The control cohort included 63 patients without coronary atherosclerosis, and the second one included 76 patients with CAD and stenosis of coronary arteries confirmed by coronary angiography. The hemodynamically significant stenosis was diagnosed in a case of more than 50% luminal narrowing in at least one vessel. The number of stenosis in CAD patients varied from 1 to 4 lesions. The patients in the control cohort were without any visible stenosis or with hemodynamically insignificant (20%) stenosis. For both cohorts, inclusion criteria were the absence of hypertension and diabetes, not alcoholics, and the absence of treatment with corticosteroids or lipid-lowering drugs for at least 3 months. The subjects in both cohorts were divided into three groups: low, normal and high HDL-C. The lower threshold of 1.0 mM between low and normal HDL-C levels is well known for men, while the upper threshold of 1.4 mM was chosen based on the recent findings on the U-shaped dependency between HDL-C and cardiovascular disease mortality [12] with a 1.4 mM within the reference interval [18]. The study was approved by the Ethics Committee of the M. F. Vladimirsky Moscow Regional Research and Clinical Institute (MONIKI), Russia (Protocol No. 12479/2019, February 17, 2019) and carried out in accordance with the WMA Declaration of Helsinki (2004). Written informed consent was obtained from patients or their families/legal guardians included in the study. Apolipoprotein and lipid measurements based on standardized protocols with reference materials provided by manufacturers were previously described [14].

PBMC preparation, total RNA isolation, cDNA synthesis, gene-specific primers, and real-time PCR conditions were described in our preceding paper [14]. Two gene clusters were selected to follow HDL metabolism and atherogenesis [13]. HDL cluster (24 genes) included ALB, A2M, AMN, APOA1, APOE, ABCA1, ABCA5, ABCG1, BMP1, CETP, CUBN, HDLBP, HMGCR, LCAT, LIPC, LPL, LDLR, PLTP, PRKACA, PRKACB, PRKACG, SCARB1, SOAT1, and ZDHHC8 genes. Atherogen cluster (41 genes) included ASGR2, CD14, CD36, F5, CSF1R, CSF2RB, CXCL5, CYBA, ITGA2B, ITGAM, ITGB3, ICAM1, IL1B, IL1R1, IL18, IL18R1, IL18RAP, JAM3, LTA, MMP9, MGST1, NPC1, NPC2, NR1H2, NR1H3, OLR1, PCTP, PLA2G7, PRKCQ, S100A12, S100A8, S100A9, SLPI, SLC7A11, SREBF1, SOD2, TNFRSF1A, TNFRSF1B, TLR5, TLR8, and VEGFA genes. Housekeeping genes included GAPDH, LDHA, and RPL3 genes.

Functional enrichment analysis was done with the Gene Ontology (GO, http://geneontology.org) [19, 20] and Reactome [21] pathway databases. GO categories and Reactome pathways were detected with the online tool Database for Annotation, Visualization and Integrated Discovery (DAVID) [22] (https://david.ncifcrf.gov/). The protein-protein interactions (PPI) were analyzed with the Search Tool for the Retrieval of Interacting Genes (STRING) database [23] (https://string-db.org/) and subsequent visualization with Cytoscape [24]. The CytoHubba plugin, with the maximal clique centrality (MCC) algorithm as the best node ranking method [25], was used to find the hub genes. The p value for PPI enrichment is an indicator of interaction efficiency.

Statistica 13 software (StatSoft, Inc. STATISTICA (data analysis software system), version 13. https://www.statsoft.com) was used. The Kolmogorov-Smirnov test was used to check for normality. Differences in continuous variables between groups were analyzed by the Mann-Whitney test. The statistical significance limit was accepted as p 0.05. Real-time PCR data were analyzed with the REST 2009 version 1 software [26] (https://www.gene-quantification.de/rest-2009.html). The relative expression of all target genes normalized by the geometric mean of three reference genes [27] was used in the logistic regression and receiver operating characteristic (ROC) analyses. The false discovery rate (FDR) for multiple comparisons was controlled by the Benjamini–Hochberg procedure [28], with an FDR value of 0.05.

The prevalence of the overall differentially expressed gene (DEG) sets in CAD and the performance of the hub genes as gene signatures for CAD patients with low, normal, and high HDL-C values were evaluated with logistic regression and ROC analysis. The Z score [29] was used for the calculation of the odds ratio per unit standard deviation (SD) increase to predict CAD prevalence with the proposed gene signature in dependence on HDL-C level. The raw expression values for two genes, which were considered specific gene signatures for groups with different HDL-C levels, were initially transformed into average composite Z scores and then used in logistic regression analysis together with individual Z scores for covariates. The area under the curve (AUC), sensitivity, and specificity were the measures of the overall performance of the procedure. These calculations were performed with MedCalc Statistical Software version 22.016 (MedCalc Software, Ostend, Belgium).

The subjects were divided into Group 1, Group 2, and Group 3 with low, normal, and elevated HDL-C. Anthropometric and laboratory data for each group are given in Table 1. For the control cohort, the numbers of patients in the three groups were equal to 19, 25, and 19 patients, respectively. These groups did not differ from each other by body mass index (BMI) or age. However, the mean HDL-C and apoA-I values significantly increased, while the atherogenicity index decreased with the increase in HDL-C, as expected. The mean levels of total cholesterol, LDL-C, non-HDL-C, and apoB for Groups 2 and 3, as compared with Group 1, were significantly higher. Patients with hyperalphalipoproteinemia possessed the lowest triglyceride (TG) and very low-density lipoprotein cholesterol (VLDL-C) levels (Table 1). For the CAD cohort, three groups of patients with low, normal, and increased HDL-C levels included 34, 23, and 19 patients, respectively. The above-mentioned changes in lipid and anthropometric data for between-group comparison for the control cohort generally existed for between-group comparison for CAD patients as well (Table 1). However, the identical groups in two cohorts differed significantly from each other by age, HDL-C, apoA-I, and atherogenicity index, LDL-C, non-HDL-C, and TG. Importantly, CAD patients with hyperalphalipoproteinemia possessed increased HDL-C and apoA-I levels. Also, the significant decrease of total cholesterol and LDL-C in Groups 1 and 2 for CAD patients, together with a tendency for Group 3 in comparison with the analogous groups from the control cohort, may originate from the prevalence of small, dense LDL in the blood of CAD patients with the concomitant decrease of the atherogenicity index. Small, dense LDL are characterized by a decreased cholesterol level per single apoB molecule [30].

Logistic regression and ROC analysis with CAD presence as the output (CAD = 1, control = 0) were used to estimate the prevalence of transcript levels with CAD in relation to HDL-C. The values of the relative gene expression for different genes varied by several orders of magnitude, which excluded the correct comparison of odds ratios (OR) within and between patient groups. For a correct comparison, the Z scores for all included variables in various models were used in the logistic regression and ROC analyses. Thus, the odds ratio for a particular transcript was adjusted for lipids and age. With this adjusted OR, the odds of the outcome increase at OR $>$1 and decrease at OR 1 when the value of the transcript level as an independent variable is increased by one SD unit. The area under curve (AUC), sensitivity, and specificity serve as measures of fitting accuracy and predictive power. Data for a single predictor with adjusted OR values higher than 1.05 or lower than 0.95 as the most meaningful OR values are included in Table 2. The calculated AUC values within the 0.73–0.98 range, sensitivity values within the 56–100% range, and specificity values within the 59–100% range indicate the high performance of the analysis. We further evaluated the prognostic significance of selected hub genes described in Section 3.2.1. and Fig. 2A–C as possible CAD gene signatures in the HDL-C context. The data for the best models of the differential expression of hub genes, with the highest AUC values in multiple logistic regression for three groups with low, normal, and high HDL-C, are included in Table 3.

For the binary classification of patient groups with low HDL-C (Table 2), the four genes with OR $>$1 (CXCL5, LPL, NR1H2, and NRH3) are independent predictors of having CAD, and nineteen genes with OR 1 (ABCA1, ABCG1, ALB, CD14, CUBN, F5, HDLBP, IL18R1, IL1R1, ITGAM, LCAT, pCTP, pRKACG, PRKCQ, S100A8, SCARB1, SREBF1, S100A12, and TNFRSF1B) are independent predictors of CAD absence. By multivariate logistic regression with the expression data for hub genes adjusted for age and lipids, the composite expression of SREBF1 and ABCG1 genes was an independent predictor of CAD absence for patients with low HDL-C (Table 3). We further suggest the validity of selected hub genes for three HDL-C groups as candidate CAD gene signatures in the context of HDL-C levels.

For the binary classification of patient groups with normal HDL-C, the five genes with OR $>$1 (APOE, CXCL5, LDLR, NR1H2, and NR1H3) are independent predictors of having CAD, and the four genes with OR 1 (ALB, CD14, F5, and ITGAM) are independent predictors of CAD absence (Table 2). By multivariate logistic regression of the expression of hub genes adjusted for age and lipids, the composite expression of ITGAM and ALB genes was an independent predictor of CAD absence for patients with normal HDL-C (Table 3).

For the binary classification of patient groups with high HDL-C, the six genes with OR $>$1 (APOE, CUBN, HDLBP, ITGB3, pRKCQ, and NR1H3) are independent predictors of having CAD, and eleven genes (ABCA5, APOA1, F5, HMGCR, IL1R1, LCAT, NPC2, PCTP, SLPI, S100A12, and S100A8) with OR 1 are independent predictors of CAD absence (Table 2). By multivariate logistic regression of the expression of hub genes adjusted for age and lipids, the composite expression of APOA1 and HMGCR genes was an independent predictor of CAD absence for patients with high HDL-C (Table 3). Of note, one standard deviation higher average gene expressions were associated with 5.1-, 48.8-, and 38.9-fold fewer CAD cases for three patient groups, thus evidencing the strong predictive power of CAD with the proposed gene signature in dependence of HDL-C level.

For the binary classification of CAD patients with high HDL-C versus CAD patients with low HDL-C (Table 2), the odds for a positive outcome in cases with high HDL-C are higher than in cases with low HDL-C for the eight genes with OR $>$1 (ABCG1, ALB, CUBN, HDLBP, IL18RAP, pRKACG, pRKCQ, and SREBF1). In contrast, the odds for a positive outcome in cases with high HDL-C are lower than in cases with low HDL-C for the nineteen genes with OR 1 (BMP1, CD14, CD36, CYBA, HMGCR, LCAT, NPC2, MGST1, OLRI, PRKACA, pRKACB, S100A12, S100A8, S100A9, SLPI, SOAT1, TLR5, TLR8, and VEGFA). By multivariate logistic regression of the expression of hub genes adjusted for nonHDL-C, TG, and age, the composite expression of LCAT and HMGCR genes as a candidate gene signature was inversely associated with prevalent HDL-C in CAD patients (Table 3).

The expression of 65 genes involved in HDL metabolism, inflammation, and immunity reactions was compared between CAD patients (case) and patients without coronary stenosis (control), each divided into three groups with low, normal, and elevated HDL-C levels. The common feature was the increased OR values for LXR gene expression in three patient groups, thus evidencing the increased oxysterol level in CAD patients. In particular, the positive associations of the adjusted OR values for NR1H3 expression imply that, one standard deviation higher, these values were associated with 6.8-, 4.2-, and 3.9-fold higher CAD cases in low, normal, and high HDL-C groups, respectively. However, the expression of LXR-target genes ABCA1 and ABCG1 and SCARB1 as major cholesterol transporters [8] in the CAD group with low HDL-C was significantly decreased with OR 1, suggesting the impairment of cholesterol efflux. Notably, multilevel regulation of ABCA1, ABCG1, and SCARB1 expression exists [31]. The transcription of sterol regulatory element binding transcription factor 1 (SREBF1) as a master gene in fatty acid and triglyceride biosynthesis [32] was also decreased. Distinct data exist on the SREBF1 expression as a risk factor in CAD [33, 34]. The additional supply of free fatty acid as a substrate for the increased cholesteryl ester synthesis is strongly suggested by the increased expression of the LPL and SOAT1 genes in PBMC from CAD patients with low HDL-C.

Expression profiles and OR values were much more uniform in the CAD group with a normal HDL-C level. The increased OR value for LDLR gene expression in CAD observed in the present study diverges from the coordinate activation of cholesterol excretion through the LXR-controlled pathway and the depression of cholesterol uptake by the LDL receptor through the SREBF-controlled pathway [5]. However, the direct involvement of the hepatic LDL receptor in reverse cholesterol transport from macrophages to the liver has been identified [35].

For CAD patients with elevated HDL-C, the upregulation of CUBN and HDLBP genes (OR $>$1) contrasts with the decreased expression of these genes (OR 1) for CAD patients with low HDL-C. Based on the decrease in HMGCR expression, a decreased synthesis of cholesterol may be suggested. This may occur due to the accumulation of intracellular free cholesterol and the concomitant decrease in NPC2 expression [36]. Cholesterol efflux may be more balanced for CAD and control groups with elevated HDL-C, which may be suggested by the change in cholesterol transporter’s expression. However, cholesterol efflux may be hampered by the downregulated APOA1 expression, which is a primary cholesterol acceptor in ABCA1-mediated cholesterol efflux [8]. This effect may be compensated by the increased expression of the CUBN, HDLBP, and APOE genes. The upregulation of LXR$\alpha$ in CAD patients with normal and high HDL-C likely underlies the upregulation of APOE expression [37]. Also, in atherosclerosis plaques, HDLBP was found to be especially expressed in foam cell macrophages and colocalized with apoE. Their expression levels can both be elevated in vitro by high cholesterol loading. This indicates a similar area of effect for both proteins, but it could also be shown that HDLBP expression is independent of ApoE [38]. Moreover, overexpression of HBP/vigilin increases HDL binding to cultured cells [38]. However, there is an opinion that HDLBP plays an essential role in protecting cells against excessive cholesterol accumulation [39]. Also, the CUBN gene encodes a high-affinity apoA-I receptor that facilitates HDL endocytosis [40].

The possible atheroprotective effect of HDL particles in CAD may be analyzed by the adjusted OR values for CAD patients with high HDL-C versus CAD patients with low HDL-C. Based on the decrease in HMGCR and SOAT1 expression, the decreased synthesis of both cholesterol and cholesteryl ester may be suggested due to the accumulation of intracellular cholesterol in CAD patients with high HDL-C. The suggested intracellular cholesterol accumulation coincides with the increased expression of SREBF1 with the increase of HDL-C. Cholesterol efflux becomes more efficient due to the increase of ABCG1 and ALB expression as cholesterol transporters [8].

What could be a driving force for cholesterol accumulation in PBMC in CAD patients with high HDL-C? We speculate on the complex interplay between HDLBP, CUBN, and BMP gene expression. The specific Reactome pathway (R-HSA-8964043~Plasma lipoprotein clearance) indicates a HDLBP pivotal role in HDL clearance at high HDL-C levels. HDL levels in plasma from CAD patients may also be controlled by the increased and decreased OR values for CUBN and BMP1 expression, respectively. However, the decreased OR for BMP1 may be associated with the loss of the maturation of proapoA-I and increased clearance of uncleaved apoprotein [41]. The decrease of LCAT expression with the increase of HDL-C in CAD patients may be associated with the accompanying increase of HDL atheroprotective effect due to the higher efficiency of smaller HDL particles that accumulate at lower LCAT expression as promoters of macrophage cholesterol efflux by the ABCA1 and ABCG1 transporters [42]. Of note, decreased LCAT gene expression also served as a candidate gene signature in the comparison of the expression of hub genes in CAD patients with high HDL-C versus low HDL-C.

The increased expression of CXCL5 as a pro-inflammatory cytokine that accumulates at efferocytosis [43] and decreased expression of SREBF1 in CAD patients with low HDL-C levels may evidence the limited inflammation in CAD due to the linkage of SREBF1 expression to the innate immune response [44]. The decreased innate immune response with lower SREBF1 expression in CAD patients with low HDL-C may be further suggested. Besides the increased CXCL5, the pro-inflammatory effect may also be attributed to the decreased OR for ITGAM (integrin subunit alpha M or CD11b) expression. This down-regulation may suggest the decreased formation of a heterocomplex between alpha M and beta integrin 2 chains. This complex is important in the adherence of neutrophils and monocytes to stimulated endothelium. Macrophage ITGAM was significantly correlated with tissue immune infiltration within atherosclerotic tissues [45]. Also, macrophages deficient in CD11b were characterized by enhanced activation of nuclear factor-$\kappa$B (NF-$\kappa$B) and other TLR-dependent pathways and inflammatory cytokine production [46]. The suggestion of limited inflammation in CAD patients with low HDL-C is further supported by the decreased expression of the IL18R1 and IL1R1 genes. The expression of the interleukin 18 (IL-18) receptor (IL-18R) was measured in our study as it is requisite for IL-18 signaling. This receptor specifically binds interleukin 18 (IL-18). Interleukin 18 belongs to the IL-1 family of cytokines that promote the activity of the innate immune system; the complex formation between IL18 and receptor results in the activation of transcription factor NF-$\kappa$B [47]. The IL1R1 protein is a receptor for interleukin-1 alpha, interleukin-1 beta, and interleukin-1 receptor antagonist. Interestingly, interleukin-1$\beta$ has atheroprotective effects in advanced atherosclerotic lesions in mice [48], which contrasts with an observation that vaccines targeting IL1R1 in mice reduced atherosclerotic plaque area, promoted plaque stabilization, decreased macrophage infiltration in plaques, and influenced macrophage polarization [49]. Also, the TNFRSF1B silencing, which mimics the decreased OR for TNFRSF1B expression, reduces the expression of survival genes [50].

In CAD patients with normal HDL-C, the increased expression of CXCL5 also existed, thus evidencing the pro-inflammatory event. Again, the pro-inflammatory effect may also be attributed to the decreased OR for ITGAM expression. However, the expression of both IL1R1 and IL18R1 receptors, contrary to CAD patients with low HDL-C, did not differ between CAD patients and controls with normal HDL-C, suggesting normalization of the corresponding signaling pathways in CAD patients with normal HDL-C. In contrast to this observation, CD14 expression decreased in two groups of CAD patients with low and normal HDL-C. The cell surface antigen CD14 is primarily understood to act as a co-receptor for Toll-like receptors to activate innate immunity responses to pathogens and tissue injury in macrophages and monocytes [51], thus suggesting limited inflammation in CAD patients with low and normal HDL-C groups relative to controls.

In CAD patients with elevated HDL-C, the expression of some new genes became significant. Expression of ITGB3 (integrin beta chain beta 3) was upregulated, evidencing increased cell adhesion as well as cell-surface-mediated signaling. Integrin beta 3 is found along with the alpha IIb chain in platelets. Circulating monocyte-platelet aggregates are a robust marker of platelet activity and monocyte inflammation and are significantly elevated in subjects with CVD [52]. Of note, the expression of ITGAM as a component of the other integrin heterocomplex between alpha M and beta 2 integrin did not change in CAD patients versus controls, contrasting with two other groups. The pRKCQ (protein kinase C theta) gene was also up-regulated in CAD patients with elevated HDL-C. This kinase is important for T-cell activation. It is required for the activation of the transcription factors NF-$\kappa$B and AP-1 and may link the T cell receptor signaling complex to the activation of the transcription factors [53]. The decreased expression of S100A8 and S100A12 may favor the atheroprotective effect of the increased HDL-C level. Indeed, S100A8 belongs to the damage-associated molecular pattern (DAMP) protein family and is mainly produced by activated macrophages, monocytes, and neutrophils [54]. In addition, S100A12 promotes M1 macrophage polarization through the NF-$\kappa$B pathway, and LINC00894 has the potential to regulate the expression of S100A12 as a therapeutic approach by acting as a miRNA sponge [55]. Thus, both pro- and anti-inflammatory effects seem to exist in CAD patients with elevated HDL-C. A prominent hallmark of the present study was the observation of the markedly increased expression of APOE in CAD patients with normal and elevated HDL-C levels. This effect suggests apoE-mediated immunometabolic signaling through increasing cellular levels of apoE and miR-146a-5p that reduce NF-$\kappa$B signaling [56].

The comparison of gene expression profiles for CAD patients with elevated and low HDL-C revealed the increased expression of SREBF1 with a decrease in the expression of several genes connected to inflammation and angiogenesis (CD14, CD36, S100A8, S100A9, S100A12, TLR5, TLR8, and VEGFA). Of note, the profound decrease in CD36 gene expression with an OR of 0.00 (95% CI 0.00 to 0.28) may suggest the contribution of CD36 gene expression to the inflammation state of CAD patients with an increase in HDL-C. However, the expression of pro-inflammatory PRKCQ with an OR of 2.94 (95% CI 1.21 to 7.15) was significantly increased with an increase in HDL-C. Moreover, the diminished expression of the CUBN and PRKCQ genes in control patients with an increase in HDL-C level, contrasting with the increased expression of these genes in CAD patients, may attenuate the HDL-associated atheroprotective effect in CAD. Both the upregulation of the expression of genes involved in lipid efflux (ABCG1, ALB) and the downregulation of HMGCR and SOAT1 expression suggest the accumulation of free cholesterol and the decreased synthesis of cholesterol and cholesteryl ester.

The novelties of our study are, first, the use of a content-dependent transcriptome-like approach to suggest gene signatures, specific to HDL-C levels, for accurate prognosis and treatment in CAD, which seems to correct the known heterogeneity in the transcriptome approach [57]. Second, the impairment of reverse cholesterol transport and limited inflammation seem to be the major features of HDL metabolism and innate immune response in CAD patients with hypoalphalipoproteinemia. Third, both pro- and anti-inflammatory effects seem to exist in CAD patients with elevated HDL-C. Finally, the intracellular accumulation of cholesteryl ester and free cholesterol is suggested in CAD patients with low and high HDL-C groups, respectively.

The usage of only expression data, without any protein and/or activity measurements, for a limited number of genes is a major limitation of the study. However, it may be balanced by the potential wide use of real-time PCR analysis in clinical laboratories for a large number of patients with CAD to apply targeted therapy to correct underlying disturbances. Also, the ongoing measurements of cholesterol efflux capacity with cholesterol-loaded cell culture and HDL preparations as cholesterol acceptors will associate in future HDL-specific gene expression profiles with efflux efficiencies.