Our main findings are that the combination of REG4, SPINK4 and A1AT protein expression associates with CAT in an independent cohort of patients with CRC. The data that REG4, SPINK4 and A1AT protein expression associate with CAT are in line with our earlier findings that REG4, SPINK4 and SERPINA1 were the top-3 upregulated genes at mRNA level that associate with CAT [9]. In the current study, the combined protein expression of REG4 and A1AT demonstrated the strongest association with CAT.

Expression of REG4 and SPINK4 in the tumor may indicate a proinflammatory status of the tumor. In addition to being among the 21 most upregulated genes in inflammatory bowel disease (IBD), SPINK4 was the gene that most strongly co-expressed with REG4 in IBD [11]. In our CRC cohort, protein expression levels of SPINK4 also significantly correlated with REG4 (P = 0.001, Pearson r: 0.5995, data not shown). As a result, it is not surprising that combining these co-expressed genes (REG4high/SPINK4high) in a logistic regression analysis hardly resulted in an increase in the OR (OR 1.8) for developing CAT when compared with REG4high (OR 1.6) or SPINK4high (OR 2.0) alone.

Blood coagulation and the immune system of higher organisms are closely intertwined. Virtually all solid tumors induce a local or systemic inflammatory state, which may contribute to development of CAT. Indeed, inflammation was one of the pathways associated with CAT in the RNA-seq pathway analysis in CRC [9]. These findings were recently extended to lung cancer in a gene set enrichment analysis of RNA-seq data, demonstrating upregulation of genes in the inflammation and complement pathway, besides upregulation of genes associated with the KRAS signaling pathway [12].

REG4 has been shown to associate with CRC progression [13, 14]. Zhu et al. demonstrated that REG4 mRNA was increased in 40 CRC samples compared with paired adjacent normal mucosa and REG4 proteins levels assessed by immunohistochemistry associated with distant metastasis and disease-free and overall survival [14]. Furthermore, Oue et al. showed that REG4 immunostaining associated with tumor grade, liver metastasis and poor survival, and that serum REG4 levels were increased in stage IV, but not stage I–III, CRC patients, when compared with 151 healthy controls [13]. In the colon, REG4-positive deep crypt secretory cells serve as an epithelial niche for LGR5-positive stem cells [15]. REG4 upregulation in the tumor may create additional cancer stem cell (CSC) niches, facilitating the growth of the aggressive subpopulation of LGR5-positive CSCs [15, 16]. CSCs are associated with enhanced invasion and metastasis. Rather than inflammation, an alternative explanation for the potential link of REG4 with CAT, is that REG4 increases intravasation of thrombogenic CRC cells into the blood circulation.

Of the three genes identified by RNA-seq, a trend for an association with CAT on protein level was observed for A1AT. A1AT is a protease inhibitor that keeps the activity of a variety of enzymes—particularly trypsin and neutrophil elastase (NE)—under control [17]. Deficiency in A1AT leads to extensive and prolonged NE-induced degradation of elastin resulting in reduced lung elasticity and respiratory complications [17]. There are three potential pro-thrombotic effects described through which A1AT may contribute to CAT. Firstly, A1AT binds and neutralizes activated protein C (APC), a serine protease that proteolytically inactivates the activated coagulation co-factors Va and VIIIa [18]. Individuals with low levels of APC, or with some resistance to the effects of APC, are at increased risk for VTE [19, 20]. Secondly, A1AT inhibits the enzymatic function of NE, a protein secreted by neutrophils during inflammation. NE degrades cross-linked fibrin, and reduction of NE activity may therefore be suggested to impair clearance of blood clots. However, by degradation of the α-chain of fibrin, NE also reduces the stimulating effect of fibrin on plasminogen, and the net effect of NE does not appear to be fibrinolytic [21]. Thirdly, A1AT regulates fibronectin, which is covalently linked to fibrin during clot formation mediating platelet adhesion to collagen [22]. In a variety of cancers, the transcription factor Zinc finger protein SNAI1 induces an epithelial-to-mesenchymal transition (EMT) and a CSC-like phenotype. Immunohistochemical analysis of 528 CRC tumors demonstrated that not only SNAI1, but also A1AT protein expression levels were associated with tumor stage, lymph node metastasis and poor clinical outcome. Moreover, SNAI1 directly upregulated SERPINA1 (encoding A1AT protein) expression by binding its promoter region. Remarkably, the pro-metastatic effects of SNAI1 and A1AT on invasion and migration were mediated by upregulation of fibronectin [23]. In addition, Chang et al. showed that A1AT facilitates assembly of pericellular levels of fibronectin, facilitating lung metastasis [24].

The transmembrane glycoprotein TF is under physiological conditions expressed by most non-endothelial cells [25, 26]. In cancer, TF expression is regulated by both specific oncogenes and environmental factors [27, 28] and shown to regulate primary growth and metastasis formation in a variety of cancer models [29]. Yu and coworkers reported that driver mutations in colorectal cancer (KRAS and TP53) resulted in overexpression of TF via MEK/mitogen-activated protein (MAPK) and phosphatidylinositol 3’kinase (PI3K) [30]. In line, Ades et al. showed that KRAS mutational status associated with VTE in patients with colorectal cancer [31]. Interestingly, REG4 expression was recently shown to be induced by KRAS mutation in colorectal cancer cells, and act as a driver of K-RAS-induced tumorigenic effects [32, 33]. Therefore, KRAS mutational status may be the underlying genetic cause of the observed association of REG4 with VTE. Whether REG4 is also a driver—or a bystander—of thrombogenic effects remains to be investigated.

In all CAT risk prediction models, including the Khorana score [7], the PROTECHT score [34], the CONKO score [35], the ONKOTEV score [36] and the Vienna CATS score [37], the site of primary tumor is an important determinant. Consequently the discriminatory power is decreased in studies focusing on a single tumor type, e.g., in stage II–III CRC patients [38]. If tailored risk prediction modelling becomes common practice in a group of patients with single tumor type, tumor type specific biomarkers are needed to restore the predictive power.

The main limitation of the current study is the relatively small sample size, which is particularly relevant when analyzing the combined expression of proteins in the logistic regression analysis. Patients with A1AThigh/REG4high/SPINK4high tumors demonstrated an OR of 20.0 for CAT when compared with patients with A1ATlow/REG4low/SPINK4low tumors, but also demonstrated a relatively wide 95% CI (0.9–429.9). Nonetheless the high OR for REG4high/SPINK4high/A1AThigh tumors confirms the previously established association of A1AT, SPINK4 and REG4 mRNA expression with CAT. In particular, the validation cohort used demonstrated that combined expression REG4 and A1AT is associated with increased risk for CAT.

A limitation of using immunohistochemical staining as a detection method is the poor clinical translation. In general, immunohistochemical stainings are work-laborious and the results may vary largely between laboratories. Of interest, REG4 and A1AT are secreted proteins, and it would be of great interest to assess REG4 and A1AT plasma levels, as an ELISA-based assay would have far better clinical applicability as a biomarker when compared to an immunostaining. Performing ELISA-based assays in a larger cohorts would not only show whether REG4 and A1AT are good CAT biomarkers in CRC, including other cancer types in the cohort would also show whether REG4 and A1AT would be applicable as CAT biomarkers in other cancer types as well.

Besides the use of CAT biomarkers, it would be of utmost interest from a scientific point of view to perform mechanistic studies in in vitro, in vivo and thrombosis-on-a-chip models to elucidate the underlying biological mechanisms how REG4, SPINK4, and A1AT link with CAT.

In conclusion, we have shown that particularly the combination of REG4 with A1AT in tumors associate with CAT in patients with colorectal cancer. Upon validation in a large cohort, these candidate biomarkers could risk stratify colorectal cancer patients for thromboprophylaxis.