Self-renewing resident arterial macrophages arise from embryonic CX3CR1+ precursors and circulating monocytes immediately after birth

Macrophages densely populate the arterial wall, yet their origin and homeostasis are poorly understood. Robbins and colleagues show that arterial macrophages arise from CX3CR1+ embryonic precursors and adult bone marrow–derived monocytes that colonize the tissue immediately after birth. Resident macrophages densely populate the normal arterial wall, yet their origins and the mechanisms that sustain them are poorly understood. Here we use gene-expression profiling to show that arterial macrophages constitute a distinct population among macrophages. Using multiple fate-mapping approaches, we show that arterial macrophages arise embryonically from CX3CR1+ precursors and postnatally from bone marrow–derived monocytes that colonize the tissue immediately after birth. In adulthood, proliferation (rather than monocyte recruitment) sustains arterial macrophages in the steady state and after severe depletion following sepsis. After infection, arterial macrophages return rapidly to functional homeostasis. Finally, survival of resident arterial macrophages depends on a CX3CR1-CX3CL1 axis within the vascular niche.

Most tissues of the body harbor resident macrophages. Yet macrophages are phenotypically and functionally heterogeneous, a reflection of the diversity of tissue environments in which they reside. In addition to maintaining tissue homeostasis and responding to invading pathogens, macrophages contribute to numerous pathological processes, making them an attractive potential target for therapeutic intervention. This, however, will require a detailed understanding of macrophage origins, the mechanisms that maintain them and their functional attributes in different tissues and disease contexts.
Macrophage ontology has long engendered controversy 1,2 . Nevertheless, the idea that tissue macrophages develop exclusively from circulating bone marrow-derived monocytes has prevailed for nearly a half-century 3 . But accumulating evidence, including studies using sophisticated fate-mapping approaches, has determined that some tissue macrophages and their precursors are established embryonically in the yolk sac (YS) and fetal liver before the onset of definitive hematopoiesis [4][5][6][7][8][9][10][11] . Regardless of their origin, tissue macrophages can maintain themselves in adulthood by self-renewal independently of blood monocytes 12,13 .
Gene-expression profiling of macrophage populations from several tissues has established that all macrophages express only a small number of transcripts 14 , indicating the importance of the context provided by the tissue in studies of macrophage function in homeostasis and disease. The normal arterial wall contains many tissue-resident macrophages that contribute crucially to immunity, tissue homeostasis and wound healing after injury 15 . However, the regulatory networks, ancestry and mechanisms that maintain arterial macrophages remain unknown.
Using gene-expression analysis, we have found that arterial macrophages constitute a distinct population among tissue macrophages. Multiple fate-mapping approaches demonstrated that arterial macrophages arise embryonically from CX3CR1 + precursors and postnatally from bone marrow-derived monocytes that colonize the tissue during a brief period immediately after birth. In adulthood, arterial macrophages were maintained by CX3CR1-CX3CL1 interactions 1 6 0 VOLUME 17 NUMBER 2 FEBRUARY 2016 nature immunology A r t i c l e s and local proliferation without further substantial contribution from blood monocytes. Self-renewal also sustained arterial macrophages after severe depletion during polymicrobial sepsis, rapidly restoring them to functional homeostasis.

RESULTS
Phenotype and gene-expression profile of arterial macrophages Flow cytometry-based analysis of single cell suspensions from healthy aortas of 6-to 8-week-old mice revealed that 38 ± 4% (mean ± s.e.m.) of all CD45 + leukocytes were macrophages (F4/80 + CD11b + cells) (Fig. 1a). Other myeloid cell populations observed included F4/80 lo CD11c + MHCII + dendritic cells (Fig. 1a) and F4/80 lo CD64 + Ly6C hi-lo monocytes (Fig. 1a). Principal component analysis revealed a distinct transcriptome in arterial macrophages, which clustered near other macrophage populations including microglia, alveolar macrophages and splenic red pulp macrophages, as characterized by the Immunological Genome Consortium 14 ( Fig. 1b and Supplementary Fig. 1a). Stringent comparison of gene-expression profiles among arterial, brain, alveolar and splenic red pulp macrophages revealed 212 transcripts that were at least five-fold higher or lower in arterial macrophages relative to expression in the three other macrophage populations (Fig. 1c,d and Supplementary Fig. 1b,c). To gain insight into biological processes, we grouped transcripts encoding molecules with annotated functions according to Gene Ontology (GO) terms. Transcripts that were more abundant in arterial macrophages were enriched for molecular terms including translation and regulation of cell proliferation. The less abundant transcripts yielded GO enrichment of terms including homeostatic process, cell proliferation and macromolecular complex subunit organization (Supplementary Table 1). Flow cytometry analysis of specific cell surface markers confirmed that arterial macrophages share features with macrophage populations from other organs and express a unique signature. In addition to the core macrophage markers CD64 and the tyrosine kinase receptor MerTK 16 , arterial macrophages expressed Lyve1, CD68, major histocompatibility complex class II (MHCII), CD86, the class A scavenger receptor Msr1, Toll-like receptor 4 (TLR4) and T cell immunoglobulin mucin 4 (Tim-4) (Fig. 1a,e).
Zbtb46, encoding a BTB-ZF transcription factor, is selectively expressed in classical dendritic cells (cDCs) 17,18 . Analysis of mice in which a GFP reporter cassette replaces exon 2 of the Zbtb46 BTB domain (Zbtb46 +/gfp mice) revealed high expression of Zbtb46 in CD11c + MHCII + CD11b − cDCs of the aorta, liver, lung and brain (Fig. 1f). GFP was undetected in alveolar macrophages and Kupffer cells but was expressed in microglia and aortic macrophages, albeit at lower levels than in cDCs (Fig. 1f). These data extend published observations that resident macrophages in some tissues express Zbtb46 (ref. 19) and show that the level of Zbtb46 expression, as inferred from the GFP reporter, differs between macrophages and cDCs in the arterial wall. cDCs were also distinguished from macrophages by expression of the DC marker CD11c (Fig. 1g).
Intravenous administration of clodronate liposomes to mice depleted blood monocytes but not aortic macrophages (Supplementary Fig. 1d), indicating that arterial macrophages reside within the vessel wall. To examine the spatial distribution of arterial macrophages in more detail, we dissected the inner intimal and medial layers from the outer adventitia 20 and assessed the number of macrophages by flow cytometry. The adventitia contained significantly (P < 0.05) more CD45 + leukocytes than the intimal or medial layers of the vessel wall (Supplementary Fig. 1e). Consistent with this observation, macrophage number and percentage among CD45 + cells (Fig. 1h,i) were markedly higher in the adventitia than in the intima and media layers.
We next used confocal microscopy to visualize CD68 + cells in the intima, media and adventitia compartments of the aorta. On average, 29 ± 1% of the tissue area within the adventitia stained positive for CD68 (Fig. 1j). In contrast, only 2.0 ± 0.2% of tissue area in the intima and media stained positive for CD68 (Fig. 1j). Analysis at various locations in the aorta demonstrated comparable differences in macrophage content within the adventitia and intima and media (Fig. 1j). Consistent with published studies 21,22 , CD11c + DCs were most abundant in the intima of the aortic valves and arch (Supplementary Fig. 1f). We also detected macrophages in the carotid as well as femoral arteries (Supplementary Fig. 1g), indicating that resident macrophages occupy multiple arterial sites.
Arterial macrophages have embryonic and postnatal origins In many tissues, resident macrophages arise from embryonic precursors before birth [5][6][7]10,11,23 . Similarly, arterial macrophages were observed in the aorta of mice at embryonic day 16.5 (E16.5) (Fig. 2a). Flow cytometry revealed two phenotypically distinct cell populations expressing the macrophage marker F4/80 in the developing embryo: F4/80 hi CD11b lo cells that resembled YS-derived macrophages 6 and F4/80 int CD11b hi cells that appeared similar in phenotype to c-Myb-dependent fetal liver monocytes 7,23 . At birth, arterial macrophages were predominately F4/80 hi CD11b lo , although this phenotype was transient, and by 2 weeks of age the F4/80 + cells in the aortic wall were mainly F4/80 int CD11b hi (Fig. 2a). The proportion of arterial macrophages among CD45 + cells increased with age (Fig. 2b) and was associated with progressively increased expression of MHCII ( Fig. 2c), indicating a postnatal period of maturation.
The extra-embryonic YS is the main hematopoietic site in mice before E10 (refs. 4,24). YS hematopoiesis supplies erythroid as well as myeloid precursors to the embryo after the onset of blood circulation at ~E9. Because arteries contained macrophages before birth, we investigated whether YS progenitors contributed to the arterial macrophage pool using pulse labeling of CX3CR1 + progenitor cells in the YS 12 . Female Cx3cr1 CreER mice 25 , which express Cre recombinase under the control of the Cx3cr1 promoter upon exposure to tamoxifen (TAM), were crossed to male mice bearing the tdTomato red fluorescent reporter (Rosa26 Tomato mice) and injected with a single intraperitoneal dose of TAM at E8.5. This approach induces the 'preferential' and irreversible expression of the tdTomato reporter in YS-derived CX3CR1 + cells and their progeny. In Cx3cr1 CreER ; Rosa26 Tomato mice, 68 ± 8% of brain microglia showed tdTomato expression (Td tom+ ) at E16.5, indicating robust labeling efficiency of YS progenitors and their progeny (data not shown). Consistent with the hypothesis that YS progenitors give rise to arterial macrophages during embryogenesis, >40% of F4/80 hi CD11b lo cells in aortas of mice at E16.5 and on the day of birth were Td tom+ ( Fig. 2d-f). Importantly, TAM injection labeled macrophages but not monocytes in fetal liver (Supplementary Fig. 2a) and did not label F4/80 hi CD11b lo macrophages in the arterial wall (Fig. 2d). Therefore, Td tom+ macrophages probably derive from early YS erythromyeloid progenitors (EMPs) independently of a monocyte intermediate. Normalization of arterial macrophage labeling to labeling of microglia (% Td tom+ macrophages in the aorta / % Td tom+ microglia), which are entirely of YS origin 5,26 , indicated that ~60% of arterial macrophages at birth arise separate from fetal liver hematopoiesis (Fig. 2f).
To address the possibility that fetal monocytes also contribute to the generation of arterial macrophages, we analyzed aortas in Flt3 Cre × Rosa mT/mG reporter mice. Definitive hematopoietic stem and progenitor cells (HSPC) transiently augment the receptor tyrosine kinase FLT3 during differentiation to all hematopoietic lineages 27 . The approach, therefore, allowed us to identify HSPC-dependent npg A r t i c l e s GFP + macrophages (FLT3-Cre + ) and HSPC-independent GFP − macrophages (FLT3-Cre − ). At birth, approximately 10 ± 1% and 28 ± 3% of F4/80 hi CD11b lo and F4/80 int CD11b hi arterial macrophages, respectively, were GFP + and derived entirely from fetal monocytes (Fig. 2g). Fetal monocytes are generated through both FLT3-dependent and FLT3-independent pathways 10 . The contribution of fetal liver hematopoiesis to development of the arterial macrophage pool, therefore, is likely to be underestimated. The data indicate that arterial macrophages are established through differentiation of early as well as late YS-derived EMPs.
In adult mice, E8.5-labeled Td tom+ YS progenitors contributed to the macrophage pool in the aorta to a greater extent than in the liver,     ) from aorta, liver, lung and brain of Zbtb46 gfp/+ mice. One experiment of two is shown. Dashed lines, wild-type macrophages (GFP − ); dotted lines, wild-type DCs (GFP − ). (g) CD11c-eYPF expression in arterial macrophages and DCs from CD11c eYFP/+ mice. Data are from one of three animals examined. (h) Flow cytometry analysis of arterial macrophages in adventitial and intima-media compartments (mean ± s.e.m.; adventitia n = 27, intima-media n = 22). Data are pooled from four independent experiments. *P < 0.0001 (unpaired t-test). Numbers adjacent to outlined areas indicate percentages of F4/80 + CD11b + macrophages. (i) Quantification of arterial macrophages from h (mean ± s.e.m.; adventitia n = 27, intima-media n = 22). Data are pooled from four independent experiments. *P < 0.0001 (unpaired Student's t-test). (j) CD68 staining of adventitia and intima-media from ascending, descending, thoracic and abdominal aortic segments (mean ± s.d.; n = 3). *P < 0.0001 (total aorta, descending arch, thoracic arch), **P < 0.0003 (abdominal aorta), ***P < 0.0006 (ascending arch) (unpaired t-test). npg 1 6 2 VOLUME 17 NUMBER 2 FEBRUARY 2016 nature immunology A r t i c l e s lung and peritoneum (Fig. 2f,h). Moreover, the appearance of Td tom+ MHCII + arterial macrophages in adulthood (Fig. 2i) indicated that MHCII − cells gave rise to MHCII + macrophages some time after birth, because arterial macrophages in newborn mice were MHCII − (Fig. 2c). The persistence of YS-derived macrophages in the adult aorta was independently verified using Csf1r MeriCreMer × Rosa mTmG mice administered TAM at E8.5. These mice express the TAM-inducible MerCreMer fusion protein under control of the macrophage specific mouse Csf1r promoter. YS-derived macrophages appear as GFP + (Fig. 2j). The decline in YS labeling of arterial macrophages from ~60% at birth to ~20% in adulthood (Fig. 2e,f) could result from replacement of embryonic arterial macrophages by circulating monocytes 9,28 .    To address this possibility, pregnant Cx3cr1 CreER mice were injected with TAM at E18.5. The approach labeled most arterial macrophages (and microglia) in Cx3cr1 CreER ; Rosa26 Tomato mice at birth, but few blood monocytes (Fig. 2k,l). tdTomato labeling in arterial macrophages, but not microglia, decreased during the first 2 weeks of life (Fig. 2k,l), suggesting substantial turnover and replacement of arterial macrophages with unlabeled circulating monocytes. Consistent with these observations, monocyte influx was associated with increased perinatal expression in the aorta of the chemokine CCL2 and the cellular adhesion molecules VCAM-1, E-selectin and ICAM-1 ( Fig. 2m and Supplementary Fig. 2b). We investigated the postnatal contribution of definitive hematopoiesis to the adult arterial macrophage pool further using Flt3 Cre × Rosa mT/mG reporter mice. Flt3 Cre × Rosa mT/mG Ly6C hi blood monocytes were GFP + and derived entirely from HSPC precursors (Fig. 2n). In agreement with our earlier observations, arterial macrophages in mice comprised both GFP + and GFP − subsets, confirming the dual YS and HSPC origin of these cells (Fig. 2n,o). These observations suggest successive waves of arterial macrophage colonization, initially by an embryonic wave derived from early YS EMP and fetal liver monocytes 9-11,23 followed by a brief influx of bone marrow-derived monocytes immediately after birth.

CX3CL1-CX3CR1 axis determines arterial macrophage survival
Described differences in tissue-specific macrophage requirements for the growth factors macrophage-colony stimulating factor (M-CSF) and granulocyte macrophage-colony stimulating factor (GM-CSF) 29,30 led us to assess arterial macrophages in Csf1 −/− and Csf2 −/− mice, which lack the genes encoding M-CSF and GM-CSF, respectively. We found significantly (P < 0.05) fewer arterial macrophages in the arteries of Csf1 −/− , but not Csf2 −/− mice than in wild-type mice (Supplementary Fig. 3a,b), consistent with other studies demonstrating profound macrophage deficiencies in M-CSF-deficient mice 29 .
Although many tissue macrophages lose expression of CX3CR1 as they mature 13 , a large proportion of arterial macrophages in adult mice retain its expression (Fig. 3a-c and Supplementary Fig. 3c). Therefore, we investigated whether CX3CR1 contributed directly to the maintenance of arterial macrophages. Cx3cr1 −/− mice had fewer arterial macrophages than wild-type controls, as assessed by confocal microscopy (Fig. 3d) and flow cytometry (Fig. 3e). Neutralizing antibodies directed against the CX3CR1 ligand CX3CL1 also decreased arterial macrophage numbers in wild-type mice (Fig. 3f). Arterial macrophage proliferation was unchanged in Cx3Cr1 −/− mice compared to wild-type control mice (Fig. 3g), but the percentage of Fas + macrophages (Fig. 3h) and the number of TUNEL + CD68 + cells (Fig. 3i) was increased in the vessel adventitia, suggesting that CX3CR1-CX3CL1 controls the survival of arterial macrophages. We next used Cx3cl1 cherry mice 31  npg A r t i c l e s cells and CD68 + resident macrophages (Fig. 3j). Flow cytometry and immunofluorescence staining revealed two main populations of CX3CL1 + cells in the aorta of Cx3cl1 cherry mice, namely CD31 + endothelial cells and PDGFRα + mesenchymal cells (Fig. 3k,l). Therefore, arterial macrophage maintenance depends partially on a local CX3CR1-CX3CL1 axis.

Arterial macrophages self-renew in adulthood
To examine the mechanism of arterial macrophage renewal in adult mice, we examined the presence of arterial macrophages in mice deficient for the chemokine receptor CCR2 (Ccr2 −/− mice), which have reduced numbers of circulating Ly6C hi monocytes 32 . The arteries of wild-type and Ccr2 −/− mice contained similar numbers of arterial macrophages (Fig. 4a), which suggests that macrophage turnover at steady state occurs largely independently of circulating monocytes. To assess the rate of macrophage replacement in the arterial tissue, 8-week-old C57BL/6J and UBC-GFP mice, which express GFP under control of the human ubiquitin C promoter in all tissues, were joined by parabiosis for 8 months. Ly6C hi and Ly6C lo monocyte chimerism in the blood at equilibrium was high (~32% and ~41%, respectively; Fig. 4b), as expected, but macrophage chimerism in the aorta of parabiotic mice was low (~6%), suggesting that monocyte contribution to the arterial macrophage pool was limited. Macrophage chimerism was similarly low in the heart, lung and liver of parabiotic mice 12 (Fig. 4b). As shown in published studies 33 , macrophage chimerism in the aorta alone in parabiotic mice underestimates the overall contribution of circulating monocytes to the arterial macrophage pool because Ly6C hi monocyte chimerism in the blood, even at equilibrium, is only ~32% (Fig. 4b).
When we parabiotically linked Ccr2 −/− and UBC-GFP (Ccr2 +/+ ) mice for 5 weeks, monocyte chimerism (GFP + cells) in the blood (CD115 hi ) and aortas (F4/80 int CD11b + Ly6G − ) of CCR2 −/− partners was ~82% and ~73%, respectively (Fig. 4d). Despite high chimerism of partner-derived wild-type monocytes in Ccr2 −/− mice, chimerism of partner-derived F4/80 hi CD11b + arterial macrophages in Ccr2 −/− mice remained low (~9%; Fig. 4d). We also independently assessed the monocyte contribution to the arterial macrophage pool by pulse labeling adult Cx3cr1 CreER Rosa26 Tomato mice. TAM treatment induced CX3CR1-tdTomato expression in ~19% of blood Ly6C hi monocytes and ~59% of arterial macrophages (Fig. 4e). tdTomato expression remained high (~50%) among arterial macrophages at 9 and 11 months after labeling ( Fig. 4e) but was absent in blood monocytes 34 (Supplementary Fig. 4a), suggesting that maintenance of arterial macrophages depends little on blood monocytes. Persistence of pulse-labeled CX3CR1 + macrophages could also result from slow cell turnover. Therefore, arterial macrophage turnover was assessed in B6;129S4-Gt(ROSA)26Sor tm1(rtTA*M2)Jae Col1a1 tm7(tetO-HIST1H2BJ/GFP)Jae /J (H2B-GFP) mice, in which doxycycline treatment induces H2B-GFP fluorescence labeling of cells ubiquitously 35 . Cellular expression of H2B-GFP was induced in adult mice by doxycycline treatment for 4 weeks, and the loss of GFP fluorescence per cell, which is indicative of cell division, was monitored during a 2-month chase period. As expected, H2B-GFP expression in blood monocytes and arterial macrophages exceeded background by orders of magnitude after 4 weeks of doxycycline (Fig. 4f  and Supplementary Fig. 4b). Consistent with observations that myeloid precursors turnover rapidly 35 , expression of H2B-GFP in blood Ly6C hi monocytes declined below the limit of detection after 2 months (Supplementary Fig. 4b). H2B-GFP expression was also completely lost in some arterial macrophages (~25%), and mean fluorescence intensity of the GFP signal was greatly reduced in others (Fig. 4f), indicating substantial turnover of arterial macrophages within 2 months. To estimate further the turnover and loss rate of arterial macrophages, we generated a mathematical model using a three-state absorbing continuous time Markov chain (CTMC). The approach assumed two transient states of GFP + macrophages with different rates of GFP loss and one absorbing state, representing cells that have lost GFP expression (i.e., GFP − macrophages) (Online Methods). Fitting the mathematical model to npg A r t i c l e s the observed dilution of the H2B-GFP signal in arterial macrophages suggested a turnover rate of ~84% every 12 months for these cells (Fig. 4g). Independently, 5-bromodeoxyuridine (BrdU) injections into wild-type mice every other day for 9 d labeled 25% of aortic macrophages ( Fig. 4h and Supplementary Fig. 4c). Hence, arterial macrophage turnover is dynamic.
Arterial macrophages self-renew after exposure to bacteria To address how arterial macrophages are replenished during inflammation, we assessed macrophage repopulation in irradiated CD45.2 + mice transplanted with whole bone marrow from CD45.1 + mice. Recipient mice showed near complete donor chimerism among blood leukocytes 6 months after transplantation (Fig. 5a). Brain microglia excepted 36,37 , donor chimerism of resident macrophages in the liver, heart, lung and aorta exceeded 70% (Fig. 5a), consistent with reports that lethal radiation impairs the local repopulation capacity of tissueresident macrophages 38 .
To assess macrophage turnover after infection, we exposed mice to the Gram-negative bacterial cell wall component lipopolysaccharide (LPS) or subjected them to surgical puncture of the cecum. The number of F4/80 hi CD11b + CD115 + Lyve1 + resident arterial macrophages contracted immediately after LPS exposure or cecal puncture 39 (Fig. 5b and Supplementary Fig. 5a). Depletion of resident macrophages was associated with accumulation of neutrophils (Supplementary Fig. 5b), Ly6C hi monocytes (Fig. 5b) and a distinct macrophage population identified as F4/80 hi CD11b + CD115 − Lyve1 − (Fig. 5b,c and Supplementary Fig. 5c). By 1 week after LPS exposure, resident CD115 + Lyve1 + macrophage numbers rebounded to levels observed during steady-state conditions (Fig. 5b,c,e), whereas neutrophils, monocytes and CD115 − Lyve1 − macrophages were nearly absent (Fig. 5b,c and Supplementary Fig. 5b). Arterial macrophage numbers rebounded in LPS-exposed Ccr2 −/− mice as well (Fig. 5b), indicating that local expansion rather than monocyte recruitment was the dominant mechanism of recovery. In agreement with this observation, LPS exposure in wild-type mice increased the number of aortic macrophages in S, G2 and M phases of the cell cycle compared to wildtype control mice (Fig. 5f). In addition, we exposed parabiotic mice to LPS or subjected them to cecal puncture. Partner chimerism was low among arterial resident CD115 + Lyve1 + macrophages, yet chimerism of newly infiltrating CD115 − Lyve1 − macrophages was high, suggesting that monocytes were the immediate precursors of these cells (Fig. 5c,d). Intravenous transfer of Ly6C hi monocytes into LPS-treated mice showed that Ly6C hi monocytes gave rise to CD115 − Lyve1 − macrophages but not CD115 + Lyve1 + macrophages in the aorta (Fig. 5g). Moreover, CD115 − Lyve1 − arterial macrophages showed a greater in vivo capacity to phagocytose bacteria than CD115 + Lyve1 + arterial macrophages (Fig. 5h), indicating that functional differences exist between the two subsets of macrophages.
To determine whether YS-derived arterial macrophages respond differently than bone marrow-derived arterial macrophages, adult E8.5-labeled Cx3cr1 CreER ; Rosa26 Tomato mice were exposed to LPS or subjected to cecal puncture and assessed the proportions of E8.5labeled and unlabeled arterial macrophages during the macrophage recovery phase to determine whether one population was 'preferentially' expanded over the other. Neither LPS administration nor polymicrobial sepsis (cecal puncture) affected the percentage of F4/80 hi CD11b + macrophages that were tdTom + (Fig. 5i), suggesting equal self-renewal capabilities of the two subsets of arterial macrophages during infection. Microarray analysis of aortic macrophages during homeostasis and after recovery from sepsis showed that bacterial exposure had little effect on the transcriptional program of self-renewing macrophages ( Fig. 5j and Supplementary Fig. 5d). Of 10,391 genes analyzed, only 12 were differentially expressed (10 at higher and 2 at lower levels) in arterial macrophages from mice subjected to cecal puncture and steady-state control mice. In addition, in vitro, arterial macrophages could phagocytose bacteria before and after sepsis to the same extent (Supplementary Fig. 5e). Therefore, arterial macrophages return to functional homeostasis rapidly after infection.

DISCUSSION
Here we identified the molecular signature of arterial macrophages, their developmental pathways and key mechanisms that ensure their homeostasis. We have shown that arterial macrophages are distinct among tissue-resident macrophages. Multiple fate-mapping approaches demonstrated that arterial macrophages originate embryonically from CX3CR1 + precursors and postnatally from circulating monocytes immediately after birth. In adulthood, arterial macrophages were maintained by CX3CR1-CX3CL1 interactions and local proliferation rather than recruitment of circulating monocytes. Self-renewal also restored arterial macrophages to functional homeostasis after severe depletion induced by polymicrobial sepsis.
The microarray database generated by the Immunological Genome Consortium provides a valuable resource for comparing gene-expression profiles of macrophages from different organs 14 . Consistent with evidence demonstrating diversity among macrophage populations, our gene-expression and protein analyses revealed distinct patterns for arterial macrophages relative to other tissue macrophages. The data identified GO enrichment of transcripts with annotated functions that equip arterial macrophages for specialized local functions, supporting the concept that meaningful assessment of macrophage function requires careful consideration of the tissue context in which they reside.
Tissue macrophages arise from two distinct developmental programs: early YS-derived EMPs that give rise to macrophages without monocyte intermediates and fetal monocytes that derive from late c-Myb + EMPs generated in the YS 10,11 . These pathways contribute to macrophage development in several tissues, including the brain, skin, heart, liver and lung 5,8,13,23,40,41 . Consistent with these findings, F4/80 hi CD11b lo arterial macrophages and F4/80 lo CD11b hi fetal monocytes were readily identified in aortas of embryonic (E16.5) mice. Cx3cr1-, Csf1r-and Flt3-driven fate-mapping approaches indicated that arterial macrophages were derived from early YS EMPs as well as fetal monocytes. Our data also indicated that arterial macrophage colonization associates with a period bone marrow-derived monocyte recruitment shortly after birth. Development of the arterial macrophage pool, therefore, is unique. The maintenance of intestinal macrophages also depends on circulating monocytes, but renewal is constant and continues throughout adult life 28 . In arteries, the period of postnatal monocyte influx was brief, corresponding with transient expression of chemokines and cell adhesion molecules implicated in monocyte recruitment. Although adult arteries contained sizeable populations of both YS and bone marrow-derived macrophages, the relative contribution of the two subsets of macrophages to vessel homeostasis remains unknown.
In many tissues, resident macrophages lose expression of CX3CR1 during development 13 . In addition to retaining macrophage expression of the chemokine receptor in adulthood, CX3CL1 blockade and examination of Cx3cr1 −/− mice indicated that CX3CR1-CX3CL1 interactions determine the survival of arterial macrophages, possibly through Fas-Fas ligand interactions. CX3CR1-CX3CL1 similarly promotes survival of macrophages in the brain 42 , kidney 43 , solid tumors 44 and circulating Ly6C lo blood monocytes 45 . Visualization of the arterial npg macrophage niche using Cx3cl1 cherry reporter mice further showed that CD31 + endothelial cells and PDGFRα + mesenchymal cells produced CX3CL1 locally in the artery; however, the relative contribution of these CX3CL1 producers to macrophage survival is not known. The data also suggested the dependence of macrophage survival in the arterial wall on signaling pathways other than CX3CR1, as not all arterial macrophages express the receptor and Cx3cr1 −/− mice contain a moderate population of aortic macrophages. M-CSF deficiency was similarly associated with fewer arterial macrophages, although it remains to be determined whether M-CSF regulates macrophage differentiation, survival and/or proliferation.
Mathematical modeling of loss of arterial macrophages in H2B-GFP mice predicted near complete replacement of the macrophage population within ~1 year. Arterial macrophage turnover, therefore, is dynamic. Analysis of parabionts, CCR2 −/− mice and pulse-labeled adult Cx3cr1 CreER ; Rosa26 Tomato mice indicated that local proliferation rather than monocyte recruitment drives arterial macrophage renewal in the steady state and during polymicrobial sepsis. This result contrasts with a report on macrophage maintenance in the heart, where bone marrow-derived cells replace macrophages progressively with age 9 . The data further showed that the arterial macrophage response to bacteria is many-sided. Infection led first to the recruitment of Ly6C hi monocytes and their differentiation into CD115 − Lyve1 − macrophages that functioned to phagocytose bacteria. This was followed by the self-renewal and re-establishment of functional homeostasis of CD115 + Lyve1 + resident macrophages. The diversity of origins (successive contributions of YS, fetal liver and conventional hematopoiesis) of arterial macrophages highlights the importance of tissue-specific extrinsic factors, including CX3CR1-CX3CL1 interactions, in maintaining their abundance. Of note, macrophage proliferation also amplifies arterial pathology, as has been observed in atherosclerosis 33 . Therefore, future design of therapeutic strategies that target arterial macrophages will require not only elucidation of the mechanisms that maintain them, but their activities in specific disease contexts.

METHODS
Methods and any associated references are available in the online version of the paper. Accession codes. Gene Expression Omnibus: Data have been deposited under accession codes GSM1689051, GSM1689052 and GSM1689053 (aortic macrophage, naïve) and GSM1689054, GSM1689055 and GSM1689056 (aortic macrophage, sepsis). BD Biosciences, 558774) and rabbit anti-mCherry overnight. Sections were then rinsed 3 times in 1× PBS, then incubated goat anti-rat Alexa Flour 647 (1:400; ThermoFisher Scientific, A-21247) and goat anti-rabbit Alexa Flour 568 secondary antibodies for 2 h at room temperature, protected from light. Nuclei were counterstained with 10 µg/ml Hoechst 33258 in ddH 2 O for 10 min before slides were mounted with 50:50 glycerol/PBS and stored at −20 °C until imaging. Images were captured on a Zeiss Observer V spinningdisk confocal microscope.
RNA microarray. Aortic macrophages (CD11b hi F4/80 hi CD45 + MerTK + CD64 + ) were isolated from C57/B6 wild-type male mice aged 6-8 weeks before and 7 d after induction of sepsis by cecal puncture. 18-20 aortas were pooled per sample. Sorting was conducted on the MoFlo Astrios BRVY. Arterial macrophage purity was >99% (Supplementary Fig. 5). RNA was extracted using the PicoPure RNA Isolation Kit (Life Technologies). The total concentration and quality of RNA was determined using an Agilent 2100 Bioanalyzer (Agilent Technologies) according to the manufacturer's instructions. RNA transcripts were amplified using the Nugen ovation Pico WTA V2 kit. 3 ng of total RNA was amplified according the manufacturer's protocol. The amplification of RNA included amplification at the 3′ end as well random amplification throughout the transcript using Ribo-SPIA technology. 5 µg cDNA was used to generate ST-cDNA using the WT-Ovation Exon Module and 5ug of ST-cDNA was fragmented and labeled with Encore Biotin Module. RNA was hybridized on the Affymetrix Mouse Gene 1.0 ST array. Arrays were hybridized at 45 °C for 16-18 h, washed with Affymetrix fluidic station p450 and scanned with Affymetrix 7G scanner. Affymetrix gene-expression console was for QC and to generate QC reports.
Probe set expression levels were calculated using the multi-array procedure (RMA), which is available within the oligo Bioconductor package for the R statistical software project where raw intensity values were background corrected and normalized 49 . Probe sets with no gene assignments and/or raw intensities <120 were filtered out. In cases where multiple probe sets mapped onto a single gene, the probe set with the highest intensity was selected for further analysis. Principal component analysis was performed to compare naive arterial macrophage microarray data to published data sets from Immgen 46 using the prcomp command in R. RMA-adjusted values were then compared using a student's t-test with significance set as P < 0.05. qRT-PCR. RNA was extracted from whole aorta by TRIzol (Life Technologies) according to the manufacturer's protocol. RNA concentration was obtained using a Nanodrop ND100 spectrophotometer. cDNA was synthesized using qScript cDNA SuperMix (Quanta BioSciences) according to manufacturer protocol. Transcripts were then detected using specific primers (see Supplementary  Table 5) using SYBR Green Master (Roche) detected on a LightCycler 480 (Roche). Hprt was used as the housekeeping gene; values were compared using the 2 −∆∆Ct method. Samples were run in triplicate and averaged.
Mathematical modeling. MΦ turnover was predicted in mice with ubiquitous, doxycycline-inducible expression of an H2B-GFP fusion protein using a Continuous Time Markov Chain (CTMC). The system was modeled as a threestate absorbing CTMC, two transient states representing the two GFP + populations of macrophages having different rates of GFP loss and one absorbing state representing the cells that lost the GFP signal (i.e., became GFP negative) (Supplementary Table 2). This system generates a matrix of differential equations and each equation represents the transition probability rate equation of one cell moving between any two states. This matrix is the product of two other matrices, the probability matrix P, which contains the probability density function for the transitions between any two states, and the matrix of rates Q, which contains the values of the probability rates. Assuming initial conditions of [0.7 0.3 0]; the form of the Markov chain shows a representation of unconditional probabilities of reaching each state, and has the following form: From (1), a total of three simultaneous differential equations can be derived through matrix multiplication. Solving the three differential equations simultaneously gives the following three equations:  Table 3). The R 2 -value for all runs was 1.0, which confirms that the model is working; the SSE and RMSE values for each run are shown in Supplementary Tables 3 and 4. The RMSE on average was 5.75, which is acceptable, as a difference of ±5 is not significant on the range of the dependent variable in this case, which is 0-100. The Average fit was then plotted with the corresponding error bars for each calculated data point as well as the data point in the same plot for visual confirmation that the points lie within the prediction limits of the model. To confirm that our model predictions and strength were independent of the initial conditions used in the calculations; equation (7) was imported into Matlab with the average values of the coefficients substituted, and the values of x and y were changed in accordance with 0.1 steps. The functions produced from all steps were plotted on the same plot along with the Average fit. All the plots were identical, and there was a small difference between them and the Average fit, owing to the difference in the significance and number of decimal places generated during the simplification process. The difference was not significant, however (data not shown).
Flow cytometry. Antibodies used for flow cytometric analyses are provided in Supplementary Table 5. Data was acquired on an LSRII flow cytometer (BD Biosciences) and analyzed with FlowJo v8.8.6 (Tree Star, Inc.). Aorta, heart, liver, lung, and brain tissue were treated with FcBlock (BD Biosciences) for 15 min before incubation with antibody cocktail for an additional 30 min. Samples were fixed before flow analysis (BD Cytofix). Cell-cycle analysis was carried out using FxCycle violet stain (Invitrogen) on 95% ethanolfixed samples.

Statistics.
Results are expressed as means ± s.e.m. or means ± s.d. The statistical tests used included unpaired Student's t-test using Welch's correction for unequal variances and one-way analysis of variance (ANOVA) followed by Tukey's or Newman-Keuls multiple comparison test. P ≤ 0.05 was considered to denote significance. No animals were excluded from analyses. No randomization was used in these studies. Investigators were not blinded to group allocations during experiments.