Dissertation committee:
Εμμανουήλ Μικρός, Καθηγητής, Τμήμα Θετικών Επιστημών, ΕΚΠΑ
Ευάγγελος Γκίκας, Καθηγητής, Τμήμα Θετικών Επιστημών, ΕΚΠΑ
Ιωάννης Ντότσικας,Επίκουρος Καθηγητής, Τμήμα Θετικών Επιστημών, ΕΚΠΑ
Ιωάννα Τζουλάκη, Επίκουρη Καθηγήτρια, Πανεπιστήμιο Ιωαννίνων/ Καθηγήτρια, Imperial College of London
Julian Griffin, Professor, Imperial College of London
Ευάγγελος Τέρπος, Καθηγητής, Τμήμα Επιστημών Υγείας, ΕΚΠΑ
Νικόλαος Θωμαΐδης
Καθηγητής, ΕΚΠΑ (ΜΕΛΟΣ)
Summary:
In recent years Genome-Wide association studies (GWAS) have identified a plethora of genetic variants associated with complex diseases and traits. Although the molecular mechanisms implicating these variants in the associated disease may be known for a small number of them, for the vast majority they remain largely unexplored. Moreover, effect sizes of genetic associations with diseases are generally small requiring huge cohorts. For these reasons, GWAS findings have led to poor improvements in the clinical management of the associated disease. A shift has been observed from GWAS to intermediate trait associations, such as RNA transcription or metabolites that are known risk factors of disease (such as triglyceride or glucose levels), which are generally characterized by larger effect sizes as there are closer to the clinical endpoints compared to the genotype. In the last decade, metabolomics has emerged as a powerful tool for the comprehensive measurement of low molecular weight molecules in biological fluids, allowing for simultaneous detection and quantitation of a variety of small molecules with different chemical properties and structures. Another important aspect of metabolomics is that the metabolome reflects the effect of both intrinsic and extrinsic exposures. Therefore, both genetic and environmental impacts on the clinical phenotype can be explored. Recent studies have explored GWAS associations in a metabolomics setup known as Genome-wide association studies of metabolites concentrations (mGWAS). However, all of these studies have explored thousands of genetic variants with metabolic trait phenotypes in a general context and irrespectively of the associated diseases.
In this study the aim was to perform a metabolome-wide association study (MWAS) with genetic variants previously associated with a specific complex disease, such as cancer types and Cardiovascular Diseases (CVDs), in order to thoroughly investigate the gene-metabolite interactions and reveal underlying causal pathways of the disease in a GWAS agnostic context. The link between genetic variants previously associated with three complex diseases, Colorectal Cancer (CRC), Multiple Myeloma (MM) and Blood Pressure (BP), with metabolites as intermediated traits has been investigated using a Nuclear Magnetic Resonance (NMR)-based metabolomics approach in a large UK population cohort. More specifically, GWAS data, as well as NMR metabolomics data for urine and blood from 1,974 Airwave Health Monitoring Study participants were analysed. Initial method development has been performed, highlighting the urge of simultaneous implementation of both targeted and untargeted metabolomics approaches. We also tried to investigate the use of multivariate approaches, which has not yet been explored and established in this kind of studies. Utilization of multivariate approaches is of high importance in order to account for the dependency structures in omics data. Another asset of this study is the simultaneous analysis of both plasma and urine samples, with the urinary metabolome being largely unexplored in a GWAS context. We performed urine annotation in the NMR spectra of the Airwave Health Monitoring Study (AIRWAVE), using 2D NMR experiments, statistical approaches (e.g. Statistical total correlation spectroscopy-STOCSY), as well as spiking experiments. Based on our method development, for each genetic variant, Spearman partial correlation analysis was conducted against the 1H NMR spectral data. All univariate analyses were adjusted for confounders and corrected at 5% False Discovery Rate (untargeted approach) or 5% Bonferroni (targeted approach). We also made use of metabolic ratios, which have previously been found to increase the association of a product-substrate pair. Multivariate approaches, such as Knowledge Discovery by Accuracy Maximization (KODAMA) and sparse Partial Least Squares Regression (sPLS) were also employed in order to reveal more information on the relationship between genetic variants and the metabolic traits in a multivariate framework. For CRC and BP, where the environmental factors are known to pose an important role in their pathogenesis, we also investigated the interactions with dietary factors, while gene-metabolite-diet networks have also been constructed using Cytoscape and Data Integration Analysis for Biomarker discovery using Latent variable approaches for ‘Omics studies (DIABLO) respectively. In serum, the Metabolon Mass Spectrometry (MS) platform has also been utilized for targeted MS analysis complementary to the NMR approach. In CRC analysis, Mendelian randomization (MR) analysis has been performed to establish causality of the identified metabolites in increasing the risk of CRC.
Our results demonstrated a correlation between the genotype and metabolome of the examined diseases. In CRC analysis, several metabolites including sugars (sucrose, glucose), amino acids (tyrosine, leucine, lysine, glutamine), gut microbial metabolites (p-cresol sulfate, Trimethylamine-N-oxide (TMAO)), as well as lipids and lipoprotein particles have shown strong association with CRC genetic variants. Among the associated variants were rs10411210 in Rhophillin binding protein 2 (RHPN2) gene, rs78368589 in Solute Carrier Family 6 Member 18 (SLC6A18) gene and rs745213 in the Mitogen-Activated Protein Kinase Kinase 5 (MAP2K5) gene in the analysis of urinary metabolites and rs174533 in the Myelin Regulatory Factor (MYRF) gene for blood analysis. One of our major findings was the association for rs10411210 with sucrose levels. Our results suggest a relationship between the RHPN2 and sucrase-isomaltase (SI) expression through activation of Rhophillin A, which further results in decreased urinary levels of sucrose. Although the role of SI in CRC has been well-established, to our knowledge there is no study associating the urinary levels of sucrose with the risk of developing CRC. Moreover, we proposed novel molecular mechanisms implicating the gut microbial metabolites, TMAO and p-cresol sulfate, in the pathogenesis of CRC. Finally, among the lifestyle factors alcohol consumption exhibited strong associations with most of the implicated metabolites.
In MM analysis, several metabolites exhibited associations with MM genetic variants. Among the most important findings was the association of rs139371 in the Chromobox Protein Homolog 7 (CBX7) gene, which has been previously linked to both MM and thyroid cancer, with urinary tyrosine levels, providing further evidence about the effect of thyroid hormones in cancer. Moreover, association of the N,N dimethylglycine/glycine ratio with rs58618031 variant in Protection of Telomeres 1 (POT1) gene highlighted the role of homocysteine metabolism in telomere shortening. This result further demonstrated the advantageous investigation of metabolic ratios. In a subsequent study, we performed an NMR-based metabolomics in MM patients in order to seek for metabolic markers that could discriminate between MM patients that respond compared to those that not respond to treatment with monoclonal antibodies. Peripheral blood (PB) and bone marrow (BM) plasma from MM patients was collected at baseline (BL), e.g. at the time of diagnosis, and after treatment. This study demonstrated the role of glutamic acid and glutamine pathway in both PB and BM plasma of MM patients, while the alterations in BM were much stronger than those reported in the PΒ. The role of several amino acids, including tyrosine and branched chain amino acids also found in our MWAS analysis, has also been highlighted. Finally, lactic acid has emerged as a key metabolite for prognosis of MM patients, while other organic acids (acetic, pyruvic and citric acid) also posed a significant effect in their discrimination.
In BP analysis over 1000 BP genetic variants were examined for associations with urinary metabolic markers. Among the most profound association was that of the rs1047891 variant in the Carbamoyl-Phosphate Synthase 1 (CPS1) gene, encoding for an enzyme responsible for the rate-limiting step of the urea cycle, with glycine, creatine and guanidoacetic acid both in blood and urine. Although the association between CPS1 and blood levels of glycine was already known, other metabolites of the proposed pathway have been identified, while the association was also depicted in the urinary levels of these metabolites. Several other variants exhibited associations with urinary metabolites, mostly gut microbiota related metabolites, such as hippuric acid, TMAO and 3-(3-hydroxyphenyl)-3-hydroxypropionic acid, and coffee related metabolites, such as sumiki’s acid, trigonelline and N-methylpyridinium. Finally, our multivariate analysis using DIABLO that integrated GWAS, metabolomics and dietary data for the prediction of BP outcomes (systolic and diastolic BP), highlighted the role of gender and alcohol consumption in hypertension.
Overall, our analyses helped us highlight the metabolites and variants that could pose a key role in pathology of the examined diseases providing new insights on cancer and BP pathogenesis and paving the way to novel therapeutic or preventive strategies for these diseases. However, follow-up in vivo/ in vitro studies will be needed in order to provide evidence on the proposed molecular pathways, as well as replication to independent cohorts.
