The insulin gene variable number tandem repeat (INS-VNTR) is proposed to exert pleiotropic genetic effects on birth weight and diabetes susceptibility. In our study, we examined the influence of a polymorphism in tight linkage disequilibrium with INS-VNTR (−23Hph1) on birth weight and type 2 diabetes in the Pima population. A parent-offspring “trio” design was used to assess parent-of-origin effects and population stratification. The presence of the −23Hph1 T-allele was associated with lower birth weight (n = 192; −140 g per copy of the T-allele; P = 0.04), even after adjustment for effects of population stratification (P = 0.03). The effects of paternally transmitted T-alleles were greater than those of maternally transmitted alleles (paternally transmitted: −250 g, P = 0.05; maternally transmitted: −111 g, P = 0.43), but this difference was not statistically significant (P = 0.50). The −23Hph1 T-allele was associated with an increased prevalence of type 2 diabetes (P = 0.009), which family-based association analysis suggested was attributable to population structure (P = 0.04) without significant evidence of linkage disequilibrium between diabetes prevalence and genotype (P = 0.86). Thus allelic variation of the INS gene is associated with lower birth weight and increased prevalence of type 2 diabetes. Significant linkage disequilibrium was found between −23Hph1 and birth weight but not type 2 diabetes, an observation that supports a potential functional role of INS polymorphisms in the regulation of birth weight.
Low birth weight is associated with increased risk of type 2 diabetes (1). It has been suggested that this may be because of pleiotropic effects of one or more genes to both increase susceptibility to type 2 diabetes and lower birth weight (2). The insulin gene (INS) has been proposed as a potential candidate for such effects, as insulin is a promoter of growth in utero and insulin secretion and action are central to the pathogenesis of type 2 diabetes (2). Studies examining genetic variations of INS in relation to disease development have concentrated on the INS-VNTR (variable number of tandem repeats), a polymorphic region ∼0.5 kb upstream of the insulin gene, comprising a variable number of repeats of an oligonucleotide sequence (ACAGGGGT(G/C)(T/C)GGGG) (3). Depending on the number of repeats, INS-VNTR is described as class I (26–63 repeats, average 570 bp), II (64–140 repeats, average 1,200 bp), or III (141–209 repeats, average 2,200 bp) (3). Polymorphism of the INS-VNTR is associated with changes in insulin expression in the pancreas (4,5) and thymus (6), and the presence of the class I allele is associated with increased risk of type 1 diabetes in Caucasian populations (7).
Results of studies examining the association of INS-VNTR and type 2 diabetes have been inconsistent (7); however, a collation of five of these studies (carried out in Caucasians) suggested that class III INS-VNTR conferred a modest increase in risk of type 2 diabetes (7). More recently, a large study using family-based association methods reported no significant overall association of INS-VNTR with type 2 diabetes (8). The latter study did, however, find an excess transmission of class III INS-VNTR from fathers to diabetic probands, which would suggest that parent-of-origin effects, such as imprinting, might be at work (8). Similarly, an increase in the risk of polycystic ovarian syndrome has been reported with paternal transmission of the class III allele (9). The potential for parent-of-origin effects is biologically plausible; INS-VNTR lies in close proximity to genes known to be imprinted in humans, notably INS itself (generally bi-allelically expressed in pancreas, but imprinted in the fetal yolk sac) (4,10) and IGF-2 (11). Furthermore, if there is a genetic connection between lower birth weight and type 2 diabetes, then imprinting effects should be considered. The majority of known imprinted genes influence growth in early life or placental development (11). In contrast to studies of type 2 diabetes, studies of association of birth weight to INS-VNTR have not taken account of potential imprinting effects. Higher birth weight associated with the class III allele has been reported (12); however, this effect was not found in all subjects, but restricted to those changing weight centiles by <0.67 SD score between birth and 2 years.
The Pima Indians of Arizona have a marked propensity to type 2 diabetes and obesity (13,14). In Pima Indians, both lower and higher birth weight are associated with increased risk of type 2 diabetes (15). Paternal diabetes is also associated with lower birth weight of offspring (16), an observation that supports the presence of a pleiotropic genetic effect influencing birth weight and type 2 diabetes in this population and suggests a particular importance for paternally derived alleles in this relationship (16). In addition, linkage analyses in the Pima Indians indicate an important genetic contribution to birth weight, with evidence that an imprinted, paternally expressed gene on chromosome 11 influences birth weight (17); however, the peak of linkage (LOD 4.1) was 70–80 cM distant from INS.
Results of a previous study of a small group of Pima Indians did not support a relation of INS-VNTR to type 2 diabetes susceptibility (18). In the current study, we wished to assess the relation of INS-VNTR to birth weight, type 2 diabetes, and adiposity, allowing for the potential imprinting effects. To achieve this, we examined −23Hph1, an A/T polymorphism within the Hph1 restriction endonuclease site 23 bp upstream from the start site of the insulin gene. The polymorphism is in tight linkage disequilibrium with INS-VNTR in Caucasian populations (7) and has been used extensively in studies addressing variations of INS-VNTR. We used a parent-offspring trio design because, first, where an association is present, it allows an assessment of whether this arises secondary to linkage disequilibrium or to other effects, such as population stratification (19); and second, the presence of parent-of-origin effects can be addressed. We tested the following hypotheses: 1) −23Hph1 polymorphisms are associated with birth weight with or without imprinting effects, and 2) −23Hph1 polymorphisms are associated with type 2 diabetes or BMI with or without imprinting effects.
RESEARCH DESIGN AND METHODS
Subjects and phenotypes.
The subjects in this study were participants in the longitudinal study of health in the Pima population whose heritage is at least half Pima or Tohono O’odham or a mixture of these two closely related groups. Members of the community over age 5 years are invited to participate in biennial research examinations, including an oral glucose tolerance test (OGTT) after a 75-g glucose load. Diabetes was diagnosed by World Health Organization (WHO) criteria (20). Owing to our interest in imprinting effects, we used a “trio” design, selecting families for whom DNA was available from both parents and one or more offspring and whose offspring had been examined at least once for the presence of type 2 diabetes. A total of 660 subjects were included, comprising 418 children from 130 nuclear families (median 3 siblings, range 1–11). No half-siblings were included. For 18 subjects, children of one family were parents in another family; those individuals were also included.
Birth weight and gestational age were ascertained by a review of medical records. Subjects were included if they were singleton births, both their birth weight and their gestational age were available, and their gestational age was ≥33 weeks. Given that the presence of maternal diabetes influences birth weight because of environmental effects of maternal hyperglycemia on the fetus in utero (21) and, further, that a small secular increase in birth weight is described in this population (22), adjustments for maternal diabetes and year of birth were made. Children were considered offspring of diabetic mothers (DIAB) if their mother had been diagnosed as having type 2 diabetes according to WHO criteria (20), either before or during the index pregnancy; offspring of nondiabetic mothers (NON) if their mother was known to have a nondiabetic OGTT after the index pregnancy; offspring of mothers who were possibly diabetic (POSS), if maternal diabetes had been diagnosed postnatally without an intervening nondiabetic OGTT after the birth of the child, thus rendering the exact time of onset of maternal diabetes uncertain; and offspring of unknown status (UNK), where maternal diabetes status had not been assessed. Before analysis, birth weights were adjusted for sex, gestational age (as linear and quadratic terms), and birth year for all analyses, using data from 4,076 births in the Pima population by means of linear regression and either with or without further adjustment for maternal diabetes status as detailed (with indicator variables for the different categories representing 246 DIAB, 2,943 NON, 304 POSS, and 583 UNK subjects). Birth weights (mean ± SD) for the 192 offspring included in the present genetic analyses were NON 3.7 ± 0.5 kg (n = 136), POSS 3.9 ± 0.5 kg (n = 35), DIAB 4.0 ± 0.6 kg (n = 19), and UNK 3.3 ± 0.1 kg (n = 2).
Informed consent for participation in the study was obtained from all subjects, and ethical approval was received from both the National Institutes of Health and the Gila River Indian Community.
Sequencing of the −23Hph1 single nucleotide polymorphism in Pima subjects.
The A to T polymorphism that disrupts the Hph1 site CCACT at nucleotide 2,401 of the insulin gene (accession #V00565) was sequenced in the 660 members of parent-child “trios” described above and in 77 individuals previously typed for INS-VNTR (18,23). The PCR primers forward 5′-TCAGAAGAGGCCATCAAGCAGGT-3′ and reverse 5′-CGCACAGGTGTTGGTTCACAAA-3′ were used to amplify a 574-bp fragment from genomic DNA, and the forward primer was further used for cycle sequencing. Sequencing was performed using the Big Dye Terminator (Applied Biosystems) on an automated DNA capillary sequencer (Model 3700; Applied Biosystems).
Evidence of linkage disequilibrium between −23Hph1 and INS-VNTR was assessed using the EH algorithm (24), which was also used to estimate haplotype frequencies of the two polymorphisms.
The association of −23Hph1 genotype to traits was examined by a modification of the approach of Fulker et al. (25). It is important to note that in our design, the association of marker and trait could potentially be influenced by family membership (statistical nonindependence of siblings) and population stratification. A series of models were therefore assessed.
In model 1, “simple association,” the additive influence of the genotype on a trait was assessed by modeling a genotype score (calculated from the number of T-alleles) against the trait. To account for family membership, the relation of genotype to trait was examined in a mixed model in which membership of nuclear family was modeled as a random effect. This random effect accounted for nonindependence among siblings, but the results of this model may still be confounded by population stratification.
In model 2, “linkage disequilibrium” (LD), the hypothesis of joint linkage and association of polymorphisms against trait, was examined. Association effects were partitioned into between- and within-family components (25). The between-family component is potentially influenced by population stratification, whereas the within-family effect assesses the presence of joint linkage and association (linkage disequilibrium). In brief, the genotype score for each subject is broken down into orthogonal between-family (b; derived from expected number of T-alleles for offspring conditional upon parental genotype) and within-family scores (w; derived from the deviation of offspring genotype score from the between-family score). Between- and within-family scores are then modeled (with additional random effect of nuclear family) against the trait.
In model 3, “imprinting (Imp) LD,” the hypothesis of linkage and association of polymorphisms inherited from either father or mother against trait, was examined. Between-family scores were calculated as above. The within-family component was further broken down into terms reflecting paternal (wpa) or maternal (wma) origin of inherited T-alleles. Thus wpa was calculated from the number of T-alleles inherited from the father minus the expected number given paternal genotype (1/2 paternal genotype score). The maternal within-family component was similarly derived. It is important to note that wpa and wma cannot be calculated when all three members of the trio (mother, father, and offspring) are heterozygous, as the source T-allele of the offspring is ambiguous; where this was the case, the subjects were excluded from model 3. In addition, subjects can only be informative when either the father or the mother is heterozygous. Scores b, wpa, and wma are then modeled (with additional random effect of nuclear family) against the trait. The hypothesis that T-alleles exert different influences on trait depending on the parent of origin was assessed by deriving a P value (Pdiff) by two-tailed t test of the difference of the β coefficients of wpa and wma divided by its standard error (derived from the standard errors of wpa and wma and their covariance).
The above framework was used for analyzing offspring birth weight and BMI (kg/m2). Although the analysis of each trait with all three models entailed some degree of multiple testing, the models were highly correlated with one another, as any observed association was partitioned into different sources. Thus P values are presented without correction for multiple testing. Because the presence of type 2 diabetes may influence BMI (26), data from examinations at or after diagnosis of type 2 diabetes were excluded from this analysis. BMI was analyzed using 1) the measurement from the earliest adult (age >18 years) examination and 2) one measure each from five age ranges (ages 5–9, 10–14, 15–19, 20–29, and 30–39 years, truncated at age 40 years because of the low number of observations in nondiabetic individuals older than age 40 years). For subjects who had been examined more than once within an age group, only the earliest examination was included.
The association of genotype with prevalence of type 2 diabetes was examined by logistic regression within the above framework, using diabetes status at the last examination. A generalized estimating equation model was used, with the inclusion of “repeated” term to allow for covariance among members of the same nuclear family. In addition, to explore the influences of genotype on disease prevalence at different ages, the prevalence for examinations in four age ranges (ages 10–19, 20–29, 30–39, and 40–49 years) was also examined. In general, because of the requirement for parental genotype, models 2 and 3 were applied to offspring only. By contrast, model 1 (simple association with adjustment for nuclear family membership) could be applied to offspring alone and offspring plus parents. Model 1 is presented for all offspring plus parents and offspring alone in the analysis of type 2 diabetes, but not birth weight (very few parental birth weights were available) or BMI (no substantial difference in results).
Linkage disequilibrium of −23Hph1 and INS-VNTR.
In all, 77 unrelated individuals (40 I/I homozygotes, 35 I/III heterozygotes, and 2 III/III homozygotes) for whom DNA was available and who had previously been typed for INS-VNTR (18) were genotyped at −23 Hph1. INS-VNTR and −23Hph1 were in strong linkage disequilibrium (P < 0.0001), with the occurrence of haplotypes INS-VNTR class I/−23Hph1 A and INS-VNTR class III/−23Hph1 T being greater than expected under equilibrium, as previously described in Caucasian populations (7). An estimated 6.0% of haplotypes were discordant (2.0% class I/T and 4.0% class III/A) and 94% were concordant (73% class I/A and 21% class III/T).
Relation of −23 Hph1 polymorphism to diabetes.
A higher proportion of heterozygotes and T/T homozygotes had developed diabetes than A/A homozygotes (Table 1). This difference approached significance in all subjects (parents and offspring combined: χ2, P = 0.08), but not in offspring alone (χ2, P = 0.7). In keeping with this, logistic regression of all subjects (model 1; P = 0.009) (Table 2) showed a significant association of the presence of the T-allele with type 2 diabetes. Analysis of offspring alone displayed a nonsignificant trend (P = 0.23) in the same direction. An important finding was that where between- and within-family effects were modeled in all offspring either without (model 2) or with parent-of-origin effects (model 3), there was no evidence of linkage disequilibrium between −23Hph1 and diabetes. Contrasting with this, between-family effects were significant in both models (model 2, P = 0.04; model 3, P = 0.02), in keeping with the presence of population stratification. The influence of the −23Hph1 genotype on diabetes prevalence was examined in subgroups classified based on the age at examination. The simple association (model 1) of −23Hph1 genotype and diabetes prevalence was most apparent in the older age groups (Fig. 1). In offspring, significant linkage disequilibrium between the −23Hph1 genotype and diabetes prevalence was not observed in any of the age groups (model 2; data not shown). Potential evidence of imprinting effects (model 3; Pdiff <0.05) was seen only in the youngest age group, with a protective effect of paternally transmitted T-alleles (age 10–19 years; odds ratio [OR] 0.05, 95% confidence interval [CI] 0.01–0.23; P = 0.001 for effect of paternal alleles; Pdiff = 0.0005), although this group contained relatively few cases (34 of 342 subjects).
The results were not substantially altered by including terms for maternal diabetes or the fraction of Pima heritage in the above models (data not shown). Further, analysis of diabetes restricted to those of full Pima/Tohono O’odham heritage (153 events in 311 subjects from 96 families) confirmed an increased risk of type 2 diabetes associated with the −23Hph1 T-allele (model 1; OR for effect of presence of T-allele 1.54 [CI 0.94–2.54]; P = 0.06), again without evidence of linkage disequilibrium (model 2; OR for within-family effect, 1.02 [CI 0.54–1.92], P = 0.9; for between-family effect, 3.25 [CI 1.39–7.61], P = 0.006).
Relation of −23 Hph1 polymorphism to birth weight.
The presence of the T-allele was significantly associated (model 1; P = 0.04) (Table 3) with lower birth weight (adjusted for sex, gestational age, birth year, and maternal diabetes status). Family-based association analysis (model 2) supported significant linkage disequilibrium between the T-allele and lower birth weight (P = 0.03) (Fig. 2). The influence of paternally transmitted T-alleles showed a greater effect than that of maternally transmitted alleles (−250 vs. −111, respectively), but this difference was not statistically significant (Pdiff = 0.50).
Analysis of birth weight unadjusted for the presence of maternal diabetes showed a similar trend in both the simple association (model 1; difference in birth weight associated with one allele difference in number of T-alleles −116 g; P = 0.11) and the model examining linkage disequilibrium (model 2; within-family effect −178 g; P = 0.047). Again, no significant parent-of-origin effect was observed (data not shown for model 3; Pdiff = 0.69).
We examined whether the presence of maternal diabetes might influence the relation of −23Hph1 to birth weight. The number of offspring from mothers who had diabetes during pregnancy was small (n = 19). A subanalysis of subjects depending on time of onset of maternal diabetes after birth of the child suggested that the negative influence of the T-allele was greater when mothers had developed diabetes within a few years after the birth of their child (Table 4). In these groups, the influence of paternally transmitted T-alleles also tended to be greater (e.g., offspring of mothers with diabetes before or up to 5 years after birth: n = 45, β coefficient wpa −862 vs. wma 233, Pdiff = 0.04; offspring of mothers with diabetes before or up to 10 years after birth: n = 80, β coefficient wpa −596 vs. wma 17, Pdiff = 0.07).
Relation of −23 Hph1 polymorphism to BMI.
No simple association (model 1) of BMI to genotype was observed for all subjects with an examination in adulthood (P = 0.77) or in the subgroup analysis by age (data not shown). Similarly, there was no evidence of linkage disequilibrium between −23Hph1 and BMI in all adult subjects either without (model 2; effect of w, P = 0.58) or allowing for parent-of-origin effects (model 3; wpa, P = 0.75, wma, P = 0.93). No evidence of association of −23Hph1 and BMI secondary to population stratification was observed (b, model 2 or 3; data not shown).
Our data indicated that presence of a T-allele at −23Hph1 is associated with lower birth weight. The −23Hph1 polymorphism itself is not thought to be functional, but has been used as a more easily defined indicator of INS-VNTR polymorphism. This reflects the tight linkage disequilibrium between −23Hph1 and INS-VNTR class I/III in Caucasian populations (7). In the Pima Indians, we have confirmed that −23Hph1 and INS-VNTR are in linkage equilibrium, albeit with a haplotype discordance rate of 6%, which is higher than that previously observed in Caucasians (0.023%) (7). Nevertheless, −23Hph1 typing gives a reliable indication of INS-VNTR status in this population.
INS-VNTR, unlike −23Hph1, does appear to be a functional polymorphism. Transfection of rodent β-cell lines with reporter constructs of the INS promoter and either class I or class III INS-VNTR results in alteration of the level of transcription (27,28). The class III allele is associated with a reduction of insulin transcription in the fetal (4) and adult pancreas (5). The observation that INS-VNTR acts as a binding site to the transcription factor Pur1 (27) forms a potential basis of these functional relationships (29).
We have presented evidence of significant linkage disequilibrium between birth weight and the INS gene −23Hph1 polymorphism, with the T-allele being associated with lower birth weight. Although our data did not allow us to determine whether this association was attributable to a direct effect of the INS-VNTR or to linkage disequilibrium with a nearby functional polymorphism, the former hypothesis is biologically plausible. Insulin promotes growth in utero (2). Therefore, if the class III allele results in reduced pancreatic expression of insulin, it in turn might lead to reduced growth in utero. This is also in keeping with the suggestion that obese children who are class I/I homozygotes have increased insulin secretion in response to their obesity (30). INS-VNTR class III has also been associated with decreased expression of IGF-2 in placenta (31), forming a further potential mechanism whereby this polymorphism might result in altered fetal growth. Although such mechanisms show potential, it is important to ascertain whether the associations we observed are present in other populations. At least in Caucasians, little effect of INS-VNTR has been observed, and even a positive effect has been seen in selected subgroups, although only simple associations were examined (12).
Fetal insulin production is greatly increased in the presence of maternal diabetes during pregnancy, and is therefore of particular importance in the offspring of mothers with abnormal glucose metabolism during pregnancy (32). Consequently, we examined influences of the −23Hph1 polymorphism when maternal diabetes was present during or detected shortly after pregnancy (in the latter case, it being likely that abnormal glucose metabolism was present during pregnancy). We found that the relationship of birth weight to −23Hph1 appeared greater in these subgroups, and indeed that there was the suggestion of a greater effect of paternally transmitted T-alleles. As we do not have detailed information regarding the glycemia of mothers during pregnancy to ascertain whether an interaction between maternal glycemia and fetal genotype might have taken place, we must interpret these results with caution.
We did not find evidence of linkage disequilibrium between INS-VNTR and type 2 diabetes, in keeping with a previous report in a Caucasian population (8). Overall, we observed a significant association of the −23Hph1 T-allele with increased prevalence of type 2 diabetes, a finding also reported in collated results of smaller series (7). Our results, however, suggested that in this population, this association reflects population stratification rather than a functional association. One potential source of such population stratification is admixture with other ethnic groups. Using self-reported estimates of Pima/Tohono O’odham heritage (which correspond well with genetic estimates of admixture) (33), we found that our results were consistent, even when analysis was restricted to those of full Pima/Tohono O’odham heritage. This suggests that ethnic admixture, at least as represented by Pima versus non-Pima heritage, is not the source of the association of −23Hph1 genotype and type 2 diabetes. It would also suggest that the Pima population is heterogeneous both in terms of diabetes risk and −23Hph1 genotype, perhaps as a result of an older population admixture in the Pima, although this remains entirely speculative.
Both INS (in the fetal yolk sac) (10) and IGF2 are imprinted with expression from paternally derived alleles (11). In addition, results of a previous study in Caucasians (8) suggest that paternal transmission of the class III INS-VNTR allele is associated with increased risk of type 2 diabetes. Apart from the results of subgroup analyses (which are relatively small and introduce potential problems of multiple testing), we found no clear evidence of a parent-of-origin effects of the −23Hph1 locus on diabetes or birth weight.
As part of this investigation, we examined the hypothesis that INS-VNTR might provide a genetic explanation for the association of low birth weight and type 2 diabetes; more specifically, that the presence of INS-VNTR class III would be associated with lower birth weight and increased risk of type 2 diabetes. Taken together, our results only partly support this hypothesis. Although −23Hph1 T, and by inference INS-VNTR class III, does appear to be associated with lower birth weight, we did not find significant evidence to support a functional effect of the class III allele on diabetes susceptibility.
Class I INS-VNTR has also been associated with an increase in childhood obesity, an effect particularly associated with paternally transmitted class I allele (34). This association has not been apparent in adults (8) and was not apparent in our series.
We used the −23Hph1 polymorphism as a means of assessing the INS-VNTR. Although supported by linkage disequilibrium between −23Hph1 and INS-VNTR, our data suggest that a small number (∼6%) of haplotypes will be misclassified, thereby potentially weakening the association if INS-VNTR is indeed of functional importance to variations in birth weight. Of greater concern is the potential oversimplification of variations of INS-VNTR by categorization into class I and class III. This is convenient for studies such as ours and allows comparison with previous population and functional studies. Nevertheless, it is important to note that INS-VNTR is extremely polymorphic; subclassifications based on the size of INS-VNTR have been proposed (5), and sequence variation within the repeat sequences is also present, with potential functional consequences. For example, it has been reported that certain subclasses of the class I INS-VNTR are protective against type 1 diabetes (class I generally being a risk factor) (5). The reasons for this remain unclear and perhaps await a clearer understanding of the functional relationship of INS-VNTR to the transcription of INS and other genes before this can be fully explored at the population level.
In conclusion, we have presented evidence that the −23Hph1 T-allele of the INS gene (and by implication class III INS-VNTR) is associated with both lower birth weight and prevalence of type 2 diabetes. Furthermore, our assessment with family-based association analysis supported the presence of linkage disequilibrium with birth weight. These findings suggest a potential functional relationship of polymorphisms in INS with birth weight, but suggest the relationship with type 2 diabetes, if any, is modest.
We thank the members of the Gila River Indian Community for their continued support and participation, and the staff of the Diabetes and Arthritis Epidemiology Section for conducting this study.
Address correspondence to Dr. Robert L. Hanson, National Institute of Diabetes and Digestive and Kidney Diseases, 1550 East Indian School Rd., Phoenix, AZ 85014. E-mail:.
Received for publication 11 June 2002 and accepted in revised form 23 September 2002.
LD, linkage disequilibrium; OGTT, oral glucose tolerance test; SNP, single nucleotide polymorphism; VNTR, variable number tandem repeat; WHO, World Health Organization.