Type 2 diabetes is a complex disease comprising both environmental and genetic factors [1]. Low-grade inflammation is thought to play a role in the pathogenesis both of type 2 diabetes and of insulin resistance. Patients with type 2 diabetes have high mRNA expression of TNF in skeletal muscle [2], and their adipose tissue IL6 content correlates with resistance to the activation by insulin of glucose uptake in vivo and in vitro [3]. TNF is supposed to be involved in insulin resistance by being a potent inhibitor of tyrosine phosphorylation and by increasing serine phosphorylation of the insulin receptor and of insulin receptor substrate-1 [4, 5]. TNF may influence glucose metabolism through specific receptors including TNF receptor superfamily, member 1A (TNFRSF1A, also known as TNF receptor type 1 [TNFR1]), which is expressed in target tissues [6] and is also detectable in its soluble form in plasma. Previous studies have produced conflicting data on the plasma levels of soluble TNFRSF1A in subjects with type 2 diabetes and their association with insulin resistance. A study showed that intrapair differences in the AUC for glucose and insulin did not correlate with differences in serum soluble TNF receptor superfamily, member 1B (TNFRSF1B, previously known as TNF receptor type 2 [TNFR2]) concentrations in identical twins [7]. In contrast, another study showed that soluble TNFRSF1B but not soluble TNFRSF1A was associated with insulin sensitivity measured by the hyperinsulinaemic–euglycaemic clamp technique [8].

The role of IL6 in insulin resistance is controversial. Acute IL6 administration does not impair muscle glucose uptake or whole body glucose disposal in healthy humans [9], although IL6 contributes to the muscle-contraction-induced increase in hepatic glucose production [10]. IL6 induces insulin resistance in 3T3-L1 adipocytes and is, like TNF, overexpressed in human fat cells from insulin-resistant subjects [11] and in mice [12]. IL6 has been shown to increase hepatic glucose output in rat hepatocytes [13].

A current controversial issue in the aetiology of type 2 diabetes is the thrifty genotype versus thrifty phenotype hypotheses. The basis for a susceptibility to diabetes could be the result of an evolutionary advantageous genotype that promotes fat deposition and storage of calories in times of plentiful food supply [14]. In contrast, the ‘thrifty phenotype hypothesis’ links low birthweight with the risk of developing diabetes later in life, and proposes that intrauterine malnutrition leads to reduced birthweight/size and to permanently altered programming of various organ functions, which subsequently predisposes to diabetes later in life [15]. It was recently demonstrated that elderly non-diabetic monozygotic (MZ) twins were more insulin-resistant than dizygotic (DZ) twins [16], supporting a potential role of the intrauterine environment (i.e. zygosity) in the development of insulin resistance.

Twin studies represent an important tool to distinguish between the impact of genetic versus environmental components. To our knowledge no previous twin study has investigated the plasma levels of soluble TNFRSF1A, while a recent study reported low heritability of IL6 and TNF in a cohort of elderly twins [17]. However, no information about birthweight, insulin action or hepatic glucose production was available.

In the present study we investigated the heritability and impact of the prenatal environment (determined by birthweight) on the plasma levels of IL6, TNF and soluble TNFRSF1A. Furthermore, we studied the associations of these three cytokines with insulin sensitivity and hepatic glucose production (HGP), measured by the ‘gold standard’ hyperinsulinaemic–euglycaemic clamp technique, as well as NEFA plasma levels in young and elderly twins.

Subjects and methods


Subjects were identified through The Danish Twin Register as previously described [16, 18, 19]. A total of 98 same-sex twin pairs in two age groups: 22–31 years (33 MZ; 22 DZ) and 57–66 years (21MZ; 22 DZ) were included in the clinical examinations, including an oral glucose tolerance test (OGTT). Among the elderly MZ twins 76.2% had NGT, 19% had IGT and 4.8% had previously undiagnosed type 2 diabetes. Among elderly DZ twins 72.2% had NGT, 25.0% had IGT and 2.3% had previously undiagnosed type 2 diabetes. All younger DZ twins had NGT and among younger MZ twins 97% were glucose tolerant and 3.0% had IGT. There was no significant difference in glucose tolerance status between MZ and DZ twins within each age group. All twins regardless of glucose tolerance status were included in the subsequent analysis.

Zygosity was determined from polymorphic genetic markers [20]. All the subjects in our study gave informed consent. The study was approved by the regional ethics committees and it was conducted according to the principles of the Helsinki Declaration.

Clinical examinations

Body weight and height were measured with the subjects wearing light clothes and with shoes removed, and the BMI was calculated (weight [kg]/square of height [m2]). Waist circumference was measured using a soft tape on standing subjects midway between the lowest rib and the iliac crest. Hip circumference was measured over the widest part of the gluteal region, and the waist-to-hip ratio was calculated accordingly. Body composition, i.e. lean body mass and fat mass, was determined by dual energy X-ray absorptiometry scanning.

The subjects underwent a 2-h hyperinsulinaemic–euglycaemic clamp (40 mU m−2 min−1∼plasma insulin concentration at 400–500 pmol/l), with infusion of tritiated glucose (HOT-GINF). Insulin-stimulated rate of glucose disappearance (Rd) and HGP were calculated and values were expressed per kilogram lean body mass. NEFA plasma levels were measured during the basal and clamp periods. The clinical examinations have been previously described in detail [19].

Cytokine analysis

Plasma IL6, TNF and soluble TNFRSF1A concentrations were measured in one basal plasma sample from each twin by ELISA. All ELISA kits were from R&D systems (Minneapolis, MN, USA; human TNF-α Quantikine high sensitivity kit [HSTA00C], human sTNFR1 Quantikine [DRT100], human IL-6 Quantikine high sensitivity kit [HS600B]). All blood samples were analysed on the same day to eliminate inter-assay variation. Furthermore, blood samples were analysed in random order to minimise intra-assay variation. Positive and negative control samples were analysed in parallel on each plate for each cytokine. All samples were analysed in duplicate on the same plate, and measurements were repeated if there was more than 20% difference between duplicate measurements.

The intra-assay coefficients of variation for the cytokines were: IL6 11.6%, TNF 15.1% and for soluble TNFRSF1A 7.5%.

Statistical methods

Since MZ twins share 100% of their entire genome (with the exception of T-cell receptor and immunoglobulin variable region genes) and DZ twins on average share half of their segregating genes, one may question the supposition that the twins are independent observations. Accordingly, when comparing the phenotypic parameters between MZ and DZ twins, we performed an ANOVA in which we adjusted for the intra-twin-pair relationship. The full ANOVA model includes a random effect term for twin-pair membership and fixed effect term for zygosity. Data are shown as means (SD); p≤0.05 was considered significant.

Phenotypic correlations, in which the twins are included as individuals, and intra-twin-pair correlations were made in SAS using Spearman correlation. The intra-twin-pair correlations are correlations between within-twin-pair differences, allowing the elimination of common environmental effects (such as maternal and placental environment and common postnatal environmental effects) in both MZ and DZ twins. Importantly, the effects due to genotype can also be eliminated in MZ twins, while a significant intra-twin-pair correlation between two phenotypes in MZ twin pairs is of non-genetic origin. The designation of a member in a twin pair is arbitrary, i.e. there is no consistency in which of the twins in a pair is assigned A, and which is assigned B. To avoid this the intra-twin-pair correlations were calculated using 2n, as previously recommended [21]. The multiple regression analyses were performed with PROC MIXED of the SAS/STAT system (Version 8.2; SAS Institute, Cary, NC, USA). Residuals from the analyses were analysed for normality. When IL6, TNF, soluble TNFRSF1A, Rd, HGP basal, HGP clamp, NEFA basal or NEFA clamp were response variables, they were all transformed by the natural logarithm (ln) to reduce skewness in the residuals. The model was expanded, enabling adjustment for the effect of being a member of a twin pair, taking into account that this effect is different in MZ and DZ twins, respectively. Thus, the contribution to the response variable from a member of a twin pair is partly dependent on the co-twin, and this contribution is different for members of MZ and DZ twin pairs. The full regression models were reduced by stepwise elimination of the least significant variable before others. All models were adjusted for zygosity and sex.

Intraclass correlations and confidence intervals were calculated using the Mx software package [22]. Statistical comparisons of intraclass correlations were made using the Fisher z-transformation.

The impact of genes was estimated by the classic twin model (h 2=(r MZr DZ)*2) where r is the intraclass correlation coefficient. Furthermore, we applied biometric modelling to estimate the relative impact of additive and dominant genetic effects, as well as the relative impact of unique and shared environmental effects [19, 22]. Nevertheless, the application of biometric modelling was limited to plasma IL6 levels in the young twins because the remaining cytokine measurements did not fulfil the criteria for this analysis including equal means and variances within and between MZ and DZ twin pairs, as well as absence of negative correlations and higher correlations among DZ compared with MZ twins [19, 22].


Clinical characteristics

There were no differences in birthweight, birth length or adult weight between young and elderly MZ or DZ twins (Table 1). Both elderly MZ and DZ twins had higher BMI and total fat % than young MZ and DZ twins, respectively. Insulin-stimulated Rd was significantly higher in young as compared to elderly MZ twins, whereas Rd was similar in young and elderly DZ twins. Furthermore, Rd was lower among elderly MZ twins compared to elderly DZ twins (p=0.005).

Table 1 Clinical and metabolic characteristics among younger and elderly MZ and DZ twins

Elderly twins had significantly higher plasma levels of IL6 and soluble TNFRSF1A compared with young twins within both zygosity groups (Table 1).

Impact of genes on the plasma levels of IL6, TNF and soluble TNFRSF1A

Intraclass correlations for IL6, TNF and soluble TNFRSF1A plasma levels in MZ and DZ twin pairs within each age group are shown in Table 2. Higher intraclass correlations among MZ compared with DZ twins was seen solely for IL6 among the young twins and for TNF among the elderly twins, with only the latter difference reaching statistical significance (Table 2). In contrast, negative correlations among either MZ or DZ twins and/or a higher correlation among DZ compared to MZ twins were seen for the remaining cytokines. The heritability estimate (h 2=(r MZr DZ)*2) for IL6 in the young twins was 69%, indicating an impact of genetics. Biometric modelling fitted a model comprising an additive genetic component of 59% (a 2=0.59 [0.31; 0.32]) and a unique environmental component of 41% (e 2=0.41[0.23; 0.69]), consistent with the more simplistic heritability estimate.

Table 2 Intraclass correlations for IL6, TNF and soluble TNFRSF1A in MZ and DZ younger and elderly twins

The heritability estimate for TNF among the elderly twins is consistent with a predominant role of genetics in this age group. Unfortunately, the data for TNF in the elderly twins did not fulfil the criteria for standard biometric modelling because of unequal means and variances between MZ and DZ twins, so this issue could not be further clarified.

Impact of the intrauterine environment on the plasma levels of IL6, TNF and soluble TNFRSF1A

The effect of the intrauterine environment (i.e. birthweight) was analysed using intra-twin-pair correlations and regression analyses.

Intra-twin-pair correlations

There was a significant intra-twin-pair correlation between birthweight (BW) and IL6 plasma levels among both MZ and DZ twins (MZtotal: r=−0.45, p<0.001; DZtotal: r=−0.24, p=0.03) (Fig. 1a, b). This correlation was also significant when the twins were stratified according to age (MZyoung: r=−0.43, p<0.001; MZold: r=−0.50, p=0.03). In addition, a significant intra-twin-pair correlation was observed between soluble TNFRSF1A plasma levels and BW among all MZ twins and among elderly MZ twins (MZtotal: r=−0.28, p=0.004; MZold r=−0.49, p<0.001). In contrast, the plasma level of TNF was not associated with BW.

Fig. 1
figure 1

a,b Intra-twin correlations between differences in birthweight and plasma levels of IL6 among monozygotic twins (r=−0.45 p<0.001) and dizygotic twins (r=−0.24 p=0.03)

Regression analyses

Multiple regression analyses including age, sex, zygosity, BW, BMI and total fat % as explanatory variables and plasma levels of IL6, TNF and soluble TNFRSF1A as response variables, respectively, were performed. Plasma levels of IL6 were significantly influenced by BW, age and total fat %, with increasing levels of IL6 with increasing age and total fat % and decreasing BW (Table 3). The plasma soluble TNFRSF1A level was significantly influenced by zygosity with MZ having lower levels compared to DZ twins. Furthermore, age, sex and total fat % had significant effects on plasma levels of soluble TNFRSF1A, whereas, no association was seen with BW (Table 3). Plasma levels of TNF were not significantly associated with any of the explanatory variables.

Table 3 The significant explanatory variables on the cytokine plasma levels and the corresponding response variables

Metabolic parameters and plasma levels of IL6, TNF and soluble TNFRSF1A in twins

To examine the associations between cytokine levels and insulin sensitivity, intra-twin-pair correlations and multiple regression analyses were performed. Insulin sensitivity (measured as Rd) was negatively correlated with IL6 plasma levels (r=−0.25, p<0.001) and with the plasma levels of soluble TNFRSF1A (r=−0.29, p<0.001) when using phenotypic correlation analyses including all twins. A significant intra-twin-pair correlation between plasma IL6 and Rd was found in MZ twins (r=−0.22, p=0.02), supporting the notion of non-genetic association. However, after appropriate adjustments for age, sex, zygosity, total fat % and BW in the regression analysis, no significant impact of plasma IL6 or TNF levels on Rd were observed (Table 4).The plasma level of soluble TNFRSF1A, on the other hand, had a borderline significant negative influence on Rd, demonstrating that an increase in plasma soluble TNFRSF1A of 100 pg/ml results in a 2% decrease of Rd.

Table 4 The significant explanatory variables on metabolic parameters and the corresponding response variables

The plasma level of IL6 was significantly and positively correlated with NEFA levels in the basal state (r=0.29; p<0.0001). Furthermore, the plasma level of soluble TNFRSF1A was significantly and positively correlated with NEFA levels (r=0.29; p<0.0001) and with HGP (r=0.15; p=0.03) during the steady-state period of the clamp. Moreover, after adjustment for age, sex, zygosity, total fat % and BW in the regression analysis the plasma levels of soluble TNFRSF1A were still significantly and positively associated with plasma NEFA levels during insulin stimulation, and IL6 was consistently associated with plasma NEFA levels in the basal period but not during the insulin clamp (Table 4). The plasma TNF level was not associated with glucose or fat metabolism, and neither age, sex, zygosity, BMI, BW or total fat % had significant impact on plasma TNF level (Tables 3 and 4).


In the present study we report that the plasma levels of IL6, TNF and soluble TNFRSF1A in healthy twins are predominantly regulated by non-genetic factors, including an impact of the intrauterine environment as reflected by BW on plasma IL6 concentration, and to some extent as reflected by zygosity on soluble TNFRSF1A plasma levels. In addition, some evidence for a genetic component on plasma IL6 and TNF levels in young and elderly twins was observed. Importantly, the present study does not demonstrate any independent association in twins between systemic IL6 and TNF levels on the one hand, and peripheral and hepatic insulin sensitivity on the other. The level of IL6 in plasma was associated positively with plasma NEFA level in the fasting (basal) state, but not during insulin infusion. However, plasma levels of soluble TNFRSF1A tended to be associated with peripheral (Rd and clamp NEFA levels) in vivo insulin sensitivity.

The intraclass correlations for IL6 plasma levels in young MZ and DZ twins were not significantly different, which to some extent may be explained by lack of power in the present study. The heritability estimate as determined by classic twin analyses (66%) as well as biometric modelling (59%) suggested some genetic component, supporting the recent report [17] of some impact of genetics on plasma IL6 levels. Furthermore, TNF levels indicated a predominant genetic component among the elderly twins. The finding of a heritability estimate exceeding 100% (i.e. 104%) may be explained by inaccuracy of the estimate as a result of the relatively small sample size, or the fact that commom environmental factors associated with monozygosity in itself result in a greater similarity that can be explained by genetics alone. The remaining phenotypes (i.e. IL6 in elderly and TNF in young twins together with soluble TNFRSF1A in both age groups) exhibited negative correlations and/or higher intraclass correlations in DZ than in MZ twins. The somewhat surprising finding that MZ twins were less similar as compared to DZ twins indicates a minor role for genes in the aetiology of these plasma cytokine levels.

The fact that the genetic component in the plasma levels of IL6 and TNF differed between young and elderly twins indicates that age could play a role in unmasking or regulating the influence of other non-genetic aetiological components (e.g. adverse intrauterine environment) on the plasma level of IL6 and TNF in twins. In accordance, age in itself seems to influence plasma levels of several inflammatory parameters including TNF and IL6 [23], which is consistent with our findings that plasma IL6 and soluble TNFRSF1A levels increased with ageing. The recently reported heritability estimate of 17% [17] for the plasma levels of IL6 in elderly twins was rather low, and may therefore have been overlooked in this (according to elderly twins) smaller study.

Polymorphisms have been identified in both the TNF and IL6 promoter regions. Some [2427] but not all [24, 28, 29] of these polymorphisms have been associated with insulin resistance. The allelic frequencies of the polymorphisms with association to insulin resistance and type 2 diabetes were not very high in the above-mentioned studies. This may provide an additional explanation for the lack of a strong heritability on the plasma cytokines in our study, i.e. the determination of the heritability estimates is not sensitive enough to detect the impact of polymorphisms with low frequencies. Alternatively, cytokine gene expression may be predominantly regulated by environmental factors such as smoking, diet, prenatal environment or physical activity.

We found different lines of evidence of an association between plasma levels of IL6 and soluble TNFRSF1A and the intrauterine environment. The significant intra-twin-pair correlations between plasma levels of both IL6 and soluble TNFRSF1A on the one hand, and BW on the other, particularly in MZ twins, suggest that these associations are of non-genetic origin. Furthermore, when adjusting for confounding variables such as BMI, age and sex in the regression analysis, the influence of BW on the plasma level of IL6 remained statistically significant (Table 3), and therefore consistently supports a role of BW in the plasma levels of IL6.

In the present study, which included a large number of subjects, all of whom were studied using the extensive ‘gold standard’ hyperinsulinaemic–euglycaemic clamp technique to measure in vivo insulin action, the plasma levels of IL6 and TNF were somewhat surprisingly not independently associated with peripheral or hepatic insulin resistance. The discrepancy between the present study and previous studies [6, 12] may be explained by the fact that the previous studies included smaller numbers of subjects and/or studied animals. Most importantly, however, the majority of previous studies used indirect methods when investigating the relationship between plasma cytokine levels and insulin sensitivity.

TNF is supposed to be a powerful autocrine and paracrine regulator of adipose tissue and muscle metabolism. However, it may only be released systemically by adipose and muscle tissue in limited amounts [28, 30], and plasma levels may not necessarily represent a valid proxy of target tissue (muscle) TNF (or IL6) levels. Thus, the key message from this study concerning the effects of TNF on glucose and fat metabolism is that while plasma TNF levels do not seem to have any impact on insulin action in muscle, liver or adipose tissue, the effects of TNF on these parameters may be controlled locally by TNF levels in the target tissue (i.e. paracrine effects), and perhaps also by the expression and abundance of its specific receptors. Following this line of thinking, we cannot exclude the possibility that the plasma soluble TNFRSF1A levels (in contrast to the plasma TNF levels) to some extent might reflect the target tissue receptor levels and thereby indirectly the TNF activity present in muscle and adipose tissue. TNF may stimulate IL6 production and consequently IL-1 receptor antagonist (IL-1ra) and C-reactive protein levels. This subsequently may give rise to enhanced levels of soluble TNFRSF1A. Thus, chronically elevated levels of soluble TNFRSF1As, IL6, IL-1ra and C-reactive protein are likely to reflect local ongoing TNF production [31]. If so, this could explain the borderline significant negative association between soluble TNFRSF1A levels and insulin-stimulated glucose disposal, as well as the suppression by insulin of plasma NEFA levels (Table 4).

The theory of an inverse relationship between soluble TNFRSF1A levels and insulin-stimulated glucose disposal is consistent with data from a previous study of obese subjects [32]. Our results add to the previous data by showing that soluble TNFRSF1A levels besides being associated negatively with peripheral glucose disposal are also associated with impaired insulin-mediated suppression of lipolysis (NEFA).

The fact that the correlation between plasma IL6 levels and insulin sensitivity disappeared after appropriate adjustments indicates that the plasma IL6 levels are determined by other co-variates influencing both insulin action and plasma cytokine levels, such as adipose tissue mass [33, 34], BW and/or age. The relationship between plasma IL6 and basal plasma NEFA levels is consistent with the notion of a lipolytic action of IL6 [35, 36].

Intravenous infusion of IL6 increases hepatic glucose production after exercise [10]. Interestingly, this effect seems to depend on the depletion of muscle glycogen stores. Thus, intravenous IL6 infusion does not increase hepatic glucose production in subjects with normal muscle glycogen stores [10]. In the present study, we did not see any significant relationship between plasma IL6 levels and basal or insulin ‘suppressed’ rates of hepatic glucose production in the young or elderly MZ and DZ twins. The reason for this may be that muscle glycogen depots were normal in our study subjects who had not performed any exercise for more that 48 h prior to the studies. While elevated hepatic glucose production represents a key defect of metabolism in patients with overt type 2 diabetes, hepatic glucose production is commonly normal in insulin-resistant prediabetic subjects [37, 38], and is therefore considered a ‘secondary’ or non-genetic defect of metabolism in type 2 diabetes [19]. Nevertheless, the finding here of an association between low BW and elevated plasma IL6 levels may, in theory, play a role in the overt diabetic state when muscle glycogen deposits are depleted, enabling IL6 to enhance HGP, thereby contributing to the elevated level of glycaemia. Studies with intravenous infusions of IL6 in type 2 diabetic patients are needed to qualify this hypothetical role of IL6, linking low BW and IL6 with pathophysiological mechanisms contributing to hyperglycaemia in type 2 diabetes.

Besides contributing to the development of overt hyperglycaemia in patients with type 2 diabetes, the elevated IL6 levels in low BW subjects may explain (or contribute to) the association between low BW and risk of cardiovascular disease [39, 40].

In conclusion, this study demonstrates that the plasma levels of IL6, TNF and soluble TNFRSF1A are regulated primarily by non-genetic factors, including a major influence of the intrauterine environment on the plasma IL6 level, and to some extent on the soluble TNFRSF1A, but not on TNF level. Some evidence of a genetic component on plasma IL6 and TNF in younger and elderly twins, respectively, was observed. Our data do not support any major role of plasma levels of IL6 and/or TNF in themselves for either insulin resistance in skeletal muscle or hepatic glucose production in healthy non-diabetic twins. IL6 in plasma seems to increase plasma NEFA levels in the basal state, but the clinical importance of this effect is unknown. The plasma level of soluble TNFRSF1A seems to be associated with the development of peripheral insulin resistance. Further investigations to clarify the exact role and the clinical impact of soluble TNFRSF1A in the development of insulin resistance and perhaps type 2 diabetes are needed.