Thrombus leukocytes exhibit more endothelial cell-specific angiogenic markers than peripheral blood leukocytes do in acute coronary syndrome patients, suggesting a possibility of trans-differentiation: a comprehensive database mining study

As reported in our previous publications [8], we developed an experiment-generated data mining strategy (Fig. 1a) to analyze the messenger RNA (mRNA) expression of 163 of the most relevant angiogenic genes (Additional file 1: Table S1) in various human and mouse tissues based on literature. Expressed sequence tags (ESTs) from the UniGene Library in National Center of Biotechnology Information (NCBI) [27] were used to determine mRNA expression levels in tissues. Transcripts per million (TPM) of genes of interest were normalized with those of house-keeping gene β-actin in each given tissue for comparison among all tissues. These normalized values were then normalized with the median of all tissues to calculate the arbitrary units of gene expression for comparison among all genes (if the median value is 0, we would recalculate the median of tissues with TPM > 0). A confidence interval of the expression variation of house-keeping genes was generated by calculating the mean plus two times that of the standard deviations of the arbitrary units of three randomly selected house-keeping genes (GAPDH, HPRT1, and GUSB) normalized by β-actin in the given tissues (Fig. 2). If the expression variation of a given gene in one tissue was larger than the upper limit of the confidence interval of the house-keeping genes, the expression level of the gene in this tissue was considered high with statistical significance.

Gene expression profiles were collected from microarray datasets in NIH-GEO database (Fig. 1b). Specific samples were chosen as disease or treatment groups to compare with their parallel controls by GEO2R [28]. The number of samples was always greater than three. By searching for the gene symbols from the microarray data, we selected the genes in our list of 163 angiogenic genes with significant expression changes (p < 0.05). The genes of which the expression changes were greater than or equal to 2-fold were defined as the upregulated genes, while genes with their expression changes less than or equal to 0.5-fold were defined as downregulated ones.

Molecular function types of 163 angiogenic genes were categorized based on IPA database [29] (Additional file 1: Table S1). Functional interactions and potential physical interactions between molecules were illustrated using IPA Path Designer (Fig. 4a and Fig. 6b).

The subcellular localization of protein was determined using two widely used protein intracellular localization databases [30], namely COMPARTMENTS subcellular location database [31] and UniProtKB/Swiss-Prot location database (European Bioinformatics Institute). Details can be found in our previous publication [10].

We hypothesized that under physiological conditions, tissues undergo different patterns of angiogenesis, which are regulated based on their needs for oxygen and nutrients [32]. To examine this hypothesis, an experiment-generated database mining strategy that we developed (Fig. 2) [8] was used to examine experimentally verified expression profiles of mRNA transcripts of 163 angiogenic genes (Additional file 1: Table S1) in NCBI-UniGene database among 22 human tissues and 18 mouse tissues (fewer mouse tissues were examined due to the fact that gene expression data for the four tissue counterparts in mice were not available in the database). Based on the functional modes of angiogenic genes, they were categorized into four groups based on molecular types in IPA (TRs, GF/Rs, C/Cs, and P/Is). Tissues that had more genes with high expression levels were considered as having higher APs in preparation for physiological changes like wound healing and pathological challenges like ischemia and inflammation. We found that humans and mice have different tissue patterns in terms of AP (Fig. 3a), suggesting that specific tissue AP has been acquired by human tissues as a functional adaptation during evolution. Interestingly, muscle in human has the highest AP, which suggests that sufficient blood supply is highly significant for muscle function, tissue remodeling, and human survival throughout evolution. Among the 18 common tissues shared in humans and mice, muscle, eye, pancreas, and lymph node are the tissues with the highest APs (Fig. 3b). The heart has high AP in humans and moderate AP in mice. We then compared the composition (percentages of four groups) of highly expressed genes among heart and the nine other tissues shared by humans and mice in Fig. 3c. Again, we find the differences of composition patterns in these tissues between humans and mice. Interestingly, APs are positively correlated with the percentages of TRs among highly expressed genes, but not GF/Rs, C/Cs, or P/Is, emphasizing that TRs are the master genes in regulating tissue APs. These results suggest that under physiological conditions, the muscle, eye, lymph node, pancreas, and human heart have already been conditioned with high APs enabling their recovery from pathophysiological insults like wound healing, ischemia, and inflammation.

Pathological angiogenesis often involves inflammatory conditions such as tumor angiogenesis and ischemia-induced angiogenesis [25]. As aforementioned, the five most relevant signaling pathways are activated under chronic inflammation, MAPK, PI3K-AKT, NOTCH, NF-κB, and JAK-STAT. We also included two key EC-specific pathways regulating angiogenesis, HIF-VEGF and angiopoietin-tyrosine kinase with immunoglobulin-like and EGF-like domains (ANG-TIE) [33]. Based on the method we developed (Fig. 2), key molecules in each signaling pathway that are most relevant to angiogenesis were examined to determine the activity of each pathway under physiological conditions (Table 1). We determined that if tissues have more than or equal to (≥) 1/3 of pathway-specific angiogenic genes (listed in Table 1) that are highly expressed, these pathways are considered active in those tissues. Similarly, if tissues have more than two active signaling pathways out of the seven pathways, these tissues are considered as having active angiogenic signaling. As shown in Table 2, we found that the human lymph node and muscle are the two tissues having more active global angiogenic signaling in humans than other tissues. A lymph node has two important active pathways, NF-κB and JAK-STAT, which respond to inflammatory stimuli in both humans and mice, while muscle has two key EC-specific active pathways, HIF-VEGF and ANG-TIE, in humans. In mice, the eye, lymph node, spleen, and thymus have more active angiogenic signaling than other tissues. Interestingly, in both humans and mice, immune organs tend to have more active pathways responding to inflammation than others, whereas non-immune organs such as the muscle, eye, heart, and pancreas have more active EC-specific pathways.

Immunotherapy has been developed to treat chronic inflammatory diseases. Thus, we also examined the expression levels of 27 C/Cs, which are most relevant to angiogenesis [34], in different tissues under physiological conditions. Based on the C/Cs roles in modulating angiogenesis, these C/Cs were classified into three categories, pro-angiogenic, anti-angiogenic, and bi-functional (Tables 3 and 4). C/Cs highly expressed in more than median numbers of tissues among all are considered as abundant C/Cs. We found that human CCL5, IL-15, and IL-6 are among the most abundant pro-angiogenic C/Cs in tissues. In contrast, CXCL9 and CXCL14 are the most abundant anti-angiogenic C/Cs. IL-18 is the most abundant human bi-functional C/C. In mice, Ccl2 and Il-15 are the most abundant pro-angiogenic C/Cs; Cxcl11, Cxcl14, and Pf4 are the most abundant anti-angiogenic C/Cs; and Cxcl16 is the most abundant bi-functional C/C. We consider that tissues with more than median numbers of C/Cs being highly expressed among all have high angiogenic C/C responses. We found that the human heart, muscle, eye, pancreas, bone, kidney, liver, and lung are the tissues more likely to develop high angiogenic C/C responses (Table 3); while the mouse eye, lymph node, muscle, spleen, liver, and thymus are the tissues with high angiogenic C/C responses (Table 4). Interestingly, we also found that in humans, more pro-angiogenic C/Cs are highly expressed, while mice have more anti-angiogenic C/Cs highly expressed. This may explain the potential differences between human diseases and mouse models studying angiogenesis.

Taken together, these results suggest that the tissues with high APs based on tissue expression of 163 angiogenic genes also have more active pro-angiogenic pathways and stronger C/Cs responses; these angiogenic C/Cs probably are the mediators signaled via the seven angiogenic pathways.

We found that TRs serve as master genes in determining tissue APs in physiological conditions (Fig. 3c). To determine how these seven angiogenic pathways regulate angiogenic C/Cs under pathophysiological conditions, we used the IPA Path Designer to illustrate the interactions between 19 TRs in the seven pro-angiogenic pathways and 27 angiogenic C/Cs based on published literatures (Fig. 4a). We found three major groups of TRs that regulate angiogenesis differentially in various pathophysiological settings: first, inflammatory TRs [35] are the main regulators of all three groups of C/Cs, and on average, each inflammatory TR can regulate 43.6% of pro-angiogenic C/Cs, 47.1% of anti-angiogenic C/Cs, and 51.7% of bi-functional C/Cs, suggesting that inflammatory TRs may function differently in different phases of inflammation; second, homeostatic TRs can also regulate all three groups of angiogenic C/Cs, albeit to a lesser extent compared to inflammatory TRs (on average, each TR can regulate 20.0% of pro-angiogenic, 14.3% of anti-angiogenic, and 35.0% of bi-functional C/Cs); and third, EC-specific TRs in HIF-VEGF pathway mainly regulate pro-angiogenic C/Cs and bi-functional C/Cs (on average, each TR can regulate 19.3% of pro-angiogenic, 4.3% of anti-angiogenic, and 21.7% of bi-functional C/Cs) (Fig. 4b). These results demonstrated that pro-angiogenic TRs are specifically regulating angiogenic C/Cs. Temporally, during acute ischemic conditions, inflammatory regulators initially dominate the local environment. After intense inflammation fades, tissue repair and regeneration, including angiogenesis, can be eventually observed [36]. Thus, inflammatory TRs may regulate different angiogenic C/Cs in angiogenic cells during the mobilizing phase and vessel-building phase during pathogenesis.

To address the question of what factors could determine tissue APs, we hypothesized that tissue APs are positively correlated with the tissue expression levels of cell oxygen sensors. To examine this hypothesis, simple linear regression analyses were performed by plotting mRNA relative expression levels of PHD2 (EGLN1), HIF1A, HIF2A (EPAS1), and HIF1B (ARNT) against the percentages of highly expressed genes in each tissue (based on total 163 genes). We found that the mRNA relative expression levels of all four genes could indicate tissue APs in humans (Fig. 5a). Interestingly, expression levels of PHD1 (EGLN2) and PHD3 (EGLN3) did not correlate with tissue APs (data not shown). Since VEGF pathway components are key effectors in mediating angiogenesis [3], in addition to the oxygen sensors, we also examined the correlation of VEGFs/VEGFRs mRNA expression levels with tissue APs (Fig. 5a). We found that the expression levels of VEGFB, PGF, VEGFR1 (FLT1), VEGFR2 (KDR), and VEGFR3 (FLT4) are positively correlated with tissue APs. Since tissue repair and regeneration processes largely rely on angiogenesis, we also determined whether the expression of stem cell master genes including CD34, KIT, MYC, KLF4, OCT4 (POUSF1), and SOX2 are positively correlated with tissue APs. Of note, MYC, KLF4, OCT4, and SOX2 are the induced pluripotent stem (IPS) cell transcription factors or Yamanaka’s re-programming transcription factors [37]. Interestingly, the results showed that among the seven stem cell master genes examined, the expression levels of KIT (r ² = 0.2724; p < 0.05) and SOX2 (r ² > 0.4; p < 0.05) are positively correlated with tissue APs (Fig. 5b). Based on the different correlation factor r ² between the expression levels of angiogenic genes and tissue APs, we categorized three types of angiogenic regulators into three levels of correlation factor r ² , high, middle, and low (Fig. 5b). The results suggest that the APs among human tissues have high correlation with the tissue expression levels of cell oxygen sensors PHD2 and HIF1B, VEGF pathway component VEGFB, and stem cell master gene SOX2, moderate correlation with the tissue expression levels of PGF and FLT4, and low correlation with the tissue expression levels of HIF1A, HIF2A, KDR, FLT1, and KIT.

To further understand how these angiogenic regulators indicate tissue APs, we categorized 163 genes into four groups of angiogenic genes (TRs, GF/Rs, C/Cs, and P/Is) and found that (1) the expression of VEGFB positively correlates with APs of all four groups of genes; (2) the expression levels of HIF1B and PHD2 positively correlate with APs of TRs, GF/Rs, and P/Is; and (3) the expression level of SOX2 positively correlates with APs of TRs and GF/Rs (Additional file 1: Figure S1). These results suggest that different AP master genes are associated with the expression of specific groups of angiogenic genes in regulating tissue angiogenesis.

Anti-angiogenic therapy has been suggested as an approach to treat cancers decades ago [32]. The underlying rationale, stating that solid tumor expansion is dependent on blood supply, has been universally accepted. Many drugs targeting angiogenesis have been developed or are under clinical trials. However, the clinical outcomes from patients treated with anti-angiogenic drugs are less satisfactory than expected [38]. There are no good mechanistic explanations on why this discrepancy exists. We hypothesize that cancers from different tissues have distinct angiogenic pathways and C/C responses. To test this issue, we examined seven most relevant pro-angiogenic pathways including MAPK, PI3K-AKT, NOTCH, NF-κB, JAK-STAT, HIF-VEGF, and ANG-TIE using the method that we developed (Fig. 1b). We searched angiogenic gene expression data in 11 GSE datasets collected in the NIH-GEO Database (microarray experimental data) on most common cancers (six types of cancers in digestive system, three types of cancers in reproductive system, lung cancer in respiratory system, and lymphoma in lymphoid system) to better understand the molecular mechanisms underlying cancer/tumor angiogenesis (Table 5). In each GSE dataset, we compared the mRNA expression levels in cancer/tumor tissues to their adjacent normal tissues. We found that digestive cancers have significant gene expression changes enriched in HIF-VEGF and ANG-TIE pathways, especially in colorectal, gastric, and intrahepatic cancers, while in reproductive cancers and lung cancer, most of the pathways are regulated. Interestingly, pancreatic cancer did not show any changes in these seven pathways, suggesting that the high AP in pancreas (Fig. 3b) are equipped to match the high demand for angiogenesis in both physiological and tumor/cancer pathological conditions. As we know, most cancers have increased angiogenesis to supply blood and nutrition for cancer/tumor growth [38]. Surprisingly, here we found that prostate cancer, lung cancer, and lymphoma show significantly decreased gene expression among most changed pathways. This could be explained by sufficient oxygen supply for cancer cells in the lung, T cell-mediated immune tolerant/anti-inflammatory environment in the lung [39], and in circulating prostate cancer cells [40]. Many C/Cs play indispensable roles in communicating between immune cells, vascular cells, as well as tumor cells in regulating angiogenesis [25]. Thus, we further looked into angiogenesis-regulatory C/Cs in different disease conditions. We found that most digestive cancers and ovarian cancer have both pro- and anti-angiogenic C/Cs being upregulated, indicating that the C/Cs system is in a “fighting” mode (Table 6). However, in prostate cancer and lung cancer, where we have shown that they present an anti-angiogenic environment in the pathway analysis (Table 5), pro-angiogenic C/Cs are significantly downregulated and anti-angiogenic C/Cs are upregulated (Table 6). These results suggested that angiogenic pathways in various cancers are differentially regulated in different tissue environments and pathological conditions. In order to specifically target key molecules during the disease conditions, we would need more precise and accessible detection systems in identifying the personalized angiogenic gene expression profiles for timely and accurate anti-angiogenic treatments for cancer patients.

Angiogenesis in thrombus has been shown to play a key role in facilitating thrombus resolution [41, 42]. Hypoxia and HIF1A are induced in the naturally resolving thrombus and correlate with increased angiogenic factor expression. Upregulation of HIF1A enhances thrombus resolution and angiogenesis [43]. Leukocyte infiltration is one of the hallmarks during this process. To examine an important question of whether immune cells can directly promote angiogenesis, we compared the angiogenic gene expression profile of leukocytes isolated from thrombi from the plaque of acute coronary syndrome (ACS) patients with that in circulating blood leukocytes. We found that in ACS (unpublished GEO microarray dataset GSE19339, deposited on Dec. 4, 2009, and updated on Feb. 24, 2017), thrombus-derived white blood cells showed significant increase of angiogenic pathways with as many as 15 angiogenic regulators compared to peripheral leukocytes (Table 7). Although the research paper associated with this GEO dataset has not been published from Dr. Luscher’s team in the University Hospital Zurich, Switzerland, based on similar publications on isolation of Tregs [44], CD14⁺ monocytes and CD66B⁺ granulocytes [45] in coronary thrombi of patients with ACS and microarray profiling [46] from this highly productive team, we were fully convinced that the high quality of GSE19339 dataset and related technologies in coronary thrombus leukocyte isolation and microarray profiling are trustable. In addition, to determine the functional significance of upregulated genes, we examined the subcellular localization of the 15 upregulated genes including FLT1, NRP1, ANGPTL4, NRP2, VEGFA, EPAS1, JAG1, TEK, JUN, NOTCH3, TIE1, KDR, HIF1A, ANGPT2, and PGF using two highly reliable databases (COMPARTMENTS subcellular location database and UniProtKB/Swiss-Prot location database) as we reported [10] (Fig. 6a). As shown in Table 7, we found that among the upregulated genes, seven genes encode EC-surface receptors including FLT1 (vascular endothelial growth factor receptor 1), NRP1 (vascular endothelial cell growth factor 165 receptor), NRP2 (vascular endothelial cell growth factor 165 receptor 2), KDR (cell surface receptor for VEGFA, VEGFC and VEGFD) [47], TIE1, TEK, and NOTCH3. Moreover, EPAS1 and HIF1A are two transcription factors (Fig. 6a) directly involved in hypoxia-induced gene expression pathway; homeostatic transcription regulator JUN (Fig. 6a) is a subunit of the transcription factor AP-1 that was reported to collaborate with HIF1 to increase gene expression in response to hypoxia [48, 49]. The remaining five molecules upregulated in thrombus-derived leukocytes were all pro-angiogenic factors, including ANGPT2, ANGPTL4, VEGFA, PGF, and JAG1, which we propose can stimulate thrombus leukocytes to phenotypically switch into EC-like angiogenic cells, and thrombus-trapped circulating ECs via an autocrine and paracrine manner, to create a pro-angiogenic niche. However, peripheral blood leukocytes from ACS patients did not show the similar changes compared to healthy control (GSE 61144) (Table 7). We also found that a significant amount of C/Cs in thrombus-derived leukocytes are upregulated (Table 8), including 8 out of 14 pro-angiogenic C/Cs, 3 out of 7 anti-angiogenic ones, and 3 out of 6 bi-functional ones. The information shown in Fig. 6b suggests that the most upregulated C/Cs can be directly regulated by the three upregulated TRs (EPAS1, HIF1A, and JUN). Taken together, these results have demonstrated that thrombus-derived leukocytes exhibit potent pro-angiogenic phenotype by upregulating (i) EC-specific cell surface markers and (ii) hypoxia-induced transcription factor signaling pathways and secreting (iii) angiogenic factors and (iv) angiogenic C/Cs. These findings suggest that thrombus leukocytes promote angiogenesis not only through increasing secretion of angiogenic C/Cs and angiogenic factors but also via phenotypical switching into EC-like angiogenic cells.

We also found that retinopathy and rheumatoid arthritis (RA) have significant changes in many pro-angiogenic pathways. However, in the conditions of chronic inflammation, atherosclerosis, PAD, metabolic syndrome (MS), type 1 diabetes (T1D), and type 2 diabetes (T2D), there are no significant changes in any of these seven pathways (Table 7), suggesting that low-dose chronic inflammation does not always have a strong systemic angiogenic stimulation. We also examined the C/C expression changes among these diseases. We found that a significant number of C/Cs is upregulated in retinopathy and RA. However, in the chronic inflammatory diseases, there are little to no changes in C/C expression (Table 8). Taken together, these results suggest that diseases with active angiogenic pathways also have highly active C/C responses for cell-cell communication.