Characteristics of the study population

In the whole cohort, the prevalence of participants with normal weight (BMI 18.5–24.9 kg/m2), overweight (25–29.9 kg/m2) and obesity (30–39.9 kg/m2) was 50%, 37% and 13%, respectively. The overall prevalence of IGT was 8.6%, with the expected gradient across classes of BMI (5.3, 10.7 and 16.5%, respectively). As shown in electronic supplementary material (ESM) Table 1, in both men and women, lean individuals (i.e. in the first two or three BMI deciles) were slightly younger (by 2–3 years), while age remained similar across higher BMI deciles. In men, height did not differ across BMI deciles, whereas in women, higher BMI deciles were associated with a slightly shorter stature (by 2–3 cm). The weight gradient between the extreme BMI deciles was 37 kg in men and 40 kg in women. The fat% in women was almost double that in men across all BMI deciles, showing a twofold increase between the extreme deciles in both sexes. Fat mass was similar in men and women, with a fourfold increase between the extreme deciles. WHR was higher in men than in women but both sexes displayed a similar gradient across BMI deciles. FFM increased linearly in both men and women but only to a limited extent, reaching a maximum increase of 22%. Alanine aminotransferase (ALT) levels increased progressively across BMI deciles, while aspartate aminotransferase (AST) remained unchanged. As expected, the typical features of the insulin resistance syndrome (hyperinsulinaemia, lower insulin sensitivity, higher BP, lower HDL-cholesterol and higher triglycerides) gradually emerged across deciles of BMI.

Relationship between indices of obesity and FIS

FIS normalised for BSA (FISn, Fig. 1) increased with BMI and WHR, in a continuous quasi-linear fashion, similarly in men and women, with no clear threshold. Overweight, expressed as fat%, displayed a continuous and steep relationship with FISn in men, while in women the relationship was initially flat and increased above 30% (fat% × sex interaction, p<0.0001, Fig. 1). The curves between FIS and all obesity indices (BMI, fat%, and WHR) were steeper when insulin secretion was not normalised for BSA (Fig. 1); across deciles of BMI, non-normalised FIS increased by 2.4-fold while FISn increased by 1.9-fold.

Fig. 1

Polynomial third-order fit of the mean data of FISn (a–c) and non-normalised insulin secretion (FIS, d–f) across sex-specific deciles of three different obesity indices (BMI [a, d], fat% [b, e] and WHR [c, f]) in men (blue) and women (red) without diabetes. Vertical error bars indicate the 95% CIs of the mean FIS indices; horizontal error bars indicate the SDs of the obesity indices. The shadowed areas represent the 95% CI of the fit. The bivariate (obesity index, sex and interaction) analysis is presented as an insert in each plot, while the results of multivariate analysis are presented in the tables

In a multivariate analysis using only the primary anthropometric variables (fat mass, FFM, waist and hip circumferences, height) together with sex and age (Fig. 1), fat mass and waist circumference were the strongest independent predictors of FIS, along with male sex alone and in interaction with fat mass. The pattern was similar regardless of normalisation, although, a larger fraction of the overall variance was explained when insulin secretion was not normalised (Fig. 1).

The close relationship between fat mass and FIS, and the sex-related differences, can also be appreciated from simple linear regression analysis, which yielded similar r2 values (0.37 in both men and women), but different slopes (0.065 in men and 0.040 in women) (ESM Fig. 1). Among women, 167 (13%) were postmenopausal. The slopes of the fat mass–FIS relationship did not differ between pre- and postmenopausal women (ESM Fig. 2), indicating that menopausal status did not significantly modify this association.

Mechanisms through which overweight and obesity increase FIS

As insulin secretion is almost exclusively controlled by plasma glucose and by the ability of the beta cell to respond to glucose, we analysed the relationship between obesity and both FPG and fasting beta cell function cross-sectionally and also in prospective data.

As shown in Fig. 2, progressively higher BMI values were associated with a biphasic response of FPG, rising linearly until 26 kg/m2 and then reaching a plateau. The relationship between FPG and fat% differed between men and women, with women displaying a steeper curve (fat% × sex interaction, p<0.0001) and an increase in FPG across the entire range of fat% values, with no indication of a plateau. The impact of body fat distribution, as estimated by WHR, on FPG did not differ between men and women, despite the expected different range of values. In multivariate analysis (Fig. 2), among the primary obesity indices only waist circumference was an independent predictor of FPG, coming below age and sex, which were the strongest predictors (r2=0.17).

Fig. 2

Polynomial third-order fit of the mean data of FPG (a, b) and ISR@5 (c, d) across sex-specific deciles of two different obesity indices (BMI and fat%) in men (blue) and women (red) without diabetes. Vertical error bars indicate the 95% CIs of the mean of the variables on the y-axis, and horizontal bars indicate the SDs of the obesity indices. The shadowed areas represent the 95% CI of the fit. The bivariate (obesity index, sex and interaction) analysis is presented as an insert in each plot, while the results of multivariate analysis are presented in the tables

To assess beta cell function in the fasting condition, we analysed insulin secretion at a plasma glucose concentration of 5 mmol/l (ISR@5, expressed in U/h), derived from OGTT beta cell modelling. This measure represents the beta cell’s ability to respond to a fixed plasma glucose level (5.0 mmol/l), allowing for comparison between participants with different fasting values. ISR@5 proved highly sensitive to BMI and fat%, especially in men (Fig. 2). In women, the dose–response curve was less steep, with a clear rise only evident above 35% fat or a BMI of 30 kg/m2. In multivariate analysis over the primary variables, age had the strongest (negative) impact on ISR@5, followed in order of relevance by fat mass (positive) and the interaction fat mass × male sex, although they explained only a small fraction of ISR@5 variability (r2=0.11).

Among the 1016 individuals that completed the 3.5 years of follow-up, 390 participants gained more than 2 kg from baseline (Δ weight=+5.1 ± 3.8 kg, Δ fat mass=+3.4 ± 3.9 kg, Δ fat%=+2.7 ± 3.8%), 427 participants lost more than 2 kg (Δ weight=−4.7 ± 2.8 kg, Δ fat mass=−2.6 ± 4.3 kg, Δ fat%=−2.0 ± 5.4%) and the weight of 202 remained stable (Δ weight=0 ± 1 kg, Δ fat mass=+0.4 ± 2.6 kg, Δ fat%=+0.7 ± 3.9%). The proportion of women and men was similar in the three groups. At baseline, FIS was similar in the three groups and, as expected, it increased in weight gainers, did not change in those who were weight stable and decreased in weight losers. Notably, these changes closely mirrored those predicted by the cross-sectional analysis conducted on baseline data (Fig. 3). FPG increased in all the three groups but its rise was greater in those who gained weight (+0.20 ± 0.63 mmol/l) than in those who remained weight stable (+0.11 ± 0.55 mmol/l, p<0.03) or lost weight (+0.06 ± 0.55 mmol/l, p<0.006) (Fig. 3). ISR@5 declined in both weight losers and those with stable weight (−0.17 ± 1.9 and −0.16 ± 1.0 U/h, respectively; p<0.002 for both) but not in weight gainers (−0.06 ± 1.1 U/h). Interestingly, with respect to cross-sectional data, the decrease of ISR@5 in those with stable weight was greater while the increase in weight losers was smaller than expected (Fig. 3). Across the entire dataset, the change in FIS (expressed in U/h) was positively correlated with the change in fasting glucose (Stβ +0.23) and negatively correlated with the change in ISR@5 (Stβ −0.16).

Fig. 3

(a, c, d) Mean values and SEM of non-normalised FIS (a), FPG (c) and ISR@5 (d) at baseline and at 3.5 years of follow-up in the participants from the follow-up cohort (n=1016) who gained (≥2.0 kg; red line), maintained (dotted black line) or lost (≤−2.0 kg; blue line) weight relative to baseline. (b, e) The FIS (b) and ISR@5 (e) data at baseline and follow-up were plotted vs BMI together with the curves based on cross-sectional data (grey lines) and their 95% CI (grey area)

Aetiology of obesity-induced insulin hypersecretion

In search of factors that might explain the link between overweight/obesity and insulin hypersecretion, we built a multivariate model with the metabolic variables that are affected by the degree of adiposity and also have the potential to directly modulate insulin secretion, namely insulin sensitivity, plasma leptin and NEFA, along with height, waist circumference, fat mass (kg), sex and the interaction between sex and fat mass. As shown in Table 1, all these variables contributed to explaining the inter-individual variability of FIS but, compared with the model with only anthropometric variables, the increase in r2 was modest (changed from 0.43 to 0.49). The impact of fat mass, though reduced with respect to the model without metabolic variables (Stβ changed from 0.56 to 0.27), remained statistically significant and substantial, ranking among the highest. Interestingly, plasma leptin ranked second in terms of strength of the association, followed by insulin sensitivity. This pattern of association was not modified by adding family history of diabetes (present in 27% of participants, Stβ −0.04) or IGT (Stβ 0.09) into the model. We also verified whether the fat-related metabolic variables could explain the association between obesity and the major direct mechanisms that sustain insulin secretion, plasma glucose or ISR@5. Age was the strongest determinant of FPG, followed by fat mass and male sex and leptin (Table 1). ISR@5 bore strong and independent associations with leptin (positive), fat mass × male sex (positive), age (negative) and insulin sensitivity (negative) (Table 1). To further account for possible heterogeneity introduced by altered glucose tolerance, we performed a sensitivity analysis excluding individuals with IGT. The results of this subgroup analysis (ESM Table 2) were highly consistent with the findings in the whole cohort, confirming fat mass × male sex, fat mass, waist circumference, leptin and insulin sensitivity as the strongest independent determinants of FIS. We further repeated the analysis including ALT, which increased progressively across BMI deciles (ESM Table 1), as a surrogate marker of possible subclinical fatty liver disease (ESM Table 3). The results were unchanged, with fat mass × male sex, fat mass, waist circumference, leptin and insulin sensitivity remaining the main predictors of FIS, while ALT was not a significant contributor.

Table 1 Multivariate analysis of anthropometric and metabolic variables affecting fasting whole-body FIS, plasma glucose and ISR@5 Glucose production and obesity-induced insulin hypersecretion

Considering that beta cell activity is finely regulated to achieve a sinusoidal insulin concentration that regulates hepatic glucose production in response to the two major positive stimuli (i.e. glucagon and NEFA), we analysed these data. Although EGP was only measured in a subset (n=368) of the whole cohort, the relationship between FIS and fat mass in men and women was consistent with that observed in the full cohort (ESM Fig. 3). The analysis was performed over sex-specific quintiles of fat mass, pooling men and women together to increase the sample size (and the accuracy of the mean estimates); we also used the EGP values normalised per kg of FFM on the basis of our previous evidence that lean body mass is its major determinant [22]. As shown in Fig. 4, whole-body EGP (Fig. 4a), glucagon (Fig. 4c) and NEFA (Fig. 4d) did not change across deciles of fat mass, while FIS showed a progressive linear increase (Fig. 4b). When EGP was plotted against estimated sinusoidal insulin concentration (Fig. 4e), we observed a flat dose–response curve, indicating the presence of either severe hepatic insulin resistance or a fully operating homeostatic system that counteracts an increased EGP. In this cohort, we had no evidence of reduced insulin sensitivity of the liver as estimated through the per cent of clamp-induced EGP suppression, which was independent of both BMI and fat mass, (p=0.699 and p=0.130, respectively).

Fig. 4

(a–d) Polynomial third-order fit of the mean data of EGP (a), FIS (b), plasma glucagon (c) and NEFA concentrations (d) across BMI deciles in the subset of men and women with EGP data (n=368). Vertical error bars indicate the 95% CIs of the mean of the variables on the y-axis, and horizontal bars indicate the SDs of the obesity indices. The shadowed areas represent the 95% CI of the fit. The bivariate (fat mass, sex and interaction) analysis is presented as an insert in each plot. (e) EGP (mean and SD) is plotted vs estimated fasting sinusoidal insulin concentration (mean and SD)

To look deeper into this issue, we examined the linear fit of EGP vs log-transformed estimated sinusoidal insulin levels, both fasting and during the clamp (Fig. 5) across quintiles of fat mass. The curves displayed similar slopes, indicating a preserved EGP response to the clamp-induced insulin gradient. In each participant, the individual slope and the intercept were calculated. The mean of individual values across quintiles of fat mass are presented in Fig. 5 along with the predicted EGP at a fixed fasting sinusoidal insulin concentration (EGP@SI86) corresponding to the first quartile mean value (86 pmol/l). While no statistically significant difference was present in the slopes, both the intercepts and EGP at fixed insulin increased across fat mass quartiles suggesting the presence of a primary EGP increase with a preserved response to insulin.

Fig. 5

Plot of EGP vs the estimated sinusoidal insulin concentration both in fasting conditions and during the last 20 min of the euglycaemic insulin clamp in men and women grouped in sex-specific quintiles of fat mass. The data points marked by the dashed rectangle represent predicted log fit values of EGP for the sinusoidal insulin concentration corresponding to the first quartile value (86 pmol/l; EGP@SI86). Mean ± SEM values of fat mass and the data generated by the individual regression analysis per fat mass quartiles are indicated in the table