Macrophages Switch to an Osteo‐Modulatory Profile Upon RANKL Induction in a Medaka (Oryzias latipes) Osteoporosis Model

ABSTRACT In mammals, osteoclasts differentiate from macrophages in the monocyte lineage. Although many factors driving osteoclast formation are known, the detailed processes underlying precursor recruitment, differentiation, and interaction of macrophages with other cell types involved in bone remodeling are poorly understood. Using live imaging in a transgenic medaka osteoporosis model, where ectopic osteoclasts are induced by RANKL expression, we show that a subset of macrophages is recruited to bone matrix to physically interact with bone‐forming osteoblast progenitors. These macrophages subsequently differentiate into cathepsin K‐ (ctsk‐) positive osteoclasts. One day later, other macrophages are recruited to clear dying osteoclasts from resorbed bone by phagocytosis. To better understand the molecular changes underlying these dynamic processes, we performed transcriptome profiling of activated macrophages upon RANKL induction. This revealed an upregulation of several bone‐related transcripts. Besides osteoclast markers, we unexpectedly also found expression of osteoblast‐promoting signals in activated macrophages, suggesting a possible non‐cell autonomous role in osteogenesis. Finally, we show that macrophage differentiation into osteoclasts is dependent on inflammatory signals. Medaka deficient for TNFα or treated with the TNFα‐inhibitor pentoxifylline exhibited impaired macrophage recruitment and osteoclast differentiation. These results show the involvement of inflammatory signals and the dynamics of a distinct subset of macrophages during osteoclast formation. © 2020 The Authors. JBMR Plus published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research.


Introduction
S ince their initial description in 1883, (1) macrophages have been intensively studied for their involvement in various physiological processes including embryogenesis, infection, tissue regeneration, and bone homeostasis (reviewed in Wynn and colleagues (2) ). Bone marrow-derived monocytes and macrophages have been identified as precursors of cathepsin K-(ctsk-) positive osteoclasts, which digest mineralized matrix and thereby function in bone resorption. (3,4) On the other hand, osteal macrophages, a subpopulation of specialized bonehoming macrophages, were shown to have essential roles in bone formation during regeneration and fracture healing. (5,6) This cell type also modulates cytokine expression and performs clearance of apoptotic cells in a process known as efferocytosis. (7) During an inflammation episode, the activation of macrophages often includes dynamic regulation of proinflammatory cytokines, especially IL1β and TNFα. These cytokines trigger an immune response and recruit other immune cells to inflammation sites. (7,8) TNFα, a member of the tumor necrosis factor superfamily, is produced by different cell types such as monocytes, neutrophils, and adipocytes. (9,10) It plays important roles in cell proliferation, wound healing, and cancer progression. (8,11) TNFα is also found predominantly expressed in activated macrophages in inflammatory diseases such as rheumatoid arthritis. (12) Inhibition of this cytokine has been an effective therapy to alleviate bone inflammatory symptoms. (13) Patients with osteoporosis, who also have elevated serum levels of IL1β and TNFα, often experience severe bone loss and fractures that result from hyperactivity of osteoclasts. (14) Several studies have shown that in the presence of macrophage colony stimulating factor (M-CSF), TNFα and RANKL synergistically or independently stimulate osteoclast formation via their respective receptors. (15)(16)(17) However, how these cytokines control the dynamics of osteoclast precursor recruitment and differentiation in vivo remains unclear, mostly because commonly used animal models have limited accessibility for live imaging.
In this study, we used a transgenic medaka osteoporosis model, where excessive osteoclast formation is triggered by inducible RANKL expression. (18) Upon RANKL induction, macrophages migrate to the vertebral column where they eventually mature into ctsk-positive osteoclasts. (19) In the present study, we describe the dynamics of macrophage migration and their close interaction with bone-lining osteoblasts. We report that tnfa is upregulated in activated macrophages during recruitment and that deletion of tnfa or treatment with the tnfa transcription inhibitor pentoxifylline (PTX) blocks macrophage recruitment and osteoclast differentiation. Transcriptome profiling further revealed that the cytokine receptor gene il22ra2b was downregulated when macrophages differentiate into osteoclasts. Together, these findings suggest that inflammatory signals are initially required for macrophage recruitment towards bone matrix under osteoporotic conditions. Subsequently, these signals are downregulated to allow differentiation of recruited cells into osteoclasts. Interestingly, we also find that upon RANKL induction, macrophages upregulate genes that are known to control chondrocyte and osteoblast differentiation in a non-cell autonomous manner. This suggests an involvement of activated macrophages in bone remodeling and homeostasis under pathological conditions.

Materials and Methods
All experiments were performed according to protocols approved by the Institutional Animal Care and Use Committee (IACUC) of the National University of Singapore (NUS; protocol numbers R14-293, R18-0562, and BR15-0119).

Transgenic and mutant medaka lines
Transgenic rankl:HSE:cfp/ctsk:GFP medaka embryos were grown and subjected to RANKL induction by heat-shock as previously described. (18) Generation of the mpeg1:mCherry-F macrophage reporter line was described before. (19) Briefly, a 2.1-kb sequence of the mpeg1 promoter including the endogenous ATG was amplified from genomic DNA of WT medaka using primers mpeg1.2F and mpeg1.2R (Supplementary Table S1). The PCR product was digested with NotI and BamHI and then ligated inframe to a farnesylated mCherry in a pI-SceI plasmid. The plasmid was microinjected into one-cell stage medaka embryos together with meganuclease I-SceI enzyme as previously described. (20) Injected embryos were screened for mCherry-F transgene expression in macrophages, and stable lines were established and verified by in situ hybridization. For the generation of tnfa mutants, guide RNAs (gRNAs) targeting exons 1, 3, and 4 of tnfa were designed using CCTop (21) and synthesized by IDT (Singapore; Fig. S8; Supplementary Table S1). The three gRNAs were mixed with tracrRNA and HiFi Cas9 protein (IDT) and injected into one-cell stage medaka embryos. Injected fish and their offspring were genotyped using primers listed in Supplementary Table S1, and stable mutant lines were established.

Macrophage depletion and drug treatment
For macrophage depletion, rankl:HSE:cfp/mpeg1:mCherry-F/ctsk: GFP embryos at 9 days post fertilization (dpf) were injected intravenously with 15-nL lipo-clodronate (5 mg/mL) (Lipo-Clo) or Lipo-PBS (Encapsula NanoSciences, Brentwood, TN, USA). Macrophage-ablated embryos were selected at 10 dpf based on the reduced fluorescence signal of mCherry-F reporter. For inhibition of tnfa expression, RANKL-induced embryos after heat-shock were immediately transferred to fish medium (30% Danieau's solution containing 19 In situ hybridization Sense and antisense riboprobes spanning 790 nucleotides of the mpeg1 cDNA (nt1029-nt1818; ENSORLT00000006101.2; the Ensembl database project, http://www.ensembl.org/) were synthesized using a DIG RNA labeling kit (Roche Diagnostics, Mannheim, Germany). In situ hybridization was performed as previously described. (18) Bone staining Alizarin Red staining of mineralized bone matrix in fixed embryos was performed as previously described. (22) Briefly, larvae at 3 days post heat-shock (dphs) were fixed with 4% paraformaldehyde, washed three times with PBS solution plus Tween-20 (PBST), incubated for 15 min in 0.5% potassium hydroxide (KOH), then transferred to 0.001% Alizarin Red solution in 0.5% KOH for at least 4 hours with agitation. Samples were washed for 2 hours with 0.5% KOH, followed by depigmentation with 2% H 2 O 2 in 0.5% KOH for 1 hour. Embryos were washed two times with 0.5% KOH and mounted in 100% glycerol for imaging. For live bone staining, embryos were incubated in 0.01% calcein (Sigma-Aldrich) and kept in the dark for at least 1 hour at 30 C. Embryos were washed twice with fish medium and mounted in low-melting agarose for imaging.
Imaging of live and fixed samples For live imaging, embryos were anesthetized in 0.016% tricaine (MS-222; Sigma-Aldrich) and embedded in 1.2% low melting agarose (Bio-Rad, USA) in 35 mm glass-bottom dishes (Eppendorf, Hamburg, Germany), and covered with 2-mL fish medium containing tricaine. A Nikon SMZ18 epi-fluorescence microscope was used for capturing whole-embryo images. Z-stack images and time-lapse movies were recorded using a Nikon FV3000 confocal microscope (Nikon, Minato City, Tokyo, Japan) equipped with a ×30/1.05, WD 0.8 silicon oil objective. Time-lapse files were processed using Imaris (Bitplane AG, Zurich, Switzerland) and Fiji (ImageJ; NIH, Bethesda, MD, USA; https://imagej.nih.gov/ij/) software to generate movies. Z-stack images were processed into maximum intensity projections and channels merged using Fiji. Alizarin Red-stained bone samples were imaged using a Nikon Eclipse 90i upright microscope equipped with NIS-Elements version BR 3.0.

Cell quantification and statistics
Fluorescent images of whole embryos and confocal z-stacks were used for the quantification of macrophage and osteoclast numbers as previously described (23) with slight modifications. Using Fiji, fluorescent images were converted to greyscale, thresholded, and the area of a cell (averaged from 15 cells from three embryos), as well as the total area of cells in a selected region of interest, were derived. Cell numbers were determined by dividing the total area of cells with the average cell area. For statistical analysis, Student's t test (two-tailed, unpaired) was performed using GraphPad Prism 8.0 (GraphPad Software, La Jolla, CA, USA).
Fluorescence-activated cell sorting, RNA sequencing, and bioinformatics analysis FACS and RNA extraction were performed using previously described protocols (24) with some modifications. RNA library construction, Illumina sequencing, and bioinformatics analysis were performed by Novogene (Singapore) following company procedures. Briefly, RANKL expression was induced in rankl:HSE: cfp/mpeg1:mCherry-F/ctsk:GFP transgenic medaka embryos at 9 dpf. At 1 day post heat shock (dphs), 23 larvae per sample were dissociated in 500 μL of 0.25% trypsin EDTA (HyClone; HyClone Laboratories, South Logan, UT, USA) plus collagenase (0.2 mg/ mL) at 30 C and pipetted every 10 min for 40 min. The digestion was stopped by adding 50-μL sterile-filtered FBS. Cells were washed with PBS supplemented with 2% FBS and resuspended in 400-μL Leibovitz's L-15 medium (Gibco; Thermo Fisher Scientific) for FACS using a BD FACSAria II platform (BD Biosciences, San Jose, CA, USA). Fluorescent macrophages and osteoclasts were separated and collected into 400-μL TRIzol (Thermo Fisher Scientific). RNA extraction was done from three independent biological replicates using PureLink RNA Micro kit (Thermo Fisher Scientific). The RNA quality was determined with an Agilent Bioanalyzer 2100 using Agilent RNA 6000 Pico kit (Agilent Technologies, Santa Clara, CA, USA). Samples with RNA integrity number (RIN) values >7 were submitted for sequencing. Clean reads were aligned to the reference genome for Japanese medaka HdrR, version ASM223467v1 (http://ensembl.org/) using TopHat v2.0.12. Differential expression analysis of two biological replicates per condition was performed using Deseq R package (1.18.0). P values were adjusted using Benjamini and Hochberg's approach and genes with P values <0.05 were assigned as differentially expressed. Gene ontology (GO) enrichment of differentially expressed genes was analyzed using GOseq R package and only GO terms with a P value <0.05 were considered significantly enriched. ShinyGO V0.61 was used to analyze networks of functional groups of up-and downregulated genes. Input data were matched to human genes, the P-value cutoff was 0.05, and the 30 most significant terms were used (http:// bioinformatics.sdstate.edu/go/).

Phylogeny and synteny analysis
Amino acid sequences were retrieved from the National Center for Biotechnology Information (NCBI; www.ncbi.nlm.nih.gov/) and ENSEMBL (www.ensembl.org/Multi/Tools/Blast) databases using BLAST searches with default parameters. Sequences were aligned using Clustal W (25) and trees were constructed using the maximum likelihood method in MEGA version X (26) with the JTT model for aa substitutions and 1000 bootstrap iterations. Conserved synteny analysis was done using GENOMICUS version 100.01 (https://www.genomicus.biologie.ens.fr/genomicus-100. 01/cgi-bin/search.pl). (27) Quantitative polymerase chain reaction For whole embryos, RNA was isolated from 10 embryos per sample using NucleoSpin RNA kit (Macherey-Nagel, Duren, Germany), and total mRNA was reverse transcribed using the RevertAid First Strand cDNA Synthesis kit (Thermo Fisher Scientific). For RNA isolation from FAC-sorted macrophages and osteoclasts, the NucleoSpin RNA XS kit (Macherey-Nagel) was used. RNA isolated from three independent biological samples was converted into cDNA and preamplified using Fluidigm Reverse Transcription Master Mix and Preamp Master Mix, respectively (Fluidigm, South San Francisco, CA, USA). All steps were performed following the manufacturers' standard protocols. PowerUp SYBR Green Master Mix (Applied Biosystems, Foster City, CA, USA) was used for qPCR conducted in a CFX96 Touch system (Bio-Rad Laboratories, Hercules, CA, USA). Data analysis was performed using Bio-Rad's CFX Maestro 1.0 software with β-actin as the loading control for normalization. Two-tailed Student's t tests were performed for statistical analysis.

RANKL triggers macrophage recruitment, proliferation, and differentiation at bone matrix
To study the dynamic behavior of macrophages by live imaging in medaka, we generated a reporter line expressing farnesylated mCherry under control of the medaka mpeg1 promoter (Fig. 1A). At 10 dpf, mpeg1:mCherry-F positive macrophages were found throughout the body with increased numbers in heart, liver, aorta-gonad-mesonephros (AGM) and caudal fin (Fig. 1B). The mpeg1 transgenic line faithfully recapitulated endogenous mpeg1 transcription as confirmed by RNA in situ hybridization ( Fig. 1C-G). Next, mpeg1:mCherry-F reporter fish were crossed with rankl:HSE:cfp/ctsk:GFP transgenic fish, which harbor a RANKL transgene under control of a bidirectional heat-shock promoter that also drives cyan fluorescent protein (CFP) expression, and express green fluorescent protein (GFP) in osteoclasts under control of the ctsk promoter (for details, see To and colleagues (18) ). Confocal time-lapse imaging showed that in the absence of ectopic RANKL expression (-RANKL), the majority of mpeg1: mCherry-F-positive macrophages reside within the AGM with only few migratory cells found in other regions of the body, including the vertebral column (Supplementary Fig. S1A; Movie S1). Shortly after RANKL induction (4 hours post heat-shock [hphs]), mpeg1-positive macrophages started to accumulate in the vertebral column ( Fig. 2A,B). Time-lapse analysis revealed that the majority of mpeg1 cells in the vertebral column originated from the AGM, but also cells from other regions were recruited ( Supplementary Fig. S1). Macrophages were almost exclusively found at the mineralized matrix of vertebral bodies, but not in the nonmineralized intervertebral discs ( Starting from 35 hphs, the first osteoclasts started to undergo cell death and became cleared by newly recruited phagocytic macrophages (Supplementary Fig. S2; Movies S3 and S4). Quantification of mpeg1:mCherry-F and ctsk:GFP cells showed high numbers of recruited macrophages and ectopic osteoclasts along the vertebral column of RANKL-induced embryos compared with controls ( Supplementary Fig. S3). Osteoclast numbers gradually declined after 2 dphs, while new macrophages continued to migrate towards the vertebral column over the course of 1 week (data not shown). The total number of macrophages increased partly by steady-state hematopoiesis, but also because of cell proliferation that occurred both in the AGM and around the vertebral column (Supplementary Fig. S3; Movie S7). Our findings suggest that in medaka ectopic RANKL expression triggers the activation of macrophages, stimulates their recruitment towards mineralized matrix in a directed manner, and promotes their differentiation into osteoclasts.

Macrophages reduce their dynamics during osteoclast maturation
To quantitate the dynamics of macrophages during osteoclast differentiation, we analyzed time-lapse movies of RANKLinduced macrophages and osteoclasts recorded before and after differentiation. In an early phase at 4 hphs, macrophages in RANKL-induced embryos exhibited higher speed and displacement when compared with -RANKL controls. A more angular and less spherical cell morphology was observed in +RANKL macrophages, indicating their motile state ( Supplementary  Fig. S1A). However, at 34 hphs, the motility of differentiating macrophages became reduced as they started to express ctsk: GFP, while undifferentiated macrophages remained highly mobile (Supplementary Fig. S1B; Movie S4). This suggests that the reduction of macrophage dynamics is a prerequisite for  osteoclast differentiation, possibly to allow tighter attachment of forming osteoclasts to bone matrix for subsequent cell fusion to generate multinucleated osteoclasts.

Depletion of macrophages prevents osteoclast formation
As in mammals (3) , medaka macrophages also differentiate into osteoclasts. We next determined whether osteoclasts also form in the absence of macrophages, indicating possible alternative cellular sources. We depleted macrophages by Lipo-Clo injection into medaka embryos at 9 dpf and induced RANKL expression by heat-shock at 10 dpf. In the absence of RANKL induction, Lipo-Clo treatment did not cause any obvious changes to bone development ( Fig. 3A-D). After RANKL induction, at 1 and 2 dphs, Lipo-PBS control-injected embryos showed abundant macrophage recruitment and osteoclast formation along the vertebral column as expected (Fig. 3C-F). In contrast, Lipo-Clo treatment resulted in a 75% ablation of macrophages ( Fig. 3C-E). Only a few macrophages were recruited to the vertebral column, and osteoclast formation was strongly impaired (Fig. 3C-F; Supplementary Movies S5 and S6). The effect of macrophage depletion on bone protection was then assessed using Alizarin Red staining at 3 dphs. After RANKL induction, the bone matrix of Lipo-Clo-injected embryos was efficiently protected with only few minor defects, which was in stark contrast to the severe lesions formed along the vertebral columns of Lipo-PBS-injected embryos (Fig. 3D). This observation indicates that the integrity of a macrophage population is important for osteoclast formation.

Transcriptome profiling reveals similarities between medaka and human osteoclasts
The behavior and dynamics of macrophages change rapidly during differentiation into osteoclasts (Supplementary Movies S2-S4). To identify molecular signatures underlying these changes, we performed RNAseq analysis and compared transcriptome profiles of FACS-purified macrophages (mpeg1 + / ctsk − cells) with those of osteoclasts (mpeg + /ctsk + doublepositive cells) at 1 dphs. Analysis of differentially expressed genes (DEGs) revealed gene sets that are similarly regulated in human and mouse osteoclastogenesis. Conventional markers of osteoclastogenesis in mammals such as trap, ctsk, siglec15, nfatc1, tgfb1, and dap12 were highly upregulated in mpeg + /ctsk + double-positive medaka cells, similar to the situation in human and mouse osteoclasts (Table 1). Our RNAseq datasets also indicated that genes encoding the medaka chemokine/cytokine receptors Cx3cr1, Il22ra2a, and Il22ra2b were downregulated in medaka osteoclasts, but significance could only be validated by qPCR analysis for il22ra2b (Supplementary Figs. S5-S7; Table 1). The downregulation of cytokine receptor genes in our medaka model suggests a possible reduction of the motility of RANKLinduced macrophages as they differentiate into osteoclasts.

RANKL-induced macrophages acquire a bone-remodeling signature
Our DEG analysis identified more than 1,000 genes that were either commonly or uniquely expressed in macrophages before and after differentiation into osteoclasts (Fig. 4A). Of the commonly expressed genes, 88 genes were significantly upregulated and 39 genes downregulated with adjusted P values of <0.05 (Fig. 4A,C). A KEGG analysis of DEGs revealed an enrichment of the terms phagosome, focal adhesion, extracellular membranereceptor interaction, and cytokine-cytokine receptor interaction (Fig. 4B). Apart from genes for which the corresponding human orthologs have important functions in osteoclastogenesis, such as fes, sh2dp1, socs1b, tspan5, and mcoln3a ( Fig. 4D; Table 1), we observed other highly regulated genes with so far uncharacterized functions in osteoclastogenesis. This includes upregulation of emilin2a, which is known to block cancer cell proliferation; (29) and sfrp2, which controls normal osteoblast differentiation and exerts an antiapoptotic effect. (30,31) Among downregulated genes, we identified gimap1-like, which is essential for development of T and B lymphocytes; rad18, which is a key player in controlling DNA damage tolerance (32,33) ; and ogdh, which is involved in glucose oxidation and cancer cell viability (34,35) (Fig. 4D). Next, we performed a GO analysis using Shi-nyGO v0.61. Individual lists of up-and downregulated genes were converted into human gene IDs and UniProtKB IDs using BioMart (Ensembl.org). UniProtKB IDs were later used as input for the analysis.
The analysis of upregulated genes surprisingly showed a strong enrichment of pathways involved in chondrocyte and cartilage development, bone remodeling, and osteoblast differentiation. This included the collagen-coding genes col6a2, col12a1, as well as the secreted osteoblast differentiation inducers tgfb1 and sfrp2 ( Supplementary Fig. S4, Table 2). (36) Consistent also with earlier studies that reported expression of collagen genes in macrophages, (37)(38)(39) qPCR analysis confirmed the upregulation of col6a2 and col12a1, as well as integrin subunit beta 3 (itgb3) in RANKL-induced macrophages and validated our RNAseq analysis. Interestingly, also genes encoding the respective complexing subunits, ie, col6a1, col6a3, itga5, and itga2.2, were upregulated (Fig. S5B).
A larger fraction of downregulated genes was immunerelated and is known to be involved in the activation of the immune response of macrophages ( Supplementary Fig. S4A). Our data suggest that macrophages are highly dynamic in regulating their gene expression, not only to favor bone-resorbing activity, but also potentially to promote bone formation by secreting factors that trigger osteoblast differentiation.
Macrophage recruitment and differentiation are impaired in tnfa mutants Treatment of mouse bone marrow-derived macrophages with the proinflammatory cytokine TNFα in the presence of M-CSF and RANKL stimulates osteoclast differentiation in vitro. (15)(16)(17) We therefore investigated the requirement of this proinflammatory cytokine for macrophage recruitment and osteoclast differentiation in vivo in the medaka osteoporosis model. First, we performed quantitative PCR analysis to test whether tnfa transcription is affected by heat-shock induction and RANKL overexpression. Compared with non-heat shocked controls, heat shocked −RANKL embryos showed a significant reduction of tnfa expression at 1, 2, and 3 dphs (Supplementary Fig. S8A). This is consistent with a temperature-induced reduction of Tnfa transcription in mouse macrophages. (40) Importantly, however, RANKL induction (+RANKL) resulted in a significant upregulation of tnfa transcription suggesting that tnfa is involved in the osteoporotic response. To test this, we generated a tnfa medaka mutant using CRISPR/Cas9 ( Supplementary Fig. S8B-D). Similar to Tnfa mouse mutants, which show normal development, (41) stable CRISPR/Cas9 medaka tnfa mutants were viable and developed normally with no obvious alterations (data not shown).   (78) LOC101163603 C-C chemokine receptor type 1 341 0 inf 0.020 Osteoclast recruitment, cell motility (79) socs1b Suppressor of cytokine signaling 1b 348 1 −9.4

0.010
Osteoclast formation (85) coro1a Coronin, Actin binding protein, 1A  Mutants were crossed with rankl:HSE:cfp/mpeg1:mCherry-F/ctsk: GFP transgenic fish, and macrophage behavior and osteoclast formation were tracked by live imaging as described above. In the absence of RANKL induction (-RANKL), tnfa mutants exhibited normal distribution and density of macrophages when compared with nonmutant siblings (Fig. 5A-C). Upon RANKL induction (+RANKL), tnfa −/− mutant macrophages were recruited towards the vertebral column at 1 dphs, similar to WT macrophages. Also, the total number of macrophages was initially similar in both WT and mutant embryos at 1 dphs. At 2 dphs, however, both total and recruited macrophage numbers were significantly reduced in homozygous mutants compared with that of WT siblings (Fig. 5B,C). Furthermore, tnfa −/− mutants exhibited significantly reduced osteoclast numbers, while osteoclasts were abundant in RANKL-induced nonmutant siblings, (Fig. 5A,D). Next, to investigate whether TNFα inhibition is sufficient to protect bone matrix from osteoporotic insult, RANKLinduced larvae were stained with calcein at 3 dphs to visualize mineralized matrix in the vertebral column. Upon RANKL induction, heterozygous tnfa +/− carriers showed similar bone defects as nonmutant siblings, which included bone loss in the neural arches and large lesions in the vertebral centra (Fig. 5E). Homozygous tnfa −/− embryos, in contrast, had fewer and smaller lesions in the centra indicative of improved bone integrity. This suggests that TNFα is required for macrophage activation and osteoclast formation after RANKL induction in vivo.
To confirm this, we finally treated nonmutant embryos with the chemical inhibitor PTX, which inhibits TNFα synthesis. Importantly, in the absence of ectopic RANKL, PTX treatment did not cause any obvious changes to the distribution and density of macrophages (Fig. 6A). Moreover, upon RANKL induction, macrophages were recruited to the vertebral column to a similar extent in DMSO-and PTX-treated embryos, as quantitated at 1 and 2 dphs (Fig. 6B,C; white arrowheads). The total number of macrophages was slightly lower in PTX-treated embryos at 1 dphs but increased to control levels at 2 dphs (Fig. 6D). Thus, other than in tnfa −/− mutants, macrophage recruitment was not affected by PTX, which we attribute to a lower efficacy of the drug compared with the genetic deletion. Importantly, however, osteoclast differentiation was strongly impaired in PTX-treated embryos when compared with DMSO controls at both 1 and 2 dphs (Fig. 6B,E, green arrowheads). We then analyzed the resorptive activity of RANKL-induced osteoclasts in control and PTX-treated embryos using Alizarin Red staining. In the absence of ectopic RANKL, both PTX-and DMSO-treated embryos showed normal mineralization with intact neural arches and vertebral bodies ( Supplementary Fig. S9). Upon RANKL induction, DMSO control embryos exhibited resorption of neural arches and severe lesions in the centra. In contrast, PTX treatment resulted in considerable bone protection (Supplementary Fig. S9). Together, our data show that in medaka TNFα is required for macrophage activation and osteoclast formation.

Discussion
Small teleost fish, such as zebrafish and medaka, are widely used to model human bone diseases given their unique experimental advantages, as well as genetic and cellular similarities of medaka and human bone. (42) In the present study, we used live imaging Integrin subunit beta 3 0 3016 inf a 0.000 Transmembrane Corticalization for bone growth (87) col6a2 Collagen alpha-2(VI) chain 0 880 inf 0.000 Secreted Chondrocyte proliferation and cartilage generation (88) foxc1 Forkhead box C1-A 0 362 inf 0.007 Nuclear Osteoprogenitor proliferation; osteogenesis (89) sfrp2 Secreted frizzled related protein 2 3 854 8 0.000 Secreted Osteogenic differentiation (90) cthrc1 Collagen triple helix repeat containing 1 90 6684 6.21 0.000 Secreted Osteoblast proliferation, bone cell coupling (91) myoc Myocilin 61 1070 4.14 0.009 Secreted Osteoblast differentiation; bone remodeling (92) col12a1  to demonstrate that in medaka, similar to mammals, (3) osteoclasts are derived from the monocyte/macrophage lineage in RANKL-induced conditions. Our macrophage depletion data suggest that in medaka no alternative cellular sources exist for osteoclasts, and that the formation of osteoclasts requires a high integrity of the macrophage population. Transcriptome profiling of medaka macrophages and osteoclasts identified sets of genes that are similarly regulated in human and mouse osteoclastogenesis. It also revealed the induction of genes in differentiated macrophages that are known to be involved in osteoblast formation and bone remodeling. We show that TNFα is essential for macrophage activation and osteoclast differentiation. Together, these results strengthen the medaka as a unique in vivo model accessible to live imaging and suitable for osteoporosis-related studies.

Targeted migration and bone cell interaction of macrophages in medaka
Macrophages express membrane-bound receptors that are needed for their recruitment to target sites in response to pathogens or instructive signals coming from other cell types. After RANKL induction in medaka, macrophages are recruited towards bone matrix, where they interact with osteoblasts and osteoblast progenitors and differentiate into osteoclasts (Fig. 2). Importantly, although massively recruited to the vertebral column, macrophages only accumulate at mineralized matrix of vertebral bodies, but not at the nonmineralized intervertebral discs. This suggests that signals triggering macrophage recruitment could be released either by resident cells lining the bone surface, ie, osteoblasts and osteoblast progenitors, or the matrix itself. Such signals appear to be absent from intervertebral discs or alternatively are distinct from those that induce chronic macrophage recruitment seen in intervertebral disc-degeneration diseases. (43) Previous studies indicated that osteoclasts can form from mouse spleen or bone marrow cells in the absence of bone matrix when cocultured with osteoblasts and supplemented with calcitriol (1 α,25(OH)2 vitamin D3). (44) This suggests that ossified matrix is dispensable for osteoclast formation and that osteoblasts instead produce the necessary differentiation signals. However, whether osteoblasts also release signals to recruit macrophages is still unknown. In mammals, osteoblasts express RANKL, which is important for osteoclast differentiation. In the medaka model, however, heat-shock induced RANKL is ubiquitous, yet only mineralized tissue attracts macrophages. This suggests that ectopic RANKL potentially acts directly on osteoblasts to trigger a release of attractants needed for macrophage recruitment. Consistent with this idea, we recently reported that medaka osteoblasts express the chemokine ligand Cxcl9l, which controls migration of macrophages towards bone matrix. (19) Future characterization of osteoblasts is likely to identify additional factors controlling macrophage recruitment.
Macrophages are the exclusive source for osteoclasts in medaka Macrophages are highly dynamic and have a variety of functions ranging from the clearance of foreign effectors to mediating communication across different cell types under developmental and pathological conditions. (45,46) By time-lapse analysis, we found that all ectopically induced ctsk-expressing osteoclasts in the vertebral column originate from recruited mpeg1 macrophages, suggesting that these macrophages are the exclusive cellular source for osteoclasts. Consistently, the transient depletion of mpeg1 cells by Lipo-Clo significantly reduced osteoclast numbers upon RANKL induction (Fig. 3). The few remaining osteoclasts that escaped Lipo-Clo inhibition also originated from macrophages, but not any other cell type. These results appear in contrast to earlier studies that reported persistent osteoclast formation in macrophage-depleted mice. (47) This difference may imply that mice potentially have osteoclast precursors other than macrophages, and that these precursors are absent in medaka. The other possibility, however, is that the macrophage ablation in mice obtained by a tamoxifen-inducible Cre-lox apoptosis system was incomplete, allowing remaining macrophages to differentiate into osteoclasts, similar to our findings in medaka. Consistent with our findings, Csf1r mouse mutants exhibit a significant reduction of macrophage numbers in different tissues and also a decrease in osteoclast formation. (48,49) Thus, other more efficient ablation techniques are needed to confirm that remaining osteoclasts in mice are differentiated from nondepleted macrophages or other cellular sources. In medaka, however, our live-imaging approach confirmed that macrophages are the exclusive source for RANKL-induced osteoclasts.
TNFα is critical for macrophage recruitment and differentiation Earlier cell culture studies showed that TNFα induces osteoclast formation from mouse bone marrow-derived macrophages. (15,17) Serum levels of TNFa are also increased in humans with low thyroid-stimulating hormone (TSH) levels and in mice carrying mutations in TSH receptor (Tshr −/− ). (50,51) Both experience low BMD and an increased risk of bone fractures. Interestingly, in Tshr −/− mouse mutants, deletion of Tnfa partially rescued the low bone mass phenotype. (52) Also, blocking TNFα by injecting a binding protein or deletion of Tnfa inhibited bone loss induced by ovariectomy. (53,54) Conversely, TNFα addition promoted osteoclast survival and prevented apoptosis. (55) Thus, these findings in mice strongly suggest an osteoclast-promoting effect of TNFα. Consistent with this, we showed that in the medaka osteoporosis model, genetic or chemical inhibition of tnfa expression reduced osteoclast formation upon RANKL induction (Figs. 5 and 6). In vivo imaging further revealed that in the absence of ectopic RANKL, TNFα deficiency did not cause any obvious effects on macrophage populations. It was only after RANKL induction that tnfa mutants exhibited significantly reduced macrophage numbers overall and in the vertebral column. This strongly suggests that TNFα selectively contributes to osteoclast differentiation under osteoporotic conditions, but not in normal development. Consistently, our qPCR analysis revealed that tnfa transcription reached a maximum at 1 day after RANKL induction and was then reduced at 2 and 3 dphs ( Supplementary Fig. S8). This implies that TNFα may participate in the initial phase of macrophage recruitment. Its high levels at the beginning appear critical for macrophage differentiation into osteoclasts. Blocking TNFα has been an efficient treatment for rheumatoid arthritis, which is characterized by strong inflammation mediated by macrophages. (13) Our results suggest that targeting TNFα signaling might be a promising therapeutic strategy for osteoporosis.
Dynamic transcriptome changes in activated macrophages suggest multiple functions in bone remodeling A dynamic regulation of cytokines and their receptors is critical for immune cells to function accurately in response to foreign stimuli or microenvironmental changes. (56) Uncontrolled overproduction of cytokines can lead to destructive chronic inflammation, which results in severe tissue defects. (57) Certain cytokines receptors, such as IL22Ra and IL20R, have been shown to function as drivers for the migration and recruitment of different immune cell types including T and B cells, dendritic cells, and neutrophils. (58,59) The expression of these genes is also found upregulated in patients suffering from bone loss caused by rheumatoid arthritis and osteoporosis. (60)(61)(62) Interestingly, our transcriptome profiling of FAC-sorted cells showed that these genes were downregulated in macrophages undergoing osteoclast differentiation (Supplementary Fig. S5). The reduced cytokine receptor expression coincided with a reduction of macrophage dynamics after they reached the bone matrix and started differentiating. We speculate that it is critical for macrophages to reduce motility before they attach tightly to bone matrix and form actin rings to facilitate bone resorption. Although cytokine receptor expression was downregulated under osteoporotic conditions, pathways involved in chondrocyte and osteoblast differentiation surprisingly were upregulated in RANKL-induced macrophages ( Table 2 and Supplementary Fig. S5). This could reflect a critical role for activated macrophages in bone cell-coupling: by promoting formation of chondrocytes and osteoblasts to achieve bone homeostasis. Our findings of upregulated col6a1, col6a2, col6a3, and col12a1b expression in RANKL-induced macrophages are suggestive for a role in osteogenesis. Such a role was earlier proposed by Izu and colleagues, who found that collagens VI and XII form a complex to mediate osteoblast interaction during osteogenesis. (63) In the medaka model, we also found upregulation of different integrin subunits and other adhesion molecules (Supplementary Fig. S5). This is in line with our observation that RANKL-induced macrophages interact dynamically with osteoblasts at the bone surface when maturing into osteoclasts (see Supplementary Movie S7). We speculate that these matrix and adhesion proteins are critical for the tight interaction of preosteoclasts with osteoblasts and facilitate osteoclast differentiation. In an alternative scenario, our observation of upregulated osteogenic transcripts in osteoclasts could reflect the uptake of osteoblast-released exosomes, a process that was recently shown to occur in regenerating zebrafish scales. (64) These osteoblast-derived exosomes could not only contain factors driving osteoclast differentiation, as reported by Kobayashi-Sun and colleagues, (64) but also osteogenic mRNAs. However, we observed that the majority of upregulated osteogenic genes (i) is already expressed at basal levels in noninduced macrophages, and (ii) encode secreted or transmembrane proteins. This observation favors the idea of a non-cell autonomous role for activated macrophages in chondro-and osteogenesis. Clearly, further studies are needed to investigate the complex roles of macrophages in bone remodeling. Together, our study highlights the dynamic nature of macrophages and the complex intercellular interactions implicated in bone homeostasis.

Disclosure
The authors declare that they do not have any conflict of interest. All data presented in this article will be shared upon request.