Tese de Doutoramento Mecanismos de Doença e Medicina Regenerativa 2022 Faculdade de Ciências Médicas, Universidade NOVA de Lisboa

The alarming prevalence of non-communicable diseases (NCDs), and particularly diabetes prevalence, continues to increase. International Diabetes Federation estimated an increase of 50% of diabetes prevalence by 2040. Since the discovery of insulin and the availability of various classes of oral antidiabetic drugs, the impact of diabetes is mainly due to the onset of its chronic complications. The latter remains one of the populations' major public health problems despite all the investment made each year, contributing to the threat of healthcare sustainability. In the last 15 years, the increase in health care costs attributable to diabetes reached 315%. Importantly, diabetes has common features with other NCDs and they also share risk factors (dyslipidemia, hypertension, obesity). Diabetes is defined as an alteration in the metabolism or action of insulin that progresses with increased blood glucose values. Thus, the diagnosis is based on operationally defined blood glucose cut-offs. Nevertheless, there are people with diabetes who never develop complications while people without diabetes may already have them. One of the factors that has been pointed out to explain the above-mentioned worrying scenario is diabetes heterogeneity. It is increasingly recognized that there are several subtypes of this condition in addition to the classically recognized ones. Still, there may be other factors contributing to this situation: 1) insulin has several actions in our body intervening in the metabolism of other substrates, thus impacting on diabetes complications risk factors other than glycemia; 2) glucose metabolism occurs in an integrated manner with other mechanisms as well as the metabolism of other substrates, such as lipids. Therefore, one might ask, is diabetes just about hyperglycemia? Resolving the heterogeneity of diabetes and identifying subgroups for which health care interventions can be defined in a targeted and specific manner is an enormous challenge. In recent years the concept of precision medicine has raised the opportunity to give the right medicine to the right person at the right time. The recent development of memory computing power and mathematical algorithms that can solve problems involving the analysis of large amounts of data is helping to make precision medicine a reality. A data-driven approach has the advantage of allowing us to look at these problems without a pre-assumption and thus we can aim to better understand these conditions. In this work we apply System Medicine methodology to analyze cohorts based on detailed phenotyping, to stratify the population with respect to NCDs, more specifically diabetes and their risk factors, to be useful to precision medicine approaches. In this way, a model of integrated diagnosis and approach can be built that has application in current clinical practice and supports reversing the impact that currently NCD’s have on the affected persons, their families and ultimately in society. We first hypothesize that the pathophysiological mechanisms of metabolic diseases quantitatively determine clinical parameters related to glucose, insulin, and free fatty acid metabolism, explaining distinct dysmetabolic profiles. We aimed at stratifying blood glucose within the metabolic multidimensionality, using cluster analysis, and extract pathophysiological insights from the metabolic profile. We performed cluster analysis to stratify 974 subjects (PREVADIAB2 cohort) with normoglycemia, pre-diabetes, or untreated diabetes. The algorithm was informed by age, anthropometry, and metabolic means (glucose, insulin, C-peptide, and free fatty acid (FFA) during oral glucose tolerance test (OGTT)). For cluster profiling, we additionally used indices of metabolic mechanisms. We found prominent heterogeneity within two optimal clusters, mainly representing normometabolism (Cluster-I) or insulin resistance and non-alcoholic fatty liver disease (NAFLD) (Cluster-II), with higher granularity. This was illustrated by subclusters showing a similar prevalence of NAFLD but differentiated by glycemia, FFA, and glomerular filtration rate (GFR) (Cluster-II). Subclusters with similar glycemia and FFA showed dissimilar insulin clearance and secretion (Cluster-I). Importantly, this work revealed that type 2 diabetes (T2D) heterogeneity can be captured by a thorough metabolic milieu and mechanisms profiling the metabolic footprint. Considering that insulin also affects lipid metabolism, and that glucose and lipid metabolism are highly interconnected, we hypothesized that subjects stratified by three key pathophysiological mechanisms of diabetes (insulin resistance, insulin secretion, and insulin clearance) have distinct glucose and lipid profiles. We aimed to assess the impact of mechanisms associated with dysmetabolic conditions on both glucose and lipid metabolism. To this end, we performed a cluster analysis with the PREVADIAB cohort, informed by surrogate indexes of the abovementioned key mechanisms and profiled by other organ-specific indices of insulin resistance, as well as several key lipid species. We found that subjects homogenous in their mechanism have distinct glucose but also lipid profiles. Together these profiles might better explain diabetes's complications. Interestingly, we additionally found that men and women had similar clusters patterns; nonetheless the cluster centroids were different. This might represent a distinct risk for diabetes and diabetes complications of both genders. Nonetheless, it might additionally suggest that one might have to approach women and men separately, considering different cut-offs for the disease. These hypotheses should be evaluated in future work. Finally, we aimed to predict genetic mutations by knowing few parameters of the metabolic milieu using Machine Learning (ML) models. We hypothesize that ML models can outperform the Dutch Lipid Score, the gold standard for screening familial hypercholesterolemia (FH) cases, to detect genetic mutations of familial hypercholesterolemia. We used 3 ML algorithms informed by age, low density lipoprotein cholesterol (LDL-c), high density lipoprotein cholesterol (HDL-c) and triglycerides (TG) and compared them with the results obtained using the Dutch Lipid Score. Our results showed that by considering age and some elements of the milieu it is possible to predict a gene mutation, suggesting that we can predict an etiological factor (gene mutation) by knowing the metabolic milieu. Altogether this work suggests that diabetes is an heterogenous condition. Nonetheless it should be approached in a broader context, considering other pathological factors that are intimately connected with glycemic levels and are also impacted by insulin absolute or relative deficiency. We propose the Integrative Model for approaching diabetes. This model allows not only to address the heterogeneity of the condition but furthermore it contextualizes diabetes, or hyperglycemia resulting from altered insulin metabolism, in a broader perspective of dysmetabolism. It considers, along with glycemia, other factors that may also be affected and contribute to the onset of diabetes and diabetes complications. Inspired by the Palette Model, we propose that three multidimensional planes (etiology, milieu, and mechanisms) should be considered through time. We propose that we can predict where a person is in each of the planes by knowing the others, and that they ultimately place a subject on a path to dysmetabolism, and will allow to predict diabetes complications, and identify the most appropriate interventions for each person. In conclusion the Integrative Model will allow a precision medicine approach to diabetes and NCDs.