1. Introduction
In the human body, macrophages perform fundamental functions such as maintaining hemostasis and resisting pathogen invasion [
1]. Upon exposure to various environmental stimuli, macrophages polarize into distinct phenotypes within different tissues, including M1 macrophages and M2 macrophages [
2]. For example, lipopolysaccharide (LPS) can drive polarization toward the M1 phenotype, while interleukin-4 (IL-4) promotes the polarization of the M2 subtype [
3]. M1 macrophages are primarily responsible for triggering inflammatory responses, whereas M2 macrophages serve to balance inflammation and promote wound healing and tissue repair [
4]. In inflammatory tissues, macrophages often undergo M1 polarization initially and are subsequently repolarized into the M2 phenotype for tissue repair. This transition from one phenotype to another is referred to as repolarization or reprogramming.
Numerous studies have detailed memory T and B cells[
5,
6,
7], yet the concept of innate memory and specifically memory macrophages was only established about a decade ago. Recent research challenges the traditional view that relegates innate immune cells merely to the first line of defense, suggesting instead that these cells can also develop and maintain immunologic memory or hysteresis[
8,
9]. Multiple independent clinical and biological studies have confirmed such macrophage hysteresis[
10,
11,
12,
13,
14]. This de-facto innate immune memory, referred to as trained immunity, is mediated through extensive metabolic rewiring and epigenetic modifications[
15]. Energy and lipid metabolism, along with dietary compounds, influence enzymes that regulate chromatin compaction and structure, whereby lipid metabolites play a critical role in mediating epigenetic modifications that impact macrophage immune memory, contributing significantly to trained immunity[
16]. Multiple experiential evidence underscore the significant influence of lipid metabolism on innate immune memory and macrophage hysteresis[
17,
18].
In our previous research, we reported significant hysteresis in bone marrow-derived macrophages following M1 stimulation and subsequent de-/repolarization, attributed to the regulation of various chromatin sites by factors such as AP-1 and CTCF. This sustained M1 hysteresis induces functional heterogeneity in macrophages, affecting inflammatory responses, cell cycle regulation, and cell migration. Remarkably, we observed a hysteresis phenomenon in a set of interferon-stimulated genes (ISGs)[
9]. Research by Lisa Willemsen et al. indicates that lipid accumulation in mouse and human macrophages leads to differential expression of type-I interferons[
19]. Furthermore, studies suggest that lipid metabolism is crucial for mounting effective inflammatory responses[
20]. We hypothesize that the observed M1 hysteresis may be directly linked to macrophage lipid metabolism. In this study, through RNA-seq analysis of a total of 162 polarized and lipid-loaded macrophages (
Figure 1), we demonstrate that the scavenger receptor Marco (SR-A6) sustains high expression during de-/repolarization, leading to persistent lipid intake. Concurrently, the low expression of ATP Binding Cassette Subfamily A members (Abca1 and Abca2) results in reduced lipid efflux, collectively creating an environment of high intracellular lipid levels. This environment compensates for the suppressed expression of lipid and cholesterol biosynthesis-related genes. The elevated lipid levels provide sustained energy to macrophages, prolonging the inflammatory state associated with M1 hysteresis. These findings significantly enhance our understanding of the regulation of macrophage hysteresis.
3. Discussion
Extensive prior research, as well as our studies, have confirmed that macrophages, like adaptive immune cells, can develop an immunological memory, which is reflected not only in gene expression but also in epigenetic modifications. In previous studies, we utilized in vitro data from BMDM polarization and depolarization to ascertain significant M1 hysteresis in vitro, a phenomenon closely associated with the regulation of different chromatin regions by the AP-1 family and CTCF. We are aware that sustained M1 inflammatory responses require external lipid intake or endogenous biosynthesis to continue as a source of energy. Concurrently, we observed a hysteresis phenomenon in the gene expression of numerous ISGs, which may be linked to lipid metabolism. We propose that macrophage hysteresis may be associated to some extent with macrophage lipid metabolism.
In this study, we reanalyzed RNA-seq data from macrophages during polarization and depolarization over a period of 0 to 96 hours, incorporating RNA-seq from lipid-loaded macrophages for comparative validation. Our findings include 1. Due to the substantial plasticity of macrophages and their extreme sensitivity to stimulation duration. 2. Through WGCNA analysis, we identified two gene clusters significantly related to M1 hysteresis, MEblue and MEbrown. Compared to MEblue, most MEbrown genes exhibited a late onset of gene upregulation or downregulation after 12 hours of stimulation. Functionally, MEblue genes are enriched in M1 immune functions and interferon-mediated signaling pathways, whereas MEbrown genes are highly related to lipid metabolism, cholesterol biosynthesis, and secondary alcohol biosynthesis. 3. Further exploration within MEbrown identified 22 core driver hub genes within the gene co-expression network generated through WGNCA, including those regulating cholesterol biosynthesis such as Dhcr24, Idi1, Fdps, Fdft1, and Cyp51, which exhibit either significant positive or significant negative correlations and robust PPI relationships. These genes were also found to be significantly downregulated in lipid-loaded samples. 4. Further analysis of the correlation between hub genes which regulate cholesterol biosynthesis and ISGs revealed significant negative relationships. 5. We observed that during the M1 hysteresis process, the scavenger receptor Marco, which regulates macrophage lipid intake, exhibited a hysteretic RNA expression pattern. Regulators of lipid efflux, such as Abca1 and Abca2, along with their upstream regulators, showed persistently low expression levels. These expression patterns compensate for the reduction in intracellular cholesterol biosynthesis, maintaining lipid homeostasis within macrophages and providing the necessary energy supply for the sustained activation of pathways like IFN signaling.
Integrating these findings, we can conclude our hypothesis that following depolarization or repolarization, the sustained high expression, i.e. hysteresis, of the scavenger receptor Marco on the macrophage surface membrane leads to higher intracellular lipid uptake. Concurrently, the continuous expression of the upstream mediator Lxra leads to the suppression of genes such as Abca1, Abca2, Cd5l, and Apoc1, which are in charge of lower lipid efflux in macrophages. Studies have shown that lipids uptaken by macrophages are broken down into free cholesterol (FC) and fatty acids (FA), which are then transported to the endoplasmic reticulum (ER) membrane for further processing. When deregulated lipid uptake occurs, the excessive accumulation of FC and other lipids in the ER activates sensors of the unfolded protein response, triggering ER stress, which in turn regulates the gene expression on the lipid uptake, efflux, and biosynthesis [
30]. So it is highly likely that the increase in lipid uptake, combined with the reduction in lipid efflux, leads to lipid accumulation within the cells. Elevated lipid levels subsequently influence gene expression, resulting in the suppression of genes involved in lipid biosynthesis, such as Dhcr24, Idi1, Fdps, and Fdft1, to balance overall intracellular lipid levels. This phenomenon was also confirmed in our validation analysis using data from lipid-loaded samples. The accumulated intracellular lipids then serve as an energy source for the inflammatory responses generated by M1 hysteresis. Ultimately, this results in the activation of regulators such as IRFs, STATs, AP-1, and CTCF in M1 hysteresis, contributing to cellular functional heterogeneity, which has been extensively explored in our previous research [
9] (
Figure 8E).
Over recent years, the role of lipid metabolism has been extensively studied within the classic M1/M2 macrophage polarization paradigm. Numerous studies underscore the critical role of lipid metabolism in inflammatory macrophage polarization, supporting proper inflammatory responses through membrane remodeling and as precursors for various inflammatory mediators [
35,
36,
37,
38]. Lipids, sourced from diverse origins such as native LDL, oxidized LDL particles, or free fatty acids, and existing in different forms, are internalized by macrophages through various scavenger receptors, including CD36, MARCO, and SR-A1. Once inside the cell, most lipids—including those bound to lipoproteins and fatty acids—are transported to the lysosomal-endosomal compartment. Here, lysosomal acid lipase (LAL) catalyzes the conversion of LDL-derived cholesteryl esters into fatty acids and free cholesterol, which are then transferred to the endoplasmic reticulum where excess lipids are stored in lipid droplets [
39,
40,
41]. These lipids continuously provide energy to the macrophage. Additionally, macrophages can actively export lipids through ATP-binding cassette (ABC) transporters. Both lipid scavenging receptors and ABC transporters play pivotal roles in dictating the metabolic fates of scavenged lipids [
42,
43,
44].
Multiple lines of evidence substantiate the concept that lipid remodeling following macrophage activation is crucial for maintaining appropriate inflammatory responses and host defense [
45,
46,
47]. Certain lipids also function as precursors for the production of inflammatory lipid mediators [
48]. MARCO, a class A scavenger receptor, is widely expressed across various myeloid populations, including peritoneal, alveolar, splenic macrophages, Kupffer cells, and dendritic cells [
49,
50,
51]. It is highly induced by stimuli such as LPS or granulocyte/macrophage colony-stimulating factor (GM-CSF) [
52], which are associated with lipid accumulation during M1 hysteresis. Rapid induction of MARCO expression in response to infectious stimuli plays a pivotal role in mediating appropriate Toll-like receptor (TLR)-dependent inflammatory responses[
51,
53]. Further, evidence indicates that lipid-loaded macrophages exhibit heightened MARCO surface expression. Notably, murine and human tumor-associated macrophages (TAMs) show elevated MARCO expression, potentially linked causally to their increased lipid accumulation[
54,
55,
56]. Experiments demonstrate that the accumulation of modified and oxidized lipids is a primary driver of inflammatory responses and is implicated in numerous diseases including atherosclerosis, steatohepatitis, obesity-induced insulin resistance, and neurodegeneration[
57,
58,
59]. While these studies collectively validate our findings, several aspects still require attention, such as potential batch effects due to variations in experimental conditions across different datasets, and the limitation of RNA-seq data capturing only the endpoints of de-/repolarization at 96 hours. Therefore, further data and experimental validation are needed in the future to deepen our understanding of the hysteresis phenomenon in macrophages.
Author Contributions
Conceptualization, Y.Z. and K.N.; methodology, Y.Z., W.Y, Y.K and Y.Y; software, Y.Z.; validation, Y.Z.; formal analysis, Y.Z.; investigation, Y.Z.; resources, Y.Z.; data curation, Y.Z.; writing—original draft preparation, Y.Z.; writing—review and editing, Y.Z., L.M., S.J.P and K.N.; visualization, Y.Z.; supervision, K.N.; project administration, K.N.; funding acquisition, K.N. All authors have read and agreed to the published version of the manuscript.
Figure 1.
Flowchart illustrating the study design. Workflow for data processing and quality control, weighted gene co-expression network analysis (WGCNA), functional analysis, and downstream analysis in this study.
Figure 1.
Flowchart illustrating the study design. Workflow for data processing and quality control, weighted gene co-expression network analysis (WGCNA), functional analysis, and downstream analysis in this study.
Figure 2.
Batch effect correction and quality control in the macrophages gene expression data. (A)Relative gene expression levels (Z-score) among the RNA-seq data before and after the batch effect correction and (B)Trajectory PCA plot of macrophage RNA-seq expression across all phenotypes. The points within the red dashed box represent lipid-loaded samples. (C) Hierarchical clustering based on RNA expression of all samples. The red arrow indicates the M1-like phenotype direction, while the blue arrow indicates the M2-like direction. Lipid-loaded macrophage samples are enclosed in a red dashed box.
Figure 2.
Batch effect correction and quality control in the macrophages gene expression data. (A)Relative gene expression levels (Z-score) among the RNA-seq data before and after the batch effect correction and (B)Trajectory PCA plot of macrophage RNA-seq expression across all phenotypes. The points within the red dashed box represent lipid-loaded samples. (C) Hierarchical clustering based on RNA expression of all samples. The red arrow indicates the M1-like phenotype direction, while the blue arrow indicates the M2-like direction. Lipid-loaded macrophage samples are enclosed in a red dashed box.
Figure 3.
Re-grouping of samples based on macrophage stimulation time. The horizontal axis of the table represents the grouping of macrophages based on different polarization stimulation times and types, while the vertical axis displays the transcriptome distribution tree using hierarchical clustering for all samples. The red and blue arrows in the table indicate trends toward M1-like and M2-like macrophages, respectively. The red arrows below represent M1 repolarization stimulation, the blue arrows represent M2 repolarization stimulation, and the gray arrows represent depolarization. The points on the scale below indicate the approximate positions of each depolarized/repolarized sample, with the M0, M1, and M2 phenotypes serving as critical points.
Figure 3.
Re-grouping of samples based on macrophage stimulation time. The horizontal axis of the table represents the grouping of macrophages based on different polarization stimulation times and types, while the vertical axis displays the transcriptome distribution tree using hierarchical clustering for all samples. The red and blue arrows in the table indicate trends toward M1-like and M2-like macrophages, respectively. The red arrows below represent M1 repolarization stimulation, the blue arrows represent M2 repolarization stimulation, and the gray arrows represent depolarization. The points on the scale below indicate the approximate positions of each depolarized/repolarized sample, with the M0, M1, and M2 phenotypes serving as critical points.
Figure 4.
Co-expression network analysis based on WGCNA. (A) Analysis of scale-free fit index and mean connectivity for best parameter screening. (B) Cluster dendrogram of co-expression genes with co-expression modules. (C) Heatmap of associations among module eigengenes with all identified co-expression modules. On the right side, bar plots are used to display the correlation coefficients between MEgreen, MEblack, and MEbrown with all phenotypes. (D) The heatmap shows the expression of MEbrown genes and MEblue genes across all samples. The portion highlighted by the red dashed circle indicates the occurrence of a late on-site phenomenon for brown genes in the 4_96M1 group and the 26_96reM1_M2 group.
Figure 4.
Co-expression network analysis based on WGCNA. (A) Analysis of scale-free fit index and mean connectivity for best parameter screening. (B) Cluster dendrogram of co-expression genes with co-expression modules. (C) Heatmap of associations among module eigengenes with all identified co-expression modules. On the right side, bar plots are used to display the correlation coefficients between MEgreen, MEblack, and MEbrown with all phenotypes. (D) The heatmap shows the expression of MEbrown genes and MEblue genes across all samples. The portion highlighted by the red dashed circle indicates the occurrence of a late on-site phenomenon for brown genes in the 4_96M1 group and the 26_96reM1_M2 group.
Figure 5.
GO and KEGG analysis of all identified gene modules.
Figure 5.
GO and KEGG analysis of all identified gene modules.
Figure 6.
Identifying core sub-network and hub genes for module MEbrown. (A) Scatter plot of module membership and gene significance in the brown module to 24_48h_deM0_M1 and reM2_M1 phenotypes. (B) Venn diagram representing number of hub genes of MEbrown to 24_48h_deM0_M1 and reM2_M1 and their common hub genes (C) Pairwise correlation analysis of 22 common hub genes shows a significant positive or negative correlation between each hub gene in heatmap. (D) Protein-protein interaction (PPI) network of hub genes was identified. Yellow nodes represent uncommon hub genes and red nodes represent common hub genes (E) Experimental based pairwise gene co-expression correlation identified in the STRING database. (F) Bar plots illustrating differences in RNA expression (TPM) of hub genes after treated by AcLDL, GW3965, and KLA. The asterisks indicate that the differences are statistically significant. ‘*’ p < 0.05, ‘**’ p < 0.01, ‘***’ p < 0.001. ‘-’, not significant.
Figure 6.
Identifying core sub-network and hub genes for module MEbrown. (A) Scatter plot of module membership and gene significance in the brown module to 24_48h_deM0_M1 and reM2_M1 phenotypes. (B) Venn diagram representing number of hub genes of MEbrown to 24_48h_deM0_M1 and reM2_M1 and their common hub genes (C) Pairwise correlation analysis of 22 common hub genes shows a significant positive or negative correlation between each hub gene in heatmap. (D) Protein-protein interaction (PPI) network of hub genes was identified. Yellow nodes represent uncommon hub genes and red nodes represent common hub genes (E) Experimental based pairwise gene co-expression correlation identified in the STRING database. (F) Bar plots illustrating differences in RNA expression (TPM) of hub genes after treated by AcLDL, GW3965, and KLA. The asterisks indicate that the differences are statistically significant. ‘*’ p < 0.05, ‘**’ p < 0.01, ‘***’ p < 0.001. ‘-’, not significant.
Figure 7.
Strong gene expression correlation between ISGs and hub genes. (A) Correlation coefficients between ISGs and hub genes selected from MEbrowm. The values and transparency represent Pearson’s correlation coefficients of each gene pair. (B) Correlation coefficients between ISGs and lipid biosynthesis related genes. Heatmap of top 30 genes with the highest connectivity identified in MEmagenta and MEred across all samples. (C-H) Gene expression levels of the ISGs, negative correlation hub genes, and positive correlation hub genes under (C,D,E) M0->M1->M0 and (F,J,H) M0->M1->M2 from 0 h to 96 h.
Figure 7.
Strong gene expression correlation between ISGs and hub genes. (A) Correlation coefficients between ISGs and hub genes selected from MEbrowm. The values and transparency represent Pearson’s correlation coefficients of each gene pair. (B) Correlation coefficients between ISGs and lipid biosynthesis related genes. Heatmap of top 30 genes with the highest connectivity identified in MEmagenta and MEred across all samples. (C-H) Gene expression levels of the ISGs, negative correlation hub genes, and positive correlation hub genes under (C,D,E) M0->M1->M0 and (F,J,H) M0->M1->M2 from 0 h to 96 h.
Figure 8.
Strong gene expression correlation between ISGs and hub genes. (A-D) Gene expression levels of (A) Lox-1, Cd68, SR-A1, Cd36, (B) Marco (SR-A6) (C) Lxra, Abca1, Abca2, (D) Cd5l, Apoc1 under M0->M1->M0 between 0 to 96 hours (E) Schematic representation of the relationship between macrophage M1 hysteresis and macrophage lipid metabolism.
Figure 8.
Strong gene expression correlation between ISGs and hub genes. (A-D) Gene expression levels of (A) Lox-1, Cd68, SR-A1, Cd36, (B) Marco (SR-A6) (C) Lxra, Abca1, Abca2, (D) Cd5l, Apoc1 under M0->M1->M0 between 0 to 96 hours (E) Schematic representation of the relationship between macrophage M1 hysteresis and macrophage lipid metabolism.