Introduction

Thrombospondin-2 (TSP2) is a matricellular protein that interacts with various ligands such as extracellular matrix (ECM) structural proteins and is implicated in the pathogenesis of cardiovascular diseases (CVD)[1, 2]. Circulating TSP2 levels have been associated with adverse cardiovascular outcomes in patients with heart failure (HF) with preserved and reduced ejection fraction (HFpEF and HFrEF, respectively) [3, 4]. Recently, plasma proteomic profiling had demonstrated that circulating TSP2 level was associated with incident HF in a population-based cohort, as well as the presence of acute HF among individuals attending the emergency department, independent of N-terminal pro-hormone BNP (NT-proBNP) and B-type natriuretic peptide (BNP), respectively [5, 6]. In patients with advanced HF, circulating TSP2 levels decreased significantly from baseline following heart transplantation [5, 6]. However, in all these cohorts, patients with type 2 diabetes only constituted a small proportion, and the prognostic significance of circulating TSP2 levels with regard to HF in type 2 diabetes remains undefined. This is a significant knowledge gap because while patients with type 2 diabetes are at high risk of HF and its adverse outcomes [7, 8], the use of several anti-diabetic agents such as glitazones and sodium glucose co-transporter 2 inhibitors (SGLT2i) has been associated with altered risk of hospitalization for heart failure (HHF) [9, 10]. Most importantly, hyperglycemia induces TSP2 expression and increased tissue expression has been observed in patients with type 2 diabetes [11,12,

Results

Baseline serum TSP2 levels were higher in patients with type 2 diabetes who developed HHF

In Part 1, a total of 4949 patients with type 2 diabetes were included in the study. At baseline, serum TSP2 levels positively correlated with age, BMI, WC, systolic BP and serum hsCRP levels, and negatively correlated with eGFR levels of the study participants (all p < 0.001). (Supplemental Table S1) Moreover, serum TSP levels were significantly higher in men (p < 0.001), and among participants who had hypertension (p < 0.001), atrial fibrillation (p < 0.001), CVD (p = 0.004) and albuminuria (p < 0.001) than those who did not. (Supplemental Table S2)

Over a median follow-up of 7.8 years, 330 (6.7%) of them developed incident HHF, with a rate of 8.7 per 1000 person-years. Those who developed HHF had significantly longer duration of diabetes (p < 0.001), higher BMI (p = 0.002), WC (p < 0.001), systolic BP (p < 0.001), prevalence of dyslipidemia (p = 0.007), atrial fibrillation (p < 0.001), CVD (p < 0.001) and albuminuria (p < 0.001) at baseline than those who did not. Participants with incident HHF also had significantly higher glycated hemoglobin (HbA1c) and hsCRP, but lower eGFR levels (all p < 0.001) at baseline than those without. Moreover, there were significantly more users of insulin, angiotensin converting enzyme inhibitors (ACEI), angiotensin receptor blockers (ARB), beta-blockers, loop-diuretics, statin and aspirin (all p < 0.001) but less users of metformin (p = 0.021) among participants who had incident HHF than those who did not. Importantly, baseline serum TSP2 levels were significantly higher in participants with incident HHF than those without (Men: 3.70 ng/ml vs. 3.32 ng/ml; women: 4.51 ng/ml vs. 3.71 ng/ml; p < 0.001) (Table 1).

Table 1 Baseline characteristics of the participants by incident HHF in Part 1 of the study (N = 4949)
Full size table

Baseline serum TSP2 levels were independently associated with incident HHF in type 2 diabetes

In multivariable Cox regression analysis, baseline serum TSP2 levels were independently associated with the development of HHF (HR 1.27, 95%CI 1.03–1.58, p = 0.028), together with age (HR 1.04, 95%CI 1.03–1.06, p < 0.001), duration of diabetes (HR 1.01, 95%CI 1.00–1.03, p = 0.034), BMI (HR 1.04, 95%CI 1.02–1.07, p = 0.003), systolic BP (HR 1.008, 95%CI 1.003–1.010, p = 0.009), HbA1c (HR 1.14, 95%CI 1.05–1.23, p < 0.001), eGFR (HR 0.98, 95%CI 0.98–0.99, p < 0.001), presence of atrial fibrillation (HR 1.96, 95%CI 1.44–2.67, p < 0.001) and albuminuria (HR 2.18, 95%CI 1.65–2.88, p < 0.001), in a model also consisting of sex, smoking status, dyslipidemia, CVD and serum hsCRP levels at baseline (Table 2).

Table 2 Multivariable Cox regression analysis showing the associations of serum TSP2 level and other clinical variables with incident HHF (N = 4949)
Full size table

The association between baseline serum TSP2 levels and incident HHF remained significant after adjustments for the use of insulin, metformin, ACEI/ARB, beta-blockers, loop-diuretics and aspirin (HR 1.29, 95%CI 1.04–1.60, p = 0.02), as well as the time-dependent cDDD of SGLT2i (HR 1.31, 95%CI 1.06–1.62, p = 0.014) (Table 3). When study participants were stratified by the presence of CVD at baseline, serum TSP2 levels were significantly associated with incident HHF among those with baseline CVD (p = 0.029) but not in those without. Notably, 9.4% and 5.3% of participants with and without CVD at baseline developed HHF, respectively (Table 3).

Table 3 Multivariable Cox regression analysis showing the associations of serum TSP2 level with incident HHF stratified by presence of CVD at baseline (N = 4949)
Full size table

Since SGLT2i was not locally available until 2015, a sensitivity analysis was further conducted involving only participants who had survived and remained free of HHF in 2015. Among these 4812 participants, 296 developed HHF after 2015. In multivariable Cox regression analysis, baseline serum TSP2 levels remained significantly associated with incident HHF (HR 1.32, 95%CI 1.06–1.64, p = 0.014), together with the use of SGLT2i with cDDD ≥ 180 (HR 0.42, 95%CI 0.20–0.90, p = 0.025) (Supplementary Table S3).

Discrimination and reclassification performance of the addition of baseline serum TSP2 levels in predicting HHF in type 2 diabetes

The addition of serum TSP2 levels into a clinical model, which consisted of age, sex, duration of diabetes, smoking status, BMI, hypertension, dyslipidemia, atrial fibrillation, CVD, HbA1c, CKD, albuminuria and serum hsCRP level at baseline, led to a significant improvement of c-statistics in predicting HHF from 0.79 (95%CI 0.78–0.81) to 0.81 (95%CI 0.79–0.83) (p = 0.003). This was accompanied by improvement in both NRI (0.33, 95%CI 0.26–0.41; p < 0.001) and IDI (0.02, 95%CI 0.01–0.03; p < 0.001) (Table 4),

Table 4 Discrimination and reclassification performance of the addition of circulating TSP2 level in predicting HHF in diabetes
Full size table

Baseline serum TSP2 levels were associated with deterioration in average E/e’ and LAVi in patients with type 2 diabetes but without CVD

In Part 1, the association between baseline serum TSP2 levels and incident HHF seemed to be more readily observed among patients who had type 2 diabetes and CVD. To evaluate whether serum TSP2 levels were still associated with changes in ventricular function in patients with type 2 diabetes but without CVD, a longitudinal echocardiographic analysis was conducted as Part 2 of the study in 146 participants with type 2 diabetes who did not have CVD at baseline. Among them, participants who were in the highest serum TSP2 tertile had significantly higher baseline BMI than those in the lower tertiles (p < 0.01). Otherwise, their baseline clinical characteristics were comparable between the two groups (Supplemental Table S4).

Table 5 summarizes the echocardiographic parameters of these 146 participants at baseline and follow-up after a median interval of 27 months (Range 12–46 months). Compared to participants in the lower baseline serum TSP2 tertiles, those in the highest serum TSP2 tertile had significantly higher LVPWd and LV mass at baseline (both p < 0.05), as well as lower LVEF (p < 0.05), higher average E/e’ (p < 0.01) and LAVi (p < 0.01) at follow-up. Importantly, the increase in average E/e’ and LAVi from baseline were both significantly greater among participants in the highest serum TSP2 tertile than those in the lower tertiles (Both p < 0.05).

Table 5 Echocardiographic parameters of the study participants in Part 2 stratified by baseline serum TSP2 tertiles (N = 146)
Full size table

In multivariable linear regression analysis, baseline serum TSP2 levels were positively and independently associated with both changes in average E/e’ (beta 0.75, 95%CI 0.02–1.49, p = 0.04) and LAVi (beta 4.12, 95%CI 1.48–6.77, p < 0.01) (Table 6).

Table 6 Univariate and multivariable linear regression analyses showing the associations between baseline characteristics and change in echocardiographic parameters (N = 146)
Full size table

Discussion

To our knowledge, the current study is the first demonstration of a clinically significant association between circulating TSP2 levels and incident HHF in an exclusively diabetic population, known to be at increased HF risk, and was independent of the use of SGLT2i during the study period. Moreover, using an echocardiographic study, we found that baseline serum TSP2 levels were significantly associated with left ventricular remodeling and the subsequent deterioration of diastolic function in patients with type 2 diabetes but without CVD, highlighting the potential of circulating TSP2 level as a novel biomarker of HF in type 2 diabetes, regardless of the presence of CVD.

Although the association between circulating TSP2 and HF has been demonstrated previously in studies involving community-dwelling individuals and patients across the various stages of HF, only a minority of the participants had type 2 diabetes in all these studies [3,4,5,6, 27]. Notably, type 2 diabetes significantly elevates HF risk and is also associated with increased TSP2 tissue expression [7, 11,12,27]. Similarly, another recent study demonstrated that circulating TSP2 levels were inversely correlated with LVEF measured at 4 months post-myocardial infarction [28]. Therefore, our findings that circulating TSP2 levels were also associated with indices of LV hypertrophy and subsequent development of diastolic dysfunction in patients without pre-existing CVD or HF at baseline provided further clinical support that TSP2 could possibly be implicated in adverse left ventricular remodeling earlier on, even before the development of clinical HF.

TSP2 is a pro-fibrotic and anti-angiogenic protein that plays a major role in matrix assembly during tissue injury and remodelling processes [29, 30]. In preclinical studies, lack of TSP2 expression in mice after fetal murine cardiomyocyte grafting was shown to promote graft integration, vascularization leading to better graft survival [30]. On the other hand, mice with genetic ablation of TSP2 had difficulty to cope with increased cardiac loading, with augmented matrix metalloproteinase (MMP)-2 and MMP-9 activities which disrupted myocardial matrix integrity, resulting in cardiomyopathy and fatal cardiac rupture [31]. Subsequent studies also demonstrated that the absence of TSP2 predisposed mice to age-related dilated cardiomyopathy, doxorubicin-induced cardiomyopathy and heightened cardiac inflammation in viral myocarditis [24,25,26,27,28,29,30,31,32, 34]. Interestingly, TSP2 expression in the heart was paradoxically elevated during cardiac injury in both rodents and humans, especially among those that would further progress to HF with worse outcomes [31, 34]. Moreover, TSP2 expression was significantly higher in human hypertrophied hearts that were accompanied by LV systolic dysfunction than those with preserved LVEF [31]. Consistently, in our echocardiographic study, participants in the highest serum TSP2 tertile had significantly greater LV mass and LVPWd at baseline indicating LV hypertrophy. Furthermore, they also experienced significantly more deterioration in average E/e’ and LAVi which were two indices of diastolic dysfunction. Taken together, it is likely that TSP2 activation forms part of an important early-stage molecular program that precedes the development of HF. In other words, the resultant high cardiac TSP2 expression, and possibly its circulating levels in HF, may represent a futile adaptive response that was initially a bodily attempt to overcome HF development. Indeed, recent in-vitro studies demonstrated that pro-inflammatory and pro-fibrotic stimuli could increase THBS2 gene expression, which encodes TSP2, in cardiac fibroblasts [28]. Therefore, in patients with established CVD, this compensatory protective mechanism against cardiac dysfunction might even be more pronounced leading to their higher circulating TSP2 levels. Furthermore, this could also explain the significant reduction in circulating TSP2 levels from baseline following heart transplantation, as demonstrated in recent studies [5, 6], and the observation in the current study that a significant association of serum TSP2 levels with incident HHF was found only amongst those with CVD at baseline.

Our study has a few limitations. First, the observational study design in both parts of our study precludes any inferred causal relationship between high circulating TSP2 levels and incident HF. Secondly, echocardiographic data is not available in Part 1 of our study, and hence rendering it difficult to evaluate the relationship between baseline serum TSP2 levels and longitudinal changes in ECHO parameters in patients who had type 2 diabetes and CVD. Thirdly, serum TSP2 levels were measured only once in all participants and it was therefore impossible to investigate if changes in serum TSP2 levels would also contribute to the development of HHF. Furthermore, external validation of our findings using an independent cohort was not performed. Lastly, neither BNP nor NT-proBNP level was measured in our study. Interestingly, in a preclinical study of hypertensive renin-overexpressing rats, although the cardiac expression of BNP and TSP2 were both up-regulated with pressure overload, it was found that TSP2, but not BNP, more reliably identified those that were prone to develop HF progression [31]. Nonetheless, as ELISA kits for measuring serum TSP2 levels are now commercially available from various sources, further studies, preferably together with the measurements of these established HF markers, are clearly required to validate our current findings, as well as to assess the cost-effectiveness of using this novel biomarker for early HF risk stratification in type 2 diabetes.

Conclusion

Over the years, reductions in cardiovascular mortality due to coronary heart disease and stroke have been reported especially among older individuals with type 2 diabetes [35]. However, mortality related to HF did not change significantly and remained high. On the other hand, the advent of SGLT2i has introduced a paradigm shift in the management of type 2 diabetes and HF, with several landmark randomized controlled trials showing consistent benefits in reducing the rates of HHF regardless of the presence of CVD and the status of LV function [14,15,16,17,18, 36,37,38,39,40] While we recently demonstrated that circulating TSP2 had the potential to become a novel prognostic marker of liver fibrosis in type 2 diabetes [41], our current findings further proposed that it might also be usefully employed for HF risk stratification in patients with type 2 diabetes.