A module of human peripheral blood mononuclear cell transcriptional network containing primitive and differentiation markers is related to specific cardiovascular health variables

Abstract

Peripheral blood mononuclear cells (PBMCs), including rare circulating stem and progenitor cells (CSPCs), have important yet poorly understood roles in the maintenance and repair of blood vessels and perfused organs. Our hypothesis was that the identities and functions of CSPCs in cardiovascular health could be ascertained by analyzing the patterns of their co-expressed markers in unselected PBMC samples. Because gene microarrays had failed to detect many stem cell-associated genes, we performed quantitative real-time PCR to measure the expression of 45 primitive and tissue differentiation markers in PBMCs from healthy and hypertensive human subjects. We compared these expression levels to the subjects' demographic and cardiovascular risk factors, including vascular stiffness. The tested marker genes were expressed in all of samples and organized in hierarchical transcriptional network modules, constructed by a bottom-up approach. An index of gene expression in one of these modules (metagene), defined as the average standardized relative copy numbers of 15 pluripotency and cardiovascular differentiation markers, was negatively correlated (all p<0.03) with age (R2 = -0.23), vascular stiffness (R2 = -0.24), and central aortic pressure (R2 = -0.19) and positively correlated with body mass index (R2 = 0.72, in women). The co-expression of three neovascular markers was validated at the single-cell level using mRNA in situ hybridization and immunocytochemistry. The overall gene expression in this cardiovascular module was reduced by 72±22% in the patients compared with controls. However, the compactness of both modules was increased in the patients' samples, which was reflected in reduced dispersion of their nodes' degrees of connectivity, suggesting a more primitive character of the patients' CSPCs. In conclusion, our results show that the relationship between CSPCs and vascular function is encoded in modules of the PBMCs transcriptional network. Furthermore, the coordinated gene expression in these modules can be linked to cardiovascular risk factors and subclinical cardiovascular disease; thus, this measure may be useful for their diagnosis and prognosis.

Conflict of interest statement

Competing Interests: The authors have declared that no competing interests exist.

Figures

Figure 1. qRT-PCR quantification of gene expression in PBMCs from a sample of a healthy human population.

A. The average expression levels (indicated as the relative copy number, RCN = 2−ΔCq×100) of the tested genes ordered based on the strength of their covariation (compare to Fig. 2A). The data are expressed as the means ± SD. Inset. Cq values for housekeeping genes used as endogenous controls. Of note, the large SD displayed by CXCR4 was due not to outliers but to the skewness of the data distribution. B. Actual RCN values of the 45 tested genes in 26 healthy subjects, indicating the coordinated expression of the majority of the genes (conventional color coding).

Figure 2. Regression analysis of genes co-expressed in PBMCs isolated from healthy human adults.

A. Correlations between several cardiovascular-specific genes. B. Correlations between selected vascular and primitive genes in the same population. C. Correlations between other tissue-specific and primitive genes. D. Inverse correlation between the expression of vascular genes and a leukocyte gene (PTPRC/CD45). In all the graphs, the number of subjects is n = 26, R2 is the linear regression coefficient, and p indicates significance; these coefficients were placed in mirror positions across the diagonal with their corresponding graphs. The data represent RCNs.

Figure 3. Correlation and clustering analysis of data.

A. A heat map of the bivariate correlation matrix of gene expression levels representing Pearson's correlation coefficient, r, in descending order, beginning with KDR/VEGFR2 (red: positive correlation; blue: negative correlation; gray: no correlation). B. The corresponding probability values, p, after Bonferroni correction (red: p<0.05; gray: not significant). C. An unsupervised hierarchical clustering analysis (complete linkage on standardized data) representing the associations between the genes as distances (Y-axis). In A and C, the corresponding main gene groups are indicated by brackets.

Figure 4. Modular organization of a PBMC gene sub-network.

A. Network representation of the genetic covariation. The thickness of each connecting line is proportional to the absolute value of the respective Pearson's correlation coefficient. Genes that were significantly correlated with the age, AIx, AoPP and BMI of the subjects are encircled (cf. Table 3). Color coding identifies the participation of genes in separate underlying clusters . B. Scaling of nodes' clustering coefficient C(k) with their connectivity degrees k, as a signature of hierarchical networks. Note that the data spontaneously split into two subpopulations, suggesting distinctly organized modules (for clarity, the leukocyte genes were omitted). Members of Module 1 (right), corresponding to the functionally filtered group in A (same color convention), had higher clustering values for similar k values than those in Module 2 (left), indicating stronger transcriptional coupling. C. Genes connected to the KDR node. Note that these connections perfectly overlap those of Module 1, while PROM1/CD133 serves as a link with Module 2, arguing that KDR is a hub node of Module 1. D. Connections of another hub node, NES. The images in A, C and D are based on Pearson correlation coefficients r>|0.8| and were obtained using the software Gephi 0.8 beta (www.gephi.org). The data shown in B were also obtained using Gephi, based on the network analysis in A.

Figure 5. Modular index (MI) of Module 1 (metagene) associated with the age and cardiovascular parameters of blood donors.

A. Age-dependent variation in MI in the population. B. Regression analysis of MI on AIx. C, D. Correlations of MI with AoPP and BMI (in women). MI represents the 15-gene average of the standardized (mean = 0, SD = 1) RCN of each gene within the tested population, +/− SD. n = 26 for A–C and n = 14 for D. Note the apparently bimodal distribution of MI with age in this population.

Figure 6. Personalized representation of gene expression as radial graphs.

A. The reference population level (RCN median, n = 26, 100%) is shown in blue, and the corresponding individual percentage values for the specified genes are shown in red in the order used in Fig. 2A. Note the pattern differences in Module 1 genes between females with normal BMIs (upper row) and those with higher BMIs (lower row). B. Association of aortic stiffness (left-side images) with radial gene profiles (right-side graphs) of two subjects: left, female, 56 years, average stiffness of 5.25 kPa; right: male, 52 years, average stiffness of 6.17 kPa. Aortic stiffness was measured by MRE, as described the Methods section; the local elasticity distribution is color coded, as shown in the scale at right (in kPa).

Figure 7. Comparison of gene expression in healthy and hypertensive subjects.

A. The median RCN in healthy controls (CT, blue) vs. hypertensive patients (HT, red); all of the differences are significant (p<0.05), with the exception of those labeled by arrows. Inset. Cq values of housekeeping genes in this patient population. B. Box plots of the aggregated MI (averages of normalized RCN values of Module 1 genes) in the control subjects and patients. (Box plots show the median values, 1st and 3rd quartiles, and the interquartile range; symbols are as in A).

Figure 8. Changes in the modular organization of genes in hypertensive patients (n = 20; see Figs. 3 and 4 for details).

A. A heat map of intergenic covariation. B. The corresponding matrix of significance values following Bonferroni corrections. C. Dendrogram of hierarchical gene clustering. D. The network structure of the patient genes, indicating the fusion of Modules 1 and 2 of the network found in the healthy subjects (red). E. The relationship between the gene clustering coefficient C(k) and node degree (connectivity) k; the collapsed sub-network shows a very strong and nearly uniform connection between nodes (inset), suggestive of transcriptional primitivity. The data analysis and representation were performed as in Figs. 3 and 4.

Figure 9. Radial representation of relative gene expression in hypertensive patients.

A. Representative radial diagrams of patients with hypertension. Note the overall reduction in expression, with the exception of a subgroup of genes (e.g., ALPL, ITGAM, and PTPRC/CD45; compare with Fig. 7A). B. Radial diagrams of hypertensive patients treated with thiazide; in these cases, the gene patterns in the two modules were closer to normal (i.e. closer to the blue reference line).

Figure 10. Origins of marker gene co-expression within individual cells: in situ hybridization (ISH).

A. ISH analysis of the Module 1 hub genes KDR and NES and the node gene FSHR, identified by their fluorescent signals in a given microscopic field (brown masks were added to positive cells by the CellProfiler image analysis software; blue represents DAPI staining of nuclei). Arrow: a triple-positive cell. B. Four-color confocal images of cells that are positive for (a) all three markers; (b) NES and FSHR only; or (c) KDR and NES only (white: KDR; red: NES; green: FSHR; blue: nuclei). Bars: 5 µm. C–E. Linear regression of the integrated pixel intensity of the mRNA of each marker gene (KDR, NES, and FSHR) detected using ISH in triple-positive cells (n = 66 cells pooled from 8 individuals; r: Pearson's correlation coefficient; r2: regression coefficient; log-log scale). Inset graphs show the lack of correlation between mRNA expression (also measured as the integrated pixel intensity) in single-positive cells for each respective pair of markers. F. Nuclear area and several texture features calculated using the CellProfiler analysis significantly separated the triple-positive cells from the other cells (*, p<0.05 for single and double expressers vs. negatives; #, p<0.05 for single and double expressers vs. triple positives; ¶, p<0.05 for triple expressers vs. negatives) (a total of 2094 cells from 8 subjects were analyzed). Abbreviations: Comp., compactness; Int. Int., integrated intensity; Mean Int., mean intensity; Med. Int., median intensity; TDV, texture difference variance; TC, texture contrast; TV, texture variance; TSA, texture sum average. The data represent the means of standardized values ± SEM.

Figure 11. Detection of cells expressing representative Module 1 node proteins using immunocytochemistry (ICC).

A. Cells expressing various levels of KDR (white), NES (green), and/or FSHR (red). Nuclei are blue (DAPI). a, b: NES-FSHR double-positive cells; c: a triple-positive cell. Bars: 5 µm. B. A nuclear morphology analysis revealed alterations in the triple-positive cells detected using ICC that were comparable to those found using ISH (see Fig. 10 for abbreviations). The data represent the means of standardized values ± SEM; a total of 1655 cells were analyzed. C. Frequencies of cells positive for the three marker genes, detected using ISH (gray bars) and ICC (black bars) (n = 2094 and 1655 cells, respectively; none of the inter-method comparisons were significant, demonstrating that they detected the same cell populations). The data represent the means of individual blood donors ± SEM.