Genetics of Diabetes Subtypes. Characterization of novel cluster-based diabetes subtypes

Abstract: BACKGROUND: Type 2 diabetes (T2D) has been reproducibly clustered into five subtypes based on six-clinical variables; age at diabetes onset, body mass index (BMI), Glutamic acid decarboxylase autoantibodies (GADA), glycated hemoglobin (HbA1c) and insulin secretion and resistance estimated as HOMA2B and HOMA2IR derived from fasting glucose and Cpeptide. These subtypes have different disease progression and risk of complications. The newly defined subtypes are called Severe Autoimmune Diabetes (SAID), Severe Insulin Deficient Diabetes (SIDD), Severe InsulinResistant Diabetes (SIRD), Mild Obesity-related Diabetes (MOD), and Mild Age-Related Diabetes (MARD). AIM: The main aim of the thesis was to characterize the subtypes using genetics and biomarkers to investigate potential etiological differences, identify subtype-specific genetic associations and determine the underlying mechanisms of kidney complications in the subtypes.METHODS: The project included individuals with diabetes (cases) from the Swedish cohort All New Diabetics In Scania (ANDIS, n=10927) and the Finnish cohort Diabetes Registry Vasa (DIREVA, n=4754) as well as diabetes-free individuals (controls) from the Swedish Malmö Diet and Cancer cohort (MDC,n=2744) and the Finnish Botnia cohort (n=1683). Clusters defined in Ahlqvist et al, 2018, were used for all analyses. The number of individuals in the subtypes were as follows: SAID (n=452, n=327), SIDD (n=1193, n=394), SIRD (n=1130, n=453), MOD (n=1374, n=596) and MARD (n=2861, n=1178), in ANDIS and DIREVA respectively. In Paper I and III, genome-wide association studies (GWAS) and genetic risk score (GRS) analyses were performed to compare underlying genetic drivers in the Swedish cohorts and replicated in the Finnish cohorts. In Paper III, the primary phenotype was estimated glomerular filtration rate (eGRF) reflecting chronic kidney disease. In Paper II, epidemiological and genetic analysis was performed using clustering, Cox regression models and GRS to compare GADA negative individuals with diabetes of Iraqi (n=286) and Swedish origin (n=10641) with respect to new diabetes subclassification and complications. In Paper IV, the proteomic profiles of the subtypes were studied using 1161 biomarkers measured on Olink panels. Machine learning algorithms were applied to prioritize biomarkers, followed by Menedelian Randomization. RESULTS: In Paper I, the HLA rs9273368 variant was significantly associated with SAID (OR=2.89,P=6.5x10-40), the TCF7L2 rs7903146 variant was significantly associated with SIDD (OR=1.56, P=8.6x10-15), MOD (OR=1.40, P=3.1x10-10) and MARD (OR=1.42,P=6.1x10-16). The rs10824307 variant near the LRMDA gene was uniquely associated with MOD (OR=1.35, P=1.3×10-09). GRS for fasting insulin showed a unique association with SIRD (OR=1.855, P=5.91x10-09). GRSs for BMI were associated with SIDD, SIRD and MOD but not MARD (OR=1.046, P=0.099). Paper II concluded thar individuals with diabetes from Iraq present with a more insulin-deficient subtype than native Swedes. They have a higher risk of coronary events but a lower risk of CKD. In Paper III, in ANDIS, eGFR was strongly associated with the A allele of rs77924615 in the well-established PDILT-UMOD locus (beta=0.126, p=6.61x10-13) in all T2D; MARD and SIDD but not in MOD or SIRD (p>0.05). In the SIRD subtype, eGFR was associated with the C allele of rs3770382 in the CTNNA2 gene at near genomewide significance (beta=-0.219, p=5.5x10- 08), but was not associated in any of the other subtypes. In DIREVA, the PDILTUMOD locus replicated in T2D, MARD, and SIDD, and was also associated in SIRD (beta=0.24, p=0.001) but not in MOD (beta=0.076, p=0.109). The CTNNA2 locus did not replicate in DIREVA. Paper IV, the diabetes subtypes were shown to have different proteomic profiles and a list of prioritized biomarkers was generated for future follow-up. CONCLUSION: The newly defined subtypes are partially distinct with genetically different backgrounds and SIRD is suggested to have more beta-cell independent pathogenesis. There is some suggestive support for different genetic backgrounds of DKD in diabetes subtypes. Biomarkers could be valuable for better discrimination of subtypes and cross cohort comparisons in larger datasets. The diabetes subclassification approach paves the way for individualized patient management and the development of new therapeutic targets.