Metabolomics for Translational Research

Overview

As the closest omic to the phenotype, metabolomics has revealed important insights into the relationship between causative factors and disease. Unlike genes and proteins, biochemical pathways are conserved across species. Thus, metabolomics has been instrumental in characterizing disease biomarkers and pathways in cells or animal models that translate to humans. In this chapter, we will discuss key studies where metabolomics played an integral role in identifying an important biomarker signature or therapeutic target, characterizing a disease mechanism that was particularly relevant in human patients, or developing a novel disease model significantly closer to human pathophysiology than existing models. The case studies in this chapter cover these topics as follows:

• Disease models: Case Studies 1, 2, and 4
• Biomarkers: Case Studies 2, 3, and 4
• Mechanisms: Case studies 1, 2, 3, and 4

Disease Models, Biomarkers and Mechanisms

Case Study 1: Developing Animal Models of Dyslipidemia that Accurately Reflect Human Biology

Introduction. Cardiovascular disease (CVD) is a leading cause of death worldwide, and dyslipidemia is a major risk factor for this condition. Although statins can be highly effective at lowering LDL cholesterol and reducing CVD risk, many patients experience adverse events, which highlight the need for new therapeutic targets. However, efforts to develop novel treatments have yieldedlowclinicalsuccess, whichareoftenattributedtoinadequatepreclinical animal models that do not accurately reflect human lipid metabolism.

Preliminary Data and Study Goals. To address the limitations of today’s animal models of lipid metabolism, study investigators used metabolomics to analyze and compare plasma lipid profiles across multiple animal species and humans with the goal of determining which models most closely resemble human dyslipidemia, both at baseline and in response to statin treatment [12].

Methods. Multiple mouse strains, other non-primate species, and non-human primates were analyzed alongside dyslipidemic humans. Animals were fed either standard laboratory chow or a diet high in fat or high in cholesterol. Most species received simvastatin treatment or vehicle control for two weeks to allow comparison between baseline and drug-responding lipid profiles. Plasma samples were analyzed for lipids and lipoproteins using metabolomics. Lipids were also measured using traditional clinical measurements including total cholesterol, LDL, HDL, and triglycerides. Distance-based methods were used to quantify how closely each animal model resembled human dyslipidemia across multiple lipid fractions.

Results. Substantial differences in baseline plasma lipid profiles across animal models were observed, with only a subset closely resembling dyslipidemic humans. Nonhuman primates showed the greatest similarity to humans regarding total cholesterol, LDL/HDL balance, and overall lipoprotein distribution (Figure 22). Many commonly used models including mice, rabbits, and rats differed significantly from humans. For example, many rodent models carried most cholesterol in HDL (an atheroprotective profile), unlike humans, whereas nonhuman primates and some modified models such as CETP-expressing mice exhibited more human-like lipid patterns. Fatty acid composition across major lipid classes was broadly conserved in some species but diverged in others. In humans, most fatty acids came from nonessential and omega-6 pathways, and this pattern was best replicated in nonhuman primates, dogs, and a few rodent models. When comparing overall lipid composition across eight major lipid fractions nonhuman primate models ranked closest to humans (Figure 23). In response to simvastatin treatment, nonhuman primates and dogs showed lipid changes most consistent with humans, while most rodent models showed little or even opposite responses.

Figure 22. Fatty acid (FA) composition in plasma cholesterol esters (CEs), triglycerides (TGs), and phosphatidylcholine (PC) across animal species. FAs are shown within three categories: nonessential FAs (light blue), omega-3 (maroon) pathway FAs, and omega-6 (light yellow) pathway FAs. Measured FAs were summed for each pathway. Left panel shows the absolute amounts (nmol) of FAs in each pathway and right panel shows the relative amounts (%) for plasma CEs, (A) TGs (B), and (C) PCs. Image reproduced from Yin et al., J Lipid Res, 2012, licensed under CC BY 4.0.

Figure 23. Dendrogram comparison of baseline plasma lipid similarity based on eight major circulating lipid fractions (CE, TG, DAG, PE, FFA, LPC, PC, and FC) across species. The difference of any given lipid fraction between the means of each animal model and dyslipidemic human was calculated and weighted according to the proportion the same lipid fraction over total lipid in humans. The overall weighted distance of each animal model from dyslipidemic humans was calculated, and the models were sorted by distance. Image reproduced from Yin et al., J Lipid Res, Licensed under CC BY 4.0.

Study Conclusions

• This study shows that baseline and treatment-responded lipid profiles are crucial for evaluating how well animal models reflect human disease.
• Nonhuman primates are the most representative models of human dyslipidemia, and dogs are similar in regards to drug response. By contrast, many widely used animal models including ApoE-/- and LDLr-/- mice were significantly different from humans.
• Not only did metabolomics enable a comprehensive comparison of lipid profiles between several animal models and humans, but it also showed that integrating comprehensive lipid profiling (lipidomics) with functional drug response data provides a more accurate framework for evaluating disease similarity and improving target validation.

Case Study 2: Characterizing Mechanisms of Niemann-Pick Disease that Mirror the Phenotype of Human Patients

Introduction. Niemann-Pick type C1 (NPC1) disease is a rare, progressive, neurodegenerative disorder caused primarily by mutations in the NPC1 gene that lead to impaired intracellular cholesterol trafficking and accumulation of lipids in lysosomes. The disease typically presents in childhood with ataxia and cognitive decline, and without timely intervention, patients usually succumb to it in adolescence. A major challenge in managing NPC1 disease is the lack of a simple, non-invasive diagnostic test, which can delay diagnosis often by several years.

Preliminary Data and Study Goals. NPC1 disease is strongly associated with oxidative stress, as demonstrated by increased reactive oxygen species (ROS) and lipid peroxidation in cellular and animal models, and as reduced antioxidant capacity in patients. The oxidative environment, combined with excess cellular cholesterol, promotes the formation of cholesterol oxidation products (oxysterols). Based on these previously reported findings, the investigators of this study theorized that circulating cholesterol oxidation products could serve as sensitive and specific blood-based biomarkers for NPC1 disease [13]. The goal of this study was to determine whether oxysterols in plasma could distinguish NPC1 disease patients from healthy individuals and whether oxysterol levels correlate with disease severity.

Methods. Npc1-/- mice and a feline model of NPC1 disease were used to evaluate oxysterols as biomarkers for NPC1 disease. Plasma, liver and brain tissue, and cerebrospinal fluid (CSF) were collected from animals across their lifespan. In parallel, human plasma, CSF, and fibroblast samples were collected from NPC1 disease patients, healthy controls, heterozygous carriers of a NPC1 mutation, and individuals with other lysosomal storage diseases to assess biomarker specificity. Oxysterols were measured using targeted metabolomics. Differences in oxysterol levels and correlation with disease severity were determined using ANOVA and correlation analyses.

Results. Multiple oxysterols were significantly elevated in plasma and tissues of Npc1-/- mice, with a moderate increase seen before the onset of neurological symptoms and increasing further as disease progressed. In human subjects, the non-enzymatic oxysterols 3β,5α,6β-triol and 7-ketocholesterol (7-KC) were significantly increased in NPC1 patients compared to controls and heterozygous carriers, which detected cases with high sensitivity and specificity. These biomarkers clearly separated NPC1 patients from healthy individuals and those with other lysosomal storage disorders (Figure 24). Higher concentrations of 3β,5α,6β-triol and 7-KC were associated with earlier disease onset and greater disease severity, and an index combining these markers improved predictive power. In the NPC1 feline model, treatment with cyclodextrin significantly reduced oxysterol levels, showing that these biomarkers can also track therapeutic response (Figure 25).

Figure 24. Comparison of plasma oxysterol concentrations in NPC1 disease and other lysosomal storage diseases. (A) 3β,5α,6β-triol and (B) 7-KC concentrations in fasting plasma samples from control, NPC1, infantile neuronal ceroid lipofuscinosis (INCL), GM-1 gangliosidosis (GM-1), GM-2 gangliosidosis (GM-2) and Gaucher disease (GD) subjects. Image reproduced from Proter et al., Sci Transl Med, 2010, licensed under CC BY 4.0.

Figure 25. Circulating oxysterol biomarkers are decreased in response to cyclodextrin therapy. (A) Serum 7-KC and (B) 3β,5α,6β-triol concentrations were measured in untreated WT (4 16 weeks) and NPC1 (16 weeks) cats, and in NPC1 cats (16–18 weeks) treated with a single subcutaneous injection of 4000 or 8000 mg/kg at 3 weeks (n=2–4/group). Image reproduced from Proter et al., Sci Transl Med, 2010, licensed under CC BY 4.0

Study Conclusions

• This study shows that oxysterols are highly sensitive and specific biomarkers for NPC1 disease, which addresses a significant unmet need in diagnosis and monitoring of this disease.
• Unlike other diseases where oxysterol increases are modest, NPC1 shows large, disease-specific elevations that enable clear discrimination from controls and other disorders.
• Since oxysterols correlate with disease severity and age of onset, they may also be useful for tracking disease progression. The observed reduction in oxysterol levels following treatment in animal models suggests they could serve as surrogate endpoints in clinical trials to help evaluate therapeutic efficacy.

Case Study 3: Characterizing Gut-Brain Interactions Unique to Persons with Autism Spectrum Disorder

Introduction. Currently, Autism Spectrum Disorder (ASD) is diagnosed based on behavioral criteria rather than biological markers, which contributes to delayed diagnosis and limits early intervention opportunities. The increasing prevalence of ASD and lack of FDA-approved treatments for core symptoms calls for objective molecular biomarkers to enable earlier detection and support personalized therapies. ASD is increasingly associated with metabolic abnormalities, including mitochondrial dysfunction and oxidative stress. However, findings from prior studies have been inconsistent due to small sample sizes, limited metabolite coverage, and variability in study design, which has hindered attempts to identify consistent metabolic signatures. Additional complexity arises from the diet and gut microbiome, which can significantly shape metabolite profiles and may interact with genetic risk factors to influence ASD biology.

Preliminary Data and Study Goals. To address these gaps in knowledge investigators sought to perform comprehensive metabolomics analyses of both plasma and fecal samples from a large cohort of children with ASD alongside a neurotypical control group [14]. The goal of this study was to identify robust metabolic signatures associated with ASD and characterize associations between metabolism and behavioral symptoms.

Methods. Participants were aged 3–12 years. ASD diagnoses were confirmed using standardized clinical tools and neurotypical controls were screened to ensure typical development. A subset of participants also underwent evaluation for gastrointestinal (GI) symptoms to stratify ASD individuals into groups with and without GI dysfunction for additional analyses. Plasma and feces were analyzed using global metabolomics and a complex lipid panel. Metabolic differences between study groups were determined using ANOVA-based comparisons. Random Forest machine learning was used to determine whether metabolic profiles could distinguish ASD patients from controls. Additionally, fecal microbiota from ASD and neurotypical donors were transplanted into germ-free mice to examine whether microbiome-associated metabolic signatures could be transferred and provide insight into gut-brain interactions.

Results. Hundreds of metabolites differed between groups and metabolite profiles could modestly distinguish ASD fromneurotypical individuals, with key discriminatory metabolites including lipids, steroid hormones, and microbially derived compounds. Altered levels of acyl-carnitines and energy-related metabolites suggested impaired cellular energy metabolism and disruptions in amino acid and glutathione pathways consistent with increased oxidative stress were observed (Figure 26). Many of these altered metabolic pathways correlated with behavioral severity scores, linking metabolic abnormalities to clinical features of ASD. Several microbially derived metabolites, including 4-ethylphenyl sulfate, were elevated in ASD and some were transferable to mice via fecal microbiota transplantation, suggesting a causal contribution of gut microbes to metabolic changes (Figure 27).

Figure 26. Plasma and Fecal Metabolomes Differ between ASD and neurotypical groups. Top 30 most distinguishing metabolites between each group in plasma (A) and feces (B) by random forest analysis, with mean decrease accuracy along the x-axis. Image reproduced from Needham et al., Biol Psychiatry, 2021, licensed under CC BY 4.0

Figure 27. Transfer of human fecal microbiota into mice. (A) Scaled intensity values indicating relative levels of 4-ethylphenyl sulfate (4EPS) levels in mice colonized with neurotypical or ASD donors, colored according to donor. (B) Scaled intensity values indicating relative levels of 4EPS levels in mice colonized with TD or ASD donors, colored according to donor. Image reproduced from Needham et al., Biol Psychiatry, 2021, licensed under CC BY 4.0.

Study Conclusions

• This study revealed diverse metabolic profiles that were linked to gastrointestinal symptoms and behavioral scores, suggesting that ASD cannot be explained by a single mechanism, but rather, it is governed by genetic risk, diet, and the gut microbiome.
• Many altered metabolites were derived from or impacted by gut microbes. Some of those metabolic signatures were shown to be transferable to mice through microbiota transplantation, suggesting that the gut and immune system can impact brain function through circulating metabolites.
• Overall, this study suggests that integrating metabolomics with clinical and environmental data could improve ASD diagnosis and potentially identify novel therapeutic targets.

Case Study 4: Elucidating a Novel Mechanism that Explains why Standard of Care Treatments for Lymphangioleiomyomatosis are Partial and Reversible

Introduction. Lymphangioleiomyomatosis (LAM) is a rare, progressive lung disease that primarily affects women and is characterized by abnormal proliferation of smooth muscle–like cells, leading to cystic lung destruction and respiratory failure. LAM is linked to mutations in the tuberous sclerosis complex genes (TSC1/TSC2), which result in hyperactivation of the mTORC1 signaling pathway, a key regulator of cell growth and metabolism. However, mTORC1 activation alone does not fully explain LAM pathogenesis because clinical benefits from mTOR inhibitors like rapamycin are incomplete and reversible. The striking female predominance of LAM points to a role for estradiol in disease progression.

Preliminary Data and Study Goals. Previous studies have shown that estradiol can enhance tumor growth and metastasis in TSC2-deficient models, suggesting a hormonal contribution to disease biology. Additionally, pathways that involve inflammatory mediators have also been suggested as important mediators of disease biology. Investigators hypothesized that prostaglandin biosynthesis, driven by the COX-2 enzyme, may be a key downstream pathway linking estrogen signaling and TSC2 loss. The goal of this study was to determine how estradiol, TSC2 deficiency, and mTOR signaling interact to regulate COX-2 and prostaglandin production, and whether this pathway could serve as a therapeutic target in LAM [15].

Methods. Mechanistic studies were conducted using TSC2-deficient cell lines and alongside TSC2 “add-back” controls. Cells were treated with estradiol and various pathway inhibitors. Outcomes including COX-2 expression, signaling pathway activation, and prostaglandin production were measured using immunoblotting, ELISA, and RT-PCR. Gene knockdown (e.g., Rictor shRNA) was used to assess the role of mTORC2 signaling. Global metabolomics identified changes in prostaglandin and related lipid metabolites following estradiol treatment. TSC2-deficient and TSC2-restored cells were implanted into Tsc+/- mice to assess tumor growth, prostaglandin production, and treatment responses. Mice were treated with aspirin or celecoxib to evaluate the therapeutic impact of COX-2 inhibition. Lung tissue, serum, urine, and exhaled breath condensate from LAM patients were analyzed to validate findings and assess the relevance of COX-2 activity and prostaglandin signaling in human disease.

Results. Estradiol significantly enhanced prostaglandin production in TSC2-deficient cells, revealing a distinct metabolic signature characterized by increased levels of prostaglandin and related metabolites. In TSC2-deficient A GUIDE TO ME TA BO LO MIC S cells MAPK and PI3K–Akt signaling pathways were activated by estradiol, which elevated COX-2 expression to suggest a mechanistic link between estrogen signaling and inflammatory lipid metabolism in LAM. Increased COX-2 expression and prostaglandin biosynthesis occurred independently of mTORC1 signaling. Interestingly, rapamycin inhibited mTORC1 activity but did not reduce COX-2 levels or prostaglandin production while inhibiting mTORC2 significantly reduced COX-2 expression and downstream signaling. In mouse models, pharmacologic inhibition of COX-2 with aspirin or celecoxib reduced prostaglandin levels, suppressed tumor growth, and increased apoptosis in preclinical models (Figure 28). In human LAM samples, COX-2 expression and circulating prostaglandins were elevated, and aspirin treatment increased levels of anti-inflammatory lipid mediators, which reduced cell proliferation.

Figure 28. Inhibition of COX-2 suppresses renal tumorigenesis and inhibits the progression of xenograft tumor of TSC2-deficient cells in preclinical models. Tsc2+/2 mice were treated with either vehicle or Celecoxib (0.1% in mouse chow) for one month and then sacrificed for analysis at the end of treatment. Renal cystadenoma histology and microscopic kidney tumor scores were assessed. (A) Microscopic kidney tumor scores are plotted on a linear scale (P = 0.0002). Data are analyzed from 16 vehicle and 11 Celecoxib treatment groups. (B) Two cystadenomas are shown. Results are representatives of 11 or 16 mice per group. Image reproduced from Li et al., J Exp Med, 2014 licensed under CC BY 4.0.

Study Conclusions

• This study identified a novel link between estrogen signaling, TSC2 loss, and prostaglandin metabolism in LAM.
• Therapies that inhibit COX-2 suppressed tumor growth and increased apoptosis in a mouse model of LAM. Reduced cell proliferation was also observed in response to COX-2 suppressing treatments in human cells collected from LAM patients.
• Although LAM was previously thought to be primarily driven by activation of mTORC1, these findings showed that COX-2 mediated prostaglandin production is an mTORC1-idependent mechanism that contributes to disease progression, explaining why standard of care treatments such as rapamycin provide only partial clinical benefits.

Chapter Takeaways

• Owing to its position as the omic closest to the phenotype, metabolomics is particularly useful for characterizing biological mechanisms and identifying biomarkers and therapeutic targets in preclinical models that translate remarkably well to humans.
• Applying metabolomics to translational studies may improve the success rate of clinical trials, particularly for conditions for which many therapies have been proposed but eventually failed in clinical studies due to the lack of robust translational models.