Emerging Effects of Early Environmental Factors over Genetic Background for Type 1 Diabetes Susceptibility: Evidence from a Nationwide Italian Twin Study

Abstract

Context:

The incidence of type 1 diabetes has been increasing over time.

Objective:

We estimated the genetic and environmental components of type 1 diabetes susceptibility in a twin cohort of recent-onset cases to explore the sources of changing disease epidemiology.

Design:

We linked the population-based Italian Twin Registry with 14803 type 1 diabetes records from 36 pediatric diabetes care centers throughout Italy, except Sardinia, and identified 173 diabetic twins. Patients were positive for at least one autoantibody to islet cell, glutamate decarboxylase, tyrosine phosphatase, insulin, or zinc transporter 8 and were insulin dependent since their diagnosis. Zygosity was determined by DNA genotyping or by questionnaire.

Outcome Measures:

We estimated proband-wise concordance, cotwin recurrence risk with Kaplan-Meier method, and genetic and environmental proportions of susceptibility variance by structural equation models.

Results:

We recruited 104 diabetic twins (53 males) from 88 pairs (34 monozygotic, 54 dizygotic) and one triplet. The mean age at diagnosis was 8.1 yr (range 1.1–20.5 yr), and the median year of diagnosis was 2002. Proband-wise concordances were 45.5 and 16.4% in monozygotic and dizygotic pairs (P = 0.01). Recurrence risks in monozygotic and dizygotic cotwins were 37 and 12% after 10 yr from the proband's diagnosis (P = 0.005). Genetic contribution to type 1 diabetes susceptibility was 40% (95% confidence interval 8–78), and the shared and individual-specific environmental components were 51% (14–77) and 9% (4–19), respectively.

Conclusions:

In addition to the moderate genetic effects on type 1 diabetes susceptibility, our results draw attention to the substantial shared environmental effects, suggesting that exposures in fetal or early postnatal life may contribute to the increasing incidence of the disease.

Several environmental determinants operating from gestational period to adulthood have been suggested to affect type 1 diabetes risk (1, 2). On the other hand, more than 50 genes have been discovered in the past 2 decades (3), but they do not fully explain the disease susceptibility. Altogether, little knowledge has been gained about the gene-environment architecture of the disease, possibly because it may have changed over time.

Studies comparing disease concordance in monozygotic (MZ) and dizygotic (DZ) twins provide estimates of the relative weight of environmental and heritable sources of susceptibility. Furthermore, by using the twin design, the environmental contribution can be partitioned into components due to effects that can be either shared within pairs (i.e. intrauterine, early postnatal, or familial) or individual specific. Because these estimates depend on the population and period investigated, they can assist in detecting etiological heterogeneity among different ethnic groups and in interpreting changes of disease incidence within a population.

Twin studies on type 1 diabetes have been conducted with different identification and recruitment procedures in the past decades in North American, Danish, Finnish, and Australian populations (4–7). The latter three studies basically agree on suggesting genetic factors as the main determinant of susceptibility, whereas familial environmental contribution is predominant in the North American report (4).

There is consistent evidence for an increase of type 1 diabetes incidence (8–10) and for a decrease of high-risk human leukocyte antigen genotypes prevalence in new-onset cases over time (11, 12). Because these changes have been attributed to an increased environmental pressure on type 1 diabetes susceptibility, it is worth investigating further the environmental factors that may explain the changing disease epidemiology.

We used the twin design to estimate the genetic and environmental contributions to the disease susceptibility in a population-based cohort of Italian twins recently diagnosed with type 1 diabetes.

Materials and Methods

Identification and enrollment procedures as well as zygosity and disease ascertainment of the twins are schematically depicted in Fig. 1.

Fig. 1.

Identification, enrollment, zygosity, and disease ascertainment of Italian type 1 diabetic twin pairs. The triplet was counted as two discordant opposite gender pairs.

Twin identification and enrollment

From April 2006 to April 2010, pediatric diabetologists of the Italian Society of Pediatric Endocrinology and Diabetology (ISPED) provided data of 14803 type 1 diabetic subjects referring to 36 public diabetes care clinics located all over Italy, except Sardinia; these clinics cover 70% of yearly admissions for type 1 diabetes in the 0- to 14-yr age range. As in our previous studies (13, 14), twins among patients were identified through a record linkage procedure with the Italian Twin Registry (ITR) database, which includes approximately 700,000 records up to April 2011 (Ref. 15 and Stazi, M. A., unpublished data); the linkage was blind: ITR staff had no access to identification data, each clinical center checked the matching of their own patients, and contacted them for enrollment. One hundred seventy-three diabetic twins (from156 pairs/triplet) were identified. This gave a twin rate among patients of 2.2/100. Fifty-one twins (50 pairs) were no longer approachable, mostly because they had moved to clinics for adult diabetic patients. Of the 122 diabetic twins (106 pairs/triplet) that could be informed of the study, 104 twins (88 pairs and one triplet) agreed to participate and 18 twins (17 pairs) refused. Pair participation rate was 83.9%.

This study has been approved by the Ethics Committee of the Italian National Institute of Health (Istituto Superiore di Sanità, Rome, Italy). Twins (or their parents) gave consent for each of the following items: health data collection, zygosity determination by DNA test, feedback on their zygosity status, saliva collection, biobanking, and the use of DNA for research on type 1 diabetes and associated diseases (partially restricted consent) (16).

Disease status

The twin in a pair who first received diagnosis of type 1 diabetes is henceforth referred to as the index or proband twin; the unaffected twin or the second twin diagnosed with type 1 diabetes is the cotwin. All diabetic patients were positive for at least one autoantibody to islet cell, glutamate decarboxylase, tyrosine phosphatase, insulin, or zinc transporter 8 and were insulin dependent since diagnosis. Of the 89 index twins, 87 were diagnosed from year 1989 to 2009 (range 1977–2009, median 2002). All twin pairs (or their parents), regardless of the disease status, answered to a simple questionnaire, administered by the diabetologist, concerning twins physical resemblance, family composition, type 1 diabetes onset, parental or siblings' recurrence of type 1 diabetes, or associated diseases (i.e. Addison's disease, pernicious anemia, rheumatoid arthritis, celiac disease, autoimmune chorea, Crohn's disease, Basedow-Graves' disease, Hashimoto's disease, systemic lupus erythematosus, autoimmune thrombocytopenia, psoriasis, ulcerative colitis, multiple sclerosis, vitiligo). For some parents, information on the occurrence of type 1 diabetes or other diseases was missing.

Zygosity test

In 52 same-gender pairs who donated saliva, zygosity was assigned comparing genotypes of nine tetranucleotides (accuracy 99.98%). Accuracy of the questionnaire method to classify zygosity resulted to be 82% (nine pairs misclassified as DZ and MZ at the DNA test, but eight of them gave incoherent answers on physical similarity). For 14 same-gender pairs who refused donation, we used only questionnaire data on physical resemblance (17). Among these 14, only one disease discordant male-male pair provided incoherent answers and might have been misclassified as DZ.

Statistical analyses

We performed disease concordance and survival analyses and estimated genetic and environmental variance components of type 1 diabetes liability using 88 pairs plus one triplet (one affected female and two unaffected males) who was considered as two discordant opposite-gender pairs.

Concordance

Proband-wise concordance was estimated in MZ and DZ twin pairs as:

$$\text{Proband-wise}\,\text{concordance}=2*\text{C}/\left( 2*\text{C}+\text{D} \right)$$

where C and D are the numbers of concordant and discordant pairs, respectively. A higher concordance in MZ compared with DZ pairs revealed that genetic factors are involved in disease susceptibility.

Survival analysis

Survival analysis by zygosity was performed using the Kaplan-Meier method as implemented in the software SPSS (version 17.0; SPSS Inc., Chicago, IL) to estimate the recurrence risk of type 1 diabetes in cotwins. The terminating event was the diagnosis of type 1 diabetes in the cotwin. The health status in unaffected cotwins were last ascertained between July 2009 and April 2010.

Genetic and environmental variance components of type 1 diabetes liability

Assuming a liability-threshold model (18), tetrachoric correlations in MZ and DZ pairs were estimated; a higher correlation in MZ compared with DZ pairs points to genetic effects on liability to type 1 diabetes, with the difference between MZ and DZ revealing the magnitude of these effects. Subsequently, genetic and environmental variance components of liability to type 1 diabetes were estimated by structural equation modeling. We considered an ACE model incorporating parameters for additive genetic (A), shared environmental (C), and unshared environmental influences (E) (19). Additive genetic influences originate from the additive effects of alleles at all contributing genetic loci, without allelic or gene-gene interaction; these influences are completely correlated in MZ twins who are genetically identical and correlate to 0.5 in DZ twins, who share on average 50% of their segregating genes. Shared environmental influences relate to exposures that are common to both twins in a pair, regardless of zygosity, and that usually occur in intrauterine life, the early postnatal period, or family environment. Unshared environmental influences are to an individual and are therefore responsible for less than perfect concordance between MZ twins; measurement error is also included in this component. Total variance (V) in liability to type 1 diabetes can be decomposed as the sum of genetic and environmental components (V = A + C + E), and heritability can be estimated as the proportion of total variance that is explained by genetic variance (heritability = A/V). Tetrachoric correlations and variance components were estimated via the maximum-likelihood method as implemented in the software Mx (20). This study relied on complete ascertainment of twins, that is the case when concordant and discordant affected twins are ascertained and concordant unaffected pairs are not observed. In such a scenario, disease prevalence (i.e. liability threshold) cannot be estimated from the study sample and has to be fixed a priori in model-fitting analyses. In our analyses, we used a disease prevalence of 1:1000, based on the Italian data (21, 22).

Results

Zygosity and type 1 diabetes status

Eighty-eight pairs and one triplet entered this study: 34 were MZ (18 males), 54 were DZ (13 males, 19 females, and 22 opposite gender), and the triplet was trizygotic. Enrolled pairs were not different from nonparticipating ones with respect to gender (P = 0.6), geographic origin (northern, central, and southern Italy, P = 0.7), and proportions of same-gender vs. opposite-gender pairs (P = 0.4). Participating twins are significantly younger than nonparticipants as a consequence of our ascertainment procedure involving only pediatric centers that lose diabetic patients as they become adults. Possible differences in zygosity or concordance between the two groups could not be tested because information on these two variables was not available for nonparticipants.

Overall, there were 104 affected twins, almost equally distributed between genders (53 males, 51 females). Mean age at enrollment was 15.8 yr (range 1.4–36.5 yr; median 16.1 yr). Type 1 diabetes was diagnosed at a mean age of 8.1 yr (range 1.1–20.5; 7.8 yr in proband twins, 10.1 yr in cotwins) with no gender differences (males/females: 8.0/8.3 yr). Median age at onset was 6.9 yr in index twins, 9 yr in cotwins, and 7.6 yr in all affected twins (Table 1). Ninety-four percent of index twins (84 of 89) were diagnosed at a pediatric age (<15 yr). The test for heterogeneity of mean age at diagnosis in index twins by zygosity, gender, and type 1 diabetes concordance did not detect any significant difference.

Table 1.

Characteristics of the type 1 diabetic twins and the twin pairs

	Monozygotic twin pairs		Dizygotic twin pairs		All twins
	Concordant (n = 10)	Discordant (n = 24)	Concordant (n = 5)	Discordant (n = 50)
Mean age (yr)
Diagnosis in the index twin	7.2 (1.3–13.3)	9.3 (1.2–15.8)	7.8 (3.2–16.3)	7.2 (1.1–20.5)	7.8 (1.1–20.5)
Diagnosis in the cotwin	9.1 (1.4–15.8)		12.2 (4.3–19.9)		10.1 (1.4–19.9)
Enrollment	14.7 (1.4–24.4)	16.4 (2.1–29.3)	16.3 (9.8–21.2)	15.6 (2.6–36.5)	15.8 (1.4–36.5)
Median age (yr) at diagnosis
Index twin	6.8	9.4	6.8	6.8	6.9
Cotwin	8.8		13.8		9.0
Median time (yr)
Type 1 diabetes discordance	1.5 (0–5.2)a		4.7 (1.1–7)a		2.8 (0–5.2)
Unaffected cotwin observation		6.3 (0.3–21.1)b		7.6 (0.4–32.7)b	7.3 (0.3–32.7)

Numbers in parentheses are year ranges.

^a

P = 0.04 (Mann-Whitney test of medians).

^b

P = 0.59 (Mann-Whitney test of medians).

Type 1 diabetes concordance estimates

None of the discordant pairs became concordant during the enrolment period (end of April 2010). There were 10 of 34 MZ and 5 of 56 DZ concordant pairs. This resulted in significantly different (P = 0.01) proband-wise concordance estimates in MZ (45.5%) compared with DZ pairs (16.4%). Concordant pairs were similarly distributed between genders (Table 2).

Table 2.

Concordance rates for type 1 diabetes in monozygotic and dizygotic twin pairs

Type of pair	Concordant pairs	Discordant pairs	All pairs	Proband-wise concordance (%)	95% CI
MZ males	5	13	18	43.5	18.1–68.8
MZ females	5	11	16	47.6	21.3–74
All MZ	10	24	34	45.5a	27.2–63.7
DZ males	2	11	13	26.7	0–56.1
DZ females	2	17	19	19	0–41.6
DZ opposite gender	1	23b	24	8	0–22.7
All DZ	5	51b	56	16.4a	3.8–29
All pairs	15	75	90	28.6	17.3–39.9

^a

P = 0.01.

^b

Includes two discordant pairs from the triplet.

Survival analysis

The cumulative probability of type 1 diabetes was significantly higher in MZ than in DZ cotwins (P = 0.005, log-rank test). After 1 yr from the diagnosis in the first twin, 18% [95% confidence interval (CI) 4–32%] of MZ cotwins developed type 1 diabetes as opposed to 2% (95% CI 0–6%) of DZ cotwins; at 10 yr the corresponding figures were less far apart, being 37% (95% CI 17–57%) for MZ and 12% (95% CI 2–22%) for DZ (Fig. 2). Indeed, when we compared recurrence risks of MZ vs. DZ cotwins over time, we found borderline evidence for violation of the proportional hazards assumption (Schoenfeld residuals test; P = 0.06). In concordant pairs, median discordance time was significantly shorter in MZ than in DZ cotwins; in discordant pairs, median observation time was similar between MZ and DZ unaffected cotwins (Table 1).

Fig. 2.

Recurrence risk of type 1 diabetes in the cotwins of diabetic twins according to zygosity. Elapsed time is from the diagnosis of type 1 diabetes in the index twin. The terminating event is the diagnosis of type 1 diabetes in the cotwin. N_MZ and N_DZ are numbers of monozygotic and dizygotic cotwins entering each time interval.

Genetic and environmental variance components of type 1 diabetes liability

We estimated tetrachoric correlations and genetic and environmental variance components under complete ascertainment by fixing type 1 diabetes population prevalence at 0.1% (21, 22). Correlation estimate was higher in MZ [0.91 (95% CI 0.81–0.96)] compared with the DZ pairs [0.71 (95% CI 0.53–0.83)], suggesting a moderate genetic contribution to disease liability. Furthermore, the high correlation in the DZ pairs is consistent with substantial shared environmental effects. Indeed, according to the quantitative genetic theory, shared environmental effects are suggested when the twice DZ correlation exceeds the MZ correlation, and the difference of the twice DZ correlation − the MZ correlation can be used as an estimate of these effects (19). Unshared (individual specific) environmental influences were also suggested by a MZ correlation significantly lower than 1. In agreement with the correlation pattern, structural equation modeling provided estimates for heritability and shared environmental effects of 40% (95% CI 8–78) and 51% (95% CI 14–77), respectively. This suggests that genetic background and intrauterine or early postnatal environment may similarly contribute to type 1 diabetes liability. Moreover, environmental factors not shared within pairs explained the remaining proportion of variance [9% (95% CI 4–19)] (Table 3).

Table 3.

Twin correlations by zygosity and genetic and environmental proportions of variance for type 1 diabetes

Correlations		Genetic and environmental proportions of variance
MZ	DZ	A	C	E
0.91 (0.81–0.96)	0.71 (0.53–0.83)	0.40 (0.08–0.78)	0.51 (0.14–0.77)	0.09 (0.04–0.19)

Numbers in parentheses are 95% CI.

Case representativeness

To evaluate whether our cases were representative of the type 1 diabetes population, we compared the prevalence of other autoimmune diseases in twins with published Italian data and also explored whether a seasonal pattern of diagnoses was detectable. Among diabetic twins, 12 (11.5%) were affected by Hashimoto's disease and five (4.8%) by celiac disease; moreover, diagnoses were more frequent in winter than in summer (P = 0.14).

Discussion

We found that in our Italian cohort of relatively recent-onset cases, genetic background and environmental exposures shared by the twins explain substantially and to a similar extent the population variance of disease susceptibility. The contribution of individual-specific, nonheritable factors is less important and consistent with that reported in other twin studies (4–7); these factors are responsible for disease discordance in MZ pairs and onset time difference in MZ concordant pairs, and may include both environment-induced and stochastic epigenetic changes (23).

The finding of a significant shared environmental contribution is new compared with previous twin studies and may be interpreted in the light of the changing epidemiology of type 1 diabetes. Over the past 60 yr, the incidence of type 1 diabetes worldwide has been increasing by 3–5% per year (8–10). Such a rapid increase can be due to a growing environmental influence acting on a rather common genetic susceptibility (2); moreover, an acute environmental change in the mid-1980s or in the early 1990s has been suggested as an explanation for the increase of recent-onset cases with lower-risk human leukocyte antigen genotypes (11). On the other hand, a genetic model has also been postulated for the rising incidence of type 1 diabetes (24).

Because our cohort is population based, ascertainment bias is a minor issue in this study, and estimates of variance components of disease susceptibility can be considered quite reliable. Twin pairs were identified through affected probands, regardless of their zygosity and disease status of the cotwin, using a record-linkage between the ITR database and the type 1 diabetic patients lists provided by the ISPED Study Group.

Our cohort was homogenous from a diagnostic perspective because most patients had pediatric disease onset and all of them were antibodies positive and were insulin treated since the diagnosis; in this way, we avoided the inclusion of cases of nonautoimmune diabetes and of latent autoimmune diabetes of adults that have different phenotypes and pathogenetic origins (25). Moreover, we observed a seasonal-onset pattern that is in keeping with worldwide data (26), and the prevalence of Hashimoto's and celiac diseases in our diabetic twins is in line with previous reports in Italy (27, 28).

The concordance for type 1 diabetes was significantly higher in the MZ than in DZ pairs, in agreement with the well-established evidence of many genes contributing to the disease susceptibility. The concordance estimates in Italian MZ and DZ pairs are not significantly different from those reported in other populations, although the point estimate of the DZ concordance is slightly higher than in the other population-based studies (16 vs. 11% in Danish, 7% in Finnish, and 12% in Australian populations) and lower than in the volunteer-based North American report (Table 4). We can rule out that this difference is due to zygosity misclassification of our concordant same-gender DZ pairs because they were all typed for DNA markers.

Table 4.

Synopsis of twin studies on genetic and environmental components of type 1 diabetes liability

Study reference	Kumar et al. (4)	Kyvik et al. (5)	Hyttinen et al. (6)	Condon et al. (7)	This study
Population and sample size (pairs)	North American; n = 224 MZ = 132; DZ = 86; unclassified = 6	Danish; n = 95 MZ = 26; DZ = 69	Finnish; n = 228 MZ = 44; DZ = 183; unclassified = 1	Australian; n = 46 + 12 unmatched twins; MZ = 14; DZ = 32	Italian; n = 88 + 1 triplet MZ = 34; DZ = 56
Diabetic twins identification	Volunteer based	Danish twin registry (ad hoc questionnaire )	Finnish twin registry (record linkage with discharge, disease and drug registries)	Australian twin registry (analysis of previous surveys)	Italian twin registry (record linkage with pediatric diabetes care center databases)
Diagnostic criteria for type 1 diabetes	BMI, insulin within 1 yr from diagnosis, age diagnosis <30 yr (typical cases)	Age onset <40 yr, start insulin treatment within 2 yr from diagnosis, no overweight at diagnosis; clinical examination	Not reported	Treatment, BMI, age onset	Positivity to islet autoantibodies and insulin dependence since diagnosis
Years of birth	Not reported	1953–1982	Before 1958 to 1986	Reported for some of the cohorts studied: before 1945, before 1965, 1964–1970	1971–2007 (median 1992)
Years of diagnosis	Up to May 1992	Not reported	Before 1964 to 1998	Before 2000	1977–2009 (median 2002)
Age (yr) at diagnosis	Index mean (se): MZ = 13.6 (0.7), DZ = 13.1 (0.8)	Index + cotwins mean (range): MZ = 18.3 (2–38), DZ = 15.1 (1–35)	Index + cotwins mean (range): 13.1 (0.6–40.5); median = 11.4	Index + cotwins mean (sd): 18.4 (11.5); median = 15	Index + cotwins mean (range): 8.1 (1.1–20.5); median = 7.6
Discordance time (yr) in concordant pairs	Mean (se): MZ = 3.3 (0.6), DZ = 6.1 (1.5)	Not reported	Median (range): MZ = 2 (0–6.9), DZ = 6 (1.5–23.6)	Not reported	Median (range): MZ = 1.5 (0–5.2), DZ = 4.7 (1.1–7), mean: MZ = 1.8; DZ = 4.4
Time of observation of disease discordant cotwins	Not reported	Range: <1 and >15 (derived from Table 4)	Mean = 21.8; median (range) = 19.8 (0.1–40.9)	Not reported	Median (range): MZ = 6.2 (0.3–21), DZ = 7.4 (0.4–33)
10-yr disease progression rate in cotwins (95% CI)	Not reported	Not reported	MZ = 33 (21–43), DZ = 3.2 (0.5–6)	Not reported	MZ = 37 (17–57), DZ = 12 (2–22)
Proband-wise concordance (95% CI)	MZ = 45, DZ = 25	MZ = 53 (33–73), DZ = 11 (5–21)	MZ = 43 (27–59), DZ = 7 (2–13)	MZ = 61 (30–83), DZ = 12 (2–34)	MZ = 45 (27–64), DZ = 16 (4–29)
Correlation (95% CI)	MZ = 0.88, DZ = 0.73	MZ = 0.96 (0.78–1), DZ = 0.58 (0.44–0.71)	Not reported	MZ = 0.96 (0.86–0.99), DZ = 0.61 (0.32–0.81)	MZ = 0.91 (0.81–0.96), DZ = 0.71 (0.53–0.83)
Heritability (95% CI)	29%	72% (30–100)	88% (78–94)	56–96% according to different models and cohorts	40% (8–78)
Shared environment (95% CI)	58%	28%a	0	0–39% according to different models and cohorts	51% (14–77)
Unshared environment (95% CI)	13%		12% (6–22)	4–5% according to different models and cohorts	9% (4–19)

BMI, Body mass index.

^a

Includes shared and unshared environment.

Recurrence risks as obtained from Kaplan-Meier analysis, rather than crude proband-wise concordance rates, are useful in clinical settings because they are predictive risk estimates for cotwins of newly diagnosed subjects. According to our data, among MZ, approximately one of five cotwins is diagnosed with type 1 diabetes within 1 yr from the diagnosis in the first twin, and approximately two of five are diagnosed within 10 yr. Instead, among DZ, approximately one of 50 cotwins develops type 1 diabetes within 1 yr and six of 50 within 10 yr. The 10-yr recurrence risk in MZ cotwins is comparable with that described in the Finnish (Table 4) and British (29) populations, whereas the progression rate in DZ is higher in the Italian cotwins. This result is coherent with both concordance and variance components analyses, which provide new evidence of a role of shared environmental factors in type 1 diabetes pathogenesis.

In studies from North European and Australian populations, heritability seems to contribute the largest proportion of variance, whereas a predominant shared environmental role is shown in North Americans (Table 4). The significant role of environmental effects shared within twin pairs, emerging in our study, may reflect a higher relative weight of nonheritable phenomena occurring during intrauterine life and/or the early postnatal period, as also supported by the observation that none of the 65 nontwin siblings of our twins were affected by type 1 diabetes; it is unlikely that this is due to an underascertainment of type 1 diabetes cases among siblings because they were representative with respect to the prevalence of other type 1 diabetes-associated diseases, including Hashimoto thyroiditis, Addinson's disease, and vitiligo. Possible effects of shared environment on liability to type 1 diabetes can also be seen in the shorter discordance time observed in DZ concordant pairs compared with those of North European cohorts (Table 4). On the other hand, Redondo et al. (30) found no difference in the expression of β-cell autoimmunity between potential DZ twins and siblings of type 1 diabetes patients; furthermore, the Finnish study (6) showed a DZ concordance rate that is similar to that generally reported for siblings. These observations may provide evidence against the role of shared environmental factors.

The larger shared environmental component we detected in our study, compared with Danish, Finnish, and Australian populations, could be due to a cohort effect of a changing pressure of environmental factors, which modify the epigenetic regulation of type 1 diabetes-related genes during embryonic and fetal development or the neonatal period (31). Indeed, differently from other twin populations, this cohort is composed by patients diagnosed at a younger age and in more recent years (Table 4). Therefore, the Italian cohort may have been affected to a greater extent by the changing environmental factors that have been involved in the incidence trend of autoimmune diabetes, especially for youngest children; among these factors, those shared within twin pairs may be, for instance, gestational infections (32), chemical exposure (33), cesarean section (34), maternally derived gut microbiota (35), the mother's age and jaundice (36), and early feeding pattern (37). The hygiene hypothesis (38) has also been cited to partly explain the lately increasing incidence of allergic and autoimmune conditions, including type 1 diabetes: both uterus and early-life exposures are critical periods for T cell maturation and cytokine gene expression, with epigenetic modification playing an important role in immune system development and function (31). In terms of twin data, all these exposures may increase disease concordance in individuals who share 50% of genetic susceptibility (DZ twins) and may match disease-onset time in genetically identical individuals.

With regard to the high shared environmental estimate (58%) reported in the North American study, it may be partly due to the volunteer-based recruitment procedure, which may have led to an overrepresentation of concordant pairs (Table 4).

Between-studies differences in heritability estimates can be partially explained by type 1 diabetes prevalence that substantially varies among the populations, with values of 0.3–0.4% in Denmark (5), 0.5% in Finland (6), and 0.09–0.1% in Italy (21, 22). The higher estimates of the genetic contribution in Northern Europe are in agreement with the hypothesis that disease prevalence mirrors genetic penetrance which, in turn, is related to heritability.

Identification of twins with type 1 diabetes also differed between our study and previous studies, and this may have influenced heritability estimates. We looked for diseased twins linking records of clinical series of previously diagnosed type 1 diabetic patients to the ITR database; this may have implied a failure in identifying some twins. Danish and Australian studies relied on questionnaire to identify diabetic subjects within twin cohorts and therefore may have been prone to miss diabetic cases. Finnish authors searched for diabetic patients in a population-based twin cohort using linkage with national databases, possibly making this study the most accurate with respect to the identification of type 1 diabetic twins (Table 4).

It is important to point out that our results are not in contrast with previous observations supporting the genetic basis of type 1 diabetes, including Italian data on migrants (39). Our heritability estimate of 40% suggests that, in the Italian population, genetic factors explain a moderate proportion of interindividual differences in type 1 diabetes liability.

Finally, it is worth underlining two study limitations. It has been described that MZ pairs may become more concordant if they are followed up for a very long time (40); we cannot rule out that concordance could preferentially increase in our MZ over DZ pairs, and thus, a higher heritability could be estimated as a consequence of a longer observation time. In addition, it is important to recognize that genetic modeling is based on the simplifying assumption that there are only additive effects of genes and environment on the phenotypic variance, whereas in practice there may be interactions that can be tested when measures of a specific environmental factor are available. If not modeled, interactive effects become incorporated either into the heritability estimates (interactions between shared environmental and genetic risk) or into the unshared environmental estimates (interactions between unshared environmental and genetic risk).

In conclusion, our study of a cohort of mostly pediatric-onset patients shows a significant and substantial role of shared environmental factors in type 1 diabetes susceptibility. Beyond suggesting a possibly different disease architecture in Italy compared with North European populations, the results appear in agreement with the hypothesis that changing environmental determinants are responsible for the well-documented increasing trend of disease incidence. Studies aimed at unveiling susceptibility genes for type 1 diabetes should consider cohorts of patients diagnosed within a relatively narrow time frame, taking into account information on environmental exposure, especially in intrauterine or early postnatal life.

Appendix

Co-authors of ISPED Study Group on Diabetes include the following: Valentino Cherubini, Antonio Iannilli, Anna Maria Paparusso (Ancona); Luciano Cavallo, Clara Zecchino (Bari); Gianpaolo de Filippo (Benevento); Luigi Gargantini (Treviglio, Bergamo); Silvana Salardi, Stefano Zucchini, Giulio Maltoni (Bologna); Bruno Pasquino, Peter Kaufmann (Bolzano); Fabio Buzi, Elena Prandi (Brescia); Francesco Gallo (Brindisi); Mario Cicchetti (Campobasso); Enzo Castaldo (Caserta); Felice Citriniti (Catanzaro); Franco Chiarelli, Stefano Tumini, Alessia Di Stefano (Chieti); Domenico Sperlì, Rosaria De Marco (Cosenza); Patrizia Banin (Ferrara); Sonia Toni, Lorenzo Lenzi (Firenze); Maurizio Del Vecchio (San Giovanni Rotondo, Foggia); Renata Lorini, Giuseppe D'Annunzio (Genova); Dario Ingletto (Lecce); Andrea Scaramuzza, Gian Vincenzo Zuccotti (Milano Ospedale Sacco); Giuseppe Chiumello, Franco Meschi, Riccardo Bonfanti, Giulio Frontino (Milano Ospedale San Raffaele); Filippo de Luca, Fortunato Lombardo, Giusy Salzano (Messina); Lorenzo Iughetti (Modena); Adriana Franzese, Pietro Buono, Ilaria De Simone (Napoli Università Federico II); Francesco Prisco, Alessandra Cocca (Napoli Seconda Università); Franco Cadario (Novara); Carla M. Monciotti, Valentina Savio (Padova); Francesca Cardella (Palermo); Maurizio Vanelli, Giovanni Chiari, Katrin Errico, Brunella Iovane (Parma); Valeria Calcaterra (Pavia); Francesco Citro (Potenza); Sandro Cantoni (Reggio Emilia); Alberto Marsciani (Rimini); Marco Cappa, Patrizia I. Patera, Riccardo Schiaffini (Roma Ospedale Bambino Gesù); Nicoletta Sulli, Marialuisa Spoletini (Roma Università Sapienza); Franco Cerutti, Ivana Rabbone, Sabrina Sicignano (Torino); Vittoria Cauvin, Maria Bellizzi (Trento); Giorgio Tonini, Elena Faleschini (Trieste); Alessandro Salvatoni (Varese); and Leonardo Pinelli, Claudio Maffeis, Giovanna Contreas (Verona).

^*

L.N. and D.I. contributed equally to this study.

^†

Investigators of ISPED Study Group on Diabetes are coauthors and are listed in the Appendix.

Abbreviations

 
• A

  Additive genetic influences

 
• C

  shared environmental influences

 
• CI

  confidence interval

 
• DZ

  dizygotic

 
• E

  unshared environmental influences

 
• ISPED

  Italian Society of Pediatric Endocrinology and Diabetology

 
• ITR

  Italian Twin Registry

 
• MZ

  monozygotic

 
• V

  total variance.

Acknowledgments

We warmly thank the twins and their parents for their participation and Monica Vichi, M.Sc., for estimating admissions coverage of type 1 diabetes patients in Italian diabetes care centers. M.A.S., R.C., D.I., and L.N. designed the study; A.G. implemented the study protocol; L.N. and D.I. coordinated the study; R.C. and V.T. designed the data management and supervised the record linkage; L.N., A.G., and the ISPED Study Group on Diabetes collected and checked the quality and completeness of patients' data; C.F. and R.C. performed the statistical analysis; V.T. managed the ethical and legal issues; L.N., D.I., C.F., A.G., and M.A.S. interpreted the results; L.N. and C.F. wrote the manuscript with substantial contribution from D.I., A.G., R.C., V.T., and M.A.S. All the members of ISPED Study Group on Diabetes discussed the final results. All the authors read, revised, and accepted the final version of the manuscript.

This work was supported by intramural funding of the Istituto Superiore di Sanità to the Italian Twin Registry.

Disclosure Summary: The authors have nothing to disclose.