Type 2 diabetes is known to involve many different underlying mechanisms, but the considerable heterogeneity in the phenotype is mostly ignored in how it is treated.

Now, researchers at University of Dundee, U.K., have developed a method for visualizing this heterogeneity and shown how the risks of complications, such as chronic kidney disease or peripheral neuropathy, differ by phenotypes.

So for example, patients at risk of retinopathy are shown to be phenotypically different at the point of initial diagnosis of type 2 diabetes, from those at risk of major cardiovascular events (MACE).

The analysis highlights how underlying phenotypic variation drives type 2 diabetes onset and affects subsequent diabetes outcomes and drug response.

That demonstrates the need to incorporate these factors into a more personalized approach to managing type 2 diabetes, said Ewan Pearson, professor of diabetic medicine at the University of Dundee, who led the research, described in Nature Medicine, May 9, 2022.

"Clinically, we need to move away from a one size fits all approach to the management of people with type 2 diabetes, and be more precise in the care of patients," Pearson said.

There have been previous attempts to group patients into discrete phenotypic clusters that differ in terms of underlying genetic risk, and to provide insights into how phenotype and genetic etiology of type 2 diabetes can vary.

But to date, such attempts do not align with the molecular understanding of type 2 diabetes as a continuum of disease, rather than a number of discrete subtypes.

In addition, it has been shown that allocating patients to subgroups reduces the power to predict the risk of complications or response to therapy, compared to predictions based on continuous data.

In recognition of the continuum of phenotypic data and genetic heterogeneity seen in type 2 diabetes, the researchers applied reverse graph embedding, a method for visualizing heterogeneity in very large datasets.

The large dataset in question was routinely collected information from 23,137 people in Scotland who had been recently diagnosed with type 2 diabetes.

The aim was to reduce the complex phenotypic characteristics of type 2 diabetes into a nonlinear, two-dimensional tree structure, in order to visualize how diabetes outcomes and drug response vary across the spectrum.

Different reasons, different risks

The researchers factored in nine type 2 diabetes-related phenotypes: HbA1c; body mass index (BMI); total cholesterol; high-density lipoprotein cholesterol (HDL-C); triglycerides; alanine aminotransferase (ALT); creatinine; and systolic (SBP) and diastolic blood pressure (DBP).

HDL-C, SBP and DBP were most strongly distributed across the tree, followed by total cholesterol and triglycerides, and then HbA1c.

There was minimal variation in creatinine or ALT. Individuals with elevated HDL-C levels clustered in one part of the tree, those in another cluster had higher levels of blood pressure and cholesterol and moderate levels of hyperglycemia. Another cluster contained patients who were obese, hyperglycemic, with high triglyceride levels and low HDL-C.

The visual representation of the phenotypic characteristics of the 23,137 patients in the Scottish cohort acts as a reference onto which individuals newly diagnosed with type 2 diabetes can be mapped, using age at diagnosis, sex and the nine clinical phenotypes.

"Type 2 diabetes is a complex disease caused by many different mechanisms," said Anand Nair, the lead analyst on the study. "Some people develop type 2 diabetes due to different mechanisms than others, and can therefore differ dramatically in their clinical characteristics, such as body weight, blood fat, blood pressure or [genetics]. This new approach helps to greatly simplify this complexity for both clinicians and patients."

To validate the model, the researchers used primary care data from UK Biobank relating to 7,332 people newly diagnosed with type 2 diabetes, and data from GlaxoSmithKline's ADOPT (A Diabetes Outcome Progression Trial), a randomized, controlled study in which 4,150 patients were treated with either metformin, rosiglitazone or glibenclamide monotherapy.

The phenotype distribution was the same for the Scottish reference tree and UK Biobank; however people recruited to AADOPT did not map onto clusters with particularly high HbA1c or BMI, because these patients were excluded from the trial.

Different outcomes

The researchers used their model to investigate how the phenotype variation at diagnosis translates to variations in diabetes outcomes of: time to insulin requirement; time to severe diabetic retinopathy; time to chronic kidney disease; estimated glomerular filtration rate and time to MACE.

The individual probabilities of progression to each of these outcomes were calculated over 5 years for each patient. That showed individuals within the tree are strongly correlated with their neighbors for each outcome.

The probabilities of insulin initiation, MACE and CKD have a very similar pattern across the tree, with the highest risk observed in the most obese and dyslipidemic individuals.

In contrast, the risk of retinopathy was largely driven by the combination of increased blood pressure, hyperglycemia and dyslipidemia.

To investigate how drug response varied with phenotypic variation, the researchers turned to the ADOPT trial. There was a difference in failure for the three drugs in the trial, with metformin and glibenclamide failure being faster in obese, hyperglycemic patients, while failure on rosiglitazone was associated with high lipodystrophy and low obesity.

The researchers next explored heterogeneity in the genetic etiology of type 2 diabetes, using available genetic data from 3,512 of the Scottish cohort and 7,145 from UK Biobank. That showed one cluster had higher genetic obesity, another elevated genetic beta cell dysfunction and increased diabetes risk mediated via liver and lipid-mediated resistance. A third cluster was characterized by genetic lipodystrophy with lower genetic obesity.

"Our study demonstrates how we can look at an individual with type 2 diabetes and illustrate in an intuitive way the main reasons they have diabetes - and use this to manage them better to reduce their individual risks," Pearson said.

Of three women diagnosed with type 2 diabetes at the age of 60, one may only be slightly overweight and have developed diabetes due to reduced insulin production from the pancreas. There will be slow progression of diabetes and lower risk of complications.

A second woman may have particularly high blood pressure and be more prone to eye complications, while a third may be very overweight with high blood fats and be more resistant to the effects of insulin, meaning there is an increased risk of heart disease.

"They all have type 2 diabetes but for very different reasons and with very different profiles, meaning that different treatments may result in better outcomes, depending on their circumstances," said Pearson.

As a potential aid to help patients and clinicians visualize the risks of disease progression and complications, the researchers have developed an app. Using this, newly diagnosed patients can be placed in the disease continuum, with their risk of microvascular and macrovascular complications predicted for 5 years.

However, the researchers say they are not advocating use of the app to improve prediction, but rather to reduce a complex, multi-dimensional disease into a simpler, understandable, two-dimensional visual model. This can readily be used to explore how lifestyle or drugs can improve modifiable factors, such as blood pressure, lipids, BMI and HbA1c.