Clinical utility of brain-derived neurotrophic factor as a biomarker with left ventricular echocardiographic indices for potential diagnosis of coronary artery disease

Brain-derived neurotrophic factor (BDNF) plays a central pivotal role in the development of the cardiovascular system. Recent evidence suggests that BDNF has adverse subclinical cardiac remodeling in participants with cardiovascular disease risk factors. Relating serum BDNF levels with two-dimensional echocardiographic indices will provide insights into the BDNF mediated pathophysiology in coronary artery disease (CAD) that may shed light upon potential diagnostic biomarkers. For the study, 221 participants were recruited and classified based on coronary angiogram examination as control (n = 105) and CAD (n = 116). All participants underwent routine blood investigation, two-dimensional echocardiography, and serum BDNF estimation. As a result, total cholesterol, triglyceride, low-density lipid, high-density lipid, HbA1c (glycosylated hemoglobin), serum creatinine, eosinophils, lymphocyte, monocytes, neutrophils, and platelets were significantly elevated in CAD individuals compared to controls. Notably, the serum BDNF was significantly lower in individuals with CAD (30.69 ± 5.45 ng/ml) than controls (46.58 ± 7.95 ng/ml). Multivariate regression analysis showed neutrophils, total cholesterol, left ventricular mass index, mitral inflow E/A ratio, and pulmonary vein AR duration were associated with low BDNF in CAD. Four independent support vector machine (SVM) models performed to ensure the BDNF level in the classification of CAD from healthy controls. Particularly, the model with serum BDNF concentration and blood parameters of CAD achieved significant improvement from 90.95 to 98.19% in detecting CAD from healthy controls. Overall, our analysis provides a significant molecular linkage between the serum BDNF level and cardiovascular function. Our results contribute to the emerging evidence of BDNF as a potential diagnostic value in CAD that might lead to clinical application.


Results
patients characteristics and comparative analysis. The clinical and anthropometric characteristics of the participants involved in this study was illustrated in Table 1. All descriptive data expressed as mean ± standard deviation for control and CAD. All participants in control group are free from risk factors, Whereas the CAD group contains 10.2% of hypertensive, 15.3% of diabetes mellitus and 5.5% of smokers. However, most CAD participants in CAD are free from the risk factors. Further, the statistical t-test analysis showed a significant change in BMI, platelet, eosinophils, lymphocyte, monocytes, neutrophils, HbA1c (glycosylated hemoglobin), serum creatinine, low density lipid (LDL), high density lipid (HDL), triglyceride (TGL), and total cholesterol in CAD compared to control (Table 1). Mainly, serum BDNF was significantly (p ≤ 0.001) decreased in CAD (30.70 ± 5.4 ng/ml) compared to the control (46.580 ± 7.9 ng/ml). Although basophils, systolic, and diastolic pressure showed variations in CAD, but not statistically significant.

BDnf and cAD associated clinical and echocardiography parameters.
To determine the association between serum BDNF levels and clinical parameters, the CAD patients were quartile grouped as low and high (described in the method section). The BDNF concentration with ≤ 29.91 ng/ml considered as a low BDNF group, and those with levels > 29.92 ng/ml are considered a high BDNF group. The Supplementary Table A1 shows the characteristics of collected blood parameters and echocardiographic indices in low and high BDNF group. Elevated platelets, basophils, eosinophils, lymphocytes, monocytes, neutrophils, LDL, TGL, total cholesterol, HbA1c, and serum creatinine were observed in low BDNF group of CAD patients. In contrast, significant decreased BMI and HDL noticed in low BDNF groups compared to the high BDNF group. Similarly, the echocardiographic indices in low BDNF shows evaluated Biplane LVEF, PV AR (m/s), GLS LVEF, and decreased LVMI, MV E/A ratio, IVRT (ms), and PV S/D ratio compared to the high BDNF group. The statistical analysis of blood parameters and echocardiographic indices confirms the significance between the low and high BDNF groups, except LAD, MV S E/e' ratio, and L E/e' ratio (Supplementary Table A2).

Multivariate regression analysis.
We performed a multivariate regression analysis to determine the blood parameters influencing low BDNF concentration in CAD ( Table 3). The neutrophils (β = − 0.494, p ≤ 0.001) and total cholesterol (β = − 0.407, p ≤ 0.001) were noticed to contribute low BDNF concentration in CAD with a model fit measure of r = 0.725 and r-square = 0.526. Similarly, the echocardiographic indices (Table 4)  www.nature.com/scientificreports/ expected, the model-B with echocardiographic indices revealed high (98.64%) accuracy as the echocardiographic imaging is the important tool that routinely used in detecting CAD. However, the model-D with serum BDNF concentration as an additional attribute to the other echocardiographic parameters of model-B resulted 100% accuracy in detecting CAD.

Discussion
BDNF has been extensively studied to promote neurogenesis 18 and also plays a dominant role in the cardiovascular system 19 . BDNF expressed in endothelial cells, vascular smooth muscle cells, macrophages, lymphocytes, and atherosclerotic vessels 14 . Framingham Heart Study 2015 suggests that high BDNF concentration was associated with a decreased risk of CVD and mortality 15 . A recent study suggests the involvement of BDNF in oxidative stress in coronary artery vasculature and atherosclerotic plaque formation 20 . Despite various studies, the association of BDNF in CAD with non-invasive cardiovascular imaging techniques like two-dimension echocardiography imaging has not been reported. Therefore, we investigated the involvement of BDNF in CAD compared to healthy control. Elucidating the association of BDNF with CAD associated parameters may enable utilization of BDNF as a possible diagnostic biomarker for CAD. Particularly, establishing the relationship between alerted serum BDNF level and echocardiographic indices will help develop a potential alternative method for CAD diagnosis. This is the first study conducted in South Indian ethnicity to assess the association of serum BDNF levels with clinical parameters and echocardiography indices in CAD. We segmented our study design into threefold. First, we confirmed the significant changes in blood parameters, imaging indices, and BDNF levels in CAD compared to controls. Second, we use multivariate regression analysis to determine the blood parameters and imaging indices influencing low BDNF concentration in CAD. Finally, we generate SVM models to determine the influence BDNF in improving the classification of CAD from control by including and excluding the BDNF attribute while training and testing the models.
Our present study showed decreased serum BDNF in CAD which is in agreement with Eyileten et al., 2016, confirming the decreased serum BDNF that correlates with VCAM1 and soluble P-selectin in CAD 21 . Similarly, Aleksandra Sustar et al., 2019 reports lower BDNF in the CAD associated with an increased risk of cardiovascular events and mortality 22 . Additionally, BDNF showed a significant association with traditional risk factors, including diabetes, hypertension, smoking, physical activity, and obesity 23 . Similar results observed in our study relating low BDNF with a lipid profile and body mass index in CAD (Supplementary Table A1). Also, our results follows the similar outcome of Jiang H et al., 2011showing association with increased LDL, TGL, and decreased HDL levels with lower BDNF concentration in angina pectoris 24 . Interestingly, Ejiri et al., 2005 report altered BDNF in the coronary circulation between coronary sinus and aorta in patients with angina. Similarly, altered BDNF expression in human atheromatous intima, adventitia, macrophages, and smooth muscle cells in atherosclerotic coronary arteries 14 . These results suspect us to investigate the association of serum BDNF with echocardiography indices in CAD.
Multivariate regression analysis of blood parameters showed a significant contribution of neutrophils and total cholesterol with low BDNF (≤ 29.91 ng/ml) in angiogram proven CAD patients. Aleksandra Sustar et al., 2019 confirm the significant association of low BDNF with total cholesterol in coronary heart disease. Similar result was noticed in the Chinese population that relates low BDNF level with cardiovascular disease risk factors 22 . In addition to total cholesterol concentration, the increased neutrophils was noticed associated with low BDNF in CAD individuals. Neutrophils play a vital role in CAD, which promotes atherothrombotic mechanisms leading to cause myocardial infarction. Halade  We construct SVM models which showed significant improvement in detecting CAD with BDNF as one of the attributes (Table 5). Recently, Akella et al., 2020 use a variety of machine learning algorithms and achieved maximum accuracy of 93% in detecting CAD 26 . Interestingly our study showed benefit of adding the BDNF as one of the attribute to SVM models that represents blood parameters (neutrophils and total cholesterol) and echocardiography indices (LVMI, MV E/A, PV AR, and Biplane LVEF) in CAD (Tables 3, 4). Although our findings provides significant improvement in CAD diagnostic research, there are few limitations that to be considered before its clinical utility. First, this study includes only the South Indian population. Second, we did not follow-up on the CAD patients. Third, the changes in BDNF levels was not studied on the improvement after the treatment of CAD. Alternatively, strength of the this study need to be acknowledged that (1) selection of participants both control and CAD has been proven with coronary angiogram. (2) this study integrates the blood parameters and echocardiography indices to develop a diagnostic method that has been strengthened with a machine learning algorithm showing better accuracy in detecting CAD.
In conclusion, our study has presented a novel approach for determining the association of decreased serum BDNF with blood parameters and echocardiography indices of CAD. The machine learning (SVM) algorithm was developed to determine accuracy of BDNF with blood parameters and echocardiography indices for disease classification in CAD and healthy controls. Although all our SVM models showed better accuracy in disease classification, the model-C and model-D will be significantly improve detecting CAD using serum BDNF concentration. Therefore, our results, along with machine-based disease classification, has demonstrated the emerging evidence of BDNF in the prediction of CAD with the best accuracy value that may pave the way towards bench side clinical application. Quantification of serum BDNF. Overnight fasting blood (4 ml) was collected from the participants in BD vacutainer Plus Plastic Serum Tubes. The samples were immediately centrifuged at 3000 RPM for 15 min to separate serum from other cellular material and stored at − 80 °C for future analysis. Serum BDNF levels (ng/ ml) were measured using the ELISA kit (R&D SYSTEMS, USA) following the manufacturer's instructions. The BDNF concentrations were measured based on the optical density (OD) curve using known standards concentration provided within the kit. In addition, the cellular and biochemical parameters such as complete blood count, lipid profile, HbA1c, and serum creatinine were determined by standard laboratory techniques as a part of routine blood tests.
Statistical analysis. Student t-test was performed to confirm the statistical significance of variables in CAD compared to healthy controls. Serum BDNF concentration in CAD was stratified into quartiles as low and high levels to determine its relationship with clinical and echocardiography indices. In brief, we divide CAD patients into quartiles by merging first and second quartile to have cut-off ≤ 29.91 ng/ml, represented as low BDNF group (n = 58). The third and fourth quartiles are merged obtaining a cut-off above 29.92 ng/ml, designated as high BDNF group (n = 58). Further, the multivariate regression analysis was performed to confirm the major contributing clinical and echocardiography indices associated with low serum BDNF levels in the CAD. All statistical analysis was performed using SPSS software (version 21), and the significance was considered based on p value < 0.05.
Classification based on the machine learning algorithm. Support Vector Machine (SVM) is one of the efficient and widely used supervised machine learning algorithms for disease classification 27 . Prior to classification, the SVM Attribute Evaluator with ranker method 28 was used to select most (top five) contributing blood parameter and echocardiographic indices to achieve maximum accuracy in CAD prediction. Further, we use SVM in Weka software 29 to generate four SVM models (model-A, B, C, and D) with selected attributes from the data set (116 CAD and 105 control) for training and testing of each model. The predictor variables for model-A contains BMI, HbA1c, HDL, LDL, and Total cholesterol (Table 5). Whereas the model-B contains LVEF, LAD, L E/e, IVRT, and GLS LVEF, a top five echocardiographic indices for CAD prediction ( Table 5). The attributes for model-C contains BDNF concentration as an additional attribute along with the other five predictor variables of model-A (Table 5). Similarly, the model-D contains BDNF concentration as additional attribute of model-B for CAD (Table 5). prediction. Each data set was represented with a class attribute of "CAD" or "control" for 221 instances designated based on the coronary angiogram. A tenfold cross-validation method was adopted to measure an unbiased prediction of the models. The performance of each model was assessed based on the accuracy, true positive (TP) rate, false positive (FP) rate, precision, recall, F-measure, and receiver operating characteristic (ROC). Comparing the models based on accuracy would enable us to explore the importance of BDNF in the classification of CAD from controls.