The High-Risk Type 1 Diabetes HLA-DR and HLA-DQ Polymorphisms Are Differentially Associated With Growth and IGF-I Levels in Infancy: The Cambridge Baby Growth Study

Type 1 diabetes is a complex disease manifested in genetically susceptible individuals, by autoimmune destruction of pancreatic β-cells, resulting in insulin deficiency. The pathogenesis of this condition may begin in early childhood with the appearance of one or more insulin autoantibodies, IA-2 antigen, and GAD antibody (GADA) at a median age of 3 years (1). The genetic component is largely attributed to allelic variants in the HLA class II locus, with the highest risk conferred by the DRB1*04 (DR4), DRB1*03 (DR3), and DQB1*03:02 (DQ8).

Almost 40 years ago, Court et al. (2) suggested that within the diabetic spectrum of genotypes, the HLA antigens are relevant to growth and development. The observation that HLA genes are associated with fetal growth rates in studies on human pregnancy (3) provided evidence of the HLA effect on intrauterine development. Fetal growth rate is crudely represented by birth weight, and large birth cohort studies have established that high birth weight is associated with risk of type 1 diabetes, whereas low birth weight is ascribed a protective effect (4,5). In addition, studies in infancy have provided evidence, albeit heterogeneous, of associations between risk of type 1 diabetes, or its prodrome of islet autoimmunity, and postnatal measures or gains of weight and height (6–9).

The explanation for the potential relationship between risk of diabetes in childhood and early growth remains elusive. IGF-I and insulin, which share considerable sequence homology and signal through structurally similar receptors, are critical growth factors in early life. It has been hypothesized that high growth rates increase the demand for insulin production, which acts as a stress factor on the insulin-secreting cells, an effect coined “β-cell overload,” and accelerates the process of β-cell destruction (10). The overload hypothesis has recently resurfaced through The Environmental Determinants of Diabetes in the Young (TEDDY) study of newborns with HLA-conferred risk, which identified distinct patterns of infancy growth (weight and height) in children who progress to autoantibody positivity and type 1 diabetes (1). Other reports have found low IGF-I concentrations in autoantibody-positive children with HLA-conferred risk of type 1 diabetes (11) as well as in those with an established diagnosis (12), but the cause and effect remain unclear.

The putative association between early growth and diabetes-related HLA types has been directly analyzed in prospective cohorts that recruited children at genetic risk of type 1 diabetes or registry-based populations of infants stratified by HLA type, which are summarized in Supplementary Table 1. High-risk HLA haplotypes were associated with a high relative weight or length at birth in Scandinavian populations (13–15); however, other studies did not generate evidence of an association (16,17). In contrast, a retrospective study in Norway identified an inverse association, that is, lowest birth weight was observed in the high-risk HLA carriers (18). The TEDDY study raised the possibility that the HLA influence on birth size varies by country (19). Intriguingly, HLA types that protect against type 1 diabetes were associated with higher birth weight (13,17), but such direction of effect does not confirm the notion that protection relates to reduced fetal growth and ensuing low birth weight. The (DIABIMMUNE) Pathogenesis of Type 1 Diabetes - Testing the Hygiene Hypothesis study reported reduced postnatal weight and height gains in infants with the highest HLA-conferred risk (20). However, a Swedish study lent no support to the effect of HLA on infancy growth (15).

The conflicting findings could be attributed to discrepancies in study design or the diverse HLA class II constitution. Additionally, the genetic effects on growth might be modulated by other factors. A population-based study found that recurrent infections in pregnancy modulated the associations between birth weight and high-risk HLA and between birth weight and islet autoantibodies (14). The link between early growth and islet autoantibodies has also been suggested by other studies in infancy (13,21). These findings illuminate the complex interplay of factors pertaining to the environment and disease prodrome, which possibly interact with genetic effects on growth that might be masked in populations enriched with children with prediabetes.

Based on existing knowledge, it re mains unclear whether the HLA region that confers heightened genetic risk of developing type 1 diabetes influences prenatal and/or postnatal growth and the underlying endocrine factors. Our aim was to conduct a thorough investigation of associations between high-risk HLA polymorphisms and weight, length, skinfold thickness (indicator of adiposity), and concentrations of IGF-I and IGF-binding protein-3 (IGFBP-3) in a prospective cohort of infants from the general population both cross-sectionally at discrete ages and longitudinally in infancy. We envisaged that this exploratory approach would help to test the validity and probe into the mechanisms of the reported association between size at birth and/or early growth and the HLA polymorphisms that predispose to type 1 diabetes.

The Cambridge Baby Growth Study (CBGS) is a prospective observational birth cohort study conducted at a single center. The study’s aim was to investigate the antenatal and postnatal determinants of infancy growth from birth in children from the general population of Cambridgeshire, U.K. (22). Between 2001 and 2009, expectant mothers aged ≥16 years were recruited at their first routine antenatal ultrasound clinic appointment at 12 weeks of gestation and participated voluntarily in a formal 75-g oral glucose tolerance test at 28 weeks of gestation. The participants’ offspring were assessed by trained pediatric research nurses at a minimum of one study visit scheduled within 8 days of birth and at ages 3, 12, 18, and 24 months. More than 95% of the cohort was White. The study was approved by the Cambridge local research ethics committee, and all mothers gave written informed consent. The data analyzed herein were based on a subcohort of 612 children who had provided a biological specimen (cord blood or infancy capillary blood) for DNA extraction (Supplementary Fig. 1). Selection of samples to genotype was based on DNA availability.

Weight, length, and skinfold thickness were routinely measured in triplicate at each study visit except for birth weight, which was retrieved from maternity hospital records. Weight was measured to the nearest 1 g by using a Seca 757 electronic digital scale (Seca, Birmingham, U.K.). Length was measured to the nearest 0.1 cm by using a Seca 416 infantometer. Skinfold thickness was measured at four sites (triceps, subscapular, flank, and quadriceps) on the left side of the body (Supplementary Table 2) by using a Tanner/Whitehouse Skinfold Caliper (Holtain, Pembrokeshire, U.K.). Perinatal questionnaires captured maternal anthropometry and mode of infant feeding (breast milk only or formula milk) at age 3 months. Capillary blood was obtained from age 3 months onward by heel prick (Tenderfoot; ELITech, London, U.K.), part of which was transferred to a vial and the remaining blotted onto Whatman 903 filter cards (Whatman, Maidstone, U.K.). IGF-I and IGFPB-3 concentrations were measured in dried blood spot samples by radioimmunoassay as described previously (23).

DR and DQ genotypes were determined in 612 DNA samples. Genomic DNA was extracted using a chloroform-based method and quantified using PicoGreen. We used a three-tag single nucleotide polymorphism (SNP) model (N.M. Walker, unpublished data) to capture the highest risk in the HLA region: rs17426593 for DR4, rs2187668 for DR3, and rs7454108 for DQ8. The SNPs were genotyped using the Kompetitive Allele Specific PCR genotyping assay by LGC Genomics (Hoddesdon, U.K.). Genotype data passed our study-wide quality control criteria, which were as follows: missing rate <5%, minor allele frequency >1%, and Hardy-Weinberg equilibrium P > 0.05 as calculated from the goodness-of-fit χ² test. Call rates, calculated upon excluding samples of poor DNA quality as designated by LGC Genomics (n = 13), ranged from 95 to 97%. Minor allele frequency was 0.19 for DR4 (T>C), 0.13 for DR3 (G>A), and 0.10 for DQ8 (T>C). Genotype frequencies are summarized in Supplementary Table 3.

Weight and length were converted to age- and sex-appropriate SD scores (SDSs) (also known as z-scores) as well as corrected for gestational age for outcomes at birth and 3 months only on the basis of the U.K. 1990 growth reference using LMS growth software (24). Measurements of skinfold thickness at each anatomical site (Supplementary Table 4) were converted to internal SDSs adjusted for sex and age at measurement as well as for gestational age for outcomes at birth and 3 months only and averaged into an overall skinfold thickness index as an indicator of adiposity (25). The Method algorithm for collecting and preparing data for association analyses is presented in Supplementary Fig. 2.

The statistical analysis to examine the effect of high-risk HLA variants on infancy growth and IGF concentrations was restricted to 586 children born between August 2001 and February 2009, after excluding infants who were not linked to phenotype data (n = 15) or were subject to extremes of growth trajectories (n = 11) resulting from maternal type 1 diabetes, gestational age <36 weeks, and/or multifetal pregnancies. Although late preterm infants (36 weeks of gestation) were included in the analysis, the mean gestational age of the subcohort analyzed was 40 weeks at birth. Comparison of outcome measures by sex was performed by t test for growth parameters and Mann-Whitney U test for IGF concentrations. Of the 586 infants included in the analysis, only 212 (36%) had known IGF-I and IGFBP-3 concentrations at age 3 months (Supplementary Fig. 1), which reflects the proportion of the entire CBGS cohort with IGF levels measured at the same age (34%) by virtue of availability of dried blood spot samples. The proportions of children who had IGF-I and IGFBP-3 concentrations measured in the genotyped subcohort versus the entire CBGS cohort remained comparable, if not slightly higher, at subsequent ages (27–28% vs. 23% at 12 months; 23% vs. 17–18% at 18 months; 15% vs. 9–10% at 24 months).

Parsimonious models were built for each outcome measure and tag SNP fitted as a continuous variable. Cross-sectional association analyses between anthropometric parameters or IGF concentrations for every age at measurement (explained variable) and tag SNP (explanatory variable) were performed by using a multivariate linear regression model with adjustment for preselected covariates. Longitudinal analyses relating the 1) growth rate between either birth to age 3 months or birth to age 24 months or 2) the rate of change in IGF concentrations between ages 3 and 24 months and tag SNP were performed by using a linear mixed-effects (LME) model with random intercept and fixed slope and adjustment for preselected covariates. Our approach was reinforced by a report that showed that a mean-based statistic is more powerful in detecting a genetic association than a slope-based statistic (26). The aim of longitudinal analysis was to test for the null hypothesis of no SNP-age interaction effect (i.e., the growth trajectories are parallel between genotypic groups of subjects). Models were fitted using maximum likelihood estimation.

Additive genetic effects were assumed (2 = minor homozygote, 1 = heterozygote, 0 = major homozygote), with the minor allele conferring risk of type 1 diabetes. With additive genetic effects, the standardized regression coefficient is interpreted as the change in the outcome measure for each additional risk allele; the estimate of the SNP-age interaction is interpreted as the per-allele change in the outcome measure per unit time.

Predictors to adjust for were determined a priori in the entire CBGS population of infants on the basis of biological reasoning and a mix of statistical workings between outcome measures and potential phenotypic covariates: pairwise associations, multivariate linear regressions on phenotypes, and model selection based on the Akaike information criterion. Weight and length models were adjusted for parity, maternal smoking during pregnancy, mode of infant feeding at age 3 months, maternal prepregnancy weight, and maternal height. Skinfold thickness models were adjusted for parity and mode of infant feeding. IGF models were adjusted for sex and mode of infant feeding. For cross-sectional models, adjustments depended on age at measurement: maternal smoking during pregnancy was included in models at birth, parity in models at birth and 3 months, and mode of infant feeding in all models except at birth.

Statistical analyses were performed using SPSS for Windows version 23.0 (IBM Corporation, Armonk, NY) and R 3.3.2 (R Foundation for Statistical Computing, Vienna, Austria) software. Corrections for multiple comparisons were not performed because this was an exploratory study with intercorrelated multiple outcome measures.

Compared with non-genotyped offspring, genotyped newborns were similar with respect to demographics except that the genotyped subcohort had, on average, a higher occurrence of normal birth delivery (P < 0.0001), lower frequency of twins (P = 0.001), lower frequency of premature birth (P = 0.007), and longer gestational length (P = 0.002), which was not clinically significant (mean difference 0.3 weeks) as evidenced by the absence of between-group differences in weight SDS or length SDS at birth (Supplementary Table 5). Overall, 586 genotyped children (315 boys and 271 girls) were included in the analysis. Table 1 displays the outcome measures of participating children stratified by sex at each study visit. Despite modest differences in the IGF-I and IGFBP-3 concentrations between males and females, the counts of the respective samples were too small to allow for sex-specific analysis.

In cross-sectional models, there was no significant effect of the SNPs tagging DR4, DR3, or DQ8 on birth size (Table 2). Postnatally, greater skinfold thickness at the end of infancy was associated with the SNPs rs17426593 tagging DR4 (β = 0.15, P = 0.003) and rs7454108 tagging DQ8 (β = 0.12, P = 0.02).

Longitudinal models between birth and age 3 months did not yield significant associations (Table 3). Modeling across the 24 months of infancy identified a significant SNP-age interaction for rs2187668 tagging DR3 and postnatal linear growth (estimate 0.06 ± 0.02, P = 0.007) (Table 3), which is reflected in different slopes of length with age across genotypic groups, with minor homozygotes showing a significant upward trajectory (Fig. 1). SNP rs2187668 reached nominal significance with increased length at age 24 months (β = 0.12, P = 0.02). Analysis by sex showed that the SNP-age interaction for rs2187668 with linear gains was stronger in boys (estimate 0.14 ± 0.03, P = 3 × 10⁻⁷) than girls (estimate −0.06 ± 0.03, P = 0.07), and the cross-sectional association with length at age 24 months was restricted to boys (β = 0.20, P = 0.002) versus girls (P > 0.9). We tested for the interaction of rs2187668 with sex and age by introducing a SNP ∗ age ∗ sex–interaction term to an LME model for unadjusted length (with model adjustments for sex and predetermined covariates as described in Research Design and Methods) and found that it was significant for boys (estimate 0.32 ± 0.12, P = 0.009) but not girls (P = 0.6). In the cross-sectional model of rs2187668 on unadjusted length at age 24 months, the SNP-sex interaction was of borderline significance (β = −0.28, P = 0.06).

There was evidence of associations between IGF-I concentrations and the SNPs rs7454108 tagging DQ8 (β = −0.23, P = 0.003) and rs2187668 tagging DR3 (β = 0.30, P = 0.009) at ages 12 and 24 months, respectively, albeit in opposite directions. Longitudinal analysis of IGF-I concentrations strengthened the association with rs2187668 (estimate 5.62 ± 1.83, P = 0.002), which is reflected in a more pronounced increment in IGF-I levels over time in homozygotes for the minor DR3 allele compared with other genotypic groups. Similarly, rs2187668 showed a pronounced increment in longitudinal IGFBP-3 levels by addition of a copy of minor allele (estimate 106.89 ± 35.23, P = 0.003), albeit an isolated association in the opposite direction was detected at age 3 months (β = −0.21, P = 0.003).

We have taken a different approach to the investigation of the debated relationship between early growth and the high-risk type 1 diabetes HLA through analyses of longitudinal anthropometric measures (weight, length, and skinfold thickness) and circulating endocrine factors (IGF-I and IGFBP-3) in a general population-based cohort of infants by considering the SNPs tagging the HLA alleles independently. Cross-sectional analyses identified positive associations between skinfold thickness at the end of infancy and the SNPs tagging DR4 and DQ8. Longitudinal models showed significant upward linear trajectories in children carrying two copies of the minor DR3 allele, accompanied by pronounced increases of IGF-I and IGFBP-3 concentrations. Comparison of longitudinal analyses over the first 3 versus 24 months of life suggested that the timing of a discernible HLA effect on growth is the latter part of infancy.

Prior studies explored the possibility that the HLA confers susceptibility to type 1 diabetes through its association with higher birth weight, but the results have been inconsistent (13–18). The TEDDY study recently reported that adjusted birth weight is positively associated with risk of appearance of islet autoantibodies, predominantly GADA (1). Prior investigations found that GADA was negatively associated with risk of higher relative birth weight (13,14). Direction of association aside, our findings, albeit in the absence of data on autoantibody status, do not lend support inferentially to the hypothesis that the high-risk HLA mediates the putative relationship between birth weight and development of HLA-conferred islet autoimmunity. This could be explained by a small magnitude of the HLA effect on birth size or the involvement of susceptibility non-HLA loci instead.

In the CBGS, significant associations were found postnatally. The minor DR4 allele was associated with increased adiposity only at the end of infancy (P = 0.003), concurring with a small study suggesting that adiposity at birth has no influence on the development of type 1 diabetes (27). Our result raises the question of whether the relationship observed between infancy weight measures or weight gains and development of islet autoimmunity in cohorts of at-risk children (1,6,8) reflects gains in adipose tissue and is dependent on HLA genotypes.

An important finding of our study is that the minor DR3 allele was associated longitudinally with increased linear growth in infancy (P = 0.007) and cross-sectionally with greater length at age 24 months (P = 0.02). The timing of this effect may reconcile prior findings on length/height gains in type 1 diabetes cohort studies. The TEDDY study reported that risk of developing type 1 diabetes is associated with increased height growth in early childhood (1). A registry-based study found no association between gains in length from birth to 12 months and development of type 1 diabetes (9), raising the question of whether an association could have been found after mid-infancy.

The mechanisms behind the influence of the HLA on linear growth are unknown, but our study found that the minor DR3 allele was also associated with greater longitudinal gains in IGF-I concentrations in infancy, which appeared to accumulate over time and attained significant correlations with higher IGF-I levels at ages 18 (P = 0.03) and 24 months (P = 0.009). The interpretation of cross-sectional results on endocrine factors should be made with caution because of the lack of children in the subcohort that carried two copies of the minor DR3 allele and had IGF-I concentrations measured at 24 months. However, longitudinal modeling showed in parallel faster increases of IGFBP-3 concentrations by progressive addition of a copy of minor DR3 allele (P = 0.003). In contrast, the minor DQ8 allele was associated with lower IGF-I concentrations at age 12 months (P = 0.003). The role of IGF-I in the possible pathogenesis of type 1 diabetes has been revived through recent reports that IGF-I levels decrease longitudinally with autoantibody positivity (12) and are lower in carriers of high-risk HLA genotypes in infancy (11), which echoes the direction of association found here with the SNP tagging DQ8. Our approach of analyzing SNPs tagging the HLA region independently underscores the heterogeneity between the etiology of “DR4-associated” and “DR3-associated” disease that has been long proposed (28). The excursions of length and, arguably, IGF-I and IGFBP-3 concentrations in infancy in relation to DR3 open the possibility that DR3-conferred risk influences dysregulation of IGF-I production/bioactivity and is phenotypically manifested by increased linear growth.

IGF-I promotes bone formation by inducing osteoblastogenesis and reducing apoptosis of osteoblasts, which are responsible for synthesizing the bone matrix (29). It has been known that IGF-I and growth hormone (GH) stimulate production of proinflammatory cytokines in human osteoblasts (30) as well as immune cells (31,32), and children with chronic inflammatory conditions show elevated anti-inflammatory and proinflammatory cytokines in parallel with abnormalities in the GH/IGF-I axis (33). In the CBGS, the timing of the association found between DR3 and gains in length or IGF-I concentrations in mid- to late-infancy might be explained by the timing of GH receptor expression that occurs after 6 months of age. Larsen and Alper (34) expressed the view that the HLA-DRB1, DQB1 alleles are simply markers for type 1 diabetes risk rather than the susceptibility genes themselves. Evidence of linkage disequilibrium between the high-risk SNPs herein and polymorphisms in genes other than the HLA that could relate with growth was absent from public databases (including the GWAS Catalog and LDlink). This suggests that our result is unlikely explained by association of alleles on chromosome 6 but rather points to the implication of the HLA-DRB1, DQB1 molecules in a biological pathway where immune mediators, such as cytokines, regulate IGF-I (35,36) and vice versa.

Sex-specific analyses unveiled that the influence of the minor DR3 allele on postnatal linear gains was more pronounced in boys (P = 3 × 10⁻⁷), which aggregated over time to bring about material differences in increased length at the end of infancy in boys but not girls. Observational studies in diabetes cohorts reported that height gains between probands developing type 1 diabetes and control subjects were more marked for boys than girls (7,37). The bias in male incidence of type 1 diabetes has been attributed to the DR3 type (28). Male sex was further found to convey increased risk of development of islet autoimmunity and multiple autoantibodies (38). Our data from the general population reinforce the putative link between DR3 and linear growth, male sex, and, potentially, IGF-I. The well-known differences between male and female immune systems, which translate to sex-dimorphic resilience to infectious and noninfectious diseases (39), involve the HLA complex, which could account for sex-dimorphic growth and endocrine phenotypes.

The main strength of our study lies in the advantage of a longitudinal study design to gain insight into the natural history of biological processes that start early in life and progress at variable rates by HLA genotypes made possible by the integration of anthropometric, endocrine, and genetic data. Longitudinal analysis has overcome the limitation of the attrition of growth measurements collected by age by virtue of using more data and achieving statistical power gains. There are a couple of limitations worthy of mention: 1) The CBGS did not systematically collect data on immune function (e.g., viral infections), and 2) the frequencies of the high-risk HLA polymorphisms are too low in the general population to allow for conclusive findings. Future studies should seek to investigate the differentiating effect of high-risk HLA loci on growth and endocrine factors in well-powered cohorts in infancy and early childhood that allow for sex-specific analysis. Because our results are largely drawn from European populations, further studies are required to determine the implications for other populations, especially those with a high underlying risk of type 1 diabetes.

This study makes the case for the mechanistic plausibility that the high-risk type 1 diabetes HLA, or an associated immune molecule, is intricately linked to IGF-I production and systemic growth. General population-based studies conducted in infancy, when genetic effects are strongest, offer the potential to tease out mechanisms behind autoimmune diseases in childhood, which are non-Mendelian, by pinpointing changes in endocrine profiles owing to allelic variants that predispose to autoimmunity. Various autoimmune diseases in children are characterized by abnormalities in the GH/IGF-I axis that are believed to be mediated by proinflammatory cytokines (33). It is also probable that the mechanism rests with the microbiome. A recent study identified distinct associations between type 1 diabetes HLA genotypes and the gut microbiome (40), which could mediate metabolic effects on growth. Whereas aggregate gene scores are becoming the mainstay of risk stratification in prospective cohorts of at-risk children, independent genotypes of HLA loci would help to disentangle distinct mechanisms behind the coincident HLA-conferred predispositions to type 1 diabetes and early growth/IGF-I derangement, with the potential for defining endotypes for stratified medicine before the development of islet autoantibodies.

Acknowledgments. The authors acknowledge Dr. Carlo L. Acerini with the Department of Paediatrics at the University of Cambridge for contributions to the conception and design of the cohort. The authors are grateful to all the families who participated in the CBGS; research nurses Suzanne Smith, Ann-Marie Wardell, and Karen Forbes and laboratory staff Dianne L. Wingate and Karen Whitehead at the Department of Paediatrics, University of Cambridge; the staff at the National Institute for Health Research Cambridge Biomedical Research Centre, Cambridge, U.K.; Keith Burling, Peter Barker, and the staff at the Core Biochemical Assay Laboratory at the University of Cambridge; and the midwives at the Rosie Maternity Hospital, Cambridge, U.K. The authors also thank Dr. Tengyao Wang with the Statistical Laboratory of the Centre for Mathematical Sciences at the University of Cambridge for providing statistical advice.

Funding. This work was supported by the Medical Research Council (MR/K50127X/1) and the Raymond & Beverly Sackler Foundation. The CBGS has been funded by the European Union Framework 5 (QLK4-1999-01422), the Medical Research Council (7500001180, G1001995, U106179472), and the World Cancer Research Fund International (2004/03). We also acknowledge support from National Institute for Health Research Cambridge Biomedical Research Centre. K.K.O. is supported by the Medical Research Council (MC_UU_12015/2). D.B.D. is supported by funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement no. 115797 (Innodia) and no. 945268 (Innodia Harvest).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.E. contributed to the study design, designed and performed the analyses, interpreted the results, and wrote the manuscript. C.J.P. contributed to data generation and critically reviewed the manuscript. I.A.H., K.K.O., and D.B.D. contributed to the cohort conception and study design and critically reviewed the manuscript. D.B.D. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.