Association of Plasma Leucine-Rich α-2 Glycoprotein 1, a Modulator of Transforming Growth Factor-β Signaling Pathway, With Incident Heart Failure in Individuals With Type 2 Diabetes Free

Diabetes is a common comorbidity in patients with heart failure (HF), and presence of diabetes per se adversely affects prognosis of HF (1,2). Similarly, individuals with diabetes have more than twofold higher risk of incident HF compared with those without diabetes, and HF is a leading cause of morbidity and mortality in the population of individuals with diabetes (3). Therefore, HF and diabetes are two highly interrelated entities linked by shared risk factors (4), with deterioration of one condition being associated with worsening of the other (5). It should be noted that despite being a preventable and treatable complication in the population of those with diabetes, HF is often unrecognized in such patients (6–8).

Leucine-rich α-2 glycoprotein 1 (LRG1), a circulating protein characterized by marked periodicity in leucine residues, which enables its interaction with other proteins (9), has been identified as a modulator of the transforming growth factor-β (TGF-β) signaling pathway (10). LRG1 binds directly to TGF-β accessory receptor endoglin and enhances activation of downstream signaling pathways, including smad1/5/8, which plays pivotal roles in cellular processes such as fibrosis and angiogenesis (10,11). In addition, LRG1 has been considered as a potential acute-phase protein because it is regulated by mediators of acute-phase response, such as interleukin-6 (12). In clinical studies, plasma LRG1 has been associated with pathogenesis of several diseases, including cancer (13,14), inflammatory disease (15,16), and neurodegenerative disease (17). In our early work, we found that the level of plasma LRG1 predicted progressive kidney disease in individuals with type 2 diabetes, and this finding has since been validated by an independent study (18,19).

Fibrosis, angiogenesis, and inflammation are key pathophysiologic mechanisms underlying incident HF (20), especially in patients with diabetes (21). Given that LRG1 is involved in several cellular processes, including fibrosis, vascular and tissue remodeling, and inflammation, we hypothesized that LRG1 may play a role in the pathogenesis of HF. To our knowledge, the relationship between LRG1 and incident HF has not been prospectively examined in clinical studies. Therefore, we sought to examine whether the level of plasma LRG1 is associated with risk of incident HF and subsequent hospitalization attributable to HF (HHF) in individuals with type 2 diabetes.

Participants in the SMART2D (Singapore Study of Macro-angiopathy and Micro-Vascular Reactivity in Type 2 Diabetes) cohort (N = 2,057) were recruited from outpatient clinics in a secondary hospital and an adjacent primary care medical facility in the northern region of Singapore between August 2011 and March 2014 (22). Type 2 diabetes was diagnosed by attending physicians after excluding type 1 diabetes and diabetes attributable to other specific causes. Exclusion criteria were pregnancy, point-of-care fasting plasma glucose >15.0 or <4.5 mmol/L, HbA_1c >12% (108 mmol/mol), autoimmune disease, and actively treated cancer. Participants were followed during their regular clinical visits and by review of their electronic medical records. Follow-up for the current study was censored at date of death or 31 October 2019.

HF was diagnosed by trained cardiologists and cross validated by review of electronic medical records by two researchers of the current study (J.-J.L. and Y.M.S.) according to European Society of Cardiology 2016 criteria. Specifically, ascertainment of incident HF must have met both of the following criteria: 1) N-terminal prohormone B-type natriuretic peptide (NT-proBNP) >125 pg/mL and 2) echocardiographic evidence of HF obtained at the time or within 1 month of clinical diagnosis (23), with or without documentation of clinical symptoms and/or signs in medical records. HF was subtyped as reduced (HFrEF), midrange (HFmrEF), or preserved ejection fraction (HFpEF) based on left ventricular ejection fraction (LVEF) <40%, 40–49%, or ≥50%, respectively. Given that HFmrEF was often considered a mild form of HFrEF (24,25), we classified HFmrEF as HFrEF in the current study. HHF was ascertained by ICD-9 and ICD-10 codes in the hospitalization discharge summary.

We excluded 27 patients with a history of HF from the current study because we focused on risk of incident HF and HHF. We also excluded 44 patients with end-stage kidney disease (ESKD) (baseline estimated glomerular filtration rate [eGFR] <15 mL/min/1.73 m² and/or on sustained dialysis). Participant selection is illustrated in Supplementary Fig. 1.

The study was conducted in accordance with the principles of the Declaration of Helsinki and approved by the Singapore National Healthcare Group Domain Specific Review Committee. Each participant provided written consent.

Ethnicity, smoking status, and diabetes duration were self-reported. History of atherosclerotic cardiovascular diseases (ASCVDs), including acute myocardial infarction (AMI), stroke, percutaneous coronary angioplasty, and coronary artery bypass grafting, was determined by review of electronic medical records. Atrial fibrillation was diagnosed by regular electrocardiogram at cohort enrollment. Blood pressure and resting heart rate were measured three times using an automated blood pressure monitor in sitting position with a 5-min interval between measurements; the average of three readings was used. Mean arterial pressure was calculated as (systolic blood pressure + 2 × diastolic blood pressure)/3. HDL and LDL cholesterol and serum triacylglycerol were quantified by enzymatic methods (Roche Cobas Integra 700; Roche Diagnostics, Basel, Switzerland). HbA_1c was measured using a point-of-care immunoassay analyzer (DCA Vantage Analyzer; Siemens, Munich, Germany). Creatinine was quantified by an enzymatic method that was traceable to isotope dilution mass spectrometry reference, and urinary albumin was measured by a chemiluminescent immunoassay (Immulite; DPC, Gwynedd, U.K.). eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation, and albuminuria was expressed as albumin-to-creatinine ratio (ACR) (mg/g).

Blood specimens were collected after overnight fasting. Plasma was separated by centrifuge, aliquoted, and stored at −80°C. Plasma CRP was measured using an immunoassay kit according to the manufacturer’s instructions (R&D Systems, Minneapolis, MN). Plasma LRG1 concentration was measured using a sandwich ELISA kit (Immuno-Biological Laboratories, Hamburg, Germany). The reported intraassay coefficient of variation was 3.0–4.9%, interassay coefficient of variation was 4.2–5.1%, and sensitivity was 0.17 μg/mL.

Clinical and biochemical variables were presented as mean ± SD, median (interquartile range [IQR]), or percentage. Plasma LRG1, triacylglycerol, CRP, and urinary ACR were natural logarithmically transformed because of skewed distributions. Between-group differences were compared by Student t, one-way ANOVA, Kruskal-Wallis, Mann-Whitney U, or χ² test where appropriate.

We applied Kaplan-Meier plots to visualize cumulative HF incidence (1 − survival) stratified by plasma LRG1 quartiles. Differences among quartiles were examined by log-rank test. We fitted cause-specific Cox regression models to study the association of LRG1 with risk of incident HF, because these may be better suited to address etiologic questions in the presence of the competing risk of death (26). The outcome was time to incident HF or death, whichever occurred first, or censor date of 31 October 2019. Plasma LRG1 was modeled as continuous (per SD of natural log-transformed LRG1) and categorical (quartile) variables, respectively, and covariates were a priori selected according to biologic plausibility (27). We adjusted demographic and clinical risk factors in model 1 (age, sex, ethnicity, diabetes duration, active smoking, and history of ASCVD) and further adjusted cardiorenal risk factors in model 2 (BMI, HbA_1c, mean arterial pressure, resting heart rate, HDL and LDL cholesterol, triacylglycerol, eGFR, and urine ACR). Only 13 participants had atrial fibrillation at baseline; therefore, we did not adjust for this risk factor. Association of LRG1 with HHF was analyzed using the same approach as above. Proportionality assumption was examined using Schoenfeld residuals and by visualization of log − log plots. The hazard function of LRG1 on risk of incident HF and HHF violated the proportional hazards assumption. Therefore, LRG1 was modeled as a variable with a time-dependent coefficient in the current study.

In sensitivity analyses, we excluded participants in whom AMI occurred during follow-up. We also studied the association of LRG1 with incident HF by treating death as a competing risk in competing risk regression (Fine and Gray subdistribution).

In an exploratory analysis, we studied whether LRG1 may improve prediction of incident HF and HHF above the Thrombolysis in Myocardial Infarction Risk Score for Heart Failure in Diabetes (TRS-HF_DM), a simple integer- based risk engine using five clinical variables (prior HF, coronary heart disease, atrial fibrillation, eGFR >60 or <60 mL/min/1.73 m², and ACR level) that was derived from the SAVOR-TIMI 53 (Saxagliptin Assessment of Vascular Outcomes Recorded in Patients With Diabetes Mellitus–Thrombolysis in Myocardial Infarction 53) trial and validated in the DECLARE-TIMI 58 (Dapagliflozin Effect on Cardiovascular Events–Thrombolysis in Myocardial Infarction 58) trial (28). We used the Harrell C-index to assess the additive value of LRG1 for discrimination of incident HF and HHF above the TRS-HF_DM with 1,000-time bootstrap resampling to estimate 95% CIs.

Data analysis was performed using SPSS (version 22) and R software (version 3.4.2). Two-sided P values <0.05 were considered statistically significant.

A total of 1,978 participants with type 2 diabetes were included in the current study, and baseline characteristics were listed after stratifying participants according to LRG1 quartile. As shown in Table 1, participants with plasma LRG1 in higher quartiles were older, had longer diabetes duration, and had higher BMI, systolic blood pressure, serum triacylglycerol, CRP, and urinary ACR and lower eGFR. They were more likely to be women and of minority ethnicities and also more likely to be receiving statin, insulin, and antihypertensive medication therapies.

During a median of 7.1 years (IQR 6.1–7.6) of follow-up (13,084 patient-years), 191 incident HF events were identified (crude incidence rate 1.46 per 100 patient-years). Compared with those with no events, participants with incident HF occurring in follow-up were older, had longer diabetes duration, higher BMI, HbA_1c, blood pressure, and triacylglycerol, and lower HDL cholesterol. They had more pronounced kidney impairment, as manifested by lower eGFR and higher urinary ACR. In addition, they were more likely to have ASCVD history and be of Malay ethnicity (Supplementary Table 1).

Female participants had a higher level of LRG1 than their male counterparts (median 17.7; IQR 12.8–25.1 vs. median 14.3; IQR 10.6–19.9 µg/mL; P < 0.001). However, LRG1 did not interact with sex in association with incident HF in either the unadjusted (P interaction = 0.84) or multivariable (P interaction = 0.68) Cox regression model. Therefore, we combined male and female participants in the following analyses.

Participants with incident HF had a higher level of plasma LRG1 as compared with those with no incidence (median 20.4; IQR 15.1–29.0 vs. median 15.4; IQR 11.3–21.6 µg/mL; P < 0.001). As shown in Fig. 1, participants with plasma LRG1 in quartiles 3 and 4 had a significantly higher risk of incident HF compared with those in the lowest quartile (log-rank test P < 0.001).

In Cox regression, compared with those in quartile 1, participants with LRG1 in quartiles 3 and 4 had 3.60-fold (95% CI 1.63–7.99) and 5.99-fold (95% CI 2.21–16.20) higher risk of incident HF after adjustment for multiple demographic and cardiorenal risk factors, including eGFR and ACR. As a continuous variable, 1-SD increment in LRG1 was associated with adjusted 1.78-fold (95% CI 1.33–2.38) higher risk of incident HF (Table 2 and Supplementary Table 2). Further adjustment for CRP above the full model (model 2; Table 2) slightly attenuated the association of LRG1 with incident HF (adjusted hazard ratio [HR] 1.68; 95% CI 1.24–2.27; P = 0.001, per 1-SD increment in LRG1). Replacing BMI with waist circumference in the full model yielded a similar outcome (adjusted HR 1.90; 95% CI 1.37–2.63, per 1-SD LRG1) (29). Additional adjustment for use of insulin, calcium channel blocker, renin- angiotensin system blocker, β blocker, and diuretics above the full model did not materially alter the strength of association between LRG1 and incident HF (adjusted HR 1.76; 95% CI 1.31–2.37, per 1-SD LRG1).

We identified 23 AMI events occurring before date of incident HF and 53 AMI events in those without incident HF during follow-up. Excluding these 76 participants with interim AMI events from data analysis did not weaken the association of LRG1 with incident HF in the full model (adjusted HR 1.74; 95% CI 1.29–2.34, per 1-SD LRG1).

During follow-up, 165 death events were identified. Of these, 65 occurred after incident HF, whereas 100 occurred in participants with no incident HF. The Cox regression model suggested that LRG1 was also significantly associated with risk of death not attributable to HF (adjusted HR 1.27; 95% CI 1.02–1.58; P = 0.04, per 1-SD LRG1). Notably, the association of LRG1 with incident HF did not materially change after treating death as a competing risk (adjusted Fine and Gray subdistribution HR 1.83; 95% CI 1.44–2.32; P < 0.001, per 1-SD increment in LRG1).

Of 191 incident HF events, 83 were classified as HFrEF (63 with LVEF <40%, 20 with LVEF in range of 40–49%), whereas the other 108 were classified as HFpEF (LVEF ≥50%) at HF diagnosis (Supplementary Table 3). The level of plasma LRG1 did not significantly differ between participants with HFrEF and HFpEF (19.1; IQR 13.5–26.7 vs. 20.8; IQR 15.8–31.6 pg/mL; P = 0.21). In the exploratory analysis, 1-SD increment in plasma LRG1 was associated with 1.69-fold (95% CI 0.98–2.91; P = 0.06) and 1.34-fold (95% CI 0.94–1.91; P = 0.11) unadjusted risk for HFrEF and HFpEF, respectively (Supplementary Table 4).

At the time of or after ascertainment of incident HF, 119 participants experienced HHF. Compared with quartile 1, participants with LRG1 in quartiles 3 and 4 had significantly higher risk of HHF (log-rank test P < 0.0001) (Supplementary Fig. 2). The cause-specific Cox regression model showed that those with LRG1 in quartiles 3 and 4 had 5.88-fold (95% CI 1.83–18.85) and 10.44-fold [95% CI 2.37–45.98) higher risk of HHF after adjustment for multiple known traditional risk factors. As a continuous variable, 1- SD increment in LRG1 was associated with an adjusted 1.92-fold [95% CI 1.27–2.92] higher risk of HHF (Table 3).

The TRS-HF_DM demonstrated a high discrimination for 5-year risk of incident HF in our study population (Harrell C-index 0.79; 95% CI 0.75–0.82). Adding LRG1 to the TRS-HF_DM improved risk discrimination (Harrell C-index 0.81; 95% CI 0.77–0.85; difference 0.02; P = 0.02). Similarly, the TRS-HF_DM also showed high discrimination for 5-year risk of incident HHF in our cohort (Harrell C-index 0.81; 95% CI 0.76–0.85). Adding LRG1 to the TRS-HF_DM significantly improved risk prediction for HHF (Harrell C-index 0.84; 95% CI 0.78–0.88; difference 0.03; P = 0.02) (Supplementary Table 5).

In this prospective study in South East Asian people with type 2 diabetes, we found the following: 1) a high level of plasma LRG1 is associated with an increased risk of incident HF and subsequent HHF independent of traditional risk factors and 2) adding LRG1 to clinical risk factors might moderately improve risk discrimination for incident HF and HHF. To our knowledge, this may be the first study of the association of LRG1 with HF development and progression. Our data suggest that LRG1, a modulator of the TGF-β signaling pathway, may be a novel factor in the pathophysiologic network of HF pathogenesis in individuals with type 2 diabetes.

Clinical studies of the relationship between LRG1 and HF are scarce. In an early cross-sectional study in hypertensive patients, high level of serum LRG1 was found to be associated with prevalent HF independent of BNP (30). Interestingly, the study also found that LRG1 was expressed in myocardial tissues and correlated with TGF-β receptor 1 expression. Another clinical study found that plasma LRG1 was significantly elevated in those with a history of congestive HF among individuals with type 2 diabetes. In our exploratory analysis, participants with prevalent HF in our cohort (n = 27) also had a significantly higher level of LRG1 compared with those with no prior HF (median 26.3; IQR 18.3–34.2 vs. median 15.8; IQR 11.5–22.5; P < 0.001), which was concordant with observations in these two studies above. However, a recent cross-sectional study in patients with ESKD found that plasma LRG1 level did not differ significantly in patients with and without prevalent HF (31). This discrepancy remains to be reconciled, but it is reasonable to postulate that the relationship between LRG1 and prevalent HF may differ in populations of individuals with and without ESKD.

To our knowledge, the finding of a strong association of plasma LRG1 with risk of incident HF is novel. However, the pathophysiologic mechanisms linking elevated LRG1 level and increased risk of HF remain to be elucidated. We postulate that LRG1 may be involved in pathogenesis of HF in several pathways. First, activation of angiogenesis is one of the nodes in the pathogenic network of HF (20,21). LRG1 is an established proangiogenic factor in the presence of TGF-β (10). Therefore, it may play a role in pathogenesis of HF by modulating the TGF-related angiogenic pathway. Second, remodeling of the extracellular matrix is a key contributor to HF pathogenesis in individuals with and without diabetes (32). TGF-β is a master regulator of fibrosis, and LRG1 is an established modulator of the TGF-β signaling pathway (11). LRG1 may be involved in HF development and progression by modulating TGF-β–related profibrotic pathways. Third, LRG1 was causally involved in diabetic kidney disease in a preclinical study (18). LRG1 may be involved in pathogenesis of HF by promoting diabetic kidney disease. Indeed, adjustment for eGFR and ACR attenuated the strength of association between LRG1 and incident HF (HR change from 2.25; 95% 1.68–3.00 to 1.82; 95% CI 1.39–2.43, per 1-SD LRG1), partially supporting this hypothesis. Additionally, elevated inflammation tone is associated with a high risk of incident HF (20,21,33). However, adjustment for CRP, an established biomarker of systemic inflammation, only modestly attenuated the association of LRG1 with HF incidence. Therefore, our data do not support that elevated inflammation tone is a major mediator between LRG1 and pathogenesis of HF.

Biomarkers are essential for the identification of patients at high risk of HF for effective and early intervention (28). Our study showed that plasma LRG1 might improve risk discrimination above and beyond clinical risk factors for both incident HF and HHF, as demonstrated by a moderate but statistically significant increment in the C-index, suggesting that LRG1 may be explored as a potential biomarker for risk stratification in patients with type 2 diabetes. Future studies are needed to assess the value of LRG1 for HF risk prediction compared with established biomarkers, including NT-proBNP.

Data on incident HF in Asian populations with diabetes are scarce. The incidence rate of HF of 1.46% per year in our current study falls within the range of 0.7–2.3% per year reported in European countries and the U.S. (34). In multivariable analysis, we found that older age, higher BMI, higher HbA_1c, minority ethnicity, and prevalent diabetic kidney disease were independently associated with increased risk of incident HF, similar to observations in other populations with type 2 diabetes (27). These data suggest that findings from our study potentially could be extrapolated to other non-Asian populations. However, we would like to highlight that Asian people develop type 2 diabetes at a younger age, and prevalence of ASCVD and atrial fibrillation, two established risk factors for HF, is relatively low in Asian patients with type 2 diabetes compared with their Caucasian counterparts (35,36). These characteristics of our study population should be considered in the interpretation of our findings.

The strengths of the current study include the prospective study design, relatively large sample size, and 7-year follow-up. However, several important weaknesses should be acknowledged. First, as in all observational studies, residual confounding is inevitable. For example, we did not have data on potential risk factors for HF, such as anemia, chronic obstructive pulmonary disease, and obstructive sleep apnea (29). Although we considered interim events, such as AMI, which might precipitate risk of HF, we did not have data on other HF precipitators, including incident atrial fibrillation. Also, baseline cardiac function was not available in our cohort. The decision to conduct NT-proBNP assay and echocardiography examination was made by attending physicians in routine clinical practice. Therefore, some early-stage HF events might have gone unrecognized, as reported in other studies (6–8). Second, established HF biomarkers, especially natriuretic peptides, were not available in the current study. We identified only 33 participants with NT-proBNP measurements at baseline. We found that LRG1 was positively correlated with NT-proBNP (Spearman ρ = 0.52; P = 0.002) in this small analysis. However, we were unable to compare the performance of LRG1 for risk prediction of incident HF and HHF with that of NT-proBNP because of the small sample size. Third, the number of HHF events was relatively small, which resulted in a broad CI when LRG1 was analyzed as a categorical variable. Our current study was also obviously underpowered to examine the association of LRG1 with HF subtype. Fourth, we measured plasma LRG1 only at baseline. The dynamic change in circulating LRG1, especially at the time of HF rehospitalization, may have provided pivotal evidence to better suggest the involvement of LRG1 in HF development and progression. Finally, our study was conducted in a South East Asian cohort. External validation in other ethnic groups is needed to assess the generalizability of our findings.

In conclusion, we found that a higher level of plasma LRG1 is associated with increased risk of incident HF and HHF independent of known clinical risk factors in individuals with type 2 diabetes. Our findings may open an avenue to further characterize the underlying pathophysiologic mechanisms linking LRG1 and pathogenesis of HF, which may shed light on prevention and treatment of HF in patients with type 2 diabetes. The value of LRG1 as a novel biomarker for HF risk stratification in the population of individuals with diabetes warrants further study.

Acknowledgments. The authors warmly thank participants of the SMART2D cohort and all staff of the clinical research unit at Khoo Teck Puat Hospital, Singapore, for their contributions to the study. The authors also thank staff of the cardiology center at Khoo Teck Puat Hospital for assistance with echocardiographic data retrieval.

Funding. This work was funded by Singapore National Medical Research Council grants CSA-INV/0020/2017 and CS-IRG (MOH-000066) and Khoo Teck Puat Hospital STAR grant 18203.

The funders have no role in study design, data analysis, manuscript writing and decision for publication.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.-J.L. designed the study. S.L.T.P., S.L., K.A., Y.M.S., J.I.-S.T., R.L.G., S.T., W.E.T., C.F.S., and S.C.L. collected data and contributed important intellectual knowledge. J.-J.L. and J.W. researched the data. J.-J.L. drafted the manuscript, and all other authors revised the manuscript critically for important intellectual contents and approved publication of the manuscript. S.C.L. is the guarantor of this work and, as such, had full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.