An anti-inflammatory activation sequence governs macrophage transcriptional dynamics during tissue injury in zebrafish

Macrophages are essential for tissue repair and regeneration. Yet, the molecular programs, as well as the timing of their activation during and after tissue injury are poorly defined. Using a high spatio-temporal resolution single cell analysis of macrophages coupled with live imaging after sensory hair cell death in zebrafish, we find that the same population of macrophages transitions through a sequence of three major anti-inflammatory activation states. Macrophages first show a signature of glucocorticoid activation, then IL-10 signaling and finally the induction of oxidative phosphorylation by IL-4/Polyamine signaling. Importantly, loss-of-function of glucocorticoid and IL-10 signaling shows that each step of the sequence is independently activated. Lastly, we show that IL-10 and IL-4 signaling act synergistically to promote synaptogenesis between hair cells and efferent neurons during regeneration. Our results show that macrophages, in addition to a switch from M1 to M2, sequentially and independently transition though three anti-inflammatory pathways in vivo during tissue injury in a regenerating organ.


Main Text:
Innate immune cells, and more particularly macrophages, are essential during vertebrate embryonic development, tissue repair and regeneration (1,2).Ablating macrophages during, or after injury of major organs in regenerating species blocks the regeneration process (3)(4)(5)(6).Yet, the sequence of signals that controls their activity and their dynamics at a high spatio-temporal resolution are still poorly defined.Mammals can regenerate organs and tissues without scarring during embryonic and early postnatal stages but loose this ability as adults.A recent study in a mouse model of spinal cord injury showed that the crucial difference between neonatal mice and adults resides in differences in the activation state of macrophages (7).Strikingly, transplanted young macrophages allows adult mice to fully regenerate their spine.This stresses the urgent need to obtain a better understanding of embryonic macrophage activation states to develop therapeutics for adult regeneration.
Macrophages exist in various molecular activation states.They are commonly classified as proinflammatory (M1) or anti-inflammatory (M2) depending on the type of cytokines/signals they get activated with (8).However, this classification is an oversimplification.M1 macrophages can be activated by several pro-inflammatory signals such as Interleukin1, Interferon-gamma, lipopolysaccharide (LPS), while M2 are triggered by anti-inflammatory Glucocorticoids, Interleukin10, Interleukin4/13 and more (9).In addition, each of these signals triggers distinct downstream gene regulatory networks depending on the context and can have distinct functions.
Nevertheless, a switch from M1 to M2 has been observed in different contexts of tissue injury (pathogen-induced-, sterile-or mixed injury).Two models have been proposed and observed in non-regenerative species: a "phenotypic switch", where the same population switches from proto anti-inflammatory (10)(11)(12)(13)(14)(15) or the "independent recruitment" of pro-and anti-inflammatory populations sequentially (16)(17)(18).Which model is at play during tissue regeneration is still poorly documented.A recent study in a model of tailfin regeneration in zebrafish demonstrated that the same macrophages can switch from pro-inflammatory (tnfa+) to anti-inflammatory (cxcr4b+) during the course of regeneration (19) favoring the phenotypic switch model.However, if cells can sequentially transition between several pro-or anti-inflammatory states during tissue injury is understudied mainly due to the lack of high spatio-temporal resolution analyses.Indeed, the majority of injury paradigms studied thus far occurs over many days, making it challenging to observe fast transitions between different macrophage activation states.Determining both the timing of the transitions as well as the genetic programs triggered by each activation state during tissue regeneration will be unvaluable to design targeted immunomodulatory therapies.
Here we took advantage of the rapid sensory hair cell (HC) regeneration that occurs in the zebrafish lateral line (20,21) to uncover the molecular "What" (signals and genetic program) and "When" (sequence of activation) that control macrophage activity and dynamics during injury of a regenerating organ in zebrafish.This regenerating lateral line sensory system is powerful for several reasons: (i) antibiotic/neomycin treatment induces rapid HC death within minutes (21); (ii) HC regeneration occurs within hours and the first new pair of regenerated HCs is detected five hours after neomycin treatment (22); and (iii) the optical clarity of the larvae allows us to follow the behavioral dynamics of macrophages using confocal microscopy at high spatio-temporal resolution.Together, these characteristics enabled us to interrogate the transcriptional and behavioral dynamics of macrophage activation at an unprecedented spatio-temporal resolution.

The same population of effector macrophages resolves inflammation within five hours in neuromasts
Tissue injury triggers an inflammatory response that involves the recruitment of macrophages to the injury site.These macrophages will be responsible for clearing cellular debris, remodeling the extracellular matrix and in some cases, provide signals to the tissue stem cells to start the repair process (23).We designated this population as 'effector macrophages' (24).It is not clear if effectors represent a single, or several macrophage populations.A recent study of muscle regeneration in zebrafish showed that two populations of macrophages were dynamically regulated after injury (25).One population reacted rapidly to tissue injury and left the damaged area within twenty-four hours, while a second population stayed in contact with the muscle stem cells for a longer period.Thus, the first step in understanding how effectors are molecularly regulated is to describe their dynamic after HC death.To follow both macrophages and HCs over time we study Tg(mpeg:GFP) and Tg(she:lckmScarletI) transgenic larvae that label macrophages and neuromast cells, respectively (Fig1A).Neuromast HCs are superficially located all along the larval body (Fig1A, movieS1).Treatment with neomycin for 30 min rapidly kills HCs by caspase-independent cell death (movieS2) (26).Macrophages start phagocytosing dead HCs as early as 15 min after the first cells die (movieS2-S3) (27).To describe the dynamics of effector macrophages from the time they enter the neuromast to when they leave during HC regeneration, we performed a macrophage recruitment assay.We treated the larvae with neomycin for 30 min and imaged both macrophages and neuromasts 1H, 3H, 5H and 7H after treatment (Fig1B, movieS4).Quantification of the number of macrophages surrounding and inside the neuromasts shows a rapid recruitment of effectors with a peak at 1H after neomycin (Fig1C-D).Subsequently, cells progressively return to homeostatic levels which they reach at 7H after neomycin treatment.Thus, there is a window of five hours when effector macrophages interact with neuromasts, which coincides with the appearance of the first regenerated HCs (22).
To decipher if effector macrophages represent a specific population, we performed time-lapse recordings of macrophages and neuromasts during neomycin treatment and homeostasis in a large area of the larval trunk to identify where the effectors reside prior to HC death (movieS5).
Quantification of the location of effector and non-effector macrophages before HC death shows that effectors are located closer to the neuromast than non-effectors (71µm+/-14µm vs 175µm+/-15µm, respectively) (Fig1E).In contrast to non-effectors, effectors show a rapid, directional migration toward the neuromasts in response to HC death (Fig1F).Interestingly, non-effector cells show an increase in non-directed cell velocity after neomycin treatment suggesting that HC death initially triggers a global injury response (Fig1G).
To assess if a single or multiple populations of effectors are recruited to neuromasts during the five hour window, we specifically photoconverted effector cells in neuromasts 1H after neomycin and quantified the ratio of photoconverted vs non-photoconverted macrophages 3H and 5H after HC death (Fig1H).Photoconverted cells represent an average of 98%+/-2% and 100% of the macrophages inside neuromasts at 3H and 5H after HC death, respectively, demonstrating that a single population of effectors is recruited during the five hour window (Fig1I).

High temporal resolution scRNA-seq identifies a population of effector macrophages
To molecularly characterize all macrophage populations, identify the effector population and characterize their activation sequence and underlying molecular program over the course of HC death, we performed a scRNA-seq time course of Tg(mpeg:GFP) transgenic larvae.We dissociated 600 5dpf larvae each during homeostasis, and 1H, 3H and 5H after a 30min neomycin treatment and FACsorted GFP+ cells for 10x Chromium genomics scRNA-seq (Fig2A).To determine the earliest activation of macrophages we also collected GFP+ cells immediately after a 15 min treatment with neomycin (Fig2A).We integrated these five timepoints using Seurat and performed UMAP dimensional reduction (Fig2B).We downsampled the number of cells per timepoint to fourteen thousand to avoid a bias based on over-representation of a specific timepoint (FigS1A).The seventy thousand cell atlas of all mpeg:GFP cells of 5 dpf larvae revealed that mpeg:GFP, in addition to macrophages, also labels several other immune cell types, some of which have also been recently observed in 4dpf larvae (28).We identified dendritic-like cells (DC-like) based on their expression of the master regulator flt3 (29), as well as spock3 and hepacam2 (Fig2B, FigS1B-C and TableS1).Previously described antigen presenting metaphocytes (30) are marked by cldnh, epcam and prox1a (Fig2B, FigS1B-C and TableS1).Two clusters of natural-killer like cells express their master regulator eomesa (31) (NK-like) and gata3 (32) (NK-like2) (Fig2B, FigS1B-C and TableS1).A neutrophil population is labeled by mpx and lyz (33), an unidentified population expresses skin (rbp4, sparc) and collagen markers (col1a2) as well as two unidentified small clusters labelled by kng1 and acta3b respectively (Fig2B, FigS1B-C and TableS1).Our analysis revealed an unexpected heterogeneity of the macrophage population consisting of eight clusters that cluster closely together (Fig2B).Proliferating macrophages form a cluster characterized by pcna, mki67 and tubb2b (Fig2B, FigS1B-C and TableS1).Another cluster that is heavily influenced by ribosome genes we called the 'translation' cluster (Fig2B, FigS1B-C and TableS1).Cells in the small 'stat1b' cluster show an Interferon signaling response signature (stat1b, isg15, cxcl20) (Fig2B, FigS1B-C and TableS1) that is also induced in response to an endemic picornavirus in zebrafish facilities (34).Another cluster is marked by genes classically upregulated in response to bacteria (35,36) that we named 'irg1/acod1' (irg1/acod1, hamp, mxc) (Fig2B, FigS1B-C and TableS1).In addition, we identified a cluster that did not show unique markers but is broadly labelled by f13a1b, junba and btg1 (called 'fa13a1b') (Fig2B, FigS1B-C and TableS1); a cluster that expresses markers of potentially immature microglia, such as mcamb, apoeb (37) and apoc1 (called 'mcamb') (Fig2B, FigS1B-C and TableS1); and two clusters that represent unidentified macrophage states or populations expressing runx3, cxcl19, cxcl8a (called 'runx3') and tspan10, slc43a3b, illr1 (called 'tspan10'), respectively (Fig2B, FigS1B-C and TableS1).
To characterize macrophage clusters that show transcriptional changes during the neomycin time course, we performed differential gene expression for each cluster at each timepoint.
Quantification of the numbers of up-and down-regulated genes during the time course shows that several macrophage clusters respond to neomycin treatment (Fig2C, D, H, L and FigS2A-B).To determine which macrophages enter the neuromasts and represent effector cells, we performed fluorescent in situ hybridization using the hybridization chain reaction (HCR-FISH (38)) with the macrophage cluster markers irg1/acod1, f13a1b, mcamb, tspan10, runx3, as well as the nonmacrophage NK-like (eomesa) and DC-like (hepacam2) clusters as negative controls (Fig2D-O, FigS2A-D'').HCR-FISH was performed on mpeg:GFP larvae 1H after neomycin, when all effector cells have migrated into the neuromasts.This screen demonstrated that 48% of the effectors are labelled with irg1/acod1 (Fig2D-G), 18% with f13a1b (Fig2H-K) and 34% with mcamb (fig2L-O).The tspan10 marker did not label any effector cells (FigS2A-A"), while we found that 3% of the cells were labeled with runx3 (FigS2 B-B").As expected, no effector cells were labelled with eomesa (NK-like cells) (FigS2 C-C") or hepacam2 (DC-like cells) (FigS2 D-D").Additionally, we performed a macrophage recruitment assay with two newly generated transgenic reporter lines that drive the expression of a red fluorescent protein under the promoter of irg1/acod1 and stat1b.These experiments confirmed that 'irg1/acod1' cells are indeed effector cells, whereas 'stat1b' cells do not enter neuromasts (FigS3A-D).Altogether we conclude that the effector population is composed of cells belonging to the 'irg1/acod1', 'f13a1b' and 'mcamb' clusters and focused our analysis on these cells as effector macrophages.
Likewise, Pathway and Gene Ontology (GO) analysis shows 'Oxidative phosphorylation' as the most enriched term (TableS2).Interestingly, manf, a gene required for retina repair and regeneration in fly and mouse (54) is strongly upregulated at the three-hour timepoint hinting toward a switch to a repair state of macrophages (Fig3A, S4A).In contrast to the earlier timepoints, very few genes are found specifically upregulated at the 5H timepoint (Fig3A, S4A).grn2, a progranulin growth factor regulator of the anti-inflammatory macrophage phenotype (55) is highly upregulated.Interestingly, in grn knock-out mice, muscle injury leads to a persistence of macrophages at the injury site suggesting a role for grn in regulating macrophage dynamics (56).
Altogether, this analysis reveals a linear sequence of macrophage anti-inflammatory activation immediately after HC death starting with Glucocorticoid signaling, followed by IL10 signaling, and lastly a combination of Polyamine and IL4 signaling, which induces oxidative phosphorylation at the transcriptional level.To our knowledge, this is the first report of an in vivo linear sequence of three major anti-inflammatory activation pathways after injury.This suggests that in addition to a transition from a pro-inflammatory M1 to an anti-inflammatory M2 state, the same macrophages transition through different anti-inflammatory states to potentially regulate their dynamics/activity.

Each anti-inflammatory state is independently activated
The discovery of this linear anti-inflammatory sequence of effector macrophage activation raises the interesting question if epistatic relationships between Glucocorticoid signaling and IL10 signaling and between IL10 signaling and IL4/Polyamine signaling exist.
To address this question, we first inhibited Glucocorticoid signaling during HC death using the GR inhibitor RU486.HRC-FISH for the GR target dusp1 confirmed that a 10µM treatment with RU486 was efficient in blocking GR activation (Fig4A).HCR-FISH for il10ra at the 15 min timepoint, as well as for the IL10 signaling target gene fgl2a at the 1H timepoint showed no downregulation of these genes in effector macrophages after RU486 treatment ( Fig4B-E).This demonstrates that GR activation is not required for IL10 activation in effector macrophages.
To test for a possible epistatic relationship between IL10 signaling and the following activation state (Polyamine + IL4 signaling), we generated a mutant for il10ra by deleting the promoter region, as well as the first exon using CRISPR/Cas9 (Fig4F).HCR-FISH for il10ra in homozygous mutants confirmed the absence of transcript (Fig4G-H).We also performed HCR-FISH for the target gene fgl2a at the 1H timepoint and found a strong downregulation in effector cells in the mutant (64%+/-7% in homozygous vs 21%+/-5% in heterozygous) (FigS6A-B).Next, we performed a scRNA-seq time course after neomycin treatment in homozygous and heterozygous il10ra larvae (FigS7A).Integration of these six datasets (sixty thousand cells) using Seurat and UMAP dimensional reduction shows that the effector clusters ('irg1/acod1', 'mcamb' and 'f13a1b') are conserved in mutant larvae (Fig4I-J, S7B-C, TableS3).Likewise, quantification of the expression levels and dynamics of il4r.1, and odc1 shows that they are unaffected in the mutants compared to the siblings (Fig4K).The subsequent activation of genes related to oxidative phosphorylation and manf 3H after neomycin is also unaffected (Fig4K and S8).Thus, the induction of oxidative phosphorylation by IL4/Polyamine signaling is independent of IL10 signaling activity.This important result reveals that, while a sequential induction of antiinflammatory pathways underlies macrophage activation during HCs regeneration, its components are independently activated (Fig4L).Therefore, activation of a single anti-inflammatory pathway is likely not sufficient to induce proper tissue regeneration and a sequential and independent activation of the three pathways might be required.

Discussion
Macrophage molecular activation regulates the dynamic and activity of these phagocytes and modulation of their activation states and their dynamic can have dramatic effects on tissue repair and regeneration (57).Here, using a high spatio-temporal resolution analysis of macrophage activation during HC regeneration, we provide evidence that the same population of macrophages is sequentially and independently activated by three major anti-inflammatory pathways.
It is now broadly documented that macrophages adopt a pro-inflammatory activation state immediately after injury (19,28,(58)(59)(60).The role for this pro-inflammatory phase is mainly to attract more macrophages to the injury site if the organ does not possess a resident population, or if the size of the injury requires more macrophages (61).This first phase must be followed by an anti-inflammatory phase to resolve the inflammation and ensure proper tissue repair.Our analysis shows that in the lateral line, a single population of tissue resident macrophages resolves HC death induced inflammation, favoring the phenotypic switch model.Furthermore, our data show a very short pro-inflammatory phase and a rapid transition to an anti-inflammatory state marked by the strong and systemic activation of the GR pathway.A possible explanation for this lack of a strong pro-inflammatory activation of macrophages after HC death is that the tissue resident population is sufficient to resolve inflammation and the recruitment of inflammatory macrophages is not required.This hypothesis is supported by our finding that the effector macrophages reside in immediate vicinity of the neuromasts during homeostasis and that after HC loss on average only three macrophages are detected in neuromasts.
The anti-inflammatory role of the GR pathway has been extensively documented (39,(62)(63)(64).It can both directly inhibit pro-inflammatory gene transcription by direct binding of the GR receptor to their enhancers/promoters or by tethering the pro-inflammatory activators AP-1 and Nfkb (64).
GR also triggers a rapid and large-scale chromatin unwinding, potentially creating a permissive environment for drastic changes in gene expression required for macrophage activity (65).
Therefore, the strong and transient activation of the GR pathway immediately as the first HCs start to die likely turns off the transcription of pro-inflammatory cytokines.
The short phase of GR activation is immediately followed by IL10 signaling activation and a subsequent transition to a IL4/Polyamine activation state that induces oxidative phosphorylation.
A recent study in mouse Bone Marrow Derived Macrophages (BMDM) showed that part of the IL10 anti-inflammatory function is to inhibit glycolysis while promoting oxidative phosphorylation after treatment with LPS, which mimics a bacterial infection (66).Our il10ra mutant analysis demonstrates that lack of IL10 signaling after HC death does not affect the induction of oxidative phosphorylation.Co-stimulation of mouse BMDM with LPS and IL10 also leads to the upregulation of the IL4 receptor (67) whereas the loss of IL10 in zebrafish does not affect the induction of il4r.1.These discrepancies suggest that IL10 and IL4 signaling are differently regulated in response to bacterial infection versus tissue injury.
The IL4/OXPHOS state is characteristic for wound healing macrophages and responsible for the last step of inflammation resolution (2,(68)(69)(70).In addition, the induction of the pro-repair gene manf, which is required for retina regeneration (54), suggests that at the 3H timepoint macrophages transition to a wound-healing state.Altogether, our in vivo data demonstrate the sequential and independent transition of effector macrophages through three major anti-inflammatory states.This finding has important implications for the design of targeted immunomodulatory therapies.

Figure 3 .
Figure 3. Transcriptional dynamics reveal an anti-inflammatory activation sequence in

Figure 4 .
Figure 4.Each anti-inflammatory state is independently activated.(A) Representative confocal images (maximum projection of a 30µm z-stack) of an HCR-FISH for dusp1 in a neuromast (10 larvae per condition and 30 neuromasts total).(B, C) Representative confocal images (projection of a 30µm z-stack) of HCR-FISH within the effector macrophages for (B) il10ra and (C) fgl2a (12 larvae per conditions and 36 neuromasts total) (D, E) Quantifications of the ratio between HCR+ cells and GFP+ effector macrophages for (D) il10ra and (E) fgl2a.(F) Schematics representing the CRISPR/Cas9 design used to generate the il10ra mutant zebrafish.(G) Representative confocal images (projection of a 30µm z-stack) of HCR-FISH within the effector macrophages for il10ra in the il10ra homozygous mutant.(H) Quantifications of GFP+ effector cells and HCR in effector cells for il10ra in the il10ra homozygous mutant.Each dot represents the number of macrophages per neuromast (33 neuromasts from 11 larvae).P values represent non-parametric Student t-test.(I) Integrated UMAP of the six datasets for the il10ra mutant.Cluster names are labelled on the UMAP.(J) Split UMAP per condition (il10ra+/-and il10ra-/-).(K) Line plots representing the average expression for each timepoint between the il10ra mutant (cyan) and the sibling (red) from the scRNA-seq integrated dataset.(L) Model of independent activation of the three anti-inflammatory pathways in effector macrophages during the HC regeneration time course.Resting macrophages are represented in cyan.Purple arrows represent the independent induction of each macrophage activation state.