Introduction

Type 1 diabetes is an autoimmune disease in which destruction of pancreatic islet beta cells leads to insulin deficiency. In first-degree relatives of Type 1 diabetes probands, the presence of circulating autoantibodies to the islet antigens insulin, GAD or tyrosine phosphatase-like insulinoma antigen 2 (IA2) denotes underlying islet autoimmune pathology and an increased risk of diabetes. Risk has been shown to be related to the number of antigen specificities, the level of islet antibodies, specific HLA alleles, and the degree of impaired insulin secretion [1, 2, 3, 4], but prediction remains inexact. Improved prediction would facilitate the counselling of at-risk individuals and the selection of subjects for intervention trials to prevent clinical disease.

Glucose disposal and prevailing glycaemia are determined not only by insulin secretion but also by insulin action [5]. In contrast to in Type 2 diabetes, impaired insulin action or insulin resistance has not been considered to play a major role in the pathogenesis of Type 1 diabetes. Nevertheless, insulin resistance has been documented in established Type 1 diabetes [6] and physiological states of insulin resistance such as puberty [7] and pregnancy [8] are associated with the onset of Type 1 diabetes [9, 10]. First-degree relatives positive for islet antibodies who are at increased risk of Type 1 diabetes have varying degrees of insulin resistance [11], but whether insulin resistance influences progression to clinical disease has not been documented. Interestingly, in the non-obese diabetic mouse model of Type 1 diabetes, hyperinsulinaemia, a marker of insulin resistance, antedates clinical disease [12], and the incidence of diabetes has been reduced by treatment with rosiglitazone, a thiazolidinedione drug that decreases insulin resistance [13].

If insulin resistance is found to play a role in the progression to human Type 1 diabetes, this would have important implications not only for risk prediction but also for intervention to reduce insulin resistance and delay the onset of diabetes. Therefore, in a cohort of first-degree Type 1 diabetes relatives positive for islet antibodies, we prospectively examined measures of insulin resistance and insulin secretion to determine whether insulin resistance is, in itself, a risk factor for progression to clinical disease.

Subjects and methods

Subjects

A total of 3773 relatives of Type 1 diabetes probands were screened for islet antibodies in the population-based Melbourne Pre-Diabetes Family Study [14] (Fig. 1). Of these, 104 relatives (3%) confirmed as positive for one or more islet antibodies were followed annually, testing for islet antibodies and using an IVGTT to measure first-phase insulin response (FPIR). All participants provided written informed consent to participate in the study, and all procedures were approved by institutional ethics committees.

Fig. 1
figure 1

Summary of the subjects of the trial

Full size image

Over a median follow-up period of 4.0 years, 43 subjects progressed (progressors) and 61 did not progress (non-progressors) to diabetes (Fig. 1, Table 1). Diagnosis of diabetes was based on symptoms (polydipsia, polyuria, weight loss) and plasma glucose levels, as laid out in the American Diabetes Association criteria [15].

Table 1 Characteristics of progressors and non-progressors
Full size table

Islet antibodies

Insulin autoantibodies (IAA) were measured by precipitation of 125I-(A14)-labelled human insulin [16]. Antibodies to GAD (GAD Ab) and IA2 (IA2 Ab) were measured by precipitation of 35S-methionine-labelled, in vitro transcribed and translated GAD65 or IA2 [17]. These assays have shown optimal sensitivity and specificity in International Workshops and Standardization Programs conducted by the Immunology of Diabetes Society (e.g. in [2]). The specificity and sensitivity of our assays in the 2003 Diabetes Antibody Standardization Program were: IAA (95%, 26%), GAD Ab (97%, 80%) and IA2 Ab (100%, 68%). Thresholds for positivity established as the 97.5% percentile of unselected healthy school children were: IAA, 100 U/ml; GAD Ab, 5 U/ml; and IA2 Ab, 3 U/ml. Until 1995, all individuals were screened for islet cell antibodies (ICA) by standard indirect immunofluorescence, as well as for GAD Ab, IA2 Ab and IAA. The ICA assay was discontinued in 1995. Thus, all relatives positive for GAD Ab, IA2 Ab or IAA, and also for ICA prior to 1995, were included in the analysis.

Insulin secretory function and insulin resistance

The IVGTT was performed according to the ICARUS protocol [18] by one of two research nurses. Subjects were instructed to maintain their normal dietary intake and to include at least 150 g of carbohydrate per day for at least 3 days before the test. They fasted for 10 to 12 hours overnight and abstained from smoking, coffee consumption and excessive physical activity for 24 hours before the test. Intravenous glucose (0.5 g/kg body weight as a 25% solution, up to a maximum of 35 g) was injected over 3 minutes into an antecubital vein. Blood samples were collected from a cannula in situ 5 minutes before and immediately prior to injection of glucose, as well as 1, 2, 3, 5, 7 and 10 minutes after injection. Serum insulin was measured with a commercial radioimmunoassay kit (Pharmacia, Uppsala, Sweden); the inter-assay coefficient of variation was 3% at levels of 15 and 50 mU/l, and 5% at 138 mU/l. FPIR, a validated measure of beta cell function, was calculated as the sum of the serum insulin concentrations at 1 and 3 minutes after i.v. glucose [18]. Samples from each subject were all analysed in the same assay. In our study, the mean inter-assay coefficient of variation for FPIR in healthy young adults was 11.4% [19].

Insulin resistance was estimated by the homeostasis model assessment of insulin resistance (HOMA-R) [20] as fasting insulin (mU/l) × fasting glucose (mmol/l)/22.5. In a large adult population, the median HOMA-R was 2.1 [21]. HOMA-R was calculated from the mean of the fasting insulin and glucose values at −5 and 0 minutes both at baseline (from the IVGTT at recruitment) and serially over the duration of follow-up (median number of IVGTTs = 4).

Statistical analysis

Differences in age, body weight percentile, number of islet antibody specificities, islet antibody levels, follow-up time and metabolic parameters between the whole cohort of progressors and non-progressors were analysed by the non-parametric Mann–Whitney test. Differences between the frequency of HLA types were analysed by the chi square test for trend. The association between HOMA-R and FPIR was analysed by least squares regression. Potential predictive variables for progression to diabetes were analysed by a Cox proportional hazards model using baseline data. Age band (0–5, 6–10, 11–20, 21–30, 31–40, >40) at baseline and islet antibody number (at baseline, as a continuous variable) were included in all analyses to avoid confounding, although only the number of antibodies was associated with significantly increased hazard. Sex was also added to the model at various stages but appeared unrelated to progression. The ‘standard’ model was obtained by adding FPIR (discretised by quartiles). Additional potential predictive variables were then added: fasting plasma glucose, fasting serum insulin, HOMA-R, log(fasting insulin : FPIR) and log(HOMA-R : FPIR). These additional variables were evaluated, either as continuous covariates or when discretised by quartile, using the likelihood ratio (LR) test. Untransformed fasting insulin : FPIR and HOMA-R : FPIR were not included because they were heavily skewed and their inclusion in the model gave unstable estimates. The ratios were therefore log transformed and base-2 logs were used for ease of interpretation. The ability of log(fasting insulin : FPIR) or log(HOMA-R : FPIR) to supplant other metabolic variables was then tested by including them in a revised model with age band and number of islet antibodies. The LR test was then used to determine whether any of the other variables significantly improved the model.

In addition to the multivariate analysis, a clinical approach was adopted in which progressors were individually matched with non-progressors as closely as possible for age, sex, number of islet antibodies and FPIR. Most progressors with very low FPIR could not be matched, because FPIR in non-progressors was, of course, normal. Consequently, a comparison was possible between 21 progressors and non-progressors. Differences between matched progressors and non-progressors were analysed by a paired t test. The rate of change of HOMA-R values over time was calculated as the slope of the linear regression line. For follow-up measures of HOMA-R and other metabolic parameters, the mean of serial values in each subject was calculated, followed by a calculation of the median of these for the whole group.

Results

Progressors were comparable in age, sex, body weight percentile and HLA DR allele frequency to non-progressors, but were distinguished by a higher number and higher levels of islet antibodies (Table 1). Non-progressors were obviously followed for longer, the median time to diabetes in progressors being 2.7 years. Although baseline HOMA-R was correlated with baseline FPIR (r=0.53, p<0.0001), the progressors had greater insulin resistance for their level of insulin secretion (Fig. 2). Thus, the ratio (median [interquartile range]) of fasting insulin : FPIR at baseline was significantly higher in progressors than in non-progressors (0.146 [0.091–0.236] vs 0.061 [0.050–0.074], p<0.0001), as was the ratio of HOMA-R : FPIR (0.033 [0.012–0.057] vs 0.013 [0.011–0.017], p<0.0001). Similarly, during follow-up, fasting insulin : FPIR and HOMA-R : FPIR were greater in progressors (Table 1).

Fig. 2
figure 2

Baseline HOMA-R and FPIR in progressors (filled circles) and non-progressors (open circles)

Full size image

Kaplan–Meier survival curves according to HOMA-R : FPIR quartiles demonstrated that higher baseline HOMA-R : FPIR ratios were associated with an increased risk of progression to diabetes (Fig. 3). Cox proportional hazards analysis was performed to determine whether the fasting insulin : FPIR and HOMA-R : FPIR ratios were independent risk factors for progression to diabetes. The analysis was initially based on a standard model incorporating islet antibody number, FPIR and age. Islet antibody number and FPIR were confirmed to be independently associated with progression to diabetes. Age, included to avoid confounding, was not consistently associated with progression. Metabolic variables, namely fasting glucose, fasting insulin, HOMA-R, log(fasting insulin : FPIR) and log(HOMA-R : FPIR), added separately to the standard model, were each associated with progression (Table 2). Given that log(HOMA-R : FPIR) is derived from HOMA-R and FPIR, it was substituted for FPIR in the standard model to determine whether it supplanted the other metabolic variables. In the revised model (Table 2), islet antibody number (hazard ratio 3.16 per antibody, p<0.001) and log(HOMA-R : FPIR) (hazard ratio 2.57 per doubling, p<0.001) were independently associated with progression. When the other metabolic variables, fasting glucose, fasting insulin, HOMA-R and FPIR were added separately to the revised model they were no longer independently associated with progression. Thus, log(HOMA-R : FPIR) supplanted the other metabolic variables for association with progression to diabetes (Table 2). Log(fasting insulin : FPIR) in place of log(HOMA-R : FPIR) in the revised model also supplanted fasting glucose (hazard ratio 1.74, p=0.098), fasting insulin (hazard ratio 1.00, p=0.90), HOMA-R (hazard ratio 1.04, p=0.74) and FPIR (hazard ratio 2.84 per doubling, p<0.001) (p=0.17).

Fig. 3
figure 3

Kaplan–Meier survival estimates according to baseline HOMA-R : FPIR quartile

Full size image
Table 2 Cox proportional hazards analysis
Full size table

In the matched pair comparison, progressors and non-progressors were similar for age, sex ratio, body weight percentile, frequency of HLA DR3,4 alleles and number of islet antibody specificities, levels of islet antibodies and baseline FPIR (Table 3). Despite being within the population reference range, fasting plasma glucose and fasting serum insulin were higher at baseline in matched progressors. In addition, progressors had greater insulin resistance relative to insulin secretion at baseline and during follow-up (Table 3).

Table 3 Comparison of matched progressors and non-progressors
Full size table

Three other findings require comment. Firstly, HOMA-R declined with time to diabetes in progressors but remained unchanged in non-progressors, the rate of change (mean of the slopes of individual regression lines) being −0.40 compared with 0.05 (p=0.010). Because the decrease in HOMA-R over time was lower than that in FPIR, the ratio of HOMA-R : FPIR increased over time in progressors (Table 3). Secondly, 22 of the 104 relatives were positive for ICA only, but none progressed to diabetes. Their HOMA-R, FPIR and HOMA-R : FPIR were not significantly different from those of the other non-progressors, but the subjects were significantly older (mean age 35.1±2.9 vs 17.5±1.8 years, p<0.0001). Thirdly, although IAA could theoretically interfere with the measurement of insulin, we found there to be no relationship between IAA and HOMA-R either in the whole cohort (r=0.07) or separately in progressors (r=0.23) and non-progressors (r=−0.34).

Discussion

This is the first study to investigate longitudinally both insulin secretion and insulin resistance in the progression to clinical Type 1 diabetes. Progression to diabetes in first-degree relatives positive for islet antibodies was associated with an increase in insulin resistance relative to insulin secretion, independently of known risk markers. While fasting insulin has been used in the past as an index of insulin resistance, HOMA-R has been shown to correlate better with results from hyperinsulinaemic–euglycaemic clamp studies [20]. HOMA-R has been widely applied as an index of insulin resistance in people with NGT [20, 22, 23] and in those with IGT or diabetes [24, 25, 26], and has also been used to document insulin resistance in longitudinal studies [27]. This is an important consideration, as invasive techniques for direct measurement of insulin resistance would not be feasible or ethically acceptable in longitudinal studies involving children.

Relatives positive for islet antibodies who progressed most rapidly to clinical diabetes had evidence of disturbed insulin–glucose homeostasis years before diagnosis. Their increased fasting plasma glucose and serum insulin levels compared with the non-progressors, although still within the normal ranges, account for their higher baseline HOMA-R values. Further evidence for this subtle disturbance was the increase in HOMA-R relative to FPIR, which increased over time to diabetes, indicating progressive dysregulation of the relationship between insulin action and insulin secretion. The significance of these findings lies in the fact that the most rapid progressors can be identified early, even when metabolic parameters such as fasting glucose, fasting insulin and FPIR are still within the normal ranges. It is important to note, however, that a single estimation of HOMA-R could be misleading, as HOMA-R decreases as individuals approach clinical disease, most probably because fasting insulin levels decline. Nevertheless, as we demonstrate here, HOMA-R is best interpreted in the context of insulin secretion; the HOMA-R : FPIR ratio was increased in progressors not only at baseline but also during follow-up and the progression to diabetes. Furthermore, log(HOMA-R : FPIR) was superior to HOMA-R or FPIR alone for estimating risk. The use of an alternative ratio, log(fasting insulin : FPIR), yielded similar results.

Insulin resistance in individuals with islet autoimmunity could have a genetic–constitutional basis, or could be secondary to the autoimmune disease process itself. The commonest association of insulin resistance is with obesity, and children who develop diabetes have been shown to have an increased BMI in the first year of life [28]. In our cohort, body weight percentile was not significantly different between progressors and non-progressors. However, this measure may not accurately reflect visceral adiposity and insulin resistance [29], and subtle differences in body composition between the progressors and the non-progressors cannot be excluded. Alternatively, insulin resistance may reflect a more aggressive form of autoimmune disease mediated, for example, by immuno-inflammatory factors that also mediate beta cell destruction. Such factors include TNF-α and IL-6 [30], which have been implicated in insulin resistance [31, 32], and, conceivably, autoantibodies to the insulin receptor, which have been reported to be associated with IAA [33]. Further studies are required to determine whether autoimmune mechanisms in Type 1 diabetes contribute to insulin resistance. Because plasma glucose is determined by the interplay of insulin secretion and insulin action [5], one would expect insulin resistance to lead to an earlier appearance of hyperglycaemia when beta cell function is compromised by autoimmune disease. Furthermore, an increase in plasma glucose secondary to insulin resistance could augment islet inflammation and beta cell death. Glucose has been shown to induce the pro-inflammatory cytokine IL-1β in islets [34], to up-regulate expression of the Fas death receptor [35] and autoantigens [36, 37] in beta cells, and to induce beta cell apoptosis [38].

The finding that insulin resistance in individuals positive for islet antibodies is a risk factor for progression to clinical diabetes has several important implications. Firstly, it should improve risk prediction, currently based on FPIR and islet antibody parameters, and should therefore assist in counselling individuals and families. Secondly, it could allow at-risk individuals to be stratified into more uniform risk groups for recruitment into intervention–prevention trials, reducing the number of participants and thereby increasing the efficiency and decreasing the cost of such trials. Finally, reduction of insulin resistance, with insulin-sensitising drugs, diet or exercise, could delay the onset of clinical Type 1 diabetes.