Background
Diabetes is a heterogeneous group of diseases resulting in hyperglycemia due to insulin secretory dysfunction as well as insulin resistance. A substantial proportion of type 1 diabetes (T1D) cases present in adulthood, and despite the presence of diabetes-associated autoantibodies, the majority of these patients do not initially require insulin [
1,
2]. The manifestation of this ‘latent autoimmune diabetes in adulthood’ (LADA) is clinically defined by (1) an adult age of onset, (2) at least one diabetes-associated autoantibody, and (3) the lack of requisite insulin treatment for at least 6 months after diagnosis. This definition overall represents approximately 5–10% of all cases of adult-onset diabetes, potentially the most frequent form of autoimmune diabetes [
3,
4].
However, classifying adult-onset autoimmune T1D, including LADA, remains challenging. The need for insulin treatment is a clinical decision, while diabetes-associated autoantibodies are neither pathogenic nor categorical features of LADA. Decisions are further confounded by false positives when large numbers of patients are screened [
5]. Since LADA has intermediate features between T1D and type 2 diabetes (T2D), there are limits to the current classification of diabetes. New paradigms are needed to distinguish LADA and ensure appropriate disease treatment and management.
Recently, several studies have used genetic information derived from diabetes-associated risk variants across the genome to reclassify diabetes [
6]. To date, comprehensive genetic studies of T1D and T2D have uncovered dozens of distinct susceptibility loci for each of these two diseases [
7‐
9]. Initial analyses of T1D loci in relatively small LADA cohorts have consistently shown an association with the T1D locus
HLA-DQB1, which resides in the major histocompatibility complex (MHC) [
3,
10,
11], as well as at
PTPN22 and
INS [
12,
13]. Similar analyses of T2D loci have suggested an association in LADA with the strongest T2D locus harboring
TCF7L2 [
12,
14,
15] and the
ZMIZ1 locus [
16]. A significant challenge of these studies has been the lack of statistical power due to the small number of LADA patients included. Thus, the genetic etiology of LADA remains largely unresolved.
To quantify the genetic liability to LADA contributed by genetic risk factors for T1D and T2D, we amassed the largest LADA cohort to date. By assessing the association of these variants in LADA, our objective was to place LADA along the etiological diabetes spectrum and reshape our understanding of the relationship between LADA and classic diabetes phenotypes.
Discussion
Defining LADA as a distinct form of T1D has two broad benefits. First, it highlights the potential to understand what determines both the degree and rate of disease progression. Second, it helps define differences between adult-onset autoimmune diabetes, including LADA, and T2D in terms of co-morbidities and putative therapy [
22]. Leveraging children whose future diabetes risk is unknown represents the most conservative setting in which to conduct this study given they serve as excellent population-based controls in which to contrast the cases; however, the conservative nature of the approach may result in some false negative results.
To shed light on the genetic etiology of LADA, we tested the impact of established T1D and T2D risk loci in the largest set of LADA cases collected to date. Our study differs from a previous association study with GWAS-implicated loci in adult-onset autoimmune diabetes by Howson et al. [
23]; first, our LADA cases are distinguished by the fact that they were not treated with insulin upon diagnosis. Furthermore, our study looked at a larger set of T1D and T2D loci, as well as comparing their roles in LADA against T1D and T2D, including taking population substructure into account. As with Howson et al. [
23], we observed significant association of the T1D loci
PTPN22,
INS,
HLA, and
SH2B3. However, we did not observe significant association with the
CLEC16A,
IL2RA,
CTLA4, and
STAT loci
. Despite published data observing the association of T2D locus
TCF7L2 with a subset of T1D patients [
24,
25], our study did not observe an association of this locus with LADA; one possibility could be that we used population-based controls, while previous studies may have used a different control strategy where the difference in the risk allele was more evident due to its under-representation in relatively disease-free controls. Our study goes further by leveraging GRS to offer a further line of evidence for the classification of diabetes subtypes, complementing standards for clinical decision-making and additional standardized (antibody) testing, each with their strengths and weaknesses.
LADA shows the MHC risk found in adult-onset T1D [
23] with a reduced genetic susceptibility at this locus compared with childhood-onset T1D. Less clear is whether T2D loci play a role in adult-onset autoimmune diabetes. Our results show that genetic signals implicated in T1D or T2D both play a role in LADA, with four T1D loci and one T2D locus significantly associated with this form of diabetes. LADA is genetically more similar to T1D, especially when cases are constrained on both GADA+ and IA2A+, although LADA shares part of its genetic etiology with T2D. When constrained on GADA+ only, LADA cases became less distinct from T2D, highlighting the importance of IA2A in discriminating LADA within the T1D-T2D spectrum. By implication, a GRS derived from T1D can discriminate, to a degree, non-insulin requiring adult-onset diabetes patients with either autoimmune diabetes or T2D.
Regarding the loci implicated in T1D, our results are consistent with previous studies showing a major role for the MHC,
PTPN22, and
INS loci in LADA [
10,
12,
13]. Interestingly, the risk allele frequency at
INS (rs689) was even more strongly associated with LADA than with T1D. Therefore, our data strongly points to common insulin-related pathways underpinning autoimmune diabetes irrespective of the age at onset of the disease. Given the evidence that age at diagnosis is genetically determined [
26], these loci may play a key role in determining the age at disease onset and the rate of disease progression.
While our results suggest LADA is genetically closer to T1D than to T2D, we observed an association at one T2D locus,
HNF1A, known to be associated with T2D and ‘maturity-onset diabetes of the young’; strikingly, the
HNF1A signal remained significantly associated with LADA even in the cohort enriched for both T1D autoantibodies. Nevertheless, the nature of the role of
HNF1A in LADA is unclear, although any gene compromising insulin secretory function could predispose to diabetes. This is the first report describing an association between this T2D-associated risk allele and LADA, although this locus has been previously implicated in T1D [
16]. Additionally, the strongest T2D-associated locus,
TCF7L2, has been associated with LADA in a Finnish cohort [
14‐
16], but in our study, the risk allele frequency in LADA was very close to that of controls and lower than controls in GADA+ IA2A+ LADA. Our findings were further supported by leveraging healthy adult British controls from the WTCCC, which provided overall consistent results, including for the
HNF1A signal. However, given the borderline association of T2D loci identified and the modest power in this single study, these signals must be subjected to replication efforts by independent investigators in order to fully validate these observations.
We found that, from GRS calculated from T1D- and T2D-implicated SNPs, which distinguished LADA cases from controls, the T1D GRS performed better than the T2D GRS; this difference was particularly striking in GADA+ IA2A+ LADA cases. Comparison of GRS between the six defined groups placed LADA in between T1D and T2D but closer to T1D. GADA+ IA2A+ LADA was very similar to T1D, primarily because such constraint filters out ‘T2D-like’ cases and enriches for ‘T1D-like’ cases. The potential for clinical, immunological, or genetic filters to define forms of diabetes is emphasized by the marked overlap in GRS scores, even between T1D and controls.
This study does have limitations. First, GADA-only LADA cases had a T2D-SNP GRS distribution more similar to T2D than controls. The specific association between the T2D risk score and GADA-only LADA cases could be in part due to the fact that a fraction of these cases might be false antibody-positive T2D, though those with double antibody positivity are likely to have a very low false positive rate. Thus, larger studies may resolve whether T2D risk alleles play a role in LADA. Indeed, this study was underpowered to identify specific associations other than for HNF1A. Second, two different genotyping arrays were utilized; thus, to correct for potential batch effects due to genotype array differences, population substructure, and relatedness among samples, we used a linear mixed model, resulting in highly conservative effect estimates. Consequently, it is possible that we have missed some true positive associations since we robustly controlled for false positive results.
The current nomenclature to classify diabetes, designating it as ‘T1D’ or ‘T2D’, was adopted to foster research and appropriate therapy for different phenotypic presentations. The combination of GRS, age at diagnosis, clinical phenotype, autoantibody assays, and C-peptide estimates as a proxy for insulin secretion affords a more sophisticated approach with the potential to dissect the heterogeneity of diabetes [
6]. This study highlights the uncertainty of the current classification of diabetes [
27]. These results suggest that clinical phenotype alone is insufficient to define the major types of diabetes. To better treat the various diabetes subtypes, we need to integrate the use of clinical phenotype, metabolic status, immune changes, and underlying genetic risk.
Acknowledgements
We would like to thank the JDRF, German Research Council (DFG: SFB 518, A1), German Diabetes Foundation, and EUFP5 (Action LADA) for providing samples for this research. This study makes use of data generated by the Wellcome Trust Case Control Consortium. A full list of the investigators who contributed to the generation of the data is available from
www.wtccc.org.uk. Funding for the project was provided by the Wellcome Trust under award 076113.
The Bone Mineral Density in Childhood Study is a multicenter, longitudinal study of bone accrual in healthy children. Authors are listed below:
Heidi J. Kalkwarf, PhD1; Joan M. Lappe, PhD2; Vicente Gilsanz, MD3; Sharon E. Oberfield, MD4; John A. Shepherd, PhD5, Andrea Kelly, MD6,7, Babette S. Zemel, PhD6,8
1Division of Gastroenterology, Hepatology and Nutrition, Cincinnati Children’s Hospital Medical Center, Cincinnati; 2Division of Endocrinology, Department of Medicine, Creighton University, Omaha; 3Department of Radiology, Children’s Hospital Los Angeles, Los Angeles; 4Division of Pediatric Endocrinology, Diabetes, and Metabolism, Department of Pediatrics, Columbia University Medical Center, New York; 5Department of Radiology, University of California San Francisco, San Francisco; 6Department of Pediatrics, Perelman School of Medicine, University of Pennsylvania, Philadelphia; 7Division of Endocrinology and Diabetes, The Children’s Hospital of Philadelphia, Philadelphia; 8Division of Gastroenterology, Hepatology and Nutrition, The Children’s Hospital of Philadelphia, Philadelphia.
Authors’ contributions
Study concept and design: RM, AC, DLC, MIH, JPB, KMH, VCG, HH, DM, NCS, KBY, BFV, SS, BOB, RDL, and SFAG. Analysis and interpretation of data: RM, AC, DLC, JPB, BFV, and SFAG. Resources: MIH, HH, DM, NCS, KBY, BFV, SS, BOB, RDL, and SFAG. Drafting and critical revision of the manuscript: RM, AC, DLC, BOB, RDL, and SFAG. Obtained funding: SS, BOB, RDL, and SFAG. All authors contributed to the final version of the manuscript. RM, AC, DLC, BOB, RDL, and SFAG are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. All authors read and approved the final manuscript.