Science Spotlight

New model for early detection of high risk type I diabetes

from the Zhao lab
Heatmap representing individuals clustering based on their genotypes.
Clustered group of patients based on their genotypes. Color bar across columns (subjects) indicates diabetes cases (green) and controls (red). Similarity values of every subject with genotype profiles, ranging from 0 to 1, are color-coded by the color map. Row labels on the right side include estimated Odds Ratio (OR) for genotype profiles. Three numbered labels, on the top of hierarchical tree, indicate three clusters of subjects: Cluster 1 includes a group of subjects, mostly patients, have exceptionally high similarity to genotype profiles 1, 4 and 14 indicated on the right; subjects in cluster 2 tend to have relatively high similarities to genotype profiles 1-15 and 17, and subjects in cluster 3 tend to be normal subjects with high similarity to genotype profiles 21–24. Figure from the publication.

Diabetes type I is an autoimmune disorder resulting in the destruction of the pancreatic cells by the patient’s immune system. Pancreatic islet beta cells produce insulin, which once released in the blood facilitate glucose (a type of sugar) absorption by the cells. When the pancreatic cells are destroyed, glucose builds up in the blood causing significant health damage over time. The only available treatment is insulin injection. Improving early diagnosis and treatment will be key to decrease symptoms in high-risk individuals.

To this end a collaborative effort from Fred Hutch led by Dr. Lue Ping Zhao (Public Health Sciences Division) in collaboration with Drs. Hamid Bolouri (Human Biology Division), Chul-Woo Pyo and Daniel Geraghty (Clinical Research Division), developed a prediction tool based on sequencing of HLA genes to improve the diagnosis of type I diabetes in children. The results of this study were recently published in the journal Diabetes/Metabolism Research and Reviews.

As explained by Dr. Zhao, “The HLA genes are the blueprints for the protein complexes that present foreign antigens (virus, bacteria etc) to the immune system. By mistake in autoimmune diabetes, the HLA protein complex breaks tolerance by presenting self-antigens such as insulin or GAD65, resulting in wrongful destruction of healthy beta cells.  The HLA protein complexes are an integral part to trigger the disease pathogenesis.” This can be explained by specific nucleotide variations in the sequence of HLA genes, called single nucleotide polymorphisms (SNP), that can impact the reaction to a specific antigen. Characterizing HLA SNP associated with insulin sensitivity is critical to identify the individuals at high risk to develop the disease.

In this study, the authors combined high-resolution sequencing of an extended panel of class II HLA (HLA-DR, -DP, -DQ) genes and used object-oriented regression to establish the correlation between complex genotypes and disease phenotypes. 962 patients from the nation-wide Swedish Better Diabetes Diagnosis (BDD) study were included in the study along with 448 healthy normal individuals from the same geographic area as controls. The sequencing of HLA genes led to the identification of many SNPs in the HLA-DRB1, -DRB345, -DQA1, -DQB1, -DPA1 and -DPB1 genes whose frequencies were higher in the control populations, thus predicted as protective alleles, while other SNPs at a high frequency in the patients were potential risk alleles.

A heatmap regrouping individuals by similarity in their genotype was then used to show the associated SNPs as well as corresponding risk or protective association (see the figure). Interestingly, several groups clustered together, specifically 705 subjects out of the 962 tested. Some of the patients clustered together with SNPs previously identified as risks factors. Conversely, some controls clustered together and were associated with SNP previously identified as protective. Such analyses greatly facilitated the interpretation of datasets as complex as SNP and correlation to genotypes and diseases risk.

This approach was further tested by the utilization of four known biological markers: autoantibodies recognizing insulin, GAD65, IA-2 (islet antigen 2) or any of three variants of the Zinc transporter 8 transporter. Among the tested autoantibodies only IA-2A antibodies were significantly and positively correlated with the disease risk, supporting previous data showing this antibody as a robust biomarker and reinforcing its use as a validation tool. “When IA-2A autoantibody appears, it is associated with an acceleration of the disease process to escalate the rate by which beta cells are destroyed”, said Dr. Zhao.

From this study, it is anticipated that the prediction model will allow the identification of 80% of the individuals with high-risk to develop type I diabetes.  The benefit from this model could be an earlier diagnosis, such as in newborns, from at risk families, to allow treatment at very early onset of the disease. When asked about future directions, Dr. Zhao said, “There are multiple aspects of this study that are exciting for us to move forward.  First, through careful explorations of predictive associations, it allows us to discover and define peptides that are involved in the disease onset.  Such information potentially allows us to define the environmental factors, eg. peptides from virus that are able to bind the HLA protein complexes. In this way, it may be possible to predict which environmental factor triggers the disease. If the peptide is coming from a virus, we may seek the development of appropriate vaccine to protect high-risk subjects from the disease. Second, the predictive associations may be readily translatable to disease prevention studies or to certain specific clinical practices. For example, this predictive model may be useful for recruiting high risk subjects into longitudinal follow-up studies, in which active monitoring of A1C (a blood test to monitor the average levels of blood glucose) and timely counseling may delay the disease onset or reduce severity of the disease onset. Additionally, such a predictive model may also be helpful for risk assessment in high-risk family setting, where some family members may desire to evaluate their own disease risk. Indeed, when effective preventative strategies are developed for type 1 diabetes, our predictive models may have much broad application in the future. Lastly, the possibility of accurately predicting type 1 diabetes onset based on HLA genetics is intriguing, as an example to the personalized prevention and treatment. Lessons from exploring this model will help us to appreciate challenges and to develop novel strategies of delivering predictive models for other human diseases”.


This study was funded by European Foundation for the Study of Diabetes (EFSD), the Swedish Child Diabetes Foundation (Barndiabetesfonden), the Swedish Research Council including a Linné grant to Lund University Diabetes Centre, the Skåne County Council for Research and Development, the Swedish Association of Local Authorities and Regions (SKL), the National Institute of Health/National Institute of Diabetes and Digestive and Kidney Disease and the institutional developmental fund from Fred Hutchinson Cancer Research Center.



Zhao LP,Carlsson A,Elding Larsson H,Forsander G,Ivarsson SA,Kockum I,Ludvigsson J,Marcus C,Persson M,Samuelsson U,Ortqvist E,Pyo C-W,Bolouri H,Zhao M,Nelson WC,Geraghty DE,Lernmark A. 2017. Building and Validating a Prediction model for Pediatric Type 1 Diabetes Risk Using Next Generation Targeted Sequencing of Class II HLA Genes. Diabetes/Metabolism Research and Reviews.