet D, Nederveen AJ, et al. Cerebral dopamine deficiency, plasma monoamine alterations and neurocognitive deficits in adults with phenylketonuria. Psychol Med. 2017;47: 2854–2865. doi:10.1017/S0033291717001398 97. van Spronsen FJ, van Wegberg AM, Ahring K, Bélanger-Quintana A, Blau N, Bosch AM, et al.

Key European guidelines for the diagnosis and management of patients with phenylketonuria. Lancet Diabetes Endocrinol. 2017;5: 743–756. doi:10.1016/S2213-8587(16)30320-5

98. Ahring K, Bélanger-Quintana A, Dokoupil K, Gokmen-Ozel H, Lammardo AM, MacDonald A, et al. Blood phenylalanine control in phenylketonuria: a survey of 10 European centres. Eur J Clin Nutr. 2011;65: 275–278. doi:10.1038/ejcn.2010.258

99. Van Vliet D, Van Der Goot E, Bruinenberg VM, Van Faassen M, De Blaauw P, Kema IP, et al. Large neutral amino acid supplementation as an alternative to the phenylalanine-restricted diet in adults with phenylketonuria: evidence from adult Pah-enu2 mice ☆. 2017; doi:10.1016/j.jnutbio.2017.09.020

100. Van Vliet D, Bruinenberg VM, Mazzola PN, Van Faassen MHJR, De Blaauw P, Kema IP, et al. Large neutral amino acid supplementation exerts its effect through three synergistic mechanisms: Proof of principle in phenylketonuria mice. PLoS One. 2015;10: 1–18. doi:10.1371/journal.pone.0143833