Chapter 11

Prediction of Type 1A Diabetes:

The Natural History of the Prediabetic Period

 

Updated 09-2012

 

George S. Eisenbarth

 

 

Introduction

 

Predictive Factors in Relatives

 

Prediction in the General Population

 

Prediction/Diagnosis in Adults

 

Stage in Life Initiation

 

Are there Abnormalities that Precede Autoantibodies?

 

Environmental Factors?

 

Are Beta Cells Destroyed in a Progressive/Linear Fashion?

 

Conclusions

 

Introduction

 

The importance of understanding the natural history of immune mediated pre-diabetes1-3 lies in the development of prevention strategies4-6.  Several initial randomized clinical intervention trials have concluded and the next generation of such trials will rely upon improved and simplified identification of individuals7 who are at high risk of progression to diabetes8.  This is essential to ensure that trials will have sufficient statistical power to detect a given effect of the intervention (if it exists) within the time available for the study.  Such understanding is also needed to avoid exposing those who will not develop diabetes to the risk of adverse effects of the intervention.  In addition, it is likely that many interventions will be more effective if given early, with more extant beta cells.  In addition there is accumulating evidence that at the onset of type 1A diabetes, and in a subset of patients years after the onset of diabetes (Figure 11.1) there remains islet beta cells9, and preservation of even low levels of insulin secretion has multiple benefits in terms of improved glycemic control and prevention of complications10-14.

 

The amount of beta cell destruction at the onset of diabetes remains an open question15-20 with one estimate that a 40% reduction in beta cell mass is sufficient for diabetes of a 20 year old while 80-90% loss may be required for children presenting with diabetes less than age 521.  Given individual differences in rates of progression to diabetes and insulin resistance (as well as potential function per beta cell), it is likely that the variance of beta cell loss will be large at any age.  It is of interest that the slope of loss of c-peptide after diabetes onset decreased modestly from the immediate prediabetic phase22.  This may reflect “synchronization” with greater rate of loss associated with presentation with diabetes.  With long-term type 1 diabetes the mean beta cell loss is dramatic(Figure 10.1).

 

 

Figure 11.1.  Area of insulin containing cells in long-term patients with diabetes 23.

 

First-degree relatives of individuals with type 1A diabetes have an approximate 5% risk of developing the disease (independent of country 24) while children without a relative with diabetes in the United States have a risk of 1/300 while in Japan the risk is less than 1/3,000.  Longitudinal studies of autoantibody-positive relatives and more recently general populations25, 26 have provided a wealth of information on the natural history of autoimmunity during the pre-hyperglycemic phase of the disease (prediabetes).  These studies have established the predictive value of age27, islet cell antibodies (ICAs)28, multiple “biochemical” autoantibodies 29-32 first phase insulin release (FPIR)33, impaired glucose tolerance, C-peptide secretion (Prediction”Scores”) 34, and human leukocyte (HLA) haplotypes35.  Increasingly, combinations of markers are being used to better define the risk of diabetes 36-40 Prediction is not absolute, but can be expressed as the percentage of individuals developing diabetes within a given time period.  In addition the age at which islet autoantibodies first appear and levels of insulin autoantibodies (not antibodies reacting with GAD, IA-2 or ZnT8) predict approximate age of onset of type 1 diabetes25, 26.  This chapter will review predictive factors currently in use and discuss some of the unanswered questions on the natural history of Type 1A (immune mediated) prediabetes.

 

Predictive Factors in Relatives

 

The first large scale studies of the prediction of type 1A diabetes relied upon the detection of cytoplasmic islet cell autoantibodies (ICA, Figure 11.2).  These studies framed much of our knowledge concerning progression to diabetes but suffered from the difficulty of the cytoplasmic ICA assay, including difficulty in quanitation and standardization30.  As will subsequently be emphasized determination of autoantibodies reacting with four major islet autoantigens, insulin, GAD65, ICA512 (IA-2), and ZnT8 41, 42has become central components of studies of the natural history of diabetes.  Cytoplasmic ICA represents antibodies reacting with GAD65, ICA512, ZnT8 and other unknown antigens, but not insulin autoantibodies.  High titer cytoplasmic ICA is most often associated with the presence of multiple anti-islet autoantibodies (of GAD65, ICA512, ZnT8, or insulin) and thus is associated with a high risk of progression to diabetes43. It is important to emphasize that a positive ICA test (binding of antibodies to islets of human pancreas detected with indirect immunofluorescence assays) can represent antibodies to only the GAD autoantigen (Glutamic acid decarbosylase), only IA-2 autoantibodies, only non-GAD and non-IA-2 antigens, or a mixture of the above44, as well as antibodies reacting with ZnT8 (unpublished).  Thus detection of ICA positivity in addition to detection of biochemical autoantibodies usually increases risk of progression to diabetes given association with higher titer and multiple biochemical autoantibodies.

 

Riley and coworkers reported on the University of Florida, Gainesville, family study in 199027.  At that time 3413 first-degree relatives had been screened for ICA, of whom 3.3% were positive.  Positive sera were defined as those with 10 or more Juvenile Diabetes Foundation (JDF) units.  ICAs are more frequent among the siblings of diabetic probands than among parents.  They were also more frequent among relatives from multiplex families and among relatives less than 20 years old.  After a maximum follow-up of 10 years, (median 3.5 years) 40 relatives developed type 1 diabetes.  The risk of diabetes was significantly higher in relatives with ICAs of 20 or more JDF units at the time of initial screening, those aged less than 10 years (at initial screening), and those from multiplex families.  Each of these risk factors was independent of the others.  The presence of IAA was not an independent risk factor, after allowing for ICA.  However, a later report with further follow-up came to the opposite conclusion, with evidence that IAA add independently to prediction45.  The presence of ICA in titers of less then 20 JDF units did not confer significantly increased risk of diabetes, but increasing titers above this level were associated with progressively increasing risk.  Nevertheless, one third of the relatives progressing to diabetes were ICA negative on the first test and, of these, 62% remained ICA negative at the onset of diabetes.  With analysis of “biochemical” autoantibodies it has become evident that the presence of ICA in the absence of GAD65 or ICA512 autoantibodies is associated with a low risk of progression to diabetes.  Verge et al 36 proposed the “general” rule that presence of autoantibodies reacting with >=2 of the biochemical autoantibodies (GAD65, ICA512, insulin) are associated with greatly increased risk of type 1 diabetes,and we would now add ZnT8 autoantibodies to the group7.  The same was found with further follow up of the Gainesville natural history studies, with the concordant observation that in the absence of biochemical autoantibodies ICA alone is associated with little risk46.

 

The Bart's-Windsor family study, conducted in England from 1978 onwards, also found that higher ICA titer was associated with shorter diabetes-free survival28.  Seven hundred and nineteen first-degree relatives were followed for up to 10.5 years.  ICAs were tested every 4 to 6 months, using a sensitive assay with a detection limit of four JDF units.  Detectable ICAs were found in the sera of 3.3% of the relatives at initial testing, compared with 2.2% of 540 healthy child and adult controls.  However, only one control had ICA of 20 or more JDF units, compared with 10 relatives.  ICAs were detected in an additional 14 relatives on follow-up samples and follow-up time survival analysis was calculated from the time of ICA detection.  With increasing ICA cutoff the positive predictive value for future diabetes rose, and the sensitivity fell.  The risk of diabetes within 10 years associated with ICAs of 80 or more JDF units was 100% (95% confidence interval: 52%-100%).  In comparison, ICAs of 20 or more JDF units were associated with a 73% risk (95% confidence interval: 45%-100%), and ICAs of four or more JDF units with a 40% risk (95% confidence interval: 23%-57%).  Utilization of biochemical autoantibody assays in these cohorts also demonstrated the importance of multiple autoantibodies to identify high-risk individuals.  The cumulative risk of developing diabetes within 15 years was 47% (>=10 JDF units), 66% for >=20JDF units, but only 2.8% for those with ICA but without GAD or ICA512 (IA-2) autoantibodies versus 66% for those with ICA and either or both of GAD or ICA512 autoantibodies37.

 

Figure 11.2

 

Data from the Joslin Diabetes Center family study indicate that impaired FPIR (First Phase Insulin Release, usually analyzed as the sum of insulin at 1 and 3 minutes following a bolus of intravenous glucose) is an additional risk factor33.  Thirty-five first-degree relatives with high-titer ICAs (> 40 JDF units) underwent serial intravenous glucose tolerance testing (IVGTT).  The age of the subjects ranged from 2.6 to 66 years, the mean follow-up was 3.6 years from the first test, and 18 progressed to overt diabetes.  The FPIR was calculated as the sum of the 1- and 3-minute insulin levels after a standard bolus of intravenous glucose (0.5 g per kg body weight, infused as a 20%-25% solution over 2 to 4 minutes).  Percentiles for the FPIR were determined in 225 healthy, non-obese, control subjects.  Even in control subjects the FPIR showed wide within-subject variation (discussed in detail below).  Nevertheless, relatives with an initial FPIR below the first percentile (48 mU/l) had significantly reduced diabetes-free survival.  Importantly, the presence of FPIR below the first percentile did not signify that diabetes was already present47-50.  For most of the relatives, oral glucose tolerance tests (OGTT) were also performed during follow-up to detect asymptomatic diabetes.  For the survival analysis the onset of diabetes was defined by a diabetic OGTT or the occurrence of symptomatic hyperglycemia, or whichever came first.  A number of recent studies are analyzing impaired fasting glucose, impaired glucose tolerance (2 hour glucose on oral glucose tolerance testing)51, c-peptide secretion52 and potential correlates of insulin resistance (e.g. HOMA-R) and there is evidence of abnormalities preceding diabetes even in the subset of individuals with relatively normal first phase insulin secretion53, 54.   The average time from the discovery FPIR below the first percentile to the onset of diabetes was 1.8 years.  Sosenko and coworkers for DPT and Trialnet data have developed risk scores that primarily depend on quantitation of glucose tolerance testing.  Though there is variability between subsequent tests, in general, as glucose increases risk increases especially in children53, 54.

 

Updated data from the Joslin family study, with longer follow-up and larger numbers of relatives, have confirmed these findings.  Among 79 relatives with high titer ICAs (> 40 JDF units), those with FPIR below the first percentile on the first test had 3-year diabetes-free survival of 13% (95% confidence interval: 0%-30%) compared with a 78% (95% confidence interval: 63%-93%) for the group with higher FPIR55.  Studies at the Barbara Davis Center have confirmed the predictive value of FPIR measurements as has studies from the Melbourne family study, studies from Finland 56, and the DPT-1 (Diabetes Prevention Trial) North American study 57 and analysis of the combined ICARUS database58.

 

Data from the Joslin study also suggest that IAA add to the prediction of type 1 diabetes, but with a weaker effect than ICA59.  Forty-two ICA-positive (> 20 JDF units) relatives and 1670 ICA-negative relatives (representing a subset of all relatives found to be ICA negative) were tested for IAA.  Among the ICA-negative relatives, 2.7% were IAA positive, whereas among ICA-positive relatives, 45% were IAA positive.  IAA alone had less predictive value than ICA, but the combination of IAA and ICA was useful.  The risk of diabetes within 5 years was 17% for the IAA-positive/ICA-negative group, increasing to 42% for the ICA-positive/IAA-native group and to 77% for double antibody positive relatives.

 

Reports that identifying ICA subtypes improves the predictive value of ICA can now be put in the general context of the rule that multiple biochemical autoantibodies are associated with high risk36.  The restricted ICA subtype (reacting with human and rat islets but not mouse [mouse islets express little or no GAD65] and with antibody staining restricted to beta cells of rat islets), defined according to the pattern of sustaining on pancreatic sections, confers a significantly lower risk of progression to diabetes than a non-restricted subtype60.  Preabsorption of sera with glutamate decarboxylase (GAD) blocks the ICA staining of restricted ICA-positive sera61 and reduces the ICA staining of most non-restricted ICA-positive sera62.  This suggests that restricted ICA is due to antibodies directed against a single antigen (GAD) and is associated with lower risk.

 

The ICA assay is difficult to standardize, is labor intensive, and requires human pancreas36.  At the Barbara Davis Center, we utilize a combination of four assays employing recombinant antigens (insulin autoantibodies, anti-GAD, anti-ICA512(IA-2) and anti-ZnT8 63) and no longer determine cytoplasmic ICA.  For children the ICA assay provides only marginal additional information, compared with the combination of defined-antigen assays.  Relatives expressing two or more of IAA, anti-GAD, and anti-ICA512 have overall risk of diabetes within 5 years of more than 68% by life table analysis.  The addition of intravenous glucose tolerance testing does improve prediction of the time to overt diabetes.  Among relatives with two or more antibodies, those with an FPIR less than the first percentile have a 50% risk of diabetes within 1 year; those with higher FPIR have a 50% risk within 3 years36.

 

Genetic factors can also be considered in assessing diabetes risk64-68.  Deschamps and coworkers examined the predictive value of HLA typing in a study of 536 siblings of diabetic probands in France69.  The risk of type 1 diabetes after 8 years, estimated by life table analysis, was 10% for siblings who were HLA identical with the probands, 3%-4% for siblings with either DR3 or DR4, and 16% for those with DR3/DR4.  This compares with 56% for those with ICA greater than 4 JDF units and 70% for those with the combination of ICA and the highest risk HLA type, DR3/DR4.  In addition studies by Becker and colleagues from Pittsburgh indicate a high risk with long-term follow-up (12.5 years) for autoantibody negative relatives with the DR3/4 (DQ8) genotype (approximately 25%) compared to 6% for those lacking this genotype and autoantibody negative70.  Even greater risk can be defined genetically for siblings of patients with type 1 diabetes who are DR3/4-DQ2/DQ8 and have inherited both HLA haplotypes identical by descent with their proband sibling.  The risk for such children appears to be as high as 80% of activating anti-islet autoimmunity (by age 15) with most proceeding to diabetes with a several years delay from the appearance of autoantibodies (Figure 11.3)67.  In contrast siblings of patients with type 1 diabetes who are DR3/4-DQ2/8 who have inherited one or no HLA haplotype identical by descent with their proband have a risk of approximately 20% of progressing to diabetes by age 15.67

 

life table daisy 3 4 g

Aly et al Extreme genetic risk for type 1A diabetes67

 

 

Figure 11.3  Highest risk siblings in the DAISY study with DR3/4-DQ2/8 genotype progressing to expression of islet autoantibodies (left panel) and diabetes (right panel).

 

 In other studies, molecular typing has revealed that the HLA haplotype DQA1*0102 DQB1*0602 confers strong protection from type 1 diabetes, in a dominant fashion (Chapter 7).  In our experience, autoantibody-positive relatives with this haplotype have a very low risk of progression to diabetes71 and usually express only a single autoantibody, namely anti-GAD, although a few also express IAA.  Such protection is however not absolute, and approximately 1% of children developing type 1A diabetes72 (versus 20% of the general U.S. population) and 3% of adults with type 1 diabetes have DQB1*0602  (DQB1*0602 is usually part of the haplotype DRB1*1501, DQA1*0102, DQB1*0602).   Approximately 5% of older individuals developing type 1 diabetes are reported to have the protective HLA allele DQB1*060273.

 

In addition to HLA more than 50 loci contribute to risk of type 1 diabetes74.  Each locus has a small effect but a report by Winkler and coworkers suggests that combining loci can impact prediction68.  We have followed a set of identical triplets and now all three triplets have progressed to diabetes including the last triplet who was non-diabetic in 1983 (Figure 11.4).

 

Prediction in the General Population

 

It is likely that genetic typing will have an even greater impact on assessing diabetes risk in the general population25, 68, 75, 76.  Most studies of prediabetic subjects have involved the screening of first-degree relatives of diabetic probands, rather than the general population.  However, less than 10% of new cases of type 1 diabetes have an affected relative, so the general population will need to be screened eventually if an effective intervention is to have a major impact.  Screening the general population is likely to be more difficult than screening relatives.  Bayes' theorem states that a screening test will have a lower positive predictive value in the general population than in a selected group with a higher prevalence of disease, such as first-degree relatives.  One approach toward solving this problem is to screen the general population with markers of genetic susceptibility first, followed by autoantibody testing of susceptible individuals.  For example, among the general Denver population 2.4% of individuals express both DR3 and DR4 (with associated DQ2 and DQ8).  Of this subgroup, it is predicted that approximately 6% will develop type 1 diabetes, similar to the risk among first-degree relatives.

 

Unexpectedly, studies performed in Florida suggest that ICA have a predictive value in the general population similar to that in relatives77.  In contrast, studies in England suggest that ICA will have a lower positive predictive value in the general population.  The prevalence of ICAs of 20 or more JDF units was only two to three times higher in siblings than in the general population, compared with a 13 times greater risk of diabetes in the siblings78.  Many population studies have been studied (Table 11.1). The difference between the above two studies probably relates to differences in the ICA assays, with primarily individuals with higher levels of ICA (that associated with multiple biochemical autoantibodies) followed in the Gainesville general population studies.    With analysis of biochemical autoantibodies it appears that even general population individuals expressing multiple anti-islet autoantibodies are at very high risk of progressing to diabetes similar to first degree relatives.79  Those expressing single autoantibody (of insulin, GAD, IA-2 or ZnT8 autoantibodies) are at low risk for progression.  This very likely results from the high specificity of expression of multiple autoantibodies if the islet autoantibody assays are set at the 99th percentile.  Assuming independent “false positives” one can use the binomial theorem to calculate the probability of expressing two or more of the four autoantibodies by chance, and this probability is very low (.05%, specificity .9995; Binomial theorem Pn(k)=n!/(n-k)! times pk*(1-p)n-k where p=probability of antibodies positive in given population, n= number of trials, and k=number of antibodies positive).  With four biochemical autoantibodies to measure n=4, K=2,3,and 4 for >=2 autoantibodies, and for control population with assays set with 99% specificity, p=.01.  We believe it is important to utilize assays with high specificity as well as >=2 autoantibodies in evaluation of individuals without diabetes or individuals clinically having type 2 diabetes with attempts to diagnose LADA (Latent Autoimmune Diabetes of Adults).  For instance if assay specificity is 95% with four assays approximately 20% normals will express a single autoantibody but only 0.25% would express >=2 autoantibodies.  As shown in DASP (Diabetes Autoantibody Standardization Program of the Immunology of Diabetes Society and the CDC) the different assays vary as do the different laboratories80, 81.  In particular many laboratories have great difficulty measuring insulin autoantibodies82, while most laboratories have high sensitivity/specificity GAD65 autoantibody assays and IA-2 assays.  The difference appears to relate to the separation of signals between normal control samples and patients with Type 1 diabetes, with relatively separation for approximately ˝ of patients positive for insulin autoantibodies.  A new electrochemiluminescent assay has the potential to improve standardization of insulin autoantibodies81.

 

 

 

Figure 11.4:  Development of autoantibodies and loss of first phase insulin secretion in identical triplets of a patient with type 1A diabetes.

 

Several studies have now been initiated where children are followed from birth for the development of anti-islet autoantibodies.  The three studies with the longest follow-up are the BabyDiab study from Germany, the DAISY study from Denver Colorado, and major studies in Finland39, 83-86