Total anomalous pulmonary venous connection (TAPVC) is one of the most critical congenital heart diseases (CHD), accounting for 1 to 3% of all CHD. All four pulmonary venous of TAPVC patients cannot be properly connected to the left atrium, and abnormal connections to the right atrium or the systemic venous system may occur (1). Patients with TAPVC have a mortality rate of up to 80% in the absence of surgical intervention (2). However, postoperative pulmonary venous obstruction (PVO) occurs in 5–20% of patients resulting in increased mortality and morbidity which poses an ongoing surgical challenge (3–5).

Preoperative obstruction is a vital risk factor for postoperative obstruction, and 50% of patients with preoperative obstruction develop postoperative obstruction (6). Previous studies showed that, for preoperative obstructive patients, the use of sutureless technology can reduce the probability of postoperative PVO by avoiding pulmonary vein endothelial damage, preventing reactive intimal hyperplasia, and avoiding twisting and rotation (6, 7). In addition, some patients with primary preoperative obstruction may need heart–lung transplantation instead of repair surgery, because the existing surgical methods not only bring no benefits but will increase the pain and economic burden of multiple operations for children (8). Therefore, we need to identify preoperative obstruction to facilitate decision-making and operation processing to reduce the incidence of postoperative obstruction and improve surgical survival.

The following four conditions are currently used to confirm the diagnosis of preoperative obstruction: oxygen saturation, echocardiography data which indicated the existence of continuous pulmonary venous flow velocity >1.8 m/s, morphological features investigated with computed tomography angiography (CTA), and intraoperative assessment (9). Preoperative clarification of obstruction is more relevant for the prevention of postoperative obstruction, and CTA has limitations in its ability to detect intravenous or small venous obstruction in TAPVC and requires a high level of operator proficiency (6). Therefore, the development of an accurate and noninvasive predictive model to complement existing strategies is essential for the early prediction of preoperative obstruction. Some studies have constructed models to predict preoperative obstruction based on CTA images or pulmonary vein coefficients of variation (10, 11). However, these prediction models are not effective due to their still depend on image capture and the heterogeneity and complexity of the disease (11). Moreover, some patients may not be identified with a preoperative obstruction or have preoperative obstructive symptoms before surgery, but they already have the hypoplastic pulmonary venous which develops into progressive obstruction under the stimulation of the operation, the full length from the anastomosis to the pulmonary vein and leading to fetal morbidity (12). Therefore, we need features that can describe molecular alterations and predict preoperative obstruction independently of CTA images.

The location of the preoperative obstructive TAPVC lesion occurs in the pulmonary veins, and the presence of myofibroblasts is initially observed at autopsy (13). Further studies revealed an overgrowth of connective tissue with medial hypertrophy and intimal fibrosis, leading to extracellular matrix (ECM) deposition and subsequent obstruction (14). Therefore it can be assumed that venous vascular fibrosis leads to preoperative obstructive TAPVC. Fibrosis can affect any organ, of particular concern are the lungs, the cardiovascular system, the liver, etc (15). Leukocytes play a role in fibrosis, where stimulated tissue releases inflammatory mediators, which trigger an influx of leukocytes to the site of injury, and the activated leukocytes then further regulate pathological fibrosis (16). For example, identified alveolar macrophages specific to pulmonary fibrosis in single-cell analysis (17). Granulocytes [mast cells (18), eosinophils (19), neutrophils (20)] can partially influence fibrosis by secreting fibrogenic mediators and regulating the expression of matrixes (16); Macrophages and dendritic cells have different regulatory effects on the outcomes of fibrosis (21, 22). In addition to this, markers of elevated lipid values and systemic inflammation can be detected in the circulating blood of patients with atherosclerosis (23). Thus, blood cells circulating throughout the body are an important direction for liquid biopsies that can noninvasively characterize the molecular alterations and independent of CT images, which are ideal tissues for preoperative obstructive TAPVC. Therefore, we hypothesized that leukocyte transcripts can provide biomarkers for the prediction of preoperative obstructive TAPVC. In the current study, to test this hypothesis, we consecutively enrolled 26 preoperative obstructive and 22 non-obstructive TAPVC patients, performed and analyzed transcript profiling of their preoperative peripheral leukocytes. We observed differentially expressed mRNA and long noncoding RNA transcripts in obstruction vs. non-obstruction, the up-regulated mRNA in obstructive TAPVC were enriched in hydrogen peroxide catabolic process, response to mechanical stimulus, neutrophil degranulation, hemostasis, response to bacterium and NABA CORE MATRISOME pathways, all of them are associated with the development of fibrosis. Leukocyte subtype analysis revealed a low percentage of T-cells CD8 in preoperative obstructed TAPVC. Furthermore, NR3C2 and MEG3 were selected as molecular features and RF preoperative prediction model was constructed with an AUC of 0.95 in the training set and AUC of one in the validation set. To further verify the usefulness of the features, both GLM and support vector machine (SVM) prediction models were constructed, and the area under curve in the validation set both were one. We confirmed the hypothesis that leukocyte transcripts can evaluate obstructive-related pathophysiological alterations, providing a novel direction for the study of preoperative obstructive TAPVC.

In the current study, we performed a transcripts comparison of peripheral leukocyte profiles between patients with obstructive and non-obstructive TAPVC. We found that the up-regulated mRNA in leukocytes were enriched in the hydrogen peroxide catabolic process, response to mechanical stimulus, neutrophil degranulation, hemostasis, response to the bacterium, and NABA CORE MATRISOME pathways, all of which are associated with the development of fibrosis. Leukocyte subtype analysis revealed a lower percentage of T-cells CD8 in preoperative obstructive TAPVC patients. The mRNA NR3C2 and lncRNA MEG3 were two features screened from all transcripts, which can correctly distinguish all patients with preoperative obstruction TAPVC in the validation set. Our project developed a novel imaging-independent approach, which is the first integration of peripheral blood leukocytes transcript to construct machine learning prediction models. It is expected to complement existing imaging techniques in clinical practice to guide the surgical protocol for TAPVC.

Venous vascular fibrosis leads to preoperative obstructive TAPVC (13, 14). The release of inflammatory mediators from fibroblast-activated venous triggers an influx of leukocytes to the site of injury, and the activated leukocytes further modulate pathological fibrosis (16). We found that up-regulated mRNAs of leukocytes in patients with preoperative obstructive TAPVC enriched in 6 GO pathways. Pathway 1, hydrogen peroxide catabolic process. ROS (including hydrogen peroxide) and oxidative stress appear to be important in renal fibrosis and tissue repair/remodeling after myocardial infarction (38). Pathway 2, response to mechanical stimulus. Fibroblasts are cells that respond to mechanical stretching stimuli. They maintain the structure and function of organ tissues by altering the expression of genes and proteins in their ECM in response to external physical, chemical, and biological factors (39). Pathway 3, neutrophil degranulation. Dysregulation of neutrophil degranulation capacity in patients with cystic fibrosis (CF) (40). Pathway 4, hemostasis. Hemostasis contributes to fibrosis of the lung, liver, kidney, and heart (41). Pathway 5, response to the bacterium. Bacterial respiratory infections are the main cause of morbidity and mortality in patients with cystic fibrosis (CF (42) Pathway 6, The NABA CORE MATRISOME pathway is a collection of genes encoding the core extracellular matrix, including ECM glycoproteins, collagens, and proteoglycans. The pathways that upregulate mRNA enrichment all seem to play a role in the development of fibrosis and the formation of ECM. In addition, we found the percentage of T-cell CD8 was lower in the preoperative obstructive TAPVC with significant differences. T-cell CD8was also found to negatively regulate renal fibrosis in previous experiments with knockout CD8 mice. These data suggest that gene dysregulation in the venous is reflected in the transcriptomic data of leukocytes.

The mRNA NR3C2 and the lncRNA MEG3 were the two most important features screened from all transcripts. NR3C2, the mineralocorticoid receptor, plays important role in the modulation of blood pressure (43). MEG3, Maternally expressed gene 3, is an imprinted gene (44). According to the HCL database, NR3C2 is expressed in three subtypes of peripheral blood cells, including monocyte, T-cell CD8, and dendritic cell, while MEG3 is expressed in the macrophage subtype (33). Dendritic cells are antigen-presenting cells that present captured antigens to T cells and regulate the number and activity of T-cell CD8 (45). T-cell CD8 then control the production of monocytes (46). Activated macrophages are derived from tissue-resident macrophages or peripheral blood monocytes and induce fibroblast activation by secreting TGF-β (47). And it has also been found that in the late stages of pulmonary fibrosis, activated macrophages and myofibroblasts are even thought to cross-stimulate each other, leading to a vicious circle that ensures the propagation of fibrosis (47). MEG3 regulates TGF-β gene activity by binding to distal regulatory elements (48). We know that fibrosis is always marked by increased expression of TGF-β (49). We therefore hypothesized that MEG3 in peripheral blood macrophages activates the expression of cytokines TGF-β, which promote fibrotic secretions by recruiting more fibroblasts, enhancing their differentiation to myofibroblasts, or promoting ECM proteins. We must acknowledge the limitations of peripheral blood leukocytes compared to diseased tissue cells in explaining disease mechanisms, and we need further studies to test this hypothesis and molecular mechanisms. However, peripheral blood leukocytes are an excellent material for noninvasive monitoring.

We constructed a machine-learning model to evaluate the predictive value of NR3C2 and MEG3. The sensitivity achieved as high as one. However, the small-size dataset might lead to an overestimation of the model prediction performance. To reduce this effect, we used both RF and GLM algorithms to select features and constructed three machine learning prediction models RF, GLM, and SVM, all of which have good prediction results. However, a blinded clinical trial with a larger sample size and different races is essential before a diagnostic or screening test based on this work can be used in the clinic.

In our study, the patient information was collected from regional cardiac centers, which were internationally and nationally renowned, and the preoperative evaluation, imaging, and intraoperative exploration of patients are standardized, and the data collected are from medical records, which are more accurate and credible. But there are still limitations. First, since still no gold standard for the diagnosis, we can only define the disease using imaging and laboratory test results, as described in the methods section. It cannot be denied that there may be some preoperative obstructive TAPVC that fall outside these definitions, or that do not have a significant decrease in oxygen saturation at the time of presentation, which potentially leads to a misclassification bias of obstructive TAPVC in our study. Second, we observed variations in expression levels within groups. Although the patients with TAPVC recruited in the current study had similar baseline data, their complex etiologies and unclear and highly heterogeneous pathogenesis will unavoidably introduce variation within groups. The model lacks validation with external datasets, and we will further refine it in the subsequent study. Third, because of ethical issues, we were unable to directly obtain pulmonary vein tissue. Further validation using the pulmonary vein tissue is warranted based on ethical standards.

Our pilot studies have shown that noninvasive blood tests were able to reflect the characteristics of preoperative obstruction in patients with TAPVC. mRNA NR3C2 and lncRNA MEG3 can accurately predict obstructive patients. Peripheral leukocyte RNA can be used as a biomarker source for noninvasive surveillance, but there are limitations in the interpretation of disease mechanisms. We believe that the application of blood transcripts will provide new research directions for disease research and help promote clinical development. Future research can integrate imaging, transcripts, and clinical features to find the most suitable clinically available indicators, which may truly change the preoperative monitoring of the disease.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: http://db.cngb.org/cnsa, CNP0002105.

The studies involving human participants were reviewed and approved by the Guangdong Provincial People's Hospital Human Subjects Committee (No. GDREC2018363H). Written informed consent to participate in this study was provided by the participants or their legal guardian/next of kin.

YO and W-JW conceptualized and designed the whole study. XiaohoL extracted white blood cells and conducted qRT-PCR verification. XiaohuL and HD are cardiac surgeons and cardiologists who were responsible for blood sample collection, disease diagnosis, and clinical information verification, and then the ZL isolated RNA and constructed PALM-Seq libraries. HQ reviewed the CTA data for patients and reconstructed the 3D image based on CTA data. ZW conducted preliminary analysis and drafted the first manuscript and JS, QZ, and TY verify the underlying data. ZW and XiaohoL critically revised the manuscript and YO and W-JW refined the manuscript. All authors contributed to the article and approved the submitted version.

This work was supported by the National Key Research and Development Program [No. 2018YFC1002600], National Natural Science Foundation of China [Nos. 81903287 and 82170259], Natural Science Foundation of Guangdong Province [Nos. 2021A1515011445 and 2018A030313329], Guangdong Provincial Clinical Research Center for Cardiovascular disease [No. 2020B1111170011] and the Science, Technology, and Innovation Commission of Shenzhen Municipality under grant [No. JCYJ20180703093402288].

JS, ZL, QZ, JX, TY, and W-JW were employed by BGI-Shenzhen, Shenzhen, China. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.