Introduction

Type 1 diabetes is an autoimmune disease mediated by T helper cells. During the process preceding clinical diabetes, the immune system has a bias towards a Th1 response wherein autoimmunity exists but there are no overt clinical signs of type 1 diabetes. Vitamin D has been examined as a potentially protective factor because it plays an active role in the regulation of the immune system [1]. Mechanistically, vitamin D has been shown to shift the balance of the body’s T cell response towards downregulation of the Th1 immune response [2]. Both in vitro studies and animal studies have found that vitamin D stimulates a Th2 response [35].

A number of epidemiological reports suggest a role for vitamin D in the pathogenesis of type 1 diabetes. A case–control study found that fish oil supplementation in infancy was associated with a lower risk of type 1 diabetes [6]. As fish oils contain both n-3 fatty acids and vitamin D, it is not possible to attribute this association to one specific component. Two additional studies found that diabetic children were less likely to have received vitamin D supplements during infancy than non-diabetic children [7, 8]. Vitamin D intake from either foods [9] or supplements [10] during pregnancy was associated with a lower risk of islet autoimmunity (IA) in the offspring, although a third study was not consistent with these findings [11]. While these studies have examined vitamin D exposures very early in life, with conflicting results, it is not known whether later childhood exposures to vitamin D are associated with the risk of IA or type 1 diabetes. Moreover, reliance on intake alone to assess exposure may not be an accurate measure of vitamin D status, as this does not account for non-dietary sources of vitamin D.

The data regarding 25(OH)D levels have been inconsistent. Three cross-sectional studies have shown lower 25(OH)D levels among newly diagnosed type 1 diabetes cases compared with controls [1214], although another found no association between serum 25(OH)D and type 1 diabetes [15]. In addition, two clinical trials reported no effect of 1α,25-dihydroxyvitamin D3 supplementation on sustained insulin production among persons with new-onset type 1 diabetes [16, 17]. However, no inferences can be made from these studies about 25(OH)D before onset of disease.

The purpose of this study was to examine the association between two assessments of vitamin D: reported dietary intake of vitamin D and plasma 25(OH)D measured throughout childhood; and the time to development of IA, and the progression to type 1 diabetes in IA-positive children.

Methods

In order to investigate the association between vitamin D, IA and type 1 diabetes, we performed a number of studies using cohorts derived from the Diabetes Autoimmunity Study in the Young (DAISY) population, as described below. Figure 1 is a flow chart that describes these studies and their respective study cohorts.

Fig. 1
figure 1

Flow chart illustrating the formation of the cohorts for the vitamin D studies from the DAISY population. T1D, type 1 diabetes

DAISY population

DAISY is a prospective study of two groups of children at increased risk of developing type 1 diabetes. One group consists of first-degree relatives of patients with type 1A diabetes mellitus, recruited between birth and 8 years of age. The second group consists of babies born at St Joseph’s Hospital in Denver, CO, USA, whose umbilical cord blood was screened for diabetes-susceptibility alleles in the HLA region. The St Joseph’s Hospital newborn population is representative of the general population of the Denver metropolitan area. The details of the newborn screening [18] and follow-up [19] have been published elsewhere. Cord blood or the first available blood sample (depending on enrolment group) is sent to Roche Molecular Systems (Alameda, CA, USA) for PCR-based HLA class II typing. Recruitment took place between 1993 and 2006. Written informed consent was obtained from the parents of study participants. The Colorado Multiple Institutional Review Board approved all study protocols.

Measurement of autoantibodies

Autoantibodies were tested at 9, 15 and 24 months, and annually thereafter. Radioimmunoassays were used to measure serum autoantibodies to insulin, GAD65 and IA-2 (BDC512), as previously described [2023], with rigorous duplicate testing and confirmation of all positive and a subset of negative results. The cut-off for positivity was established as the 99th centile of healthy controls. If a participant tested autoantibody positive, they were put on an accelerated testing schedule of every 3–6 months.

Measurement of dietary intake of vitamin D

Current dietary intake of vitamin D in international units (IU) was assessed using one of two instruments. For children aged 2–9 years of age, parents filled out a food frequency questionnaire (FFQ) (Nutrition Questionnaire Service Center, Boston, MA, USA) on the child’s behalf annually. Starting between the ages of 10 and 12 years, children were asked to recall their own diets and complete the youth/adolescent questionnaire (YAQ), a food frequency questionnaire geared toward adolescents [24, 25] that is based on the FFQ. Both surveys collected the average food consumption over the course of the previous year, using structured responses regarding how often a child consumes commonly sized portions (example: one slice of bread), with the response options ranging from ‘never’ to ‘6+ servings per day’. Fluid cow’s milk is a major source of dietary vitamin D in US children, because it is routinely fortified to a level of 400 IU per 0.95 l. We conducted an instrument-comparison study in our DAISY population, in which we determined that data from these two instruments may be combined when an instrument indicator variable (i.e. type of food frequency questionnaire [FFQ vs YAQ]) was included in the model [26]. The FFQ has been validated against multiple 24 h recalls [27] and has been found to correlate with micronutrient [28] and fatty acid [29] biomarkers in our population. Dietary data were linked to an autoantibody measurement if the 1-year time period of the questionnaire encompassed the time directly preceding the clinic visit at which the autoantibody was measured.

Measurement of plasma 25(OH)D levels

Blood was drawn and kept from light at all times during processing. Plasma was separated immediately, snap-frozen in liquid nitrogen, and stored at −70°C until sent to the University of Colorado Pediatric Clinical Translational Research Center Core Laboratory, which has a certificate of proficiency from the Vitamin D External Quality Assessment Scheme. Vitamin D3 [25(OH)D] was measured by radioimmunoassay (DiaSorin, Stillwater, MN USA), with a CV of 7.5%. Quality control was assessed via assay of 109 blinded duplicate plasma samples, and excellent agreement was observed (intraclass correlation coefficient = 0.91). 25(OH)D levels are reported in nmol/l.

Study 1: analysis of vitamin D and risk of IA

The outcome for Study 1 was the development of IA, which is defined as testing positive for at least one of three autoantibodies (described above) on at least two consecutive visits less than 6 months apart. The median number of clinical visits per child at which autoantibody status was ascertained was seven. The median time that elapsed between clinical visits was 356 days. As of February 2011, 198 children developed IA during follow-up of 2,644 DAISY children; however, not all of these children completed a diet survey or had a plasma sample for the measurement of 25(OH)D before their first positive visit (Fig. 1).

Vitamin D intake and risk of IA (Study 1a)

Study 1a investigated whether lower vitamin D intake during childhood was associated with increased risk of IA. This study was conducted using all DAISY children who had completed at least one dietary questionnaire before their first autoantibody positive visit (n = 1,875). The outcome is the development of IA (n = 123). We had multiple measures of vitamin D intake before and including the first IA positive visit, with 203 children with one measure, 210 children with two, 177 children with three, 175 children with four, 187 children with five, 183 children with six, 148 children with seven, 119 children with eight, 93 children with nine, and 380 children with ten or more measures.

All analyses were conducted in SAS for Windows Version 9.2 (SAS Institute, Cary, NC, USA). Using proportional hazards analysis, HRs and 95% CI were estimated for the risk of IA for a SD difference in vitamin D intake. All SDs used for this standardisation technique are listed in the footnote of Table 3. As vitamin D intake was measured longitudinally, we treated it as time-varying in our analyses, such that intake could vary with the clinical visits and reflect change over time in children who were still at risk of IA at a given event time.

All models were adjusted for family history of type 1 diabetes, HLA-DR3/4 DQB1*0302 genotype, average daily dietary energy intake, and type of survey completed (FFQ vs YAQ). In addition, we considered sex, ethnicity (non-Hispanic white vs other), maternal age and maternal education as potential covariates and retained them if they significantly contributed to the model (at p < 0.05).

25(OH)D and risk of IA (Study 1b)

Study 1b used a case–cohort design to investigate whether lower plasma 25(OH)D or vitamin D insufficiency during childhood was associated with increased risk of the development of IA. To conduct the case–cohort study, a representative sample of children was selected from the eligible pool via stratified random sampling based on HLA-DR genotype and family history of type 1 diabetes. On the basis of availability of biological samples and timing of entry into the study, 1,433 DAISY children were eligible for selection, and, of these, we selected 380 children for this subcohort. During follow-up of the subcohort, 20 children developed IA. We supplemented this with 68 IA-positive children who developed in DAISY outside of this subcohort, and who had 25(OH)D measurements before the development of autoantibodies. Therefore 88 children with IA and 353 IA negative children were included in this analysis. The number of IA-positive children in this analysis (n = 88) is less than the overall number in DAISY (n = 198) because 110 children developed IA before we began the protocol of collecting plasma for measurement of 25(OH)D. Electronic supplementary material (ESM) Table 1 shows that the characteristics of the IA-positive children included in the case–cohort analyses are similar to those of the IA-positive children who were not included because of missing 25(OH)D. We were able to measure 25(OH)D in these children after they became IA positive and therefore could include them in Study 2 (described below).

HR and CI were estimated for the risk of IA for a SD difference in 25(OH)D. 25(OH)D was included in the model as a time-varying covariate. To account for the sampling of the case–cohort design, the analyses were weighted using the Barlow method [30] and a SAS macro developed by Ichikawa and Barlow [31] (http://lib.stat.cmu.edu/general/robphreg, accessed 29 July 2011).

Recognising that the relationship between vitamin D status and IA risk may not be linear, we analysed 25(OH)D both as a continuous variable and as a dichotomous vitamin D inadequacy variable, using the cut-off point of 50 nmol/l, as recommended in the recent Institute of Medicine report [32].

All models were adjusted for family history of type 1 diabetes and HLA-DR3/4, DQB1*0302 genotype. Ethnicity, month of blood draw (May–October vs November–April), sex, maternal age and maternal education were considered as potential covariates and retained if they significantly contributed to the model (at p < 0.05).

25(OH)D at 9 months of age and risk of IA (Study 1c)

In order to examine whether vitamin D status during infancy was specifically associated with risk of IA, we analysed the 128 children from Study 1b who had a 25(OH)D measurement at their 9 month clinic visit. Thirty of these children subsequently developed IA, and of these, six went on to develop type 1 diabetes. We treated the 9 month 25(OH)D as fixed in the proportional hazards model. We adjusted for family history of type 1 diabetes and HLA-DR3/4, DQB1*0302 genotype. Month of blood draw (May–October vs November–April), sex, maternal age and maternal education were considered as potential covariates and retained if they significantly contributed to the model (at p < 0.05).

Study 2: analysis of vitamin D and risk of type 1 diabetes in IA-positive children

The outcome in Study 2 is physician-diagnosed type 1 diabetes, and was defined as a random blood glucose >11.1 mmol/l and/or a HbA1c > 6.3% (45 mmol/mol) with clinical symptoms of diabetes. We followed 198 IA-positive children; 62 developed type 1 diabetes during follow-up (Fig. 1).

Vitamin D intake and the risk of type 1 diabetes in IA-positive children (Study 2a)

Study 2a investigated whether lower vitamin D intake increased the risk of type 1 diabetes among IA-positive children. Of the 198 IA-positive children in DAISY, we collected vitamin D intake data on 178, of whom 35 developed type 1 diabetes during follow-up. Using proportional hazards analysis, HR and CI were estimated for risk of type 1 diabetes for a SD difference in vitamin D intake, which was treated as a time-varying covariate. Family history of type 1 diabetes, HLA-DR3/4, DQB1*0302 genotype, age at first appearance of autoantibodies, energy intake and type of survey completed (FFQ vs YAQ) were included in all models. Sex, maternal age, maternal education and ethnicity were examined and included if they significantly contributed to the model (at p < 0.05).

25(OH)D concentration and risk of type 1 diabetes in IA-positive children (Study 2b)

Study 2b examined whether lower plasma 25(OH)D levels were associated with a risk of type 1 diabetes among IA-positive children. We obtained an average of 10 measures of 25(OH)D in 185 IA-positive children during the period in which they were autoantibody positive. Of the 62 children who developed type 1 diabetes during follow-up, seven were not included as diabetes cases in the analysis because we did not have a 25(OH)D measurement within 1 year of diagnosis.

HR and CI were estimated to measure the risk of progression to type 1 diabetes associated with a SD difference in 25(OH)D levels, using proportional hazards analyses. 25(OH)D was analysed both continuously and dichotomously (inadequate vs adequate) as a time-varying covariate. Family history of type 1 diabetes, HLA-DR 3/4, DQB1*0302 genotype and age at first appearance of autoantibodies were included in all models. Ethnicity, maternal age and maternal education were included if they significantly contributed to the model (at p < 0.05).

Study 3: longitudinal analysis of 25(OH)D

In order to test the hypothesis that children who convert to IA positivity have a lower vitamin D status throughout childhood than children who remain IA negative, we created a longitudinal prediction model for 25(OH)D that included factors known to influence vitamin D status. We then used this model to test if 25(OH)D varied throughout childhood by a child’s subsequent autoantibody status.

The potential predictors of 25(OH)D levels that were considered were year and month of sample collection, age of the study participant, sex, ethnicity, vitamin D intake, maternal age and maternal education. Of the 448 children in the case–cohort population (as described in Study 1b above), 47 were excluded because they did not have dietary intake data, leaving 401 children, 85 of whom subsequently developed IA for these longitudinal analyses (Fig. 1). Data from visits after the child became IA positive were excluded.

We analysed the relationship between 25(OH)D and correlates using a linear mixed modelling approach [33]. The procedure discussed by Cnaan et al. [34] was used to determine best fit polynomials for both the fixed and random effects of month of blood draw. These models distinguish variability between participants and variability between repeated measurements over time within participants. The between-participant covariance matrices were restricted to be positive definite. We first developed a model that describes the variation of 25(OH)D on the basis of the month the sample was collected to capture seasonal variation. We then added the remaining variables to this model. The mixed model provides a regression coefficient, a standard error and p value for each variable to indicate its contribution towards explaining 25(OH)D levels.

BMI was initially included in the model and was marginally significantly associated with 25(OH)D in our population (p = 0.056). However, because DAISY did not collect height before the age of 2 years, the early visits had to be excluded from the analysis because of missing BMI data. As the inclusion of BMI in this model did not substantially alter the estimate for the IA status variable, we decided to exclude BMI from our final model so that we could examine the association between 25(OH)D and IA status at very young ages.

Results

Tables 1 and 2 describe the characteristics of the analysis populations used in each study. Figure 2 displays mean 25(OH)D levels and percentage of children who were vitamin D inadequate (<50 nmol/l) in DAISY by age. A summary of the analysis results of Studies 1 and 2 is included in Table 3.

Table 1 Characteristics of the analysis populations used to study the association between vitamin D and the risk of IA in DAISY (Study 1)
Table 2 Characteristics of the analysis populations used to study the association between vitamin D and the risk of type 1 diabetes (T1D) in IA-positive children in DAISY (Study 2)
Fig. 2
figure 2

Unadjusted plasma 25(OH)D levels and percentage of children with inadequate 25(OH)D by age, in a representative cohort of 373 DAISY children. Mean 25(OH)D (nmol/l) of children at each age are represented by a black circle, and the line is intended to connect these points. The right-hand axis lists the values for mean 25(OH)D. The grey bars represent the percentage of children with inadequate plasma 25(OH)D (≤50 nmol/l) at each age. The actual percents are listed on the left-hand axis. The values in parentheses on the x-axis represent the number of children included in each age group. As children had multiple measurements of vitamin D over time, they may be in more than one age group

Table 3 Summary of proportional hazards models of the association between vitamin D and the risk of IA and type 1 diabetes (T1D) in DAISY

Study 1: analysis of vitamin D and the risk of IA

After adjustment for having a first-degree relative with type 1 diabetes, HLA-DR3/4, DQB1*0302 genotype, energy intake and the type of survey filled out (FFQ vs YAQ), intakes of vitamin D from all sources, from foods only, and from supplements only were not associated with the risk of IA (Study 1a in Table 3).

Neither 25(OH)D level nor vitamin D insufficiency was associated with the risk of IA, after adjustment for family history of type 1 diabetes, HLA-DR3/4, DQB1*0302 genotype and ethnicity (Study 1b in Table 3).

Recognising that much of the previous literature had focused on vitamin D in infancy, we examined whether plasma 25(OH)D at 9 months was associated with the risk of IA. Neither 25(OH)D level nor vitamin D insufficiency at 9 months of age were associated with the risk of IA, after adjustment for family history of type 1 diabetes and HLA-DR3/4, DQB1*0302 genotype (Study 1c in Table 3).

Study 2: analysis of vitamin D and risk of type 1 diabetes in IA-positive children

We then examined whether vitamin D may play a role in the progression to type 1 diabetes in children who are autoantibody positive. After adjustment for family history of type 1 diabetes, HLA-DR3/4, DQB1*0302 genotype, the type of survey completed, energy intake and age at first IA positivity, vitamin D intake was not associated with the risk of type 1 diabetes (Study 2a in Table 3).

Neither 25(OH)D levels nor vitamin D insufficiency were associated with progression to type 1 diabetes in IA-positive children, after adjustment for family history of type 1 diabetes, HLA-DR3/4, DQB1*0302 genotype, age at first IA positivity and ethnicity (Study 2b in Table 3).

Study 3: longitudinal analysis of 25(OH)D

We then examined if differences in vitamin D status existed between IA-positive children before seroconversion and IA-negative children. To do this, we first plotted mean plasma 25(OH)D by age in children who became autoantibody positive and in those who remained negative (Fig. 3). We then modelled predictors of 25(OH)D in our population in a longitudinal model, and then plotted the adjusted predicted 25(OH)D by age in children who became IA positive and those who were IA negative (Fig. 3, see legend for regression equation). In summary, 25(OH)D levels followed a yearly pattern best described by a 4th degree polynomial function of month of sample collection; it also decreased by year of visit (β = −1.03, p = 0.0001), increased with higher dietary intake of vitamin D (β = 1.54, p = 0.0002), decreased with increasing age (in years) (β = −0.69, p = 0.004), and were marginally higher in non-Hispanic white children (β = 2.11, p = 0.18). 25(OH)D levels did not differ by IA status (β = 1.08, p = 0.54).

Fig. 3
figure 3

Plasma 25(OH)D levels by age in children who subsequently developed IA (n = 85) and those who remained autoantibody negative (n = 316), using data from 2,207 visits. The symbols represent adjusted mean 25(OH)D levels (in nmol/l); black symbols are mean 25(OH)D in children who subsequently became IA positive; white symbols are mean 25(OH)D in children who remained IA negative. The solid line is the predicted mean 25(OH)D in IA-positive children, and the dotted line is the predicted mean 25(OH)D in IA negative children by age, with adjustment for month and year of the blood draw, ethnicity and dietary intake of vitamin D. The grey shaded areas surrounding each line represent the 95% CIs of the predicted line. The shading of the 95% CI of the IA positive line is lighter than the shading of that of the IA negative line in order to differentiate the two intervals when they overlap. The regression equation used to generate these predicted lines is: 69.23 + 1.71 × month + 1.99 × month2–0.38 × month3 + 0.017 × month4 − 1.03 × year of blood draw–0.69 × age (years) + 2.11 × non-Hispanic white + 1.54 × vitamin D intake (SD of IU/day) + 1.08× IA positivity status. The p values for each covariate are: month (0.62), month2 (0.05), month3 (0.001), month4 (0.0002), year (0.0001), age (0.004), non-Hispanic white ethnicity (0.18), vitamin D intake (0.0002) and IA positivity status (0.54)

Discussion

In our population of children at increased risk of type 1 diabetes, neither dietary intake of vitamin D nor 25(OH)D levels in infancy or throughout childhood were associated with the risk of IA. Moreover, progression to type 1 diabetes in IA-positive children was not associated with vitamin D intake or 25(OH)D levels.

As there were no prospective studies examining 25(OH)D levels and risk of IA or type 1 diabetes, it is difficult to place our negative results in the context of the existing literature. By showing that 25(OH)D levels at 9 months of age were not associated with risk of IA, our findings are somewhat contradictory to previous epidemiological studies that have suggested a protective effect of vitamin D supplementation in infancy on risk of type 1 diabetes [68]. Some caveats to this comparison include (1) our inability to examine a comparable vitamin D supplement variable because the infant supplements taken by DAISY children contained a number of vitamins besides vitamin D, (2) our inability to examine the outcome of type 1 diabetes, and (3) the relatively small number of infants in whom we measured 25(OH)D at 9 months of age. In our analysis of 25(OH)D levels at 9 months of age, the HR for IA was 1.16 for a SD increase in 25(OH)D. The 95% CI (0.69, 1.93) of this HR suggests that our data do not support an hypothesis that increased 25(OH)D levels at 9 months are strongly associated with decreased IA risk (as defined by HRs < 0.69), although weaker associations cannot be excluded.

Negative results mandate consideration of whether noise in the exposure measure biased the association towards the null. We analysed predictors of 25(OH)D in our DAISY children and demonstrated that plasma 25(OH)D is associated with previously documented predictors of 25(OH)D in children [35], suggesting that our measure of 25(OH)D is internally valid.

While there are additional hypotheses that remain to be tested, including the exact role of supplementation during infancy, and whether only very high levels of vitamin D are protective (or conversely, only extremely low levels are a risk), our study, which uses a powerful combination of prospectively collected reports of vitamin D intake and a biomarker of vitamin D status, does not support an association between a child’s usual vitamin D intake or 25(OH)D levels throughout childhood and the risk of IA or progression to type 1 diabetes, in children at increased risk of the disease.