Macrophage gene expression determines phagocyte responses and effector functions. Macrophage plasticity has been mainly addressed in in vitro models that do not account for the environmental complexity observed in vivo. In this study, we show that microarray gene expression profiling revealed a highly dynamic landscape of transcriptomic changes of Ly6CposCX3CR1lo and Ly6CnegCX3CR1hi macrophage populations during skeletal muscle regeneration after a sterile damage. Systematic gene expression analysis revealed that the time elapsed, much more than Ly6C status, was correlated with the largest differential gene expression, indicating that the time course of inflammation was the predominant driving force of macrophage gene expression. Moreover, Ly6Cpos/Ly6Cneg subsets could not have been aligned to canonical M1/M2 profiles. Instead, a combination of analyses suggested the existence of four main features of muscle-derived macrophages specifying important steps of regeneration: 1) infiltrating Ly6Cpos macrophages expressed acute-phase proteins and exhibited an inflammatory profile independent of IFN-γ, making them damage-associated macrophages; 2) metabolic changes of macrophages, characterized by a decreased glycolysis and an increased tricarboxylic acid cycle/oxidative pathway, preceded the switch to and sustained their anti-inflammatory profile; 3) Ly6Cneg macrophages, originating from skewed Ly6Cpos cells, actively proliferated; and 4) later on, restorative Ly6Cneg macrophages were characterized by a novel profile, indicative of secretion of molecules involved in intercellular communications, notably matrix-related molecules. These results show the highly dynamic nature of the macrophage response at the molecular level after an acute tissue injury and subsequent repair, and associate a specific signature of macrophages to predictive specialized functions of macrophages at each step of tissue injury/repair.
Sterile inflammation is the response to tissue damage. After mounting inflammation to remove tissue debris, resolution of inflammation takes place to allow tissue repair. Macrophages are major players in these processes because they are involved in both mounting and resolution of inflammation, as well as in subsequent tissue repair. These multiple roles are possible due to their high versatility and plasticity (1). Macrophage plasticity has been mainly explored in vitro, where defined conditions of stimulation led to the definition of activation states. Classically activated macrophages (M1) were originally elicited by IFN-γ (then later on by LPS) and secrete proinflammatory compounds (IL-12, IL-1β, TNF-α, reactive oxygen species, and so on). Alternative activation state was first described upon IL-4 exposure. M2 macrophages notably express YM1, arginase 1, CCL24, and CCL17. However, a series of stimuli (glucocorticoids, TGF-β, IL-10, etc.) induce M2-like phenotypes, which overlap with each other in the expression of some genes and functions (2). Moreover, it has been recently shown that even in a controlled in vitro context, a variety of stimuli induce some intermediate phenotypes between the originally described M1 and M2 profiles (3).
In vivo analyses of macrophages used transgenic mice strains able to trace macrophage subsets, which fates and phenotypes depend on the tissue and the immune context (4–7). Tissue damage induces the infiltration of CCR2posCX3CR1loLy6Cpos monocytes that express proinflammatory markers. In all tissues studied so far, resolution of inflammation is associated with a shift of the inflammatory status of macrophages that become CCR2negCX3CR1hiLy6Cneg (4, 5, 7). The latter sustain dampening of inflammation and tissue remodeling (1). Based on the analysis of a series of markers, the CCR2posCX3CR1loLy6Cpos/CCR2negCX3CR1hiLy6Cneg macrophage subsets were proposed to mirror the M1/M2 in vitro paradigm. However, aligning the macrophage phenotypes observed in vivo with those described in vitro has proved difficult because of the overlapping of several markers in the in vivo populations (8). Recent studies, based on high-throughput technologies, have defined phenotypes of in vivo macrophages, but these studies have been mainly carried out in resting tissues. They defined a core signature of macrophages, helping the identification of new pan-macrophage markers, such as CD64 and merTK (9). In contrast, genomic analyses evidenced that tissue-resident macrophages isolated from a variety of tissues exhibit specific enhancer/transcriptional signature according to the tissue where they reside (10, 11). As a whole, very little information is available about the dynamics of the molecular signature of macrophages during acute inflammation and subsequent tissue repair in vivo.
We provide in this article a comprehensive analysis of the dynamics of the molecular signature of macrophages from the mounting of the immune response to the resolution of inflammation and tissue repair. For this purpose, the model of postinjury skeletal muscle regeneration, which follows a reproducible kinetics of macrophage subset transition, was used in the CX3CR1GFP/+ mouse strain, which allows to trace macrophages (4–7, 12). Macrophage subsets were isolated at various time points of muscle regeneration and their transcriptomic profile was analyzed by microarray, defining new molecular pathways characterizing in vivo dynamics of macrophage populations during tissue repair.
Materials and Methods
CX3CR1GFP/+ mice were bred and used according to French legislation. Experiments were conducted at 8–10 wk of age. Muscle injury was caused by i.m. injection of cardiotoxin (CTX) in the tibialis anterior muscle, as previously described (13). Muscles were harvested 1, 2, 4, and 8 d postinjury. In some experiments, CTX was coinjected with LPS (50 μg/ml final). Muscle fascia was removed; then muscles were either dissociated or frozen in nitrogen-chilled isopentane and kept at −80°C until use.
Cryosections were prepared for H&E staining and immunolabeling. Immunolabeling used anti-F4/80 (ab6640, ab74383; Abcam), Lcn2 (BAF1857; Biotechne), saa3 (provided by P.E. Scherer, Dallas, TX), and haptoglobin (ab35835; Abcam) Abs. Secondary Abs were coupled to FITC, Cy3, or Cy5 (Jackson Immunoresearch). HE‐stained muscle sections were recorded with a Nikon E800 microscope at 20× magnification connected to a QIMAGING camera. Fluorescent immunolabeling was recorded with a Zeiss Axio Observer Z1 microscope connected to a Coolsnap camera at 20× magnification.
Isolation of macrophages from skeletal muscle was performed as described earlier (13) after sorting of CD45pos cells and labeling with allophycocyanin-conjugated anti-Ly6C/G Abs (eBioscience), using a FACSAria III cell sorter (BD Biosciences). Cells were immediately centrifuged and RNA was extracted with the TRIzol/chloroform procedure. Validation of cell identity was assessed by flow cytometry using anti-CD64, anti-F4/80, and anti-CD103 (eBioscience) Abs.
Real-time quantitative RT-PCR
RNA was extracted from isolated CD45pos cells with TRIzol/ethanol precipitation. Transcript quantification was performed by real-time quantitative RT-PCR using SYBR green assays (Ifitm6 and Serpinb2) or PrimeTime assays from IDT (Il1f9, Saa3, and S100a9). Results were analyzed with the standard delta cycle threshold method and were normalized to the expression of ActB.
RNA integrity was checked on Agilent Bioanalyzer 2100, RNA samples with >9.0 RNA integrity number value were used in the downstream experiments. NanoDrop ND-1000 was used to determine RNA concentration. Global expression data were obtained using Affymetrix GeneChip Mouse Gene 1.0 ST arrays. Ambion WT Expression Kit (Life Technologies) and GeneChip WT Terminal Labeling and Control Kit (Affymetrix) were used for generating biotin-labeled sense strand cDNA. The biotinylated sense strand cDNA samples were hybridized at 45°C for 16 h; then standard washing protocol was performed using Affymetrix GeneChip Fluidics Station 450, and the arrays were scanned on GeneChip Scanner 7G (Affymetrix). The microarray data are publicly available (data access: http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE71152).
Expression data processing and analysis
Data quality control and analysis were carried out following the recommendations put forward on the ImmGen Web site (http://www.immgen.org/Protocols/ImmGen%20QC%20Documentation_ALL-DataGeneration_0612.pdf). Data were loaded into the Genespring GX software, and robust multi-array average summarization was carried out. Next, a set of filtering steps was applied to the data set. In brief, data distribution curve was generated and the lowest 5% of the entities with detectable signals were filtered out as not expressed. Duplicate entities, not/poorly annotated transcripts and transcripts reporting inconsistent expression values were also discarded. Further analysis was carried out on the filtered data set. Data were analyzed either based on the RAW expression values or after a “per gene” normalization (individual gene expression data normalized to the median of the gene). Further analysis of gene expression and comparisons were made either within Genespring GX or using the R software package. Two-way ANOVA tests were performed in R using functions aov and Tukey HSD of package MASS. Heat maps were drawn with package pheatmap. Statistically significant difference was considered as p < 0.05.
Clusterization of genes according to the time
Data were robust multi-array average normalized with R/Bioconductor. Data were first controlled and analyzed in an unsupervised way by principal component analysis (PCA), and a one-way ANOVA was applied to extract DEGs using PartekGS software. Genes were selected on the global p < 0.01. A cluster analysis was then applied on selection by hierarchical clustering (Pearson for similarity and average for clustering) to find correlate genes. Enrichment analyses were carried out through the use of IPA and Pathway Studio software.
Comparison with previously described modules
Gene set enrichment analysis (GSEA) was performed as described previously (14) using the KEGG pathway database (http://www.genome.jp/kegg/pathway.html) as in Jha et al. (15) and for 49 gene sets associated with macrophage activation as in Xue et al. (3).
Gene ontology analysis
Lists of genes in the clusters were analyzed using DAVID tool (http://david.abcc.ncifcrf.gov/home.jsp) using at least 10 occurrences to create a gene ontology (GO). GOs with p values <0.05 were then entered together with their p value in REViGO tool (http://revigo.irb.hr/), and results were presented according to their log10 p value.
Injection of CTX in tibialis anterior muscle of CX3CR1GFP/+ mouse triggers damage of the whole muscle, a massive infiltration of Ly6CposCX3CR1lo monocytes/macrophages, followed by a sequence of well-characterized events triggering muscle regeneration (12, 16). At day 1 after injury, myofibers were necrotic, whereas neutrophils and circulating CCR2posCX3CR1loLy6Cpos monocytes enter into the damaged muscle to create Ly6CposCX3CR1lo macrophages (Fig. 1). They represent the majority of macrophages at this time point (∼1800 macrophages/mg muscle ). At day 2, macrophages phagocyted the dead myofibers and skewing from Ly6CposCX3CR1lo (hereafter Ly6Cpos) to Ly6CnegCX3CR1hi (hereafter Ly6Cneg) macrophages had already started (17, 18) (Fig. 1). They represent ∼5000 and 5500 macrophages/mg muscle, respectively (13). Day 4 after injury, the regeneration process was visible with the appearance of the new regenerating myofibers surrounded by numerous macrophages, which were mainly Ly6Cneg at this stage (∼5100 versus 750 macrophages/mg muscle ). At day 8, muscle had almost recovered its original appearance, with neofibers characterized by the central location of their nuclei and the progressive disappearance of macrophages (600 Ly6Cneg macrophages/mg muscle ). Macrophages were sorted according to their Ly6C and CX3CR1-GFP expression. Macrophage identity of the isolated CD45posCX3CR1pos muscle cells was confirmed by the almost complete overlap of CX3CR1 labeling with two macrophage markers, F4/80 and CD64, and by the absence of overlap of CX3CR1 labeling with the DC markers CD103 (Fig. 1). Pure populations (>98%) were obtained (Fig. 1) and global gene expression analysis was performed. Because normal skeletal muscle parenchyma contains only few macrophages (12, 19), this state was not evaluated. Notably, with very few amounts of Ly6Cpos cells being recovered at day 8, only one point (and not triplicate) was analyzed, to avoid unethical sacrifice of numerous mice.
Muscle macrophages among other macrophage subtypes
PCA analysis of muscle macrophages versus monocytes, macrophages, and DCs sorted from various tissues and identified by the ImmGen Consortium (9, 20) revealed that muscle macrophages formed a distinguishable group (Supplemental Fig. 1A). Because a technical batch effect resulting in a similar pattern cannot be excluded, the batch removal utility of the Iimma R package (21) was used. The analysis showed that batch effects played a detectable but minor role in the PCA analysis, and that the two-dimensional PCA distribution of muscle macrophages was not altered (Supplemental Fig. 1B). We therefore suggest that the specific clustering of muscle macrophages is mostly due to their origin from a regenerating and not a resting tissue. Expression of Ly6C and Cx3cr1 genes reported the expected patterns in the sorted populations and validated the sorting strategy (Fig. 2A). ImmGen proposed a list of genes that are the most specific to macrophages and classical DCs (9, 20), which was compared with muscle-derived macrophages (Supplemental Fig. 2A). Three genes (Slamf7, Ass1, and Gpr132) showed a DC-like expression pattern in all muscle-derived macrophages. Eighteen of the 24 classical DC genes were either not (Rab30, Hmgn3, Gpr82, etc.) or were weakly (kit, amica1, ap1s3, pvrl1, etc.) expressed by muscle macrophages. The three remaining genes (CCR7, P2Ry10, and Adam19) were upregulated in Ly6Cneg macrophages notably at late time points. Conversely, muscle-derived macrophages clearly expressed a core macrophage signature and clustered with macrophages sorted from other origins, and apart from DCs (Supplemental Fig. 2B). Only a few genes (Pecr, Tmem195, and Pla2g15) were not or were very weakly expressed.
Unique molecular signature of muscle macrophage subsets
PCA showed that macrophage populations segregated by day, more than by their LyC6pos/neg status (Fig. 2B). Day 1–2 macrophages clustered together, whereas day 4–8 macrophages clustered together apart, corresponding to the proinflammatory phase (days 1–2) and the resolving/regenerative phase (days 4–8) of muscle regeneration, respectively. Scatterplots of gene expression changes between macrophage subsets showed a correlation coefficient r value of 0.222 at day 2, indicating that the most differential transcriptional changes occurred at day 2, which corresponds to the beginning of skewing of macrophages from Ly6Cpos to Ly6Cneg phenotype (13, 22) (r values of 0.696, 0.756, and 0.499, for days 1, 4, and 8, respectively) (Fig. 2C). Further analysis of differential gene expression involved two-way ANOVA analysis, followed by Tukey post hoc tests (p < 0.05) (Supplemental Table 1). Fig. 2D shows a scheme of the comparisons made, with the number of genes showing significant differences between any two conditions. In vertical comparisons, Ly6Cpos were compared with Ly6Cneg macrophages for each time point. Confirming the earlier analyses, the highest number of differentially expressed genes was observed at day 2 of regeneration. In horizontal comparisons, each macrophage subset was compared with its closest neighboring time points to analyze variation of gene expression longitudinally along the regeneration process. The numbers of genes differentially expressed in horizontal comparisons were higher than that in vertical comparisons, confirming the PCA analyses. Thus, although Ly6Cpos and Ly6Cneg macrophages were different, both underwent larger transcriptomic changes during the time course of muscle regeneration, and particularly between days 2 and 4 (Fig. 2D).
Comparative analysis of Ly6Cpos and Ly6Cneg macrophage subsets
Previous studies showed that Ly6Cpos macrophages express higher levels of TNF-α and IL-1β and lower levels of TGF-β and IL-10 than Ly6Cneg macrophages (12, 16, 23). Therefore, they were considered as proinflammatory and anti-inflammatory macrophages, respectively. However, the analysis of a series of markers showed that, in general, Ly6Cpos and Ly6Cneg macrophages did not exhibit high differential expression levels of proinflammatory (M1) and anti-inflammatory (M2) markers (Fig. 3A, 3B). The difference in expression was modest and reached significance only for Tlr4, IL1rap, IL15, and IL18 genes of the 23 investigated proinflammatory markers (Fig. 3A) and for 6 of the 40 anti-inflammatory genes analyzed (Irf4, Ccl17, Ccl22, Ccl12, Clec10a, and Socs2) (Fig. 3B) at day 1. These results show that Ly6Cpos and Ly6Cneg macrophages in regenerating muscle cannot be defined as canonical M1 or M2 macrophages. To refine muscle-derived macrophage subset phenotypes, we performed GSEA enrichment against the 49 modules of genes defined by Xue et al. (3), in which various combinations of these modules characterize 29 conditions of in vitro activation of human macrophages (Supplemental Table 2). Among all comparisons, only genes upregulated in Ly6Cneg macrophages at day 4 of regeneration were associated with 26 modules. The best representative combinations (Supplemental Table 2) were treatments with fatty acids (oleic, stearic, and lauric acid) and with TNF-α+Pam3CysSerLys4+PGE2, all of which triggered opposite activation status along the IFN-γ/IL-4 activation axis (3). These results indicate that, in vivo, macrophages display highly different inflammatory profiles than those previously described in vitro and from the M1/M2 nomenclature.
The expression pattern of a series of molecules secreted by macrophages and involved in myogenesis (insulin-like growth factors, urokinase plasminogen activator, platelet-derived growth factor α, hepatocyte growth factor, bone morphogenetic protein 2 [24–29]) was increased with time of regeneration, and particularly in Ly6Cneg macrophages (Fig. 3C). This shows good overlap with anticipated expression patterns of regulators of muscle regeneration. This indicated that an unbiased analysis of gene expression patterns might report yet unidentified molecular mechanisms during both acute and resolution phases of inflammation.
Comparison of Ly6Cpos and Ly6Cneg macrophage subsets during muscle regeneration
Vertical comparison between Ly6Cpos and Ly6Cneg macrophages at day 1 showed that the list of upregulated genes in Ly6Cpos macrophages (Fig. 2D, Supplemental Table 1) was enriched for GO terms associated with innate immune response and cell locomotion (Supplemental Table 3), in accordance with a role in the mounting of the inflammatory response. Strikingly, 8 of 10 of the top-ranking genes were related to inflammation (Supplemental Table 1). Thus, the “find similar entity” algorithm of the Genespring software (with similarity setting 0.5–1) was used to find genes that were regulated similarly as Il1f9, the top-ranked inflammatory gene at day 1 of muscle regeneration. Altogether, 38 genes were identified, among which 17 genes are linked to inflammation (Fig. 4A, underlined in red) and therefore constitute the best molecular signature of Ly6Cpos macrophages after a sterile muscle damage. This panel of Ly6Cpos inflammatory genes showed very good overlap with the genes identified by the GO/REVIGO approach, including alarmins (S100a8 and S100a9) and several acute-phase proteins (Lcn2, Lrg1, Hp, and Saa3). The biological relevance of this expression was confirmed by the detection of haptoglobin, lipocalin-2, and Saa3 proteins in macrophages (F4/80pos) in early regenerating muscle (Fig. 4B).
Classical activation has been originally linked to the IFN-γ–STAT1 signaling pathway (2, 30). We showed that the IFN-γ–responsive gene, Ifitm6, as well as Saa3, Il1f9, Serpinb2, and S100a9, were not differentially expressed in wild-type (WT) versus STAT1 knock-out CD45pos cells sorted from early regenerating muscle (Fig. 4C), indicating that the inflammatory profile of Ly6Cpos macrophages was not driven by IFN-γ. Nevertheless, the expression of these genes relied on a proinflammatory profile because the coinjection of CTX with LPS strongly induced their expression (Fig. 4C).
From day 2, vertical comparisons reflected the proliferative nature of Ly6Cneg muscle macrophages as they showed a high enrichment for cell cycle and cell division GOs, particularly 2 and 4 d after injury (Supplemental Table 3), confirmed by the definition of a cell cycling molecular signature of the Ly6Cneg macrophage subset (Fig. 3D), that was previously evidenced at the cell level by experiments showing that only Ly6Cneg macrophagic cells were capable of proliferation (12).
Dynamics of Ly6Cpos macrophages during the course of muscle regeneration
Ly6Cpos macrophages are the most numerous during the first steps of regeneration. From day 2, they start to switch their phenotype into Ly6Cneg cells (13) whose number gradually increases. Horizontal comparisons showed no specific GO function enrichment between days 1 and 2 of regeneration (Fig. 2D). The most important changes were observed between days 2 and 4 of regeneration when the cells showed marked differential expression of genes related to inflammation and metabolism (Supplemental Table 4). Among the upregulated genes (see top/bottom genes in Supplemental Table 1) were found a series of molecules involved in the interactions with adaptive immunity (Cd209d, Cd8b1, Cd7, Ifng, Cd4, Cd3e, Ciita, Cd8a, Gbp4, P2ry14, Cd28, Cxcl9), whereas both proinflammatory and anti-inflammatory secreted effectors were downregulated (Ppbp, Cxcl3, Il6, F10, Cxcl1, Il1f9, Il10, Itga1, Clec4d, Olr1, Sphk1, Ednrb). A strong decrease of expression of genes involved in transcription, translation, and lipid metabolism was observed (Arg1, Agpat9, B3gnt5, Ero1l, Sgms2, Car5b, Avpi1, Gyk, Ell2, Car6, Agpat4, Gpd2, Odc1, Idi1, Padi4, Eif4e, Nop58, Mboat1, Eef1e1). A complementary tool to analyze gene expression with time was to cluster the genes according to their kinetics of expression. Genes differentially expressed by Ly6Cpos macrophages gathered in only two main clusters that presented radical changes in gene expression profile between days 2 and 4 (Fig. 5). Accordingly, the GO enrichments observed in clusters 1 and 2 matched with those observed for downregulated and upregulated genes between days 2 and 4 depicted earlier. Although much less numerous from day 4, Ly6Cpos macrophages exhibited gene expression variations in some part similar to those of Ly6Cneg cells (see later) during the late steps of regeneration (Supplemental Table 4), indicating that the environment largely controls gene expression of this cell subset from day 4. Notably, because a single sample was available for Ly6Cpos cells at day 8, some variations between days 4 and 8, such as upregulation observed in some genes of cluster 1 (and exemplified in Fig. 4A), were not taken into account for this scarce population at this time point.
Dynamics of Ly6Cneg macrophages during the course of muscle regeneration
Horizontal analysis showed that Ly6Cneg cells upregulated genes, first associated with the preparation of proliferation and then with cell division, until day 4 (Supplemental Table 4), confirming the earlier vertical analysis. GO terms related to response to wounding and immune response were variably regulated in Ly6Cneg macrophages but clearly showed a downregulation of genes associated with chemotaxis and cell motility in the early regenerating phases (see top/bottom genes in Supplemental Table 1) (e.g., Fpr1, Cxcl3, Ccl24, Itga1, Il6, Arg1, Clec4n, Ccl22, Pecam1, Itgal, Selp). At later stages, GO categories including intracellular signal transduction associated with immune response and cell communication (regulation of cytokine production; Supplemental Table 4) were upregulated (e.g., Itk, Ralgds, Adora2b, Arl5b, Map2k3, Fosl2, Rab20, Gem, Dusp16, Rapgef2, and Rcan1 for signal transduction and Il6, vcam1, Il10, Cd81, Bmp2, Fasl, Ccl22, Pdgfb, Ctla4, Il1a, Malt1, and Tigit for immune response) (Supplemental Table 1). Finally, genes associated with the metabolism of ribosomes/RNA, glucose, and macromolecules were downregulated at the late steps of regeneration, indicating a metabolic shift in macrophage subsets during regeneration (Supplemental Table 4). Some of the most downregulated genes were related to the redox status (Arg1, Odc1, Car6, Mdm2, Srxn1, Ero1l, Txnrd1, Adssl1, Sod2, Rars, Gclc, Gpd2) (Supplemental Table 1).
Overall, the differentially regulated genes in Ly6Cneg cells clustered into six groups according to their kinetics of expression (Fig. 5). Clusters 4 and 5 contained genes highly expressed at days 1–2 of regeneration and were enriched for various GOs including RNA metabolism, transcription, translation, preparation of the cell to energy expenditure, and intracellular transport (Fig. 6, Supplemental Table 5). Genes showing a progressive (cluster 1) or strong (cluster 2) upregulation at day 4 showed enrichment in GOs associated with DNA packaging, macromolecule synthesis and assembly, and cell cycle (Fig. 6, Supplemental Table 5). Cluster 3 encompassed genes showing an upregulation with time, particularly between days 4 and 8, and was enriched for wound healing, regulation of cell activation, cytokine production, and cell communication GOs (Fig. 6, Supplemental Table 5). Cluster 6 gathered genes whose expression decreased until day 4 and strongly increased between days 4 and 8. Cluster 6 was enriched for GO terms associated with wound healing and cell communication including regulation of cell differentiation, response to signaling, and tissue remodeling, among which are secreted molecules (Fig. 6, Supplemental Table 5).
Metabolic shift of macrophages during skeletal muscle regeneration
The earlier horizontal analysis of macrophage profile suggested metabolic changes during muscle regeneration. Detailed metabolic regulation has been recently investigated in in vitro–activated macrophages and showed that M2(IL-4) macrophage profile was associated with glutamine metabolism, involved in activation of tricarboxylic acid (TCA) cycle and oxidative phosphorylation (15). Most of the genes (16/22) identified as being dependent on glutamine in Jha et al.’s study (15) were upregulated during the late steps of muscle regeneration (Fig. 7A), suggesting that oxidative metabolism is associated with Ly6Cneg macrophage functions that are more numerous at this time. Moreover, GSEA analysis against the KEGG database was performed for horizontal comparisons as in Jha et al.’s study. In both macrophage subsets (Supplemental Table 6), pathways that were upregulated between days 1 and 2 of regeneration were linked to oxidative metabolism (KEGG citrate cycle TCA cycle, KEGG oxidative phosphorylation, KEGG fatty acid metabolism, KEGG pyruvate metabolism, etc.). Most of these pathways were then downregulated from day 4 of muscle regeneration (Supplemental Table 6). These results indicate important metabolic changes in macrophages, with an upregulation of oxidative metabolism preceding the change of their inflammatory status.
Secretory profile of Ly6Cneg macrophages during the late steps of muscle regeneration
GO cell component enrichment of the 10% most differentially expressed genes revealed that clusters 3 and 6, encoding for genes upregulated at the end of the regeneration process, were enriched for extracellular/secreted products (Fig. 6). Among the most differentially regulated genes, several genes encoding for extracellular matrix (ECM) components and ECM remodeling proteins were observed. Bgn and Postn, encoding for two ECM proteins, were among the top-ranked upregulated genes in Ly6Cneg macrophages between days 2 and 4. Scrutinizing the entire list led to the identification of a set of ECM genes highly expressed by Ly6Cneg macrophages during the late phases of muscle regeneration (Fig. 7B). This indicates that macrophages are direct contributors of ECM reorganization during the healing phase of tissue repair.
Macrophage plasticity is a major paradigm in innate immunology. In vitro experimental systems lack both the complex milieu present in vivo and the dynamic aspects of macrophage phenotype shifts. We have addressed this issue by taking advantage of a highly dynamic sterile inflammatory model, CTX-induced muscle regeneration, in which acute inflammation is rapidly followed by the resolution of inflammation and ensuing regeneration process. In this model, we have assessed the transcriptomes of macrophage populations, which revealed unexpected highly dynamic changes of gene expression during acute inflammation and subsequent tissue repair, confirming the highly versatile nature of these cells.
We show that macrophages from regenerating muscle were distinct from resting macrophages in other tissues, while they exhibited a pronounced macrophage-like core expression signature, particularly at late time points of regeneration. They also expressed at these late stages some DC markers, especially in the Ly6Cneg subset. Expression of DC markers by monocyte-derived macrophages has already been shown (5, 8), and some DC properties have been identified for macrophages in late regenerating muscle (31).
Analysis of gene expression revealed that the elapsed time, much more than Ly6C status, showed the highest differential gene expression in macrophages during muscle regeneration. Thus, the time course of inflammation is the predominant organizing force in gene expression by macrophages, with an important time point at days 2–4 postinjury, which corresponds to the full skewing from Ly6Cpos to Ly6Cneg status (13, 16) and the starting of resolution of inflammation. This result is a clear departure from the widely held assumption that the Ly6Cpos/neg status is the predominant defining feature of these macrophages with strong predictive power for their molecular signature/functions. A highly related issue is the polarization status of the Ly6Cpos/neg subsets, which have been assumed to correspond, as in other tissues, to M1 and M2 macrophages, respectively (13, 16, 23). We show that the damage-associated Ly6Cpos macrophages exhibited a specific inflammatory profile, independent of STAT1, which only partially overlapped with the broad M1 gene expression pattern and displayed a unique inflammatory signature during the early response to muscle injury (days 1–2), likely reporting sterile inflammation (see later). Similarly, Ly6Cneg macrophages were unsuccessfully linked to canonical M2(IL-4) macrophage. Although they preferentially expressed anti-inflammatory markers, their expression profile did not match with any of the in vitro activation status tested so far (3). Only partial overlap was observed with macrophages stimulated with fatty acids (3). These results indicate that in vivo macrophages cannot be restricted to the M1/M2 nomenclature during tissue injury/repair. Instead, we evidence four specific features that account for the high dynamic changes in gene expression observed in macrophages during muscle regeneration.
The first feature is early expression of acute-phase proteins by Ly6Cpos macrophages. Ly6Cpos macrophages entering the injured muscle highly expressed genes involved in inflammation, including acute-phase proteins, such as SAA3, S100A8/9, Lipocalin-2, and Haptoglobin (32). Both pathogen-associated and damage-associated molecular pattern molecules induce macrophage expression of S100A8 and S100A9 (33), which dimerize to form calprotectin, which, in turn, triggers SAA3 expression (34) by macrophages (35–37). Interestingly, macrophage-derived calprotectin is an early contributor of skeletal muscle myositis, which is characterized by a huge inflammatory infiltrate (38) and is also upregulated after exercise (39), suggesting that these macrophage-derived proteins are crucial for launching the regeneration/remodeling process in skeletal muscle. Lipocalin-2 and haptoglobin are involved in the regulation of iron metabolism, by binding small iron-carrying molecules (siderophores) and hemoglobin (before internalization by CD163 receptor), respectively. Iron capture protects against iron-mediated toxicity (40). This is in agreement with the described iron retention phenotype of M1 macrophages (41), which also has been described in early regenerating muscle (42). It has been shown that chelatable iron release from disrupted muscle tissue represents the major cause for the oxidative stress that aggravates muscle destruction (43). Thus, high expression of iron-scavenging molecules by Ly6Cpos macrophages may favor iron elimination to protect the tissue.
Second, metabolic change of macrophages precedes and sustains the resolution of inflammation. Although metabolism is foreseen as an important regulator of macrophage function, notably in diseases (44), few studies addressed specific metabolic pathways linked to specific macrophage inflammatory status. The shift from M1 to M2 macrophages is associated with a change in glucose metabolism, with a reshuffling from glycolysis to oxidative metabolism (13, 45, 46). Accordingly, we observed at the time of resolution of inflammation (between days 2 and 4) a downregulation of genes associated with glycolysis and glucose metabolism in both Ly6Cpos and Ly6Cneg macrophages. This was preceded by an early upregulation, between days 1 and 2, of the pathways associated with oxidative phosphorylation and TCA cycle, suggesting that macrophages first started to shift their metabolic profile before shifting their inflammatory status. Finally, we observed a strong upregulation of genes involved in glutamine metabolism, associated with oxidative metabolism, which is characteristic of M2(IL-4) macrophages in vitro (15), during the late steps of muscle regeneration (days 4–8). These results suggest that metabolic shift precedes inflammatory shift in macrophages and that a specific metabolic signature characterizes recovery macrophages during the repair phase.
The third feature is proliferation of Ly6Cneg macrophages, which likely explains the high numbers of Ly6Cneg macrophages present in the muscle later on (12). Kinetic analysis showed the molecular foundation that allowed Ly6Cneg macrophages to get prepared for division at day 1 and to actively divide at days 2–4 of regeneration. Similarly, macrophage proliferation occurs at the time of resolution of inflammation in peritonitis (47). Moreover, expression of M-CSF, which is essential to macrophage proliferation (47), was highly enhanced in macrophages between days 2 and 4 of muscle regeneration (Supplemental Table 1).
The fourth feature is expression of ECM components by Ly6Cneg macrophages during the healing phase. Ly6Cneg macrophages expressed mainly secretory proteins at days 4–8 of regeneration, which are involved in intercellular communication. A specific enrichment was observed for ECM and ECM-regulating proteins, including proteoglycans, matricellular proteins, and assembly proteins. Interestingly, most of these molecules are involved in skeletal muscle regeneration and some of them are expressed by macrophages during tissue repair (Supplemental Table 7). These results suggest that macrophages directly participate in the synthesis and assembly/organization of ECM constituents.
In conclusion, this study highlights an unexpected dynamic nature of macrophage subset gene signature during skeletal muscle regeneration. Early-arriving Ly6CposCX3CR1lo macrophages, which exhibit damage-associated, highly specialized molecular signature, sense and report tissue injury by producing acute-phase proteins, alarmins, and other factors. These macrophages early modify their cellular metabolism (downregulation of glycolysis pathway, upregulation of oxidative pathway), and later on their inflammatory phenotype, to become anti-inflammatory Ly6CnegCX3CR1hi macrophages. These latter actively divide while participating in a regeneration/healing process through the secretion of molecules involved in intercellular communications, notably ECM-related molecules, which are crucial for the recovery of muscle integrity. Ly6CnegCX3CR1hi macrophages fulfill the designation of restorative macrophages given their ability to produce proregenerative factors (e.g., insulin-like growth factor-1 ) and their contribution to muscle regeneration by acting on both myogenic (13) and fibroadipogenic (48) precursors.
Abs against Saa3 were kindly provided by Prof. Philipp E. Scherer (Department of Internal Medicine, University of Texas Southwestern Medical Center, Dallas, TX).
This work was supported by grants from the Agence Nationale de la Recherche Genopat In-A-Fib, 7th Framework Programme Endostem Grant Agreement 241440 (to B.C.), Hungarian Scientific Research Fund Grants OTKA K100196, K111941 and K116855, and Grant “NR-NET” ITN PITN-GA-2013-606806 from the European Union–7th Framework Programme PEOPLE-2013 (to L.N.). T.V. is a recipient of a Bolyai Fellowship from the Hungarian Academy of Sciences. L.N. and B.C. acknowledge cooperative funding from CNRS/Hungarian Academy of Sciences Cooperation Program 26119 and Campus France, Balaton Program 817297H.
The online version of this article contains supplemental material.
Abbreviations used in this article:
gene set enrichment analysis
principal component analysis
The authors have no financial conflicts of interest.