Introduction

Type 2 diabetes is a major cause of morbidity and mortality in the industrialised and developing world [1]. There are numerous treatment options including metformin, insulin, glucagon-like peptide 1 (GLP-1) agonists, dipeptidyl peptidase IV (DPP-IV) inhibitors, partial and full agonists of peroxisome proliferator-activated receptor gamma (PPARγ), and drugs with alternative molecular targets [2]. However, there is a need for personalised healthcare (PHC) to identify those individuals who are likely to respond best to certain treatment options in order to increase the safety and benefits of a given intervention.

Recent research has revealed that the extracellular matrix (ECM) harbours properties of an endocrine organ. Its structural proteins generate signalling molecules that can modulate cellular processes at distant sites, including cell migration, differentiation and angiogenesis. These molecules include the potent anti-angiogenic peptide, endostatin, which is derived from type XVIII collagen, and tumstatin, vastatin and restin, which are released from types IV, VII and XV collagens, respectively [3].

The microflamentous interstitial type VI collagen, a triple helical molecule composed of the constituent chains α1(VI), α2(VI) and α3(VI), is expressed in most connective tissues and prominently in adipose tissue [4], where it anchors cells through its interconnections with other ECM proteins [5]. During formation of the microfilaments, the triple helical core of this type VI collagen is proteolytically released from its pro-peptide [6, 7]. Here, further cleavage of the C-terminal pro-peptide of the α3(VI) chain generates endotrophin (also known as Pro-C6), a newly identified adipokine. Endotrophin is predominantly produced by adipose tissue and induces upregulation of TGF-β, adipose tissue fibrosis, angiogenesis and inflammation [8]. In animal models, it has also been shown to unfavourably modulate several metabolic functions, such as insulin sensitivity, food intake, energy balance [8–10] and adipose tissue inflammation [11, 12]. These findings suggest that levels of endotrophin in the blood may be useful for classifying and/or monitoring patients with metabolic dysfunction, especially those with type 2 diabetes.

Thiazolidinediones (TZDs) are PPARγ agonists and have been used widely to treat type 2 diabetes because of their ability to improve insulin sensitivity, lower glucose levels and reduce the need for insulin [13, 14]. However, the use of TZDs, such as pioglitazone, has been limited substantially by associated adverse effects (AEs), such as heart failure [15], weight gain [16], peripheral oedema [17] and bone loss in women [18]. In an attempt to minimise the AEs associated with PPARγ agonists, partial activators of PPARγ, such as balaglitazone, which trigger only a subset of PPARγ downstream signals, have been developed [19, 20]. Such partial agonists achieve good glycaemic control with a reduced risk of AEs [21]. A serum biomarker that would optimally define treatment responders could further improve the efficacy and safety associated with such glitazones.

Using our recently developed serum assay for the assessment of endotrophin [22], we hypothesised that endotrophin levels may predict insulin resistance and possibly identify those patients with an optimal response to insulin sensitisers, such as balaglitazone or pioglitazone. In the present work, data from participants of the BALLET study were used to make a head-to-head comparison of balaglitazone and pioglitazone in late-stage insulin-dependent type 2 diabetes.

Methods

Study design

The BALLET study was a phase III, randomised, double-blind, parallel-group, placebo- and active comparator-controlled clinical study designed to determine the efficacy and safety of 6 months of balaglitazone or pioglitazone treatment in individuals with type 2 diabetes on stable insulin therapy. The baseline demographics, CONSORT diagram and efficacy and safety data have previously been published [23]. The current serological assessment is a secondary post-hoc analysis in which we used the per protocol population of the BALLET study (ClinicalTrial.gov registration no. NCT00515632), which consisted of 308 individuals (from whom 297 serum samples had been procured), randomised 1:1:1:1 across four groups (placebo, balaglitazone 10 mg, balaglitazone 20 mg and pioglitazone 45 mg) as previously described [23]. Existing serum samples from the BALLET study were used and reanalysed specifically for endotrophin for the current investigation. All other measures were collected previously [23] and used in the reanalysis. Study participants gave informed consent, allowing for re-analysis of their samples, and the study was approved by the local ethics committee and carried out in accordance with the Declaration of Helsinki as revised in 2008.

Endotrophin/Pro-C6 assay protocol

An endotrophin/Pro-C6 ELISA, targeting the C-terminus of the α3 chain of endotrophin, was previously designed [22] and employed for use in the present study. This is a standard competitive ELISA, using a mono-clonal mouse detection antibody. It has previously been shown to have acceptable dilution recovery, response to spiking, and intra- and inter-individual variation [22].

Statistical analysis

The analysis included individuals from the per protocol population, from whom a baseline measurement of serum endotrophin had been obtained. Participants were grouped into tertiles based on their endotrophin level at baseline (tertile 1, ≤6.2 ng/ml; tertile 2, 6.3–7.7 ng/ml; tertile 3, ≥7.8 ng/ml). The baseline characteristics of the three subgroups were compared by ANOVA and sex was compared by Fisher’s exact test.

Spearman’s ranked correlation was conducted on baseline levels of serum endotrophin, fasting serum glucose (FSG), blood HbA1c, BMI and the derived variables of HOMA-IR and the fatty liver index (FLI). The HOMA-IR was calculated according to the homeostasis model assessment including serum glucose and insulin [24], and FLI was calculated as described by Bedogni et al [25] using the equation (triacylglycerides [mmol/l], BMI [kg/m2], gamma-glutamyl transferase [GGT; U/l] and waist circumference [cm]):

$$ \mathrm{F}\mathrm{L}\mathrm{I}=\frac{\left({\mathrm{e}}^{0.953\times { \log}_{\mathrm{e}}\left[\mathrm{triacylglycerols}\right]+0.139\times \mathrm{B}\mathrm{M}\mathrm{I}+0.718\times { \log}_{\mathrm{e}}\left[\mathrm{G}\mathrm{G}\mathrm{T}\right]+0.053\times \mathrm{waist}\;\mathrm{circumference}-15.745}\right)}{\left({1+\mathrm{e}}^{0.953\times { \log}_{\mathrm{e}}\left[\mathrm{triacylglycerols}\right]+0.139\times \mathrm{B}\mathrm{M}\mathrm{I}+0.718\times { \log}_{\mathrm{e}}\left[\mathrm{G}\mathrm{G}\mathrm{T}\right]+0.053\times \mathrm{waist}\;\mathrm{circumference}-15.745}\right)}\times 100 $$

Changes from baseline in FSG, blood HbA1c and serum endotrophin were studied as a function of time and treatment in each tertile. The least squares (LS) means ± SE were estimated from a mixed-effect repeated measure model (using baseline level, ‘visit’ (after 12 weeks of treatment) and end of treatment (after 26 weeks of treatment) with change from baseline as the dependent variable, and the baseline level vs visit, and end of treatment vs visit interactions as fixed effects, with an unstructured covariance structure for participants.

For each individual the mean change from baseline was calculated as AUC by the trapezoidal method, and the LS means ± SE were estimated from an ANCOVA model with mean change as the dependent variable, baseline level as the covariate and treatment as a fixed effect. Each tertile within each of the active treatment groups was compared with the placebo group, with the level of significance adjusted for multiple comparisons using the Dunnett method. Assessment of difference between mean change from baseline compared with zero was based on the LS mean ± SE.

All statistical calculations were performed using the SAS software package version 9.3 or higher for Windows (SAS Institute, Cary, NC, USA). Graphing was performed using GraphPad Prism version 7.01 for Windows (GraphPad, La Jolla, CA, USA).

Results

Serum endotrophin is correlated with metabolic measures

The efficacy (as assessed by metabolic measures) and safety data of the treatments used in the BALLET trial have been published elsewhere [23]. The correlations of baseline levels of endotrophin with variables associated with the metabolic syndrome are presented in Table 1. Endotrophin levels were significantly correlated with HOMA-IR, FLI, triacylglycerol and BMI, but not with FSG and HbA1c (Table 1), supporting findings that endotrophin is indeed an adipokine, related to adipocyte function, fat mass and some aspects of insulin sensitivity. Endotrophin levels were not correlated with cholesterol.

Table 1 Demographic and baseline characteristics of participants grouped according to baseline endotrophin level tertile
Full size table

In the placebo group, the correlations observed at baseline between endotrophin and the metabolic variables, FSG, HbA1c, HOMA-IR, FLI and BMI (see Table 2), were maintained at the end of the 26 week treatment period (Table 3). However, in the PPARγ agonist treatment groups, the correlation between HOMA-IR and endotrophin was eliminated, while the correlation between endotrophin and BMI or FLI persisted and even showed a trend towards being stronger (Table 3).

Table 2 Spearman correlation coefficient (rho) at baseline
Full size table
Table 3 Spearman correlation coefficients at week 26
Full size table

Endotrophin identifies responders to glitazone therapy

When assessing the change from baseline in FSG over time (Fig. 1a–d) and mean per cent change in FSG over time (Fig. 1e–h), from baseline to end of treatment (week 26), it was apparent that FSG was significantly reduced by ∼2.5 mmol/l compared with baseline and placebo for participants with endotrophin in the upper two tertiles in all three glitazone treatment groups. A less marked reduction in FSG, compared with baseline and placebo, was observed for participants in the lower tertile.

Fig. 1
figure 1

Mean change in FSG over time from baseline to end of treatment (week 26) in the (a) placebo, (b) balaglitazone 10 mg, (c) balaglitazone 20 mg and (d) pioglitazone 45 mg groups. Data were analysed using a mixed-effect repeated measures model for each tertile of baseline endotrophin. Mean absolute change in FSG from baseline to end of treatment (week 26) in the (e) placebo, (f) balaglitazone 10 mg, (g) balaglitazone 20 mg and (h) pioglitazone 45 mg groups. White circles, tertile 1; grey circles, tertile 2; black circles, tertile 3. T1, tertile 1; T2, tertile 2; T3, tertile 3. All data are LS estimates (± SE). *p < 0.05, **p < 0.01, ***p < 0.001, Dunnett-adjusted level of significance of treatment effect against placebo; § p < 0.05, §§§ p < 0.001, level of significance of change from baseline

Full size image

When response to therapy was investigated, patients in the upper two tertiles for baseline serum endotrophin were more likely to show a clinically significant response to glitazone therapy at week 26, compared with patients in the lower tertile. Furthermore, compared with participants in the lower tertile, those in the upper two tertiles had an OR of 3.83 (95% CI 1.62, 9.04) (p < 001) and 3.85 (95% CI 1.94, 7.61) (p < 0.001) for achieving a reduction in HbA1c of ≥1% or ≥0.5%, respectively (Fig. 2a). The power of baseline serum endotrophin for predicting the probability of responding to treatment is illustrated in Fig. 2b. For example, at a baseline serum endotrophin level of 4 ng/ml, the predicted probability of achieving a treatment response of a reduction in HbA1c ≥0.5% is 26%, whereas an endotrophin level of 10 ng/ml at baseline is associated with a predicted probability of treatment response of ∼65%. For changes in HOMA-IR during therapy, individuals in the upper tertiles again showed the strongest numerical improvement (Fig. 3a–c), albeit non-significant. Interestingly, despite the variable efficacy of therapy between tertiles there were no marked differences in weight gain (data not shown).

Fig. 2
figure 2

(a) Responder analysis at week 26 based on a reduction (compared to baseline) in blood HbA1c of ≥0.5% (black circle, solid line), or ≥1.0% (grey circle, dashed line). ORs (95% CI) were calculated for response to therapy based on baseline endotrophin level. The analysis compares the upper two tertiles vs the lowest tertile for treatment groups combined. **p < 0.01, ***p < 0.001. (b) Power of baseline endotrophin levels for predicting the probability of treatment response. Data from treatment groups were combined and analysed in a logistic regression model including baseline HbA1c and baseline endotrophin as predictors. Based on regression estimates the predicted probability of endotrophin was calculated at a mean baseline level of HbA1c at 8.64% (70.9 mmol/mol)

Full size image
Fig. 3
figure 3

Mean absolute change from baseline in HOMA-IR during the 26 week treatment period in participants stratified into (a) tertile 1, (b) tertile 2 or (c) tertile 3 based on baseline endotrophin levels. White bars, placebo; light grey bars, balaglitazone 10 mg; dark grey bars, balaglitazone 20 mg; black bars, pioglitazone 45 mg. All data are LS estimates (± SE). *p < 0.05, Dunnett-adjusted level of significance of treatment effect against placebo; § p < 0.05, level of significance of change from baseline

Full size image

The effect on serum endotrophin as a function of treatment and time expressed as percentage change relative to baseline is shown in Fig. 4a–d. Figure 4e–h shows the mean change from baseline at 26-weeks. Overall, an increase in serum endotrophin was observed in the placebo group and in all three glitazone treatment groups, with the exceptions of the placebo and balaglitazone 10 mg groups within the uppermost tertile.

Fig. 4
figure 4

Values relative to baseline serum endotrophin level during the 26 week treatment period. (a) Relative level in the placebo, (b) balaglitazone 10 mg, (c) balaglitazone 20 mg and (d) pioglitazone 45 mg groups. Data were analysed in a mixed-effect repeated measures model. Data are presented as LS estimates (± SE), calculated at the geometric mean level of baseline endotrophin within each tertile: white circles, 5.2 ng/ml; grey circles, 6.9 ng/ml; black circles, 9.8 ng/ml. Mean change in endotrophin relative to baseline in the (e) placebo, (f) balaglitazone 10 mg, (g) balaglitazone 20 mg and (h) pioglitazone 45 mg groups. T1, tertile 1; T2, tertile 2; T3, tertile 3. All data are LS estimates (± SE). *p < 0.05, Dunnett-adjusted level of significance of treatment effect against placebo; § p < 0.05, §§ p < 0.01, §§§ p < 0.001, level of significance of change from baseline

Full size image

Adverse effects

Lower leg oedema, measured as volume increase due to water displacement, was correlated with baseline serum endotrophin level. Certain glitazone therapies led to increased lower leg volume in the lower and middle tertiles, while there were no differences between the treatment and placebo groups in the upper tertile (Fig. 5). The AEs and severe AEs (SAEs) observed within the different tertiles are presented in Table 4. There were no significant differences in the occurrence of AEs or SAEs between the three active treatment groups when stratified according to endotrophin level. The SAEs presented in Table 4 differ from lower leg oedema (reported in Fig. 5) since lower leg volume is a quantitative measure and oedema is a patient-reported outcome.

Fig. 5
figure 5

Absolute mean change in lower leg volume from baseline to 26 weeks following treatment in participants stratified into (a) tertile 1, (b) tertile 2 or (c) tertile 3 based on baseline endotrophin levels. White bars, placebo; light grey bars, balaglitazone 10 mg; dark grey bars, balaglitazone 20 mg; black bars, pioglitazone 45 mg. All data are LS estimates (± SE). *p < 0.05, **p < 0.01, Dunnett-adjusted level of significance of treatment effect against placebo; § p < 0.05, §§§ p < 0.001, level of significance of change from baseline

Full size image
Table 4 AE profile for tertiles/treatment group
Full size table

Discussion

The ECM is increasingly recognised as a dynamic structure that does not only lend stability and spatial organisation to multicellular tissue and organisms, but that also transmits signals of differentiation, growth and migration to neighbouring cells. Moreover, the ECM can serve as an endocrine organ due to the signalling potential of certain fragments released by proteolysis, creating ligands with paracrine and potential endocrine functions [3]. Endotrophin (Pro-C6) is a soluble fragment of the C-terminal α3 chain of type VI procollagen that is released by naturally occurring proteolysis [4, 8, 22, 26]. Here we demonstrate that serum endotrophin levels are predictive of a response to the insulin sensitisers, pioglitazone and balaglitazone, in patients with type 2 diabetes. Individuals within the upper two tertiles for baseline endotrophin serum levels (tertile two, 6.3–7.7 ng/ml and tertile three, ≥7.8 ng/ml) were four times more likely to respond to treatment, compared with patients in the lower tertile (≤6.2 ng/ml baseline serum endotrophin). Since the glitazones are associated with safety concerns, such as non-fatal heart failure and bone fractures [15, 27–31], identification of individuals who will gain the most benefit from treatment and also the fewest AEs is crucial for the continued use of these drugs, which are still considered highly effective insulin sensitisers. Participants in the upper two tertiles who responded to treatment with a reduction in FPG and HbA1c levels were at a reduced risk of developing lower leg oedema, a major AE associated with glitazone treatment. These efficacy and safety data are highly relevant for predicting benefit-to-risk ratios for patients treated with glitazones; this should also apply when the medications are considered for other indications, such as liver disease [32–34].

In animal models, endotrophin levels have been shown to reflect insulin sensitivity, food intake and energy balance, as well as angiogenesis [4, 5, 8]. Accordingly, suppression of endotrophin improves insulin sensitivity and attenuated adipose tissue inflammation in animals [8]. This correlates well with our findings that elevated serum endotrophin levels are indicative of a response to PPARγ agonists. Furthermore, mRNA levels of the endotrophin precursor procollagen α3(VI) are upregulated in obese adipose tissue, paralleling adipose tissue inflammation and fibrosis [9]. The ECM and its products, especially procollagen type VI and endotrophin, may be of particular relevance in fatty liver, including in the severe form of the disease, non-alcoholic steatohepatitis (NASH). NASH is a metabolic-fibrotic disorder of the liver that shows at least a partial overlap with type 2 diabetes [32–34]. Accordingly, we expect that this novel biomarker should assist in the diagnosis and management of NASH patients where insulin sensitisers may be beneficial for subpopulations, both for the treatment of insulin resistance and liver fibrosis. Here, ECM components, in particular collagens/collagen type VI, and their functional role in the transition of fatty liver to overt fibrotic NASH needs to be further investigated.

In line with this, in the current study we observed a strong correlation between serum triacylglycerol and FLI, an index that correlates with NASH inflammatory activity and predicts more severe liver fibrosis [25]. In support of a role for type VI collagen in NASH-related fibrosis, studies have demonstrated its prominent expression in areas of active scar formation [35, 36], and elevated serum levels of the collagen VI core structure (which lacks the endotrophin domain) have been shown to be associated with advanced liver fibrosis in rodents [37], humans [38–41], and with elevated portal pressure [42]. The expression of procollagen α3(VI) is regulated by PPARγ and our findings are in alignment with this. In fact, procollagen α3(VI) mRNA is suppressed by PPARγ, as demonstrated by an increase in procollagen α3(VI) mRNA expression in adipocyte cultures treated with siRNA against PPARγ [9] and by a decrease in its transcripts in subcutaneous adipose tissue of individuals with type 2 diabetes treated with the PPARγ agonist pioglitazone, especially in patients with high baseline tissue levels of procollagen α3(VI) mRNA [12]. These data may in part explain the change in correlation, from baseline to the end of glitazone treatment, between endotrophin serum levels and HbA1c or HOMA-IR, and in particular the lack of a correlation between endotrophin and the metabolic variables. Another clinical study provides further supportive evidence by showing that tissue endotrophin levels in obese individuals correlated with chronic inflammation and systemic insulin resistance [10, 26].

Generally, the ECM has, until now, been considered to be a passive scaffold. Type VI collagen has mostly been recognised through mutations in the genes COL6A1, COL6A2 and COL6A3 that encode its three constituent chains. These mutations cause muscle disorders, such as Bethlem myopathy, Ullrich congenital muscular dystrophy, limb-girdle muscular dystrophy and autosomal recessive myosclerosis [43–46]. This provides an interesting link to metabolic dysfunction, since type VI collagen mutations are associated with muscle abnormalities and muscle represents an important regulator of insulin resistance. Therefore, the available evidence strongly suggests that collagen type VI is more than a passive ECM component, but an important mediator of adipose (and liver) metabolic dysfunction related to insulin resistance, type 2 diabetes and NASH.

Of note, the measurement of only one standard biochemical variable is a weakness of the current study; because of limited sample availability we were only able to assess endotrophin levels in serum for comparisons with HbA1c, limiting the possibility of investigating other potential biomarkers for response to treatment.

In conclusion, elevated circulating levels of endotrophin, predominantly derived from adipocytes and adipose tissue, are correlated with insulin resistance and are predictive of the response to insulin sensitisers. This enables the identification and monitoring of patients who will respond optimally to an insulin sensitiser, improving the benefit-to-risk ratio for PPARγ agonists in the treatment of type 2 diabetes and possibly NASH.