Nonvalvular AF patients were prospectively and consecutively enrolled at the Chinese People's Liberation Army General Hospital between January 2021 and December 2022. Enrollment eligibility was determined based on the following criteria: patients had to be aged at least 18 years and had to have been diagnosed with AF. Individuals were excluded if they exhibited any of the following: mitral stenosis, severe mitral or tricuspid regurgitation, aortic stenosis or regurgitation, the presence of an artificial valve, a history of LAAT resection, congenital heart disease (eg, atrial septal defect, ventricular septal defect, tetralogy of Fallot), a history of catheter ablation, contraindications to TEE such as esophageal stenosis or varices, refusal to undergo TEE, or poor TEE image quality. A fasting period of at least 10hours was mandated before the TEE examination. Ethical approval for this study protocol was granted by the Ethics Committee of the Chinese People's Liberation Army General Hospital, and informed consent was obtained from all participants.

Blood samples were collected early in the morning using ethylene diamine tetra-acetic acid (EDTA) anticoagulant tubes. After centrifugation at 1500g for 15minutes, the supernatant was stored at -80°C for subsequent proteomic detection and enzyme-linked immunosorbent assay (ELISA).

Plasma samples were centrifuged at 12 000g, 4°C for 10minutes to remove cellular debris. Quantitative Sandwich ELISA kits were used to measure the levels of myosin light chain 4 (MYL4), prenylcysteine oxidase 1 (PCYOX1), and decorin in patients’ plasma using MYL4, PCYOX1, and decorin ELISA kits (LunChangShuo Biotechnology Co, Ltd, China, ED-200171, ED-200172, and ED-10690, respectively). The coating antibodies for the 3 proteins were standardized at a concentration of 2μg/mL, with 100μL applied per well, and the labeled antibodies were initially at a concentration of 1mg/mL and were then diluted 1000-fold before use. Measurements were performed using an ELISA reader (Rayto, RT-6100, Shenzhen Rayto Life Science Co, Ltd, China) at a wavelength of 450nm within 15minutes, and the mean value of duplicate wells was selected as the optical density value for the assay.

Cardiac assessments were performed using an EPIC 7C ultrasound system (Philips, The Netherlands) for transthoracic echocardiography and TEE. Transthoracic echocardiography measurements, such as left atrial diameter, interventricular septum thickness, left ventricular end-diastolic diameter, and left ventricular ejection fraction, were recorded. TEE, which was conducted after ensuring that there were no contraindications, focused on scanning the LAA for thrombi or SEC. An X7-2t adult 3D probe (Philips, Netherlands) at 2-5MHz frequency was used with synchronous electrocardiograph waveforms displayed. Ahead of the 8-hour fasted examination, probe adjustments were made to enable comprehensive 0 to 180 degrees scanning of the LAA to detect thrombi or SEC. Thrombus diagnosis was confirmed upon detection of an echo-dense mass in the left atrium or appendage, which was distinguished from the pectinate muscles through multiplane scanning (especially at 110-135 degrees). SEC, defined as characteristic swirling echoes indicative of impending thrombus formation and stroke, was differentiated from left atrial white noise artifacts. Patients with SEC were categorized as high risk for LAAT and grouped based on their imaging characteristics. Imaging was performed by experienced cardiologist sonographers, and discrepancies were resolved by a third expert.

Continuous variables are presented as mean ± standard deviation (x̅ ± s) or median (Q3-Q1), depending on their distribution. Categorical variables are expressed as frequencies (percentages). We used the independent sample t-test or Mann-Whitney U test for continuous variables and the chi-square test for categorical variables. Univariate logistic regression was used to assess risk factors for LAAT. Lasso regression refined variable selection, and model performance was evaluated via concordance c-statistics, receiver operating characteristic curves, and area under the curve (AUC). Net reclassification improvement (NRI) and integrated discrimination improvement (IDI) indices were used to determine the predictive capacity of the model. Calibration and clinical utility were examined using bootstrap-based calibration plots and nomograms, respectively. Analyses were performed in R version 4.3.0 (R Foundation, Austria), employing packages including “stats,” “MannWhitneyU,” “MASS,” “car,” “glmnet,” “pROC,” “rms,” “clusterProfiler,” and “org.Hs.eg.db.”

Plasma samples from 8 patients were analyzed and grouped based on the presence or absence of LAAT/SEC into the thrombosis and control groups, with 4 patients each. Higher levels of NT-proBNP were observed in the thrombosis group than those in the control group (764.3±489.2 pg/mL vs 63.5±27.9 pg/mL; P=.029). Significantly larger left ventricular end-diastolic diameter (47.3±8.5mm vs 33.8±3.2mm; P=.024) and notably lower LAA flow velocities (0.273±0.030 m/s vs 0.643±0.051 m/s; P<.001) were also observed in the thrombosis group. No significant differences were found between the thrombosis and control groups in terms of age, sex, CHA2DS2-VASc score, history of stroke, and anticoagulant treatment (table 1 of the supplementary data).

In our study, 8087 peptides were detected in all 8 samples. However, the number of proteins identified in each sample varied slightly, ranging from 1349 to 1353, with a total of 1351 proteins overlapping between the 2 groups. Using an absolute value of fold change greater than 1.5 and a P value less than .05 as criteria for significant difference, 7 upregulated and 23 downregulated proteins were identified between the 2 groups, totaling 30 (figure 1; table 2 of the supplementary data). The screening process, guided by proteomic results, human gene databases, and relevant literature, for these 30 differentially expressed proteins specifically targeted AF and thrombosis. Orr et al.16 discovered that the E11K-MYL4 mutation in humans and zebrafish can destabilize F-actin-Z-disk complexes, potentially impairing calcium signaling and causing atrial myopathy leading to arrhythmias. Gudbjartsson et al.17 reported that frameshift deletions in the MYL4 gene (c.234delC) can cause early-onset AF. A deficiency in PCYOX1 can lead to diminished platelet reactivity and impaired arterial thrombus formation.18 Structural features of decorin can facilitate its interaction with platelet-reactive proteins. MYL4, PCYOX1, and decorin were ultimately selected as potential biomarkers for predicting LAAT.

Figure 1.

Volcano plot comparing significant protein differences between the thrombus group and control group, based on fold change and statistical P value. Proteins with significant downregulation (fold change less than -1.5) are marked in blue, significantly upregulated proteins (fold change greater than 1.5) in red, and proteins without significant difference in green.

(0.16MB).

The enrichment analysis revealed that the identified pathways and GO functions, as presented in figures 1 and 2 of the supplementary data, might not be directly relevant to the disease under study, or, in cases such as those with KEGG analysis, did not yield substantial enrichment. This observation could be attributed to the nature of plasma as a fluid medium, which is not cellular or tissue-specific. Plasma primarily contains a mixture of proteins secreted by various tissues and cells; thus, the pathways and functions identified in plasma samples may not accurately reflect intracellular or tissue-specific processes.

The CHA2DS2-VASc score was higher in the thrombosis group than in the control group (3.2±1.2 vs 2.7±1.6; P=.019). The left atrial anteroposterior diameter in the thrombosis group was 47.0±9.5mm, which was significantly larger than that in the control group, 42.3±7.6mm (P=.001). The LAA flow velocities in the thrombosis group (0.315±0.07 m/s) were smaller than in the control group (0.510±0.228 m/s). The morphology of the LAA differed significantly between the 2 groups (P=.005). The cauliflower-shaped lesion was predominant in the thrombosis group (53.4%). In contrast, the cactus shape was the most common in the control group (42.1%). However, the type of LAA opening did not significantly differ between the groups (P=.291). No significant differences in age, sex, or opening type were observed between thrombosis and control groups. A comparison of other baseline data is shown in table 3 of the supplementary data. Three biomarkers (MYL4, PCYOX1, decorin), validated using ELISA, showed significant differences between the thrombosis and control groups (1.408±0.315 ng/mL vs 1.084±0.356 ng/mL; 2.921±0.574 ng/mL vs 4.027±0.476 ng/mL; 15.564±2.264 ng/mL vs 19.595±2.671 ng/mL, respectively; P<.0001), which is in line with proteomics testing results (figure 2).

Figure 2.

Levels of MYL4, PCYOX1, and decorin between groups. The x-axis categorizes the samples into control (blue bars) and LAAT/SEC (red bars) groups. Asterisks indicate a highly significant difference with a P value <.0001. DCN, decorin; LAAT, left atrial appendage thrombus; MYL4, myosin light chain 4; PCYOX1, prenylcysteine oxidase 1; SEC, spontaneous echo contrast.

(0.15MB).

Univariate logistic regression was performed in the collected variables to analyze the risk factors for LAAT and high-risk status. CHA2DS2-VASc, left atrial anteroposterior diameter, chicken wing, cauliflower morphology of the LAA, LAA orifice blood flow velocity, fibrinogen, prothrombin time, MYL4, PCYOX1, and decorin were associated with LAAT (figure 3 of the supplementary data). Lasso regression was performed for variable selection, yielding 5 variables (λ=0.03587), as shown in figure 3. Multivariate logistic regression analysis was performed to assess the predictive factors for LAAT. The models demonstrated that while the CHA2DS2-VASc score and cauliflower morphology of the LAA were consistently identified as independent predictors in multiple models, the inclusion of biomarkers MYL4, PCYOX1, and decorin varied, with PCYOX1 and decorin emerging as significant independent risk factors in the final models (table 1). No significant multicollinearity was found among the 5 variables used to construct the prediction model, as indicated by a variance inflation factor of less than 5.

Figure 3.

Results of Lasso regression. APTT, activated partial thromboplastin time; CRP, C-reactive protein; DCN, decorin; LAA, left atrial appendage; LAD, left atrial diameter; LVEF, left ventricular ejection fraction; MYL4, myosin light chain 4; PCYOX1, prenylcysteine oxidase 1; PT, prothrombin time.

(0.17MB).

To further appraise the model's performance, a comparison was made with the CHA2DS2-VASc score. The baseline AUC for the CHA2DS2-VASc score was 0.606. The AUC increased modestly to 0.672 with the inclusion of the LAA cauliflower morphology. Substantial enhancement was observed when biomarkers were added, with the AUC reaching 0.983 in model 3, which included LAA cauliflower morphology, MYL4, PCYOX1, and decorin. Models 4 and 5, which used different combinations of these biomarkers (MYL4+PCYOX1+decorin for model 4 and PCYOX1+ decorin for model 5), yielded AUCs of 0.971 and 0.970, respectively, indicating that while the inclusion of MYL4, PCYOX1, and decorin provided the highest AUC, the differences between models 2, 3, 4, and 5 were not statistically significant (figure 4). Model 5, which did not include MYL4, was deemed the most parsimonious and was proposed as the final model. Compared with the baseline CHA2DS2-VASc score, this final model (model 5) demonstrated an NRI of 1.544 (95%CI, 1.343-1.745; P<.001) and an IDI of 0.679 (95% confidence interval [95%CI], 0.592-0.767; P<.001), as illustrated in table 2.

Figure 4.

Comparison of Receiver Operating Characteristic (ROC) curves for the 3 models. The figure compares the ROC curves for 3 predictive models, illustrating their respective sensitivities and specificities. Model 0: CHA2DS2-VASc; Model 1: CHA2DS2-VASc + cauliflower; Model 2: CHA2DS2-VASc + MYL4 + PCYOX1 + decorin; Model 3: CHA2DS2-VASc + cauliflower + MYL4 + PCYOX1 + decorin; Model 4: MYL4 + PCYOX1 + decorin; Model 5: PCYOX1 + decorin; AUC, area under curve; MYL4, myosin light chain 4; PCYOX1, prenylcysteine oxidase 1.

(0.13MB).

A nomogram was developed to integrate and quantify independent predictors of LAAT/SEC (figure 5). The nomogram exhibited excellent discrimination with a c-statistic of 0.970. We performed 1000 bootstrap repetitions for internal validation of the performance of the nomogram. The calibration plot showed that the standard curve of the LAAT/SEC probability model was closely aligned with the 45° diagonal line, indicating good calibration of the new model (figure 4 of the supplementary data). The bootstrap c-statistic was 0.9418, with a bias of −0.0043.

Figure 5.

Nomogram for LAAT/SEC. The ‘Points’ axis at the top assigns numerical values for each variable. The ‘Total Points’ axis sums these individual points, which is then mapped to the ‘Probability’ axis to estimate the likelihood of LAAT/SEC. DCN, decorin; LAA, left atrial appendage; LAAT, left atrial appendage thrombus; PCYOX1, prenylcysteine oxidase 1; SEC, spontaneous echo contrast.

(0.1MB).

This study had several limitations. First, it was based on preliminary findings from a single-center dataset with a modest sample size. Second, although existing models consider various factors, such as clinical history, LAA morphology, and blood biomarkers, they still fall short in terms of detailed LAA morphological evaluation and the incorporation of hemodynamic parameters. Third, our study did not include further mechanistic research on the identified biomarkers. In future studies, we plan to increase the sample size and collect a more comprehensive dataset. We aim to holistically assess the risk factors of LAAT formation, thereby refining and developing new diagnostic and predictive models. Subsequent validation of these models for clinical relevance and accuracy will be performed through large-sample cohort studies and randomized controlled trials, thereby fostering the translation of our scientific findings into clinical practice.