Introduction

Iron is the most abundant metal in the body, and its high concentration in blood erythrocytes is essential for the utilisation of oxygen [1]. Excess iron is sequestered intracellularly in the body via the protein ferritin, and serum ferritin levels are commonly used for the clinical determination of the body’s iron status [1]. The reference interval for serum ferritin is wide, comprising levels between approximately 20 and 300 μg/l among men and 15 and 200 μg/l among women. Biologically, iron is essential for the utilisation of oxygen; however, it is also a strong pro-oxidant known to catalyse the formation of reactive oxygen species, potentially causing severe damage to the insulin-producing pancreatic beta cells, muscle cells and liver cells [1–3]. Moreover, ferritin not only reflects iron stores, but is also an acute-phase reactant, commonly elevated in individuals with low-grade inflammation such as is seen in obesity, the metabolic syndrome and type 2 diabetes.

A role of iron in the pathogenesis of diabetes in humans has been suggested by several previous findings. First, iron overload in the genetic disorder hereditary haemochromatosis has been associated with an increased prevalence of diabetes [4]. In studies of rodents and up to 68 humans, iron overload due to blood transfusions or the administration of ferric nitrilotriacetic acid has been shown to induce diabetes or prediabetic conditions in the form of impaired glucose tolerance or reduced insulin sensitivity [5–10]. Second, a reduction in ferritin level, by either blood-letting or iron-chelating therapy, has been associated with improved glucose tolerance and insulin sensitivity in both healthy volunteers and patients with diabetes in studies of up to 59 participants [11–15].

Furthermore, cross-sectional studies have shown that individuals with impaired fasting glucose or type 2 diabetes exhibit higher levels of ferritin [16–18]. Several studies have been conducted to investigate the association between serum ferritin levels and future risk of type 2 diabetes [19–27]. A recent meta-analysis of nine prospective studies including 26,339 participants (ranging from 560 participants in the smallest cohort to 7,827 in the largest) concluded that a high serum ferritin level is associated with an increased risk of incident type 2 diabetes [28]. However, it has not yet been examined in larger cohort-based studies whether this association is explained by impaired beta cell function, decreased insulin sensitivity or both. In the fasting state, elevated serum ferritin levels have been found to be associated with levels of insulin and glucose [16, 22, 29], and a few studies of up to 1,013 participants using OGTTs, IVGTTs or various clamps have shown an association between elevated ferritin levels and decreased insulin sensitivity [29–32].

The present study includes data from OGTTs on more than 6,000 individuals and therefore enables an elucidation of the glycaemic mechanisms that underlie the observed association between serum ferritin levels and risk of type 2 diabetes. Thus, the overall aim of this study was to test the hypothesis that an elevated fasting serum ferritin level is associated with an increased prevalence of type 2 diabetes due to its association with impaired beta cell function and decreased insulin sensitivity.

Methods

Study population

The present study is based on the Danish population-based Inter99 study (ClinicalTrials.gov ID no. NCT00289237), which was a non-pharmacological intervention study for ischaemic heart disease [33]. A random sample of 13,016 individuals living in Copenhagen County from seven different age groups was drawn from the Civil Registration System, and a total of 6,784 (52%) attended the baseline examination, which took place from March 1999 to January 2001. Of the participants, 308 were excluded due to missing serum ferritin measurements, leaving 6,476 participants; of these, 84 participants who were receiving medical treatment for diabetes were excluded, leaving 6,392 for analysis. Written informed consent was obtained from all the participants. The study was approved by the Scientific Ethics Committee of the Capital Region of Denmark (KA 98 155) and is in accordance with the principles of the Declaration of Helsinki II.

Anthropometric and biochemical measurements

Anthropometric and biochemical measures were obtained after an overnight fast. Weight (kg) and height (cm) were measured with the participants in light indoor clothes and without shoes. Plasma glucose was analysed using the hexokinase/glucose-6-phosphate dehydrogenase technique (Boehringer, Mannheim, Germany), and serum insulin (excluding des-31,32 and intact proinsulin) was measured using the AutoDELFIA insulin kit (PerkinElmer/Wallac, Turku, Finland). Serum ferritin and high-sensitivity C-reactive protein (hsCRP) were measured on serum that had been stored for approximately 13 years, using a chemiluminescence and a nephelometric assay, respectively (Siemens Healthcare Diagnostics, Eschborn, Germany). The CV was 2.6% for the ferritin assay.

Glycaemic traits

Non-diabetic Inter99 participants underwent a 75 g OGTT with plasma glucose and serum insulin measured at fasting and at 30 and 120 min after the oral glucose load. The participants were characterised as having normal glucose tolerance, prediabetes (impaired fasting glucose, impaired glucose tolerance or both) or self-reported or screen-detected type 2 diabetes according to the 1999 WHO criteria [34]. Various surrogate measures of beta cell function and insulin sensitivity were calculated. Beta cell function was assessed as the insulinogenic index and corrected insulin response, which are surrogate measures of oral glucose-stimulated insulin secretion [35]. Insulin sensitivity was assessed as Matsuda index of insulin sensitivity (ISIMatsuda) [36] and HOMA-IR. Beta cell function corrected for insulin sensitivity was expressed as the disposition index. Detailed information on the calculations of measures of beta cell function and insulin sensitivity is provided in electronic supplementary material (ESM) Table 1.

Statistical analyses

Statistical analyses were performed using Stata statistical software (version 13; StataCorp, College Station, TX, USA). A p value <0.05 was considered statistically significant. We used a non-parametric test for trend across ordered groups to test whether serum ferritin level increased across the three groups: normal glucose tolerance, prediabetes and type 2 diabetes. To estimate the ORs of type 2 diabetes or prediabetes as a function of fasting serum ferritin levels, we used a logistic regression model that included (1) age and sex, (2) age, sex and BMI, and (3) age, sex, BMI and hsCRP levels. In addition, we used a multifactor-adjusted model including age, sex, BMI, hsCRP levels, waist circumference, hypertension (systolic BP <140 mmHg, diastolic BP <90 mmHg and no antihypertensive treatment vs systolic BP ≥140 mmHg, diastolic BP ≥90 mmHg or antihypertensive treatment), family history of diabetes (no family history, one parent with diabetes or two parents with diabetes), smoking (never, former or current smoker), alcohol use based on recommendations from the Danish Health and Medicines Authority (≤14/21 vs >14/21 units for women and men, respectively, where 1 unit = 12 g of alcohol), physical activity during leisure and commuting time (0–2 h/week, 2–4 h/week, 4–7 h/week or 7–12 h/week), LDL-cholesterol, HDL-cholesterol and triacylglycerol. Fasting serum ferritin levels were entered into the model as quintiles of ferritin, and in addition we subdivided the fifth quintile into four groups (80th–90th percentile, 90th–95th percentile, 95th–97.5th percentile and ≥97.5th percentile) to examine extremely high ferritin levels.

To test whether serum ferritin levels were associated with glycaemic traits derived from OGTTs, we used linear regression models that were adjusted for the above-mentioned covariates. To examine whether the assumptions for the linear regression analysis were being fulfilled, all the variables were inspected with regard to linearity as well as homogeneity of variance and normality of the residuals. Plasma glucose was kept in its identity form, and analyses with plasma glucose were run using robust SEs to meet assumptions concerning the normality and homogeneity of variance of the residuals. As the values of serum ferritin, serum insulin, insulinogenic index, corrected insulin response, ISIMatsuda, HOMA-IR and disposition index displayed non-normal distributions of the residuals, these were transformed via the natural logarithm to approximate a normal distribution. Accordingly, the effect sizes are reported in per cent for a 1% change in serum ferritin. As serum ferritin levels are generally higher in men than women, all the analyses are presented for men and women both together and separately. Interaction with sex and age was tested for by introducing an interaction term between ferritin levels and sex or age in the linear regression model. The statistical software R version 3.1.0 (www.r-project.org, accessed 14 October 2014) was used to perform piecewise linear regression to examine whether there was a threshold of ferritin level that is associated with insulin resistance. A two-segment piecewise linear model adjusted for age, sex, BMI and hsCRP levels was fitted for ISIMatsuda and HOMA-IR, and the threshold was determined as the ferritin level at which the piecewise regression best fitted the ISIMatsuda and HOMA-IR, that is, where the residual SE was smallest.

Results

The baseline characteristics of the study population at the time of blood sampling are presented in Table 1. A total of 6,476 individuals were present at the health examination and had their fasting serum ferritin measured. The median ferritin level was 79 μg/l (IQR 35–152 μg/l) among all the participants, 140 μg/l (IQR 86–219 μg/l) among the men, and 42 μg/l (IQR 21–75 μg/l) among the women.

Table 1 Baseline characteristics of the Inter99 study participants
Full size table

Median fasting serum ferritin levels were 68 μg/l (IQR 31–134 μg/l), 113 μg/l (IQR 53–201 μg/l) and 152 μg/l (IQR 75–271 μg/l) among the participants with normal glucose tolerance, prediabetes and type 2 diabetes, respectively (p trend < 0.001). A similar pattern was observed when analysing men (p trend < 0.001) and women (p trend < 0.001) separately (data not shown). The ORs of type 2 diabetes increased in a stepwise manner with increasing fasting serum ferritin levels, and additional adjustment for BMI and hsCRP generally attenuated the ORs (Figs 1 and 2). The ORs adjusted for age, sex, BMI and hsCRP were 1.0 (95% CI 0.6, 1.8), 1.2 (95% CI 0.7, 2.1), 1.9 (95% CI 1.1, 3.3) and 4.2 (95% CI 2.4, 7.2) for the second, third, fourth and fifth quintiles of ferritin compared with the first quintile (p trend < 0.001) (Fig. 1). Stepwise increasing ORs of type 2 diabetes (p trend < 0.001) were also observed for the four subdivisions of the fifth quintile, with an especially high OR among participants with ferritin levels ≥97.5th vs <20th percentile (OR 17; 95% CI 8.9, 33) (Fig. 1).

Fig. 1
figure 1

ORs of type 2 diabetes according to fasting serum ferritin levels. The numbers of participants and events are provided for the age- and sex-adjusted analysis. There was no information on BMI for one participant and on serum hsCRP for 289 participants. p trend < 0.001 for all analyses

Full size image
Fig. 2
figure 2

ORs of type 2 diabetes according to fasting serum ferritin levels stratified by sex. The numbers of participants and events are provided for the age-adjusted analysis. There was no information on BMI for one woman, and on serum hsCRP for 122 men and 167 women. p trend < 0.001 for all analyses

Full size image

As ferritin levels are highly dependent on sex, we stratified the data in terms of sex and found that increasing levels of ferritin were associated with increasing ORs for type 2 diabetes among both men and women. However, the association was strongest among men: the OR for the fifth vs the first quintile of ferritin was 6.3 (95% CI 3.4, 12) among men and 2.8 (95% CI 1.4, 5.7) among women (Fig. 2). The corresponding ORs were 31 (95% CI 13, 71) and 7.7 (95% CI 3.1, 19) for ferritin levels ≥97.5th vs <20th percentile (Fig. 2). Increasing levels of serum ferritin were also associated with a stepwise increase in OR of prediabetes among all the participants (ESM Fig. 1) and among men and women separately (ESM Fig. 2). In addition, when analysing impaired fasting glucose and impaired glucose tolerance separately, increasing levels of serum ferritin were also associated with increasing ORs of both outcomes, and when impaired fasting glucose was defined as a fasting plasma glucose level ≥5.6 mmol/l, the results for prediabetes as well as for impaired fasting glucose were similar to the main results (data not shown).

After adjustment for age, sex, BMI and hsCRP, elevated fasting serum ferritin levels were associated with elevated plasma glucose levels at fasting, 30 min and 120 min (p < 0.001), elevated serum insulin levels at fasting and 120 min (p = 0.02 and p < 0.001), decreased beta cell function assessed as the insulinogenic index and corrected insulin response (p < 0.001), and decreased insulin sensitivity assessed as the ISIMatsuda and HOMA-IR (p < 0.001) (Table 2). Generally, adjustment for BMI and hsCRP resulted in attenuated effect sizes compared with the effect sizes from the age- and sex-adjusted analyses (Table 2).

Table 2 Associations between fasting serum ferritin levels and glycaemic traits obtained during OGTTs in the Inter99 cohort
Full size table

Interaction analyses suggested that sex modified the association between ferritin levels and 120 min plasma glucose, fasting serum insulin, 120 min serum insulin, ISIMatsuda and HOMA-IR (p int < 0.001) (Table 3). Serum ferritin levels were only associated with fasting insulin, ISIMatsuda and HOMA-IR among men but not among women. Furthermore, serum ferritin levels were more strongly associated with 120 min plasma glucose and 120 min serum insulin among men than women (Table 3). To examine whether the discrepancies in the association between ferritin levels and measures of insulin sensitivity could be explained by women having menstrual periods and thereby losing iron, or by sex hormone differences, we stratified women by age below or above 52 years, which is the median age of menopause among Danish women [37]. Women <52 years had a median ferritin level of 35 μg/l (IQR 18–62 μg/l) whereas women ≥52 years had a median ferritin level of 72 μg/l (IQR 45–116 μg/l).

Table 3 Associations between fasting serum ferritin levels and glycaemic traits obtained during OGTTs in the Inter99 cohort stratified by sex
Full size table

Generally, when stratifying by age, the effect sizes among younger women resembled the effect sizes among all women, whereas the effect sizes among older women resembled the effect sizes observed among men (Table 4). Significant interactions between serum ferritin levels and age <52 or ≥52 years were observed for fasting insulin (p int = 0.003), 120 min insulin (p int = 0.002), ISIMatsuda (p int = 0.006) and HOMA-IR (p int = 0.02). Thus, elevated serum ferritin levels were associated with a decreased fasting serum insulin level among younger women but increased fasting serum insulin among older women, whereas elevated serum ferritin levels were associated with 120 min serum insulin, ISIMatsuda and HOMA-IR among older women but not among younger women (Table 4). When stratifying by age 52 years among men, we found no major differences between the younger and older group of men, and no significant interaction between ferritin levels and age <52 or ≥52 years (data not shown).

Table 4 Associations between fasting serum ferritin levels and glycaemic traits obtained during OGTTs in women from the Inter99 cohort stratified by age
Full size table

By using a piecewise linear regression model, we found a negative association between ferritin levels and insulin resistance among individuals with serum ferritin levels ≤52 μg/l (ISIMatsuda [n = 2,010] 0.04; 95% CI 0.005, 0.07; HOMA-IR [n = 2,158] −0.04; 95% CI −0.08, −0.01) but a positive association among individuals with levels >52 μg/l (ISIMatsuda [n = 3,666] −0.13; 95% CI −0.16, −0.10; HOMA-IR [n = 3,891] 0.11; 95% CI 0.08, 0.14). However, when individuals with ferritin levels <15 μg/l (the lower limit of the reference interval for women) were excluded (n = 473 for ISIMatsuda and n = 518 for HOMA-IR), the negative association with insulin resistance observed among individuals with levels ≤52 μg/l was insignificant (p = 0.46 and 0.90 for ISIMatsuda and HOMA-IR, respectively).

Elevated levels of triacylglycerol are known to be associated with decreased insulin sensitivity, and in this study increasing levels of triacylglycerol were associated with elevated levels of serum ferritin, a decreased ISIMatsuda and an elevated HOMA-IR. However, we found no statistically significant interaction between serum ferritin and serum triacylglycerol on either the ISIMatsuda (p = 0.09) or HOMA-IR (p = 0.10).

Sensitivity analyses

Additional adjustment for waist circumference, hypertension, family history of diabetes, smoking, alcohol use, physical activity, LDL-cholesterol, HDL-cholesterol and triacylglycerol resulted in an OR of type 2 diabetes for the fifth vs the first quintile of ferritin of 3.3 (95% CI 1.7, 6.5) and decreased the effect sizes for surrogate measures of beta cell function and insulin sensitivity (ESM Fig. 3 and ESM Table 2). Notably, the effect sizes for the associations between elevated serum ferritin levels and ISIMatsuda and HOMA-IR were directionally consistent with the results from models adjusted for age, sex, BMI and hsCRP only, although they were statistically insignificant. Importantly, however, 830 participants were excluded from the fully adjusted model due to lipid-lowering treatment (n = 78) or missing information on one or more covariates (n = 752), resulting in reduced statistical power.

As serum ferritin levels are influenced by a variety of diseases or conditions, for example haematological disorders and acute inflammatory conditions, we analysed the data after excluding individuals with very high serum ferritin levels, that is, ≥97.5th percentile. Generally, the effect sizes decreased slightly, but elevated serum ferritin levels remained associated with elevated plasma glucose and surrogate measures of decreased beta cell function and insulin sensitivity (ESM Table 3). Likewise, after the exclusion of individuals with serum ferritin levels ≥97.5th percentile, the OR of type 2 diabetes for the fifth vs the first quintile of ferritin decreased to 3.0 (95% CI 1.7, 5.3) (ESM Fig. 4).

Discussion

We evaluated ferritin status and glycaemic traits obtained from OGTTs in more than 6,000 Danish individuals and observed that elevated fasting serum ferritin levels were associated with surrogate measures of impaired beta cell function and decreased insulin sensitivity. Whereas the association with impaired beta cell function was present in both men and women, the association with decreased insulin sensitivity was observed among men and older women but not among younger women. We also replicated previous findings of an association between elevated ferritin levels and increased odds of type 2 diabetes. Of interest, our examination of extremely high levels of serum ferritin showed stepwise increasing odds of type 2 diabetes, with a particularly high risk among people with very high fasting serum ferritin levels.

The precise mechanisms underlying the crosstalk between iron stores reflected in serum ferritin levels and the development of type 2 diabetes remain unclarified. Three alternative interpretations of our findings may exist.

First, elevated ferritin levels may, in concert with other pathogenic factors, play a causal role in the development of impaired beta cell function and decreased insulin sensitivity. Iron mediates oxidative stress [38], and oxidative stress has many deleterious effects [1–3]. Studies in mouse models of the genetic disorder haemochromatosis have shown that pancreatic beta cells have an extreme susceptibility to oxidative damage that translates into increased apoptosis and a desensitisation of glucose-induced insulin secretion, which together result in decreased insulin secretory capacity [39]. Moreover, oxidative stress has been shown to result in impaired insulin endocytosis in endothelial cell lines [40], and iron overload has been shown to result in disturbed metabolic activity of the liver and skeletal muscle [39, 41] together supporting a role of ferritin in insulin resistance.

Second, our findings may represent reverse causality; that is, impaired glucose regulation may lead to elevated levels of ferritin rather than the converse. In support of this hypothesis, cell studies have shown that insulin regulates the transcription of ferritin and stimulates iron uptake in fat cells [42, 43].

Third, other factors, such as low-grade inflammation, may elevate both circulating ferritin levels and the risk of type 2 diabetes, a hypothesis that is supported by the fact that serum ferritin is an acute-phase reactant that is elevated in, for example, the metabolic syndrome. In the present study, additional adjustment for hsCRP generally resulted in only slightly attenuated effect sizes, suggesting that the association between elevated serum ferritin levels and impaired glucose regulation did not simply reflect confounding by low-grade inflammation.

We confirm the findings of previous studies showing a positive association between serum ferritin levels and fasting plasma glucose and serum insulin levels [16, 18], and of previous smaller studies showing an association with 120 min plasma glucose and 120 min insulin levels [29, 31]. However, our observation of an association between elevated ferritin levels and impaired beta cell function is in contrast to a study of 538 non-diabetic persons undergoing an OGTT and 69 non-diabetic persons undergoing a hyperglycaemic clamp in which no associations were observed between ferritin levels and beta cell function [31]. Our finding of an association with decreased insulin sensitivity assessed as the ISIMatsuda and HOMA-IR is in accordance with the findings of previous smaller studies [30, 31]. A study of 257 non-diabetic people applied euglycaemic–hyperinsulinaemic clamps—the gold standard to assess insulin sensitivity—and found that elevated serum ferritin levels were associated with decreased insulin sensitivity [31]. These results were supported by a study of 138 men that used a frequently sampled IVGTT to evaluate the association between ferritin and insulin sensitivity [30].

Our observation that elevated serum ferritin levels are associated with decreased insulin sensitivity in men and older women but not in younger women is intriguing. In agreement with our findings, a study of approximately 12,000 Koreans reported that serum ferritin was associated with HOMA-IR in men but not in women (pre- and postmenopausal) [44], whereas another study of 6,311 Koreans found an association between serum ferritin and HOMA-IR in men and postmenopausal women but not in premenopausal women [45]. In contrast, a significant association between serum ferritin and HOMA-IR has been reported in Chinese women (pre- and postmenopausal) (n = 277) but not in men (n = 140) [32].

Our results from a stratified analysis based on the median age of menopause suggest that the difference in effect sizes seen between men and women may rely on differences between pre- and postmenopausal women. Thus, the association between insulin sensitivity and serum ferritin seen in older but not younger women could simply signify that younger women have physiologically lower levels of ferritin, not associated with impaired glucose regulation. In support of this notion, our results suggest that there may be a threshold of serum ferritin of 52 μg/l above which elevated serum ferritin levels are associated with surrogate measures of insulin resistance. At menopause, the production of oestrogen declines and menstrual blood loss ceases, resulting in a rise in ferritin levels. The reduction in oestrogen levels seen at menopause has also been shown to increase levels of oxidative stress [46], probably reinforcing the association between ferritin and insulin sensitivity. The fact that our analysis of men stratified by age showed no difference between age groups supports an interaction between ferritin levels and menopausal status on insulin sensitivity, and not simply an interaction with age.

This study is, to our knowledge, the largest to date to examine the association between serum ferritin levels and glycaemic traits through OGTT data. Nevertheless, some limitations need to be addressed. Although ferritin is a widely used marker of iron status, the physiological role of serum ferritin and its relation to the intracellular labile iron pool has not been fully clarified, and additional analyses of other biomarkers of iron status are needed to understand in depth the role of iron in the pathogenesis of type 2 diabetes [47]. Moreover, a full blood count profile including haemoglobin was not measured in this study, precluding an analysis of the interactions between anaemia and other factors in terms of a full blood count profile and ferritin levels. In addition, the exclusion of individuals based on information on diseases and conditions influencing ferritin levels (e.g. inflammatory conditions, haemochromatosis, transfusions and iron therapy) would have been preferable to our simplistic exclusion of individuals with very high ferritin levels. Although the results from our study suggest that confounding by low-grade inflammation cannot explain the observed association between elevated serum ferritin levels and impaired glucose regulation, it is important to underline that ferritin is also an inflammatory biomarker that is elevated in the metabolic syndrome.

Along these lines, the causality of the observed association between serum ferritin and type 2 diabetes cannot be determined from this cross-sectional study. Mendelian randomisation studies could help to inform us whether serum ferritin is likely to play a causal role in development of type 2 diabetes and consequently whether the initiation of intervention studies would be reasonable. One may speculate that a Westernised diet contributes to the risk of diabetes not only through an excess caloric intake, but also through an excess iron content in the diet and via the use of iron supplements. If Mendelian randomisation studies and interventional studies confirm an aetiological role of iron in the pathogenesis of type 2 diabetes, a reduced dietary iron intake, especially in men and postmenopausal women with additional risk factors for type 2 diabetes, could be a logical consequence.

In conclusion, we found that the observed association between elevated fasting serum ferritin levels and increased risk of type 2 diabetes is explained by an association with both surrogate measures of impaired beta cell function and surrogate measures of decreased insulin sensitivity. Interestingly, menopause seems to modify the association with insulin sensitivity since the association with decreased insulin sensitivity was only observed among men and older women but not among younger women.