Dynamic Pattern of Disease-Associated Autoantibodies in Siblings of Children With Type 1 Diabetes: A Population-Based Study

The population comprised 710 initially nondiabetic siblings of 491 children who had diabetes and had been recruited for the Childhood Diabetes in Finland Study (19). In 327 families, one sibling was included in the study, followed by 128 families with two, 27 families with three, 5 families with four, 3 families with six, and 1 family with eight siblings included. The probands’ diabetes had been diagnosed before the age of 15 years between September 1986 and April 1989, and the siblings were <20 years of age at the time (mean age 9.8 years; range 0–19 years). More than half of them (386; 54.4%) were girls. Half-sisters and half-brothers were also monitored for autoantibodies, but they were not included in the analysis related to the number of children in the family. Our requirement for inclusion was that the sibling should have a minimum of two blood samples available.

The samples for autoantibodies were scheduled to be taken at 1–3 weeks, 2–3 months, 5–6 months, 11–12 months, 17–18 months, and 23–24 months after the diagnosis of the probands’ diabetes (Fig. 1), and later samples were obtained at ∼36 months (n = 475) and ∼48 months (n = 357). A total of 183 participants had additional samples taken between or after the time points mentioned. The number of samples per participant varied from 2 to 18; 85 had 2–4 samples, 82 had 5, 111 had 6, 124 had 7, 228 had 8, and 80 had >8 samples. The first sample was taken within 1 month of the diagnosis of the proband’s diabetes in 569 siblings (80.1%), and the siblings were observed for autoantibodies for 3 years and 7 months (median range 0.1 months to 9 years and 10 months). The siblings who developed type 1 diabetes during the observation period (n = 38) were observed up to clinical diabetes, the diagnosis of which was based on clinical symptoms and an increased random blood glucose concentration (>10 mmol/l) or an elevated fasting blood glucose (>6.7 mmol/l) or random blood glucose of >10 mmol/l on two occasions in the absence of symptoms.

The samples were stored at −20°C or below until analyzed. ICA and IAA were quantified sequentially in all samples, whereas GADA and IA-2A were analyzed in at least the first and last samples available. GADA were quantified in all samples when the sibling tested positive for ICA, IAA, or IA-2A at any time point analyzed or for GADA in the first and/or last sample, and correspondingly, all samples were assayed for IA-2A when the sibling tested positive for ICA, IAA, or GADA at any time point analyzed or for IA-2A in the first and/or last sample. When the sera from a given individual were analyzed for GADA and IA-2A, all of his or her samples were with a few exceptions run in the same assay to eliminate interassay variation. A few autoantibody results were missing for technical reasons.

IA-2A were measured by a radiobinding assay as previously described (1). The results were expressed in relative units (RU) based on a standard curve run on each plate. The cutoff limit for autoantibody positivity was defined as the 99th percentile in 374 controls, i.e., 0.43 RU. Samples with initial IA-2A levels close to the limit for positivity (between 0.32 and 1.0 RU) were reanalyzed to verify the antibody status. The mean of the primary and retested results was used when the retested sample remained in the same category (negative or positive). When the results were discrepant, the sample was tested for a third time and was categorized as negative when two of three gave a value below the cutoff level and as positive when two of three assays resulted in a value above the cutoff level. In these cases, the mean of the two consistent assays was used as the final antibody level. The interassay coefficient of variation was 12% at an IA-2A level of 0.63 RU, 10% at a level of 21.3 RU, and 8% at a level of 82.6 RU.

The radiobinding assay for measuring GADA levels was first described by Petersen et al. (20). The cutoff limit for positivity (99th percentile in >370 nondiabetic children and adolescents) was 6.6 RU for the assay format described by Sabbah et al. (21) and 5.35 RU for the format described by Savola et al. (3). Samples with an initial GADA level close to the limit for positivity (between 5 and 8 RU in the assay described by Sabbah et al. and between 3 and 7 RU in the assay described by Savola et al.) were reanalyzed to verify the antibody status. The mean of the primary and retested results was used when the retested sample remained in the same category (negative or positive). When the results were discrepant, the sample was tested for a third time and was categorized as negative when two of three gave a value below the cutoff level and as positive when two of three assays resulted in a value above the cutoff level. In these cases, the mean of the two consistent assays was used as the final antibody level. The assay described by Sabbah et al. had an interassay coefficient of variation of 18% at a GADA level of 14.6 RU and 12% at GADA levels that exceeded 100 RU.

IAA were quantified with a radiobinding assay modified from that described by Palmer et al. (22). Endogenous insulin was removed with acid charcoal before the assay, and free and bound insulin were separated after incubation with mono-¹²⁵I(Tyr A 14)-human insulin (Novo Research Institute [NRI], Bagsvaerd, Denmark, or Amersham, Little Chalfont, Bucks, U.K.) for 20 h in the absence or presence of an excess of unlabeled insulin. The IAA levels were expressed in nU/ml, where 1 nU/ml corresponds to a specific binding of 0.01% of the total counts. The interassay coefficient of variation was <8%. A participant was considered to be positive for IAA when the specific binding was 55 nU/ml or more when using monoiodinated insulin from NRI and 68 nU/ml or more when using the labeled insulin from Amersham. Both of these cutoff limits represented the 99th percentile in a reference population that comprised 105 nondiabetic children and adolescents. The majority of the samples were analyzed with the NRI label, but in 1995, when NRI stopped production, we started to use labeled insulin from Amersham. Accordingly, this was used mostly in samples that were taken outside the regular sampling schedule. Samples with initial IAA levels close to the limit for positivity (between the 98th and 99.5th percentiles in controls) were reanalyzed to verify the antibody status. The mean of the primary and retested results was used when the retested sample remained in the same category (negative or positive). When the results were discrepant, the sample was tested for a third time and was categorized as negative when two of three gave a value below the cutoff level and as positive when two of three assays resulted in a value above the cutoff level. In these cases, the mean of the two consistent assays was used as the final antibody level.

ICA were detected by a standard immunofluorescence method (23), the detection limit being 2.5 JDF units. Initially ICA-positive samples were retested to confirm antibody positivity. When repeated, the new result always replaced the old one when the sample tested positive on both occasions. When there was a discrepancy, the sample was retested once more and categorized according to the two consistent results. When positive on two of three tests, the mean of the two positive tests was used. The ICA results were measured using several pancreases, which were selected to give superimposable standardization curves. Pancreas variability could not be avoided completely, however. Seroconversions that most obviously resulted from pancreas variability were eliminated by excluding 38 suspect ICA results from 11 participants, mainly in samples that were taken outside the regular sampling schedule.

HLA DR alleles were typed using conventional HLA serology as described by Tuomilehto-Wolf et al. (24). All HLA DR specificities recognized by the Nomenclature Committee of the World Health Organization in 1984 were included in the test panel (25).

We performed separate analyses for each of the four types of autoantibodies considered (IAA, ICA, IA-2A, and GADA) to describe the effects of various covariates at three stages of the process: 1) prevalence of positivity at entry to the study, 2) incidence of first positive seroconversion during the follow-up among those seronegative at entry, and 3) incidence of first inverse seroconversion among those ever positive during the observation period.

For each of the three stages, we fitted an appropriate regression model (26): 1) logistic regression for the prevalence at entry, 2) Poisson regression for the incidence rate of positive seroconversion, and 3) Poisson regression for the incidence rate of inverse seroconversion. In all of these models, we included the following factors as dichotomized covariates: 1) age of the proband at the date of diagnosis of type 1 diabetes (1 = at least 6 years at entry, 0 = <6 years), 2) age of the sibling at the diagnosis of proband (1 = at least 6 years at entry, 0 = <6 years), 3) sex of the sibling (1 = male, 0 = female), 4) HLA-phenotype of the sibling (1 = DR3/DR4, 0 = other than DR3/DR4), 5) HLA identity of the sibling with the proband (1 = identical, 0 = haploidentical or nonidentical), and 6) family size (1 = at least four children, 0 = three children or fewer).

In the analysis of the first positive seroconversion, we also considered the effect of (7) the initial antibody status (1 = at least one other autoantibody present, 0 = no other autoantibody present at entry) as an additional predictor. However, the effects of the other covariates were estimated from the model not including this apparently intermediate variable. Also, in the analysis of the first inverse seroconversion (8), the number of other autoantibodies present when positivity was first observed (1 = at least one, 0 = none) was considered as an additional predictor, but it was not included in the models from which the effects of other covariates were estimated.

When analyzing the prevalence at entry, the population comprised all siblings from whom data on relevant variables were available. The outcome cases were those siblings in whom a positive test result was observed for the antibody in question at the first observation point. The logistic regression model was fitted by the appropriate command in the SPSS program. As results from this fitting, we got the estimated relative odds (odds ratio) associated with each category of a covariate in relation to the reference category adjusting for the other covariates included in the model.

In the analysis of the incidence of the first positive seroconversion of an antibody, the population at risk included those siblings in whom the assay for that antibody gave a negative result at the first observation point and from whom data on other relevant variables were available, too. As outcome cases were those participants in the population at risk who provided a positive assay result for that antibody at any later observation point. The person-times at risk were calculated in the following way. For each child i, the total follow-up time was divided into disjoint intervals by individual time points of serum sampling t_i0, t_i1, … , t_im(i). The amount of person-time T_i contributed by an initially seronegative child who remained seronegative throughout the observation period was calculated as T_i = t_im(i) − t_i0. For an initially seronegative child who remained seronegative until time t_i,j−1 but was observed as seropositive for the first time at the time point t_ij, the seroconversion was assumed to occur between these two points, so the contribution to the follow-up time was estimated as T_i = (t_i,j−1 − t_i0) + 1/2(t_ij − t_i,j−1). When an initially seronegative sibling converted to antibody positivity more than once, only the first occurrence of positive seroconversion was taken into account

A similar approach was used in analysis 3 of inverse seroconversions. In this analysis, the population at risk comprised those participants who provided a positive assay result for the antibody in question either at entry or at any later point. In the person-time calculations, the individual starting point u_i0 was set at the sampling time t_ij when participant i was first observed to be seropositive. The end point was set either at t_im(i) or at (t_ik − t_i,k-1)/2, where t_i,k was the first occasion since t_ij when the child was observed as seronegative and t_ik-1 was the time when he or she was still observed as seropositive. Among these ever seropositive siblings, only the first occurrence of inverse seroconversion was considered in the analysis.

When modeling the incidences in analyses 2 and 3, the appropriate generalized linear model was specified such that the Poisson family was assumed as the distribution for the outcome cases, the natural logarithm was chosen as the link function, and the logarithm of the appropriate person-time was treated as an offset term in the linear predictor. These models were fitted by the method of maximum likelihood using the Genmod procedure of the SAS program. As results from this fitting, we got the estimated relative incidence rates (rate ratio [RR]) associated with each category of a covariate in relation to the reference category, adjusting for the other covariates included in the model.

In both the logistic and the Poisson models, the approximate confidence intervals for the parameters of interest were calculated mainly from the Wald statistics of the coefficients on the logit or log-linear scale. In case the estimated RR was 0, the approximate upper confidence interval was derived by the profile likelihood approach.

At the beginning, 2–9% of the siblings tested positive for any given autoantibody separately (Fig. 2). The estimated odds of DR3/DR4 heterozygosity were fourfold for IA-2A, fivefold for GADA and IAA, and threefold for ICA as compared with the other DR phenotypes (Table 1). HLA identity with the proband made a sibling more prone to positivity for any antibody as compared with HLA haplo- or nonidentity but not as strongly as DR3/DR4 heterozygosity. Some indications of an effect of age, sex, and family size on antibody positivity at entry can be seen in Table 1, e.g., boys had an increased risk of IAA positivity. Close to 90% or more of the siblings remained negative for IA-2A, GADA, IAA, and ICA throughout the follow-up (Fig. 2). Among the siblings who tested positive at least once during the follow-up for a given antibody, six had fluctuating IA-2A levels with two or more seroconversions, the corresponding number being four for GADA, 31 for IAA, and three for ICA (Table 2).

Seroconversion to autoantibody positivity occurred in 2–6% of the siblings who initially were negative for the autoantibody concerned (Fig. 2). The overall incidence rates per 1,000 years (number of cases/person-years in parentheses) were nine (19/2,123) for IA-2A, six (12/2,049) for GADA, 19 (40/2,111) for IAA, and 16 (31/1,965) for ICA. Siblings <6 years of age had a threefold estimated rate for seroconversion to IA-2A positivity as compared with older siblings, the corresponding rate being fivefold for GADA and twofold for IAA (Table 3). However, positive seroconversions for each of the autoantibodies were observed even among those aged 14 years or more (data not shown). The estimated RRs for positive seroconversion were consistently highest in the DR3/DR4 heterozygotes as compared with the other HLA DR groups, the figures of IAA and ICA being supported by the 95% confidence intervals. HLA-identical siblings seemed to have a higher relative incidence rate for seroconversion of IA-2A than haplo- or nonidentical siblings. Initial autoantibody positivity predicted strongly seroconversion to additional antibody positivity. Only weak evidence was seen for a relation between the positive seroconversion rate and age of the proband, sex of the sibling, and family size, except for an increased risk of GADA seroconversion in large families.

Inverse seroconversions occurred in 10–56% of the autoantibody-positive siblings (Fig. 2); the overall incidence rates per 1,000 years (number of cases/person-years in parentheses) were 76 (12/157) for IA-2A, 42 (10/237) for GADA, 460 (32/70) for IAA, and 27 (9/331) for ICA. No consistent pattern was observed concerning inverse seroconversions in different subgroups (Table 4). However, inverse seroconversions of IA-2A had a trend to occur more frequently in older siblings than in siblings who were <6 years of age. Boys seroconverted to GADA negativity more frequently than girls. Rates of inverse seroconversions for IA-2A, GADA, and IAA tended to be lower in the DR3/DR4 heterozygous group than in the other DR phenotypes. There was a higher relative incidence rate for inverse IA-2A seroconversions in small families compared with large ones. Inverse seroconversions for IA-2A, IAA, and ICA were more frequent in siblings with no other autoantibodies at seroconversion than in those with one to three additional autoantibodies present.

We studied here the occurrence of positive and inverse seroconversions in diabetes-associated autoantibodies in initially nondiabetic siblings of children with type 1 diabetes, providing information complementary to the previously reported initial findings on ICA and IAA in these subjects (10). The participants (n = 710) were recruited in a population-based manner, and all of the siblings were observed for all four autoantibodies from the time of diagnosis of the proband’s diabetes. Accordingly, we had a unique possibility to explore the dynamic pattern of disease-associated autoantibodies in nondiabetic siblings of children with type 1 diabetes.

When evaluating the results, we also have to consider a few limitations of the study. First, although the siblings were recruited in a close-to-optimum way, with observations beginning from the time of diagnosis of the proband’s diabetes, it may have started too late in the sense that, hypothetically, many of the siblings who would seroconvert to antibody positivity for the same reason as the proband would have already done so at the time of diagnosis of the proband’s diabetes, whereas with the present design, these siblings were excluded when the relative incidence rates for the first positive seroconversions were assessed. The results would have been more informative if the siblings had been monitored from birth, which would be possible only in the context of a very extensive survey based on the general population. With the present study design, we cannot exclude the possibility of transient seroconversions in some siblings before the current observation was initiated. Second, although the protocol laid down that serum samples should be obtained at 3- to 6-month intervals during the first 2 years, some positive and inverse seroconversions might have been missed as a result of excessively long sampling intervals, especially after the first 2 years, and in individuals from whom not all of the scheduled samples were obtained. Some transient seroconversions involving IA-2A and GADA will also have been missed because only the first and last samples were analyzed for these autoantibodies in siblings who were negative for all four. Third, as our results indicate, seroconversions seem to represent a rare phenomenon even in siblings of children with type 1 diabetes, so despite the large number of siblings monitored, the absolute number of seroconversions remained so low that we do not have enough statistical power to exclude possible differences between certain subgroups. Fourth, the estimates of relative incidence rates of first observed seroconversions may not be ideal measures for the occurrence of seroconversion, because autoantibody positivity, once it has taken place, is not a stable condition, but there seemed to be fluctuations from positivity to negativity and vice versa during the follow-up in some individuals. The relative incidence rates reported here describe only the first positive and first inverse seroconversions observed during follow-up, regardless of the subsequent dynamics of the autoantibodies. Although the performance characteristics of the autoantibody assays have improved worldwide, the assays do not give perfect sensitivity, specificity, and reproducibility figures (27). A crucial issue in a study such as this one is the assay stability over time. In the present study, the samples from the same patient were, with a few exceptions, analyzed in the same assay for IA-2A and GADA, thus minimizing the problem mentioned. Special attention was paid to assay stability in the IAA and ICA assays by analyzing regularly high and low standards, thereby assessing the assay drift over time. Samples of siblings with inverse seroconversion(s) were analyzed in the same assay whenever possible. However, when dichotomization into positive and negative samples is used, such as in this study, some misclassification is always to be expected, because biological variance and random error is next to impossible to distinguish. This misclassification is presumably nondifferential with regard to the explanatory variables, so its likely effect is dilution of the RR estimates toward 1. The problem is perhaps more severe regarding the IAA levels than with the other antibodies, because the fluctuation between negativity and positivity in IAA is more common (see Table 2) than elsewhere and because a slightly higher proportion of IAA levels fell between the 95th and 99.5th percentiles than IA-2A and GADA levels. Despite these limitations, our results provide valuable information on the dynamic pattern of disease-associated autoantibodies and on factors that affect the susceptibility to positive and inverse seroconversions in siblings of children with type 1 diabetes. The consistency of the results obtained indicates that they do reflect a biological phenomenon. We also believe that the major results would basically remain unchanged despite the choice of cutoff point. The HLA DR phenotype would remain the strongest predictor of the occurrence of seropositivity. As to the other covariates, the numbers of positive cases would still be fairly small for precise estimation of their effects, these being probably more modest than that of the HLA type anyway.

HLA genes are the major genetic determinant of type 1 diabetes, and they seem to be capable of modulating the initiation and progression of β-cell autoimmunity (28). Humoral autoimmunity against β-cells also seems to be modulated by HLA genes, because IA-2A and IAA have been shown to be associated with HLA DR4/(DQA1*0301/B1*0302) (29,30) and GADA with DQA1*0501/B1*0201 in patients with newly diagnosed type 1 diabetes (29). In the present study, the occurrence of seroconversions to IA-2A, GADA, IAA, and ICA positivity were relatively closely associated with the HLA DR3/DR4 phenotype. The positive seroconversions were also associated with HLA identity with the proband, but the HLA DR phenotypes predisposed the participants to autoantibody positivity more strongly than HLA identity. Inverse seroconversions seemed to be more common among siblings who carried an HLA phenotype other than DR3/DR4, supporting a previous observation (31). It seems logical that the siblings with the strongest HLA-defined genetic disease susceptibility will develop β-cell autoimmunity, whereas siblings with low or decreased genetic risk will tend to have transient signs of β-cell autoimmunity, if any. Because of the relatively limited number of autoantibody-positive siblings, it is not possible to reliably assess whether the same associations between given autoantibodies and HLA-defined genetic markers are present in siblings as in patients with newly diagnosed type 1 diabetes.

It has been claimed that autoantibody seroconversions occur early in childhood and that autoantibodies are stable in older children. This assumption is likely based on preliminary data on the natural history of humoral islet autoimmunity in relatives of patients with type 1 diabetes (32,33). The results of our prospective follow-up study do not support this hypothesis. We found that both positive and inverse seroconversions occurred at any age up to nearly 20 years. Positive seroconversions were, however, observed more often in children who were <6 years of age, whereas inverse ones were most frequent in older siblings.

The results point to a remarkable dynamic course for type 1 diabetes–associated humoral autoimmunity. Initial positivity for at least one autoantibody seems to be a risk factor for subsequent seroconversion to IA-2A, GADA, IAA, and ICA positivity, although positive seroconversions also occur in siblings with no autoantibodies at the time of diagnosis of type 1 diabetes in the proband. However, previous data suggest that inverse seroconversions occurred seldom in relatives with additional autoantibodies (31). In our series, inverse seroconversions were most frequent in siblings with no additional autoantibodies at seroconversion, whereas seroconversion was still possible, although rare, in siblings with at least one other autoantibody. Because of the dynamic course of humoral autoimmunity, the predictive value for future type 1 diabetes of a single serum sample obtained at any time point may be questioned. A more accurate prediction may be achieved by obtaining serum samples from siblings at or close to the clinical presentation in the proband (15). Even then, nearly half or more of the autoantibody-positive siblings do not develop type 1 diabetes when the predictive value of a single antibody is used. This may be due partly to the inverse seroconversions seen here. In intervention trials, the inclusion criteria are usually based on autoantibody positivity, and the disappearance of autoantibodies may be used as an outcome measure. This survey shows, however, that autoantibodies may disappear spontaneously without any intervention. On the other hand, the occurrence of inverse seroconversions suggests that the autoreactive immune response can be manipulated and switched off, as this also seems to occur spontaneously on occasions.

Enterovirus (34,35) and rotavirus (36) infections have been implicated in the induction of β-cell autoimmunity. The present observation that positive seroconversions seemed to be more frequent in siblings from families with four or more children than in smaller ones is in line with this idea, because exposure to infections is increased in large families. Moreover, inverse seroconversions tended to accumulate in siblings from small families.

Although the incidence of type 1 diabetes is slightly higher in boys than in girls in high-incidence countries such as Finland (37), it has been reported that the prevalence of GADA and multiple autoantibodies at the time of diagnosis of type 1 diabetes is even higher in girls than in boys (21), whereas the frequency of IA-2A is similar in both sexes (1). In our series, an increased inverse seroconversion rate for GADA was seen in boys, but possible other sex differences in seroconversion rates cannot be excluded based on the present data.

In conclusion, our observations on the dynamics of diabetes-associated autoantibodies in an extensive, representative series of initially nondiabetic siblings of children with type 1 diabetes showed that single autoantibodies were detectable in 8–13% of these siblings. Age, HLA DR phenotype, and number of other autoantibodies were the most conspicuous determinants of seroconversion rates. The complexity of individual risk assessment of type 1 diabetes can partly be explained by the phenomenon of positive and inverse seroconversions in individuals with signs of humoral β-cell autoimmunity.

Principal investigators: H.K. Åkerblom and J. Tuomilehto; coordinators: R. Lounamaa and L. Toivonen; data management: E. Virtala and J. Pitkäniemi; local investigators: A. Fagerlund, M. Flittner, B. Gustafsson, M. Häggquist, A. Hakulinen, L. Herva, P. Hiltunen, T. Huhtamäki, N.-P. Huttunen, T. Huupponen, T. Joki, R. Jokisalo, M.-L. Käär, S. Kallio, E.A. Kaprio, U. Kaski, M. Knip, L. Laine, J. Lappalainen, J. Mäenpää, A.-L. Mäkelä, K. Niemi, A. Niiranen, A. Nuuja, P. Ojajärvi, T. Otonkoski, K. Pihlajamäki, S. Pöntynen, J. Rajantie, J. Sankala, J. Schumacher, M. Sillanpää, M.-R. Ståhlberg, C.-H. Stråhlmann, T. Uotila, M. Väre, P. Varimo, and G. Wetterstrand; special investigators: A. Aro, M. Hiltunen, H. Hurme, H. Hyöty, J. Ilonen, J. Karjalainen, M. Knip, P. Leinikki, A. Miettinen, T. Petäys, L. Räsänen, H. Reijonen, A. Reunanen, T. Saukkonen, E. Savilahti, E. Tuomilehto-Wolf, P.Vähäsalo, and S.M. Virtanen.

This research was supported by the Alma and K.A. Snellman Foundation (Oulu, Finland), the Finnish Medical Foundation (Helsinki, Finland) the Foundation for Pediatric Research (Helsinki, Finland), the Juvenile Diabetes Foundation International (Grant 197032), The Novo Nordisk Foundation, and the Medical Research Council, the Academy of Finland. The Childhood Diabetes in Finland (DiMe) project has been supported by grants from the National Institutes of Health (Grant DK-37957), the Sigrid Jusélius Foundation, the Association of Finnish Life Insurance Companies, the University of Helsinki, and Novo Nordisk A/S (Bagsvaerd, Denmark).

We thank Susanna Heikkilä, Sirpa Anttila, Riitta Päkkilä, and Päivi Salmijärvi for technical assistance; Matti Ronkainen, M.Sc., for critical comments; and Marja-Leena Hannila, M.Sc., for help in statistical matters.