3.1 Identification of plasma sEVs from monocytes of CAD patients
We first extracted monocytes from plasma, and the results of the analysis of the monocytes can be found in the supplementary Fig. S1 of our published article (Cell Death Dis. 2021 Oct. 14;12(10):948. doi: 10.1038/s41419-021-04253-y). Through transmission electron microscopy (TEM), we found that the sEVs were approximately 100 nm in diameter and conformed to the known shape of sEVs. Next, we detected the sizes of sEVs using the NTA method and found that the sEVs from monocytes were between 40 and 150 nm in diameter (Figure 1C), which was consistent with the common size of known sEVs(15) and with the TEM results (Figure 1A). In a western blot assay, the protein markers CD63 and CD81 were more highly expressed in sEVs than in monocytes (Figure 1B), and the statistical data of CD63 and CD81 expression were supplemented in Figure S2. Thus, we successfully extracted sEVs secreted by monocytes isolated from CAD and healthy participants.
3.2 Baseline data
This study enrolled 243 subjects, including 123 subjects in the CAD group and 114 subjects in the control group. Age, sex distribution, body mass index (BMI), diabetes mellitus (DM), smoking, drinking, mean arterial pressure (MAP), heart rate (HR), triglycerides (TG), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), apolipoprotein a (apo-a), apolipoprotein b (apo-b), fasting blood glucose (FPG), left ventricular size, and left ventricular ejection fraction (LVEF) were not significantly different between the two groups. However, lipoprotein a (LPa), homocysteine (Hcy), uric acid, glycated hemoglobin (HbA1c), and the rate of hypertension in the CAD group were higher than those in the control group. In addition, HDL-C was lower in the CAD group than in the control group (Table 1).
3.3 Screening of differentially expressed lncRNAs using the microarray probe method
A total of 58,540 lncRNA probes for three CAD patients and three healthy controls were detected using microarray probes. The CAD1 patient was a female younger than 60, but the CAD2 and CAD3 patients were males older than 70. The result graph of PCA shows that CAD and control group can be well distinguished (Figure 2A). The results showed that the lncRNA expression profile of CAD patients was significantly different from that of the control group (Figures 2B and 2C). Compared to the control group, 89 lncRNAs were upregulated and 211 lncRNAs were downregulated in the CAD group (Table S3). Next, 10 lncRNAs with the most significant differential expression in the lncRNA array were validated in 10 CAD patients and 10 healthy controls by RT-qPCR. The results showed that SNAR-E was the most significantly upregulated lncRNA, while RPL34-AS1 was the most significantly downregulated lncRNA (Figure 2D).
3.4 Exploration of a possible mechanism underlying differentially expressed lncRNA leading to CAD
We predicted 234 and 1,212 mRNAs might be targeted by differentially expressed lncRNA by cis and trans prediction methods, respectively. The details of how the two methods of predicting mRNA and the corresponding relations are given in Tables S4 and S5, respectively. The mRNA predicted by the significantly differentially expressed lncRNAs in these two groups were included in our Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis to show the pathological pathways of the differentially expressed lncRNAs (Figure 3). The results of KEGG were classified and analyzed according to the type of disease (level 2). These results suggested that both mRNA predicted using the cis method and mRNA predicted using the trans method could be enriched in systemic diseases closely related to CAD, such as cardiovascular diseases, circulatory diseases, and lipid metabolic diseases (Figures 3A and 3B). The top 30 KEGG enrichment pathways suggest that mRNA from both prediction methods can be enriched in fatty acid metabolism, D−Glutamine and D−glutamate metabolism, apoptosis, Focal adhesion, and Endocrine and other factor−regulated calcium reabsorption pathways, which are closely related to the occurrence and development of CAD (Figures 3C and 3D). The mRNA function of the cis prediction source is also enriched in the vasopressin-regulated water reabsorption and proximal tubule bicarbonate reclamation pathways, which are closely related to the increase in front and back load in the circulatory system. The mRNA function of the trans prediction source is still enriched in inflammation-related pathways, such as TNF signaling pathway and NOD-like receptor signaling pathway, which are likely to promote the progression of atherosclerosis.
We then carried out gene ontology (GO) enrichment analysis on the mRNA predicted by the two methods. The functions of mRNA predicted by both methods were concentrated in pathways related to the pathogenesis of CAD, such as interferon−gamma−mediated signaling pathway, positive regulation of adaptive immune response, negative regulation of systemic arterial blood pressure, extracellular vesicle biogenesis, 2−oxoglutarate metabolic process, regulation of calcium ion transmembrane transporter activity, and regulation of epidermal growth factor receptor signaling pathways. The mRNA function of the trans prediction source was also found to be enriched in extracellular exosome biogenesis and exosomal secretion pathways, which suggested that these mRNA may play an important role in the occurrence and development of CAD through the secretion and transport of regulators (Figures 4A and 4B). As shown in Figure 4C, a hub lncRNA–mRNA interaction network based on validated lncRNAs was constructed using cytohubba and MCODE plugin of Cytoscape. Finally, we selected the mRNA molecules closely linked to SNAR-E and RPL34-AS1 for KEGG analysis, and the results suggested that the functions of these two lncRNA-regulated mRNA molecules are mainly concentrated in lipid metabolic and calcium regulation pathways, such as fatty acid metabolism, biosynthesis of unsaturated fatty acids, glycerolipid metabolism, PPAR signaling pathway, adipocytokine signaling pathway, phosphatidylinositol signaling system, and glycerophospholipid metabolism (Figure 4D).
3.5 Expression levels of SNAR-E and RPL34-AS1 in plasma and sEVs accurately diagnose CAD
To evaluate the application value of lncRNAs in differentiating between the CAD group and the healthy control group, the most significantly upregulated lncRNA (SNAR-E) and the most significantly downregulated lncRNA (RPL34-AS1) were selected. The lncRNA expression in plasma and monocyte sEVs was detected in the expanded samples (n = 237). As shown in Figures 5A and 5D, SNAR-E expression in both plasma and monocyte sEVs in the CAD group was significantly higher than that in the control group (P < 0.001, FDR = 0.031), whereas RPL34-AS1 expression in the CAD group was lower than that in the control group (P < 0.001, FDR = 0.015). Notably, there was a strong positive correlation between the expression of SNAR-E and RPL34-AS1 in plasma and in monocyte sEVs (n = 237, Figure 5G).
To further study the potential value of SNAR-E and RPL34-AS1 as biomarkers for CAD, the diagnostic value of SNAR-E and RPL34-AS1 for CAD was investigated using ROC analysis (n = 237). As shown in Figure 5C, the area under the curve (AUC) for SNAR-E in monocyte sEVs was 0.950 (95% confidence interval (CI): 0.920–0.980; P<0.001), the best cutoff value was 1.266, and the sensitivity and specificity were 0.992 and 0.868, respectively. The AUC for RPL34-AS1 monocyte sEVs was 0.922 (95% CI: 0.868–0.952; P<0.001), the best cutoff value was 10.351, and the sensitivity and specificity were 0.823 and 0.967, respectively (Figure 5B). The AUC for SNAR-E in plasma was 0.953 (P<0.001), the best cutoff value was 1.413, and the sensitivity and specificity were 0.992 and 0.878, respectively (Figure 5F). The AUC for RPL34-AS1 in plasma was 0.910 (P<0.001), the best cutoff value was 10.513, and the sensitivity and specificity were 0.816 and 0.951, respectively (Figure 5E). In addition, we used the lncRNA website lncSEA (http://bio.liclab.net/LncSEAv2/index.php) (16) to predict the proteins that RPL34-AS1 may bind to, and conducted enrichment analysis of these protein molecules (N = 349), from which we found that they were mainly concentrated in cell growth, differentiation, and stress-related pathways, such as RNA biosynthetic processes, cell fate commitment, and cellular responses to stress (Table S6). This suggested that RPL34-AS1 may protect the cardiovascular system by regulating cell development and the response to stress.
3.6 Abnormal expression of SNAR-E or RPL34-AS1 predicts aggressive clinical-pathological characteristics in CAD patients
Spearman’s rank correlation analysis (n = 237) was performed to examine whether the expression levels of SNAR-E and RPL34-AS1 correlated with cardiovascular risk factors and CAD-associated biomarkers. As shown in Table 2, SNAR-E expression was associated with DM, TC, HDL-C, LDL-C, HbA1c, Hcy, uric acid, urea nitrogen, and creatinine (P<0.05), whereas RPL34-AS1 correlated with DM, TC, HDL-C, Lpa, HbA1c, BNP, Hcy, uric acid, and urea nitrogen (P<0.05).
Multivariate logistic regression analysis (n = 237) was performed to confirm whether SNAR-E and RPL34-AS1 can be used as predictors of CAD development. As shown in Table 3, after correction for age, BMI, hypertension, DM, smoking, TC, and LDL-C, the odds ratio (OR) of SNAR-E for predicting CAD was 2.200 (P<0.001), and the OR for RPL34-AS1 was 0.737 (P<0.001).
3.7 Construction of accurate diagnosis model of CAD
After removing six samples for lncRNA microarray analysis, thirty-two predictors, including SNAR-E and RPL34-AS1, were analyzed using univariate logistic regression analysis (n = 237). As shown in Figure 6A, we selected 10 factors with P values under 0.1 to perform a Random Forest analysis (n = 237). We set the parameter mtry and the number of trees to 3 and 1000, respectively. Under these conditions, the error rates of OBB, CAD, and the control group were at their lowest (OBB estimate of error rate: 5.6%, Figure 6B). Next, we applied multi-dimensional scale transform (MDS) to examine the distance between samples, that is, the fitting effect. As shown in Figure 6C, our Random Forest was able to accurately separate CAD from the control group. Finally, we calculated the importance score and gini-index of each variable (Figures 6D and 6E). The importance score refers to the average reduction of the classification accuracy after the slight disturbance of the self-variable value of the data out of the bag and before the disturbance. In other words, the decrease in calculation accuracy and the decrease in average accuracy. The larger the variable, the more important it is. Therefore, we chose two variables with an importance score greater than 10 (SNAR-E and RPL34-AS1) to construct a diagnostic nomogram (n = 237, Figure 6F). The results showed that the C index of the nomogram diagnostic model can be as high as 0.956, and the calibration curve analysis showed that the diagnostic model almost completely matched with the ideal model, which indicates that our diagnostic model is very stable (Figure 6G).
Finally, we analyzed the clinical DCA (n = 237) of the diagnosis model, and the results are similar to the Random Forest analysis. SNAR-E itself showed to be more important than of RPL34-AS1, but the model that combined both molecules showed the greatest clinical benefit (Figure 7A). ROC analysis (n = 237) shows that the AUC value of the whole model can be as high as 0.993 (Figure 7B), which also shows that the diagnostic model has high sensitivity and specificity for the diagnosis of CAD.