Introduction

Lifestyle intervention has repeatedly and conclusively been shown to prevent or postpone the development of type 2 diabetes in high-risk individuals [1,2,3,4,5,6,7,8,9]. Long-term follow-up of several studies has also been reported [4, 6, 7]. The Finnish Diabetes Prevention Study (DPS) showed a relative diabetes risk reduction of up to 38% after 13 years of intervention, compared with a control group [7]. The Diabetes Prevention Program (DPP) study and the Da Qing Diabetes Prevention study also showed a significant reduction in diabetes risk (34–43%) compared with the control group, lasting for 10 to 20 years after the lifestyle intervention [4, 6].

However, the effectiveness of a standard lifestyle intervention on glucose metabolism is highly variable, and a substantial number of participants develop diabetes [1, 3, 10]. Impaired insulin secretion as well as insulin resistance are well-established pathophysiological mechanisms in the development of type 2 diabetes [11,12,13,14]. Recently, we found that, besides baseline impaired insulin secretion, the presence of an insulin-resistant fatty liver is an important predictor of the short-term response to a standard lifestyle intervention [15]. Sufficient data are still lacking to elucidate the contribution of fatty liver to the long-term effects of a lifestyle intervention [11].

It is well accepted that, besides baseline insulin sensitivity and insulin secretion, the changes in these variables during a lifestyle intervention may play a role in its success [16,17,18,19].

Perreault et al showed in the US Diabetes Prevention Program Outcome Study (DPPOS) that the regression from impaired fasting glucose (IFG) and/or impaired glucose tolerance (IGT) to a normal glucose regulation, which was strongly associated with improvement of BMI and insulin sensitivity during the lifestyle intervention, was strongly associated with a reduced incidence of diabetes [20, 21].

It is well accepted that fasting proinsulin, proinsulin to insulin conversion (PI/I ratio) and proinsulin to C-peptide ratio are predictors of the incidence of type 2 diabetes [22,23,24,25,26,27,28,29,30]. It is of great interest that lifestyle intervention in individuals with IFG and/or IGT seems to have only a slight influence on insulin secretion [31, 32]. One aspect of beta cell dysfunction in addition to insulin secretion is the conversion of proinsulin to insulin (PI/I ratio); where the conversion is adversely affected, the ratio increases and the individual develops hyperproinsulinaemia [33]. An increased fasting PI/I ratio, indicating impaired proinsulin conversion, has been observed in individuals with IGT and type 2 diabetes [34, 35]. Data on the impact of lifestyle intervention on fasting PI/I ratio are rare [36,37,38], but those available indicate that proinsulin conversion is improved after lifestyle intervention.

The Tübingen Lifestyle Intervention Program (TULIP; trial registration number KFO 114 [http://gepris.dfg.de/gepris/projekt/5396893?language=en]) was initiated in 2003 in a population at high risk for type 2 diabetes in Germany. All participants underwent a 24 month lifestyle intervention programme. One unique feature of the TULIP study is that participants repeatedly underwent whole-body MRI and 1H magnetic resonance spectroscopy (1H-MRS) before the lifestyle intervention and at the follow-up visits. This allows for the first time the precise assessment of liver fat content and body fat composition (total adipose tissue [TAT], visceral adipose tissue [VAT], subcutaneous adipose tissue [SCAT]) before, during and after the lifestyle intervention [10, 39,40,41].

We have performed a long-term follow-up of participants from the TULIP study, with the aim of identifying the risk phenotypes at baseline and changes during the lifestyle intervention that predict the long-term glycaemic non-response to the intervention. This information is important to identify early those individuals who will probably not benefit from lifestyle intervention and offer them alternative treatment.

Methods

Participants and study design

Individuals at high risk for type 2 diabetes from the southern part of Germany participated in TULIP, a prospective uncontrolled intervention study. They were included in the study when they fulfilled at least one of the following criteria: a family history of type 2 diabetes; BMI > 27 kg/m2; IGT (blood glucose ≥ 7.78 mmol/l); or previous diagnosis of gestational diabetes. Exclusion criteria were the diagnosis of diabetes mellitus or the presence of a severe critical mental or physical illness. Informed written consent was obtained from all participants and the local medical ethics committee approved the protocol.

A total of 413 individuals were included. At baseline, medical history was obtained and a physical examination performed. Phenotypic characterisation was performed using a 75 g OGTT, whole-body MRI and 1H-MRS, as well as a cycle exercise test (see below). These examinations were repeated at 9 and 24 months, and after the long-term follow-up period of about 9 years.

After the baseline examination, all participants underwent 15 sessions of standardised individual face-to-face dietary counselling by trained diabetes educators. In the first 9 months of intervention, i.e. the intensive lifestyle intervention phase of the study, monthly visits were scheduled. The moderate maintenance phase of the lifestyle intervention, comprising 3-monthly visits, lasted from the ninth month until the end of the second year.

Counselling aimed to reduce body weight by ≥ 5%. Furthermore, participants were recommended to reduce the energy intake from fat to < 30% of total energy consumed, and to increase dietary fibre intake to at least 15 g/4185 kJ (1000 kcal). They were also encouraged to reduce the intake of saturated fatty acids to < 10% of total fat. During each visit, participants presented a 3 day food diary and discussed it with the dietitians. Diet composition was estimated with a validated software, using two representative days of the 3 day diary (DGE-PC 3.0; Deutsche Gesellschaft für Ernährung, Bonn, Germany). Individuals were motivated to perform at least 3 h moderate exercise per week. A heart rate monitor (Polar, Büttelbronn, Germany) was given to the participants and they were instructed to exercise so that their heart rate was just below their individual anaerobic threshold. Individual anaerobic threshold was measured by a motorised treadmill (Saturn, HP-Cosmos, Traunstein, Germany).

OGTT

All individuals underwent a frequently sampled 75 g OGTT at baseline, after 9 months, after 24 months and at the long-term follow-up appointment. For the determination of plasma glucose and insulin levels, venous blood samples were obtained at 0, 30, 60, 90 and 120 min. Plasma glucose was measured using a bedside glucose analyser (YSI, Yellow Springs, CO, USA). AUC for glucose at 0–120 min during OGTT (AUCglucose0–120 min) was calculated with the trapezoid method. Plasma insulin and C-peptide were determined by an immunoassay with the ADVIA Centaur XP Immunoassay System (Siemens Healthcare Diagnostics, Erlangen, Germany).

Serum proinsulin concentrations were measured using a microparticle enzyme immunoassay (IBL, Hamburg, Germany) on a BEP III System (Siemens Healthcare, Erlangen, Germany). The proinsulin assay measures intact proinsulin and has 0% cross-reactivity with human insulin and C-peptide. There is cross-reactivity with the relevant proinsulin split products: des-31,32-proinsulin, 0.9%; des-32,33-proinsulin, 1.5%; des-64,65-proinsulin, 53%; and des-65,66-proinsulin, 55% [42]. Insulin sensitivity was calculated from glucose and insulin values during the OGTT as proposed by Matsuda and DeFronzo (i.e. to calculate the insulin sensitivity index [ISI]) [43]. Insulin secretion was estimated using the insulinogenic index (IGI) across 30 min (IGI0–30 min), calculated as Ins30−Ins0/Glc30−Glc0, where Ins0 and Glc0 is insulin and glucose at 0 min, and Ins30 and Glc30 is insulin and glucose at 30 min [17]. The relative insulin secretion (disposition index), which describes the insulin secretion relative to insulin sensitivity, was calculated as ISI × IGI0–30 min [17]. Fasting proinsulin to insulin conversion was determined by dividing proinsulin0 min by insulin0 min (PI/I ratio), where proinsulin0 min is proinsulin at 0 min and insulin0 min is insulin at 0 min. Fasting proinsulin to C-peptide ratio was determined by dividing proinsulin0 min by C-peptide at 0 min (C-peptide0 min) [28].

T1-weighted MRI to measure body fat

Determination of whole-body fat distribution was performed using an axial T1-weighted fast spin echo technique as previously described using a 1.5 T whole-body imager (Magnetom Sonata, Siemens Medical Solutions, Erlangen, Germany) [14, 40].

1H-MRS to measure intrahepatic fat content

Liver fat content was quantified by localised 1H-MRS as previously described [14, 40]. Non-alcoholic fatty liver disease (NAFLD) was defined as liver fat content ≥ 5.56% [44].

Maximal oxygen consumption

The individuals underwent a continuous, incremental exercise test on a cycle ergometer to volitional exhaustion. The cycle ergometer test was performed on an electromagnetically braked cycle ergometer (Ergometrics 800 S; Ergoline, Bitz, Germany). Oxygen consumption was measured using spiroergometry (Cortex MetaLyzer, Metasoft 3.9, Leipzig, Germany). Maximal oxygen consumption (\( \overset{\cdot }{V}{\mathrm{O}}_{2\max } \)) is expressed as \( \overset{\cdot }{V}{\mathrm{O}}_{2\max } \) per kg lean body mass (ml min−1 kg−1).

Statistical analyses

The primary endpoint of the study was change in AUCglucose0–120 min (response/non-response to lifestyle intervention). Data are given as mean ± SE. A Wilcoxon-signed-rank test was used to compare variables before and after lifestyle intervention and during long-term follow-up. Prior to multivariable regression analyses, variable distributions were tested for normality using the Shapiro–Wilk test. Non-normally distributed variables were transformed to their natural logarithm. To identify key model determinants, forward stepwise multivariable linear regression analyses were performed. A p value ≤ 0.05 was considered statistically significant. The software package JMP 11 (SAS Institute, Cary, NC, USA) was used for all other analyses.

Results

Longitudinal results

A total of 413 participants were enrolled in the study, with 357 undertaking 9 months of intensive lifestyle intervention, and 300 completing the full intervention in the 24 month study after (Fig. 1). We invited these 300 participants, via written invitation or telephone call, to a long-term follow-up visit; 190 individuals (63.3%) attended after a mean total follow-up time (time between baseline and long-term follow-up) of 8.7 ± 1.6 years (Fig. 1). The main reason for not participating in the long-term follow-up was unknown address or loss to follow-up (Fig. 1). Table 1 shows the participant characteristics during the four visits. After 9 months of intervention, AUCglucose0–120 min values improved in 63% of the participants (responders). This ratio of responders was decreased at the long-term follow-up to 43%. At the long-term follow-up, 46% of the 190 participants had a normal glucose regulation (normal glucose tolerance and normal fasting glucose), 43% participants had IFG and/or IGT and 21 participants (11%) developed diabetes (Fig. 2).

Fig. 1
figure 1

Trial profile of the TULIP study: screening, enrolment and post-intervention follow-up period (8.7 ± 1.6 years)

Full size image
Table 1 Characteristics of participants at different time points
Full size table
Fig. 2
figure 2

Progression of glycaemic status from baseline to long-term follow-up. Data are given as percentages. n = 190 participants. Black, baseline; light grey, 9 months of lifestyle intervention; dark grey, 24 months of lifestyle intervention; white, long-term follow-up. NGR, normal glucose regulation; T2DM, type 2 diabetes mellitus

Full size image

Among the 190 participants who attended all visits, fasting and postprandial plasma glucose levels, insulin sensitivity, the disposition index, fasting proinsulin, fasting insulin and fasting PI/I ratio improved during the 9 month intensive period of the lifestyle intervention. Body weight, liver fat content, TAT, SCAT and VAT decreased (Table 2, Fig. 3).

Table 2 Anthropometric and metabolic profile of 190 participants from baseline to the long-term follow-up
Full size table
Fig. 3
figure 3

Progression of metabolic variables and fat compartments from baseline to the long-term follow-up. Data are given as mean ± SE. AUCglucose0–120 min during OGTT (a), ISI (b), insulin secretion (disposition index [ISI × IGI0–30 min]) (c), liver fat content (d), VAT (e) and fasting PI/I ratio (f) at baseline, after 9 months and 24 months of lifestyle intervention and at long-term follow-up; LT, long-term follow-up

Full size image

Comparing the baseline with the long-term follow-up variables after 9 years, fasting PI/I ratio and fasting proinsulin improved from baseline to long-term follow-up (both p ≤ 0.0001), while post-challenge (2 h) plasma glucose, AUCglucose0–120 min during the OGTT and liver fat content did not change significantly. In contrast, fasting glucose levels increased and insulin sensitivity, as well as the disposition index, decreased significantly. Furthermore, body weight, TAT, SCAT, VAT and fasting insulin increased with the concomitant deterioration of cardiorespiratory fitness during this period (all p ≤ 0.05, Table 2, Fig. 3).

Impact of body weight change during the first 9 months of lifestyle intervention on long-term success

Decrease in body weight during the lifestyle intervention was found to associate with the success of the lifestyle intervention to improve glycaemia [3, 4, 16]. We therefore performed multivariable linear regression analysis with change in BMI during the intensive period of the lifestyle intervention as effect variable and AUCglucose0–120 min levels at long-term follow-up as the dependent variable. The model was additionally adjusted for sex, age, follow-up time, AUCglucose0–120 min and BMI at baseline. The change in BMI during the intensive period of the lifestyle intervention was associated with the AUCglucose0–120 min level in long-term follow-up (β = 0.206, p = 0.0006; Table 3, model 1).

Table 3 Prediction of the long-term success of a lifestyle intervention
Full size table

Impact of insulin sensitivity and insulin secretion on long-term success of the lifestyle intervention

Insulin secretion and insulin sensitivity are important pathophysiological factors in the pathogenesis of type 2 diabetes [11]. Therefore, these variables and the change of both during the intensive phase of lifestyle intervention were added to model 1. As shown in Table 3 (model 2), the baseline insulin secretion (β = −0.436, p ≤ 0.0001) and insulin sensitivity (β = −0.405, p = 0.0009) and the change in insulin sensitivity (β = −0.230, p = 0.0015), were associated with the AUCglucose0–120 min level at the long-term follow-up.

To visualise the effect of baseline insulin secretion on the cumulative probability of not developing diabetes, Kaplan–Meier survival curves were generated. Not a single diabetes case was diagnosed in individuals with a low disposition index compared with a high disposition index at baseline (Fig. 4a).

Fig. 4
figure 4

Cumulative probability of not developing type 2 diabetes and prediction models. (a) Cumulative probability of not developing type 2 diabetes in the TULIP study in a group with a high (dashed line) or low (solid line) disposition index at baseline, according to whether they were above or below the median. Logrank test, p < 0.0001. (b) Cumulative probability of not developing type 2 diabetes in the TULIP study in a group with (solid line) or without NAFLD (dashed line) at baseline (cut-off point, 5.56% liver fat). Logrank test, p = 0.0094

Full size image

Impact of hepatic lipid content on the long-term success of the lifestyle intervention

In a next step, hepatic lipid content at baseline and change in content during the intensive phase of the lifestyle intervention were added to the model (Table 3, model 3). Hepatic lipid content at baseline (β = 0.231, p = 0.0207) predicted AUCglucose0–120 min at long-term follow-up, in addition to baseline insulin secretion, change in insulin sensitivity and age.

To visualise the effect of baseline hepatic lipid content on the cumulative probability of not developing diabetes, Kaplan–Meier survival curves were generated. In participants with or without NAFLD, individuals without NAFLD at baseline had a significantly lower risk for developing diabetes compared with individuals with NAFLD (logrank test, p = 0.0094) (Fig. 4b).

Impact of fasting proinsulin to insulin conversion on the long-term success of the lifestyle intervention

Proinsulin conversion has been shown to have an impact on diabetes risk [34, 35]. Therefore, the fasting PI/I ratio at baseline and the change in the fasting PI/I ratio during the intensive phase of the lifestyle intervention were added to model 3. The baseline fasting PI/I ratio predicted the long-term AUCglucose0–120 min (β = 0.204, p = 0.0444), in addition to the baseline insulin secretion and the change in insulin sensitivity during the intensive phase of the lifestyle intervention. There is an inverse correlation of fasting PI/I ratio and hepatic lipid content (β = − 0.208, p = 0.0213), thus, the impact of liver fat was no longer significant in this model. Of note, model 4 gave the highest r 2 value for glucose levels at long-term follow-up, explaining 59% of variability in AUCglucose0–120 min. Furthermore, the ratio of proinsulin to C-peptide impacts similarly on long-term glycaemia as the PI/I ratio (see electronic supplementary material [ESM] Table 1).

Discussion

This long-term follow-up analysis of a standard lifestyle intervention programme aiming at preventing type 2 diabetes provides three major results: first, we show, for the first time, data on the long-term development of body fat composition and ectopic fat deposition in the liver, together with changes in insulin sensitivity and insulin secretion following a lifestyle intervention. Second, we show that the long-term success in maintaining glycaemia can be inferred from baseline variables and short-term response to the lifestyle intervention. Third, we identified a baseline high-risk phenotype for the long-term glycaemic non-response that includes, besides impaired insulin secretion and insulin sensitivity, fatty liver or impaired fasting proinsulin conversion.

Recent studies have shown that a lifestyle intervention is effective in preventing or delaying the development of type 2 diabetes during both short- and long-term follow-up [3,4,5,6,7, 45, 46]. The lifestyle changes comprised several beneficial changes in diet, as well as increased physical activity, with beneficial effects on blood glucose levels and a significant reduction in diabetes incidence [3, 45]. The TULIP intervention was most intensive during the first 9 months. Consequently, the benefit was most prominent during this time. Body weight reduction during the intensive period was moderate, as observed in two other studies including individuals with IGT [3, 45].

Using whole-body MRI, we observed a marked reduction in major body fat compartments (TAT, VAT and SCAT) between 10% and 15% during the first 9 months of intensive intervention. Importantly, liver fat content improved by more than 30% during that period. This ‘qualified’ weight reduction with decreases in visceral fat mass and liver fat content improved insulin sensitivity by nearly 15%. These risk factors are known to be associated with metabolically unhealthy obesity [47]. VAT and intrahepatic fat are the source of several hormones and cytokines that act against insulin and initiate a chronic systemic activation of the immune system. This chronic systemic inflammation is the key driver for further impairment of insulin resistance and also for the chronic progressive nature of the disease [48]. Lifestyle interventions can only modulate disease development if there is an impact on the hormonal activity of the body fat compartments and not just on the volume of body fat [49].

In contrast to the favourable short-term effects of a lifestyle intervention, the improvements in metabolic variables are lost over time when body weight and body fat composition, as well as insulin sensitivity and insulin secretion, deteriorate. Despite gain in body weight and body fat, postprandial blood glucose levels and liver fat content were comparable with baseline levels, even after 9 years (Table 2, Fig. 3). One important variable that improved during the lifestyle intervention and was improved at long-term follow-up was fasting proinsulin conversion. As fasting insulin was doubled at long-term follow-up compared with baseline and C-peptide was unchanged, this indicates declining hepatic insulin clearance over this time (9 years), suggesting improved proinsulin conversion. Fasting proinsulin levels also decreased by ~ 50%, also indicating that proinsulin conversion was improved.

In addition, it should be highlighted that a rate of 11% incident diabetes is rather low compared with similar lifestyle intervention programmes. In the lifestyle intervention group of the Finnish DPS and the US DPPOS, 29% and 40% of participants, respectively, developed diabetes during the long-term follow-up [4, 7]. This might in part be because both studies exclusively included participants with IGT at baseline, while the TULIP study included a metabolically healthier cohort. Taken together, our results are in line with earlier studies showing that healthy changes in diet and increased exercise can have beneficial long-term effects on glycaemia, and offer encouraging evidence for the effectiveness of a lifestyle intervention, even without a large reduction in body weight [4, 6, 7, 20].

Nevertheless, the problem of non-response with respect to an improvement of glycaemia during the lifestyle intervention in a substantial number of participants remains [4, 6, 7, 20]. It is, therefore, clinically important to define risk factors and risk phenotypes associated with such a non-response.

One factor might be the decrease in insulin sensitivity that is correlated with the amount of weight loss [19]. However, the correlation of weight loss and glycaemia is not very strong [15]. Paradoxically, in the US DPPOS a stronger weight loss even adversely affected diabetes risk, independent of the previous intervention [20]. Thus, other metabolic determinants may additionally affect the long-term effectiveness of a lifestyle intervention. In the present study, we addressed the pathophysiological determinants that predict non-response to the lifestyle intervention. We were able to show that low insulin secretion capacity at baseline, low insulin sensitivity and impaired improvement in insulin sensitivity during the intensive first 9 months of the lifestyle intervention predicted non-response to the intervention. These findings accord with previous studies that have shown that both baseline insulin secretion and insulin sensitivity, as well as the change thereof during a lifestyle intervention, are predictors of diabetes development [16, 19, 50] or, vice versa, regression to normal glucose tolerance [20, 21].

A new finding was that a high liver fat content at baseline or an impaired fasting relative proinsulin to insulin conversion at baseline were important determinants of impaired glycaemia at long-term follow-up, independent of insulin secretion and insulin sensitivity. Models containing these pathophysiological factors provided a better prediction (r 2 = 0.55 and 0.59, respectively) of long-term glycaemia than a simpler model containing baseline BMI as well as change in BMI (r 2 = 0.45) (Table 3).

The conversion of proinsulin to insulin is thought to be a biomarker for beta cell function [26, 51, 52]. It is important that in the present study proinsulin to insulin conversion predicted long-term glycaemia after the lifestyle intervention, independently of insulin secretion and insulin sensitivity (model 4). Our results are in line with earlier studies that showed that increased proinsulin levels, which are the consequence of impaired proinsulin conversion to insulin, predict worsening of glucose tolerance status [53, 54]. Furthermore, it has been shown that fasting proinsulin conversion and the acute insulin response independently impact on the development of type 2 diabetes [27, 55] highlighting that there are multiple aspects to beta cell dysfunction. In this context, the improvement of proinsulin to insulin conversion after the lifestyle intervention and at the long-term follow-up is an important finding. One may speculate that proinsulin conversion is a useful biomarker for both the prediction of diabetes risk and monitoring the success of the lifestyle intervention.

For the design, development and practice of improved and more successful lifestyle intervention programmes, it is important to reveal risk factors for short- and long-term success of a lifestyle intervention. Regarding short-term success, we previously found that the phenotype encompassing participants with a low disposition index or insulin-resistant NAFLD have a high risk for acute glycaemic non-response to lifestyle intervention [15]. The present study now demonstrates that such a phenotype is also associated with long-term non-response and extends the phenotype to impaired proinsulin conversion. Individuals with such a high-risk phenotype may be put on a more intensive lifestyle intervention at an earlier stage, or even receive pharmacological therapy.

One limitation of the present analysis is that the TULIP study was designed as an uncontrolled study. Therefore, we cannot compare the risk of non-response to the lifestyle intervention (i.e. improvement in glycaemia) with the risk of developing diabetes without intervention. However, a lifestyle intervention is standard care in individuals at high risk for type 2 diabetes [56] after the publication of the landmark studies, the DPS and DPP, and thus, a non-treated control group seems inappropriate. Furthermore, our study was conducted at a single academic research centre and the results cannot probably be generalised to other populations.

In conclusion, we found that, independent of the important change in body weight, an impaired insulin secretion, a fatty liver or an impaired fasting proinsulin conversion before lifestyle intervention predicted glycaemic non-response to the lifestyle intervention programme. Our current results may help to develop individualised prevention programmes targeting those with at-risk phenotypes. Furthermore, they may also aid monitoring during such programmes and may help to identify those for whom treatment intensification will be necessary.