Spatio-temporal profile, phenotypic diversity, and fate of recruited monocytes into the post-ischemic brain

Background A key feature of the inflammatory response after cerebral ischemia is the brain infiltration of blood monocytes. There are two main monocyte subsets in the mouse blood: CCR2+Ly6Chi “inflammatory” monocytes involved in acute inflammation, and CX3CR1+Ly6Clo “patrolling” monocytes, which may play a role in repair processes. We hypothesized that CCR2+Ly6Chi inflammatory monocytes are recruited in the early phase after ischemia and transdifferentiate into CX3CR1+Ly6Clo “repair” macrophages in the brain. Methods CX3CR1GFP/+CCR2RFP/+ bone marrow (BM) chimeric mice underwent transient middle cerebral artery occlusion (MCAo). Mice were sacrificed from 1 to 28 days later to phenotype and map subsets of infiltrating monocytes/macrophages (Mo/MΦ) in the brain over time. Flow cytometry analysis 3 and 14 days after MCAo in CCR2−/− mice, which exhibit deficient monocyte recruitment after inflammation, and NR4A1−/− BM chimeric mice, which lack circulating CX3CR1+Ly6Clo monocytes, was also performed. Results Brain mapping of CX3CR1GFP/+ and CCR2RFP/+ cells 3 days after MCAo showed absence of CX3CR1GFP/+ Mo/MΦ but accumulation of CCR2RFP/+ Mo/MΦ throughout the ischemic territory. On the other hand, CX3CR1+ cells accumulated 14 days after MCAo at the border of the infarct core where CCR2RFP/+ accrued. Whereas the amoeboid morphology of CCR2RFP/+ Mo/MΦ remained unchanged over time, CX3CR1GFP/+ cells exhibited three distinct phenotypes: amoeboid cells with retracted processes, ramified cells, and perivascular elongated cells. CX3CR1GFP/+ cells were positive for the Mo/MΦ marker Iba1 and phenotypically distinct from endothelial cells, smooth muscle cells, pericytes, neurons, astrocytes, or oligodendrocytes. Because accumulation of CX3CR1+Ly6Clo Mo/MΦ was absent in the brains of CCR2 deficient mice, which exhibit deficiency in CCR2+Ly6Chi Mo/MΦ recruitment, but not in NR4A1−/− chimeric mice, which lack of circulating CX3CR1+Ly6Clo monocytes, our data suggest a local transition of CCR2+Ly6Chi Mo/MΦ into CX3CR1+Ly6Clo Mo/MΦ phenotype. Conclusions CX3CR1+Ly6Clo arise in the brain parenchyma from CCR2+Ly6Chi Mo/MΦ rather than being de novo recruited from the blood. These findings provide new insights into the trafficking and phenotypic diversity of monocyte subtypes in the post-ischemic brain. Electronic supplementary material The online version of this article (doi:10.1186/s12974-016-0750-0) contains supplementary material, which is available to authorized users.


Background
Activation of brain-resident cells and mobilization of circulating leukocytes is a key feature of the inflammatory response after cerebral ischemia [1]. In animal models as in human stroke, post-ischemic brain inflammation is characterized by infiltration of hematogenous immune cells, monocyte-derived tissue macrophages (Mo/MΦ) being one of the predominant cell types [2][3][4][5]. Growing evidence suggests that Mo/MΦ invading the ischemic brain can contribute to both brain injury [6,7] and repair [8,9]. These opposing roles reflect the heterogeneity of the monocyte-macrophage lineage, which may adopt diverse phenotypes depending on environmental cues [10].
Two predominant subsets of circulating monocytes (CD115+ leukocytes) are found in the blood of rodents and humans: "inflammatory " and "patrolling" monocytes. In mice, inflammatory monocytes are characterized by the expression of the C-C chemokine receptor type 2 (CCR2), a critical molecule involved in the rapid recruitment of the cells into inflamed tissues [11]. They also express high levels of lymphocyte antigen 6 complex locus C1 (Ly6C) but low to intermediate levels of C-X(3)-C motif chemokine receptor 1 (CX3CR1). On the other hand, CX3CR1, the high-affinity functional chemokine receptor for fractalkine (CX3CL1), is enriched in patrolling monocytes, which do not express Ly6C and CCR2 or express them at low levels. Patrolling monocytes (Ly6C lo CCR2 − CX3CR1 + ) have longer half-life and crawl on the vessel luminal wall orchestrating the disposal of damaged endothelial cells [12]. In addition, patrolling monocytes can rapidly invade tissues in response to tissue damage or infection [13,14]. Patrolling monocytes were previously thought to originate from circulating inflammatory CD115 + Ly6C hi monocytes [14,15], but recent data suggest that they originate from distinct myeloid progenitors by the activity of the nuclear receptor subfamily 4, group A, member 1 transcription factor NR4A1 (Nur77) in the BM [16]. Furthermore, depletion of CD115 + Ly6C hi monocytes in the circulation does not affect the numbers of CD115 + Ly6C lo monocytes, reducing the likelihood that the sole precursors of "patrolling" monocytes are CD115 + Ly6C hi cells [17].
Little is known about the monocytes subtypes invading the ischemic brain, their morphology, and their fate once in the tissue. Monocytes infiltrating the ischemic brain acquire morphological and functional properties of tissue macrophages and can persist in the brain parenchyma for extended periods of time [7,8,[18][19][20]. However, the trafficking of inflammatory and patrolling monocytes into the ischemic brain, their relationship to the development of tissue injury, and their long-term fate remain unknown.
Here, we investigated the temporal profile, spatial distribution, and phenotype of inflammatory (CCR2 + Ly6C hi ) and patrolling (CX3CR1 + Ly6C lo ) monocyte subpopulations in transient focal cerebral ischemia. Using a mouse model to track CCR2 and CX3CR1 expression in hematogenous cells, we found that CCR2 + Mo/MΦ invade the brain in the acute phase and give rise to CX3CR1 + Mo/MΦ during the subacute and chronic phase of the injury, without the participation of circulating CX3CR1 + Mo. The findings identify CCR2 + Ly6C hi monocytes as the primary monocytic cell infiltrating the ischemic brain, where they undergo phenotypic transformation during the evolution of the damage. Such phenotypic diversity may underlie the multiplicity of roles attributed to Mo/MΦ in ischemic brain injury.

Middle cerebral artery occlusion
Transient focal cerebral ischemia was induced using the intraluminal filament model of middle cerebral artery occlusion (MCAo), as described previously [22]. Under isoflurane anesthesia (maintenance 1.5-2 %), a heatblunted nylon suture (6/0) was inserted into the right external carotid artery of anesthetized mice and advanced until it obstructed the middle cerebral artery (MCA). This was confirmed by cerebral blood flow (CBF) measured using transcranial laser Doppler flowmetry (Periflux System 5010, Perimed, King Park, NY) in the territory of the right MCA (2 mm posterior, 5 mm lateral to bregma). The filament was left in place for 35 min and then withdrawn. Only animals that exhibited a reduction in CBF of >85 % during MCA occlusion and in which CBF recovered by >80 % after 10 min of reperfusion were included in the study. Inadequate ischemia-reperfusion was found to be <4 % in the study and not different among groups. Rectal temperature was monitored and kept constant (37.0 ± 0.5°C) during the surgical procedure and in the recovery period until the animals regained full consciousness. No differences were found in mortality rate (<1 %) among groups.
Image analysis and cell counting of infiltrating CX3CR1 GFP/+ and CCR2 RFP/+ cells Mice were perfused transcardially with ice-cold PBS followed by 4 % paraformaldehyde in PBS. Brains were removed, stored overnight in the same fixative at 4°C, and then submerged in 30 % sucrose solution for 2 days and frozen. Using a cryostat, 12 serial coronal sections (thickness 16 μm) were cut every 600 μm from the frontal cortex to the caudal hippocampus (+3.14 to −3.46 bregma) and then mounted on slides (Fig. 1b). Fluorescent images were acquired by a scientific monochrome CCD camera (Retiga EXi, QImaging) using fluorescein isothiocyanate (FITC) (CX3CR1-GFP) and Texas Red (CCR2-RFP) filter sets through a 10× objective on a Nikon TE2000 microscope equipped with a motorized stage. Acquisition was controlled via software (IPLab, Biovision Technologies, Exton, PA) using a custom script. Individual image files (80 for each hemisphere) were combined using the ImageJ software (National Institute of Health, version 1.45) by invoking a stitching algorithm [24]. A custom ImageJ macro was used to semi-automatically count cells identified after thresholding the image based on pixel values. In lesions at 14 and 28 days after ischemia, we observed a densely populated lesion core where individual cells could not be unequivocally separated from neighboring cells. Therefore, these areas were excluded from the analysis. For presentation purposes, binary image masks of both green and red channel images generated for cell counting were superimposed on a grayscale image obtained from the FITC channel image by running the ImageJ "Remove Outliers" command once and the ImageJ "Smooth" command twice.
The proportion of morphological phenotypes of CX3CR1 GFP/+ cells was quantified within the ischemic hemisphere in images acquired with a 40× objective from 1.3, 0.7, −0.1, −0.5, and −1.1 bregma coronal brain sections at 14 days post-MCAO that were stained for GLUT-1 using ImageJ software. For the analysis of Ki67 + cells, coronal sections were scanned with a confocal microscope and Ki67 + and CX3CR1 GFP/+ cells were quantified from tile-scanned images using ImageJ software.

Brain mononuclear cell isolation
Mice were transcardially perfused under deep anesthesia with 30 ml of heparinized saline to remove blood cells. The ischemic hemisphere was separated from the cerebellum and olfactory bulb and homogenized with a dounce homogenizer in RPMI Medium 1640 with Phenol Red. Then, the cell suspension was subjected to discontinuous 70 %/25 % Percoll (GE Healthcare, Piscataway, NJ) density gradient centrifugation (30 min, 500 g, 4°C). Enrichedmononuclear cells were collected from the interphase, washed with Hepes-HBSS buffer (138 mM NaCl, 5 mM KCl, 0.4 mM Na 2 HPO 4 , 0.4 mM KH 2 PO 4 , 5 mM D-Glucose, 10 mM Hepes, pH 7.4), and resuspended in 2 % fetal bovine serum (FBS) in PBS FACS buffer for staining and FACS analysis.  Spatio-temporal profile of CX3CR1 GFP/+ and CCR2 RFP/+ Mo/MΦ infiltration after stroke. a Anatomical localization of CCR2-RFP (red) and CX3CR1-GFP (green) cells infiltrating the ischemic hemisphere in coronal brain sections of CCR2 RFP/+ CX3CR1 GFP/+ chimeric mice 3, 14, or 28 days after MCAo and in sham mice. Infarct volume was outlined in serial cresyl violet stained sections 3 days after MCAo and overlaid on fluorescent images (dotted line). b Quantification of infiltrating CX3CR1 GFP/+ and CCR2 RFP/+ cells in sham mice and stroke mice 1, 2, 3, 5, 7, 14, and 28 days after MCAo (n = 3-4 mice/group). *p < 0.05 and ***p < 0.001 vs. Sham. c Quantification of infiltrating CX3CR1 GFP/+ CCR2 RFP/+ double-positive cells in sham mice and stroke mice 1, 2, 3, 5, 7, 14, and 28 days after MCAo (n = 3-4 mice/group). **p < 0.01 vs. Sham. d Diagram depicting the brain-sectioning method. The brain was sectioned in 12 coronal slices at 600-μm intervals, beginning from the frontal pole and proceeding in caudal direction BioLegend (San Diego, California) and CCR2-APC (clone 475301) from R&D. For flow cytometric analysis of blood leukocytes, 100 μl of blood was withdrawn by cardiac puncture from mice under deep anesthesia immediately before sacrifice. Erythrocytes were lysed by incubating the blood three times with 5 ml erythrocyte lysis buffer (0.15 M NH 4 Cl, 1 mM KH 2 CO 3 , and 0.1 mM Na-EDTA) for 5 min at room temperature. Blood leukocytes were resuspended in FACS buffer. Cells were blocked with CD16/CD32 antibody for 10 min at 4°C, stained with predetermined optimal concentrations of antibodies for 20 min at 4°C, and analyzed on a MACSQuant (Miltenyi Biotec) or a FACSCalibur (Becton-Dickinson) cytometer. Dead cells and debris were excluded based on side and forward scatter profiles (SSC and FSC, respectively). Appropriate isotype control antibodies, "fluorescence minus one" staining and staining of negative populations were used to establish compensation parameters and sorting gates. Absolute cell numbers and frequencies were recorded. Data were analyzed using FlowJo software (FlowJo, Ashland, OR).

Data analysis
Data are expressed as mean ± SE. Intergroup differences were analyzed using a Student unpaired t test or one-way ANOVA, as appropriate. Differences were considered statistically significant for p < 0.05. Animals were randomly assigned to treatment and control groups and analysis was performed in a blinded fashion.

Results
CCR2 + Mo/MΦ accumulate in the ischemic brain before CX3CR1 + Mo/MΦ To investigate brain trafficking of monocytes after cerebral ischemia, we generated CX3CR1 GFP/+ CCR2 RFP/+ BM chimeric mice. GFP and RFP expression enables to simultaneously track CX3CR1 + Ly6C lo and CCR2 + Ly6C hi monocyte recruitment, while the generation of BM chimeras, allows for differentiation between infiltrating CX3CR1 GFP/+ monocytes and CNS-resident (CX3CR1 + ) microglia [25]. Histological analysis of brains of CX3CR1 GFP/+ CCR2 RFP/+ chimeras that underwent MCAo revealed a significant infiltration of monocytes in the ischemic brain (Fig. 1). CCR2 RFP/+ cells rapidly increased in the ischemic hemisphere 3 days after MCAo, evenly infiltrating the injured tissue, identified by Nissl stain of adjacent brain sections, with minimal penetration of the healthy tissue (Fig. 1a). During the subacute and chronic phase of injury (day 14-28 after MCAo), CCR2 RFP/+ cells were confined to the lesion core and their number decreased elsewhere (Fig. 1a). The number of CCR2 RFP/+ cells was highly dense in the lesion core 14-28 days after ischemia, which made it impossible to differentiate single cells, and therefore, this region was excluded for the quantitative analysis (see the "Methods" section; Fig. 1b). CX3CR1 GFP/+ cell numbers increased moderately in the first 7 days after MCAo and peaked at 14-28 days post-ischemia ( Fig. 1a-b). CX3CR1 GFP/+ cells delimited the CCR2 RFP/+ cell-laden core at 14 days after MCAo and were also found, albeit sparsely, throughout the entire lesion. This anatomical distribution remained unmodified 28 days after MCAo (Fig. 1a). In brains of mice undergoing sham surgery, CCR2 RFP/+ and CX3CR1 GFP/+ cells were scarce and localized mostly in the choroid plexus, olfactory bulbs and meninges (Fig. 1a). Moreover, the number of CCR2 RFP/+ CX3CR1 GFP/+ double-positive cells was homogenous among the sham and 1-2 days post-ischemia groups (20 ± 3 % CCR2 RFP/+ CX3CR1 GFP/+ of total CX3CR1 GFP/+ ) and increased 3-5 days after MCAo (33 ± 3 %) before significantly decreasing between 7 and 28 days after MCAo (6 ± 2 %, Fig. 1c).
Because irradiation of the head may induce an inflammatory response [26] that may alter immune cell infiltration, CX3CR1 GFP/+ CCR2 RFP/+ chimeric mice were also obtained by irradiation with head shielding (Additional file 1: Figure  S1). However, head shielding had no effect on brain distribution of CCR2 RFP/+ and CX3CR1 GFP/+ cells 3 and 14 days after stroke, when compared to chimeric mice without head protection (Additional file 1: Figure S1A-B).
Infiltrating CX3CR1 + cells but not CCR2 + cells displayed different morphological phenotypes CX3CR1 GFP/+ cells exhibited three distinct phenotypes 14 and 28 days after MCAo (Fig. 2a): (i) cells with an amoeboid morphology and no plasmalemmal processes, classified as amoeboid cells (left panel); (ii) cells with arborized processes, named ramified cells (central panel); and (iii) cells with elongated shape located along the vessels termed perivascular cells (right panel). Amoeboid CX3CR1 GFP/+ cells were often localized in the ischemic core and adjacent areas (95 ± 1.2 % of the total CX3CR1 GFP/+ cells), which were mainly populated by CCR2 + cells. Ramified CX3CR1 GFP/+ cells were predominant at the periphery of the lesion (85 ± 1.9 % of the total CX3CR1 GFP/+ cells) and were not seen in the core at 14 days after MCAo. Perivascular CX3CR1 GFP/+ cells were closely associated with blood vessels identified by the endothelial marker GLUT1 and were observed in both core (5 ± 1.0 %) and peri-infarct (5.8 ± 1.0 % of the total CX3CR1 GFP/+ cells) regions (Fig. 2a). These perivascular cells exhibited elongated profiles along the major axis of the vessel or encircled the entire vessels' circumference in vascular cross sections. Similar distributions were observed 28 days after MCAo (data not shown). Head shielding did not affect the phenotype distribution of CX3CR1 GFP/+ cells (Additional file 1: Figure 1D). In contrast, infiltrating CCR2 RFP/+ cells showed a uniform round shape without apparent processes, distinctive of  Blood vessels were immunostained with the endothelial marker GLUT1 (red). Scale bars represent 15 μm. Graph bars represent the frequencies for each described phenotype in the core (C) and periphery (peri-infarct; PI) of the lesion (n = 4 mice/group). b CCR2 RFP/+ cells (red), which are localized in the core of the ischemic lesion (a), have a round shape (arrowheads), while CX3CR1 GFP/+ CCR2 RFP/+ double-positive cells (arrows), have an amoeboid or spindle shape with lower level of RFP expression than CCR2 RFP/+ cells (a and b). CX3CR1 GFP/+ cells acquire a ramified phenotype as they localized in the periphery (b and c). Scale bars represent 20 μm amoeboid macrophages, and as described above, their distribution was limited to the ischemic core 14 days after MCAo (Fig. 2b). Some CCR2 RFP/+ CX3CR1 GFP/+ double-positive cells showed cellular processes (Fig. 2B), suggesting that CCR2 + cells may acquire a CX3CR1 phenotype in the tissue.
To investigate whether proliferation of brain recruited Mo/MΦ may also contribute to the increase in the number of accumulated Mo/MΦ over the time after stroke, we performed immunohistochemistry with the proliferation marker Ki67 (Fig. 3). Some CX3CR1 GFP/+ macrophages on the brain surface were positive for Ki67 (11.3 ± 2.1 % of total CX3CR1 GFP/+ cells) 3 days after MCAo (Fig. 3a) and 14 days after MCAo, Ki67 co-labeled with CX3CR1 GFP/+ Mo/MΦ macrophages localized either in the parenchyma or on the brain surface (Fig. 3a). However, at this time point, the number of CX3CR1 GFP/+ Ki67 + Mo/MΦ accounted for only 1.0 ± 0.3 % of total CX3CR1 GFP/+ cells (Fig. 3b). We did not observe CCR2 RFP/+ Ki67 + cells (Fig. 3c). Ki67 staining in head-shielded mice also revealed proliferation of CX3CR1 GFP/+ Mo/MΦ (Additional file 1: Figure S1C).

Infiltration of CCR2 + Ly6C hi and CCR2 − Ly6C lo cells in CCR2-or NR4A1-null mice after ischemic injury
To investigate whether the late surge of CX3CR1 + Mo/MΦ was due to active recruitment of circulating CX3CR1 + Ly6C lo monocytes or to transdifferentiation of invading CCR2 + Ly6C hi into CX3CR1 + Ly6C lo Mo/MΦ, we studied Mo/MΦ dynamics in CCR2 −/− mice, in which tissue recruitment of CCR2 + Ly6C hi monocytes is impaired [32] and in chimeric mice lacking NR4A1 −/− , a transcription factor required for the generation of CX3CR1 + Ly6C lo circulating monocytes [16]. These studies were conducted by flow cytometry using Ly6C expression as a surrogate marker for CCR2 + and CX3CR1 + Mo/MΦ identification. To validate this approach, we analyzed the blood of CX3CR1 GFP/+ CCR2 RFP/+ chimeric mice by flow cytometry. This analysis identified that GFP hi RFP lo cells corresponded to Ly6C lo monocytes (CD45 + Ly6G − CD11b + Ly6C lo cells), whereas GFP int RFP hi cells corresponded to Ly6C hi monocytes (CD45 + Ly6G − CD11b + Ly6C hi cells) (Additional file 4: Figure S4). NK cells and some CD8 cells also expressed RFP, as previously reported [25], but neutrophils did not express either RFP or GFP.
These results suggest that either Ly6C lo Mo/MΦ derived from infiltrating CCR2 + Ly6C hi monocytes or that parenchymal CCR2 + Ly6C hi Mo/MΦ are required for the infiltration of CX3CR1 + Ly6C lo monocytes to occur. To address this issue, we studied Mo/MΦ kinetics in mice deficient for the orphan nuclear receptor NR4A1 (Nur77). NR4A1 is essential for the generation of Ly6C lo monocytes and NR4A1 −/− mice lack circulating Ly6C lo monocytes [16]. Since NR4A1 has been implicated in neuronal function [35], we used NR4A1 −/− chimeras lacking NR4A1 in BM-derived cells to avoid confounding effects of deletion of NR4A1 in neurons. As expected, Ly6C lo monocytes were absent in the blood of  (Fig. 6a). However, the number of brain Ly6C lo Mo/MΦ significantly increased in NR4A1 −/− BM chimeric mice 14 days after MCAo compared to the number of Ly6C lo Mo/MΦ at 3 days post-MCAo (Fig. 6b), consistent with the hypothesis that infiltrating CCR2 + Ly6C hi cells transdifferentiate into CX3CR1 + Ly6C lo Mo/MΦ at later times.

Discussion
We have demonstrated that two main subsets of bloodderived macrophages accumulate in the post-ischemic brain within distinct temporal and spatial patterns. While CCR2 + Mo/MΦ spread through the ischemic lesion in the absence of CX3CR1 + Mo/MΦ at the early phase (3-5 days post-MCAo), the latter accumulated in the peri-infarct zone, encircling the lesion filled of CCR2 + Mo/MΦ at the late phase (14-28 days post-MCAo). Using mice lacking either CCR2 + Ly6C hi or CX3CR1 + Ly6C lo monocytes, we also provided evidence that CCR2 + Ly6C hi Mo/MΦ give rise to CX3CR1 + Ly6C lo Mo/MΦ independently of CX3CR1+ circulating monocytes. CX3CR1 + Ly6C lo Mo/ MΦ can adopt three different myeloid phenotypes in the ischemic brain depending on their localization: ramified phenotype reminiscent of an intermediate state of activated microglia in the peri-infarct, amoeboid phenotype consistent with phagocytic macrophages in the infarct region, and elongated perivascular macrophages associated with blood vessels.
Blood-derived macrophages accumulate sequentially in the ischemic brain The number of both CX3CR1 GFP/+ and CCR2 RFP/+ Mo/MΦ on days 1 and 2 after ischemia was low and similar to the number in brains of sham animals. These data indicate that sizable infiltration of macrophages in the brain does not occur before several days after the ischemic event, as previously described [4,36,37]. While CCR2 + Mo/MΦ predominated during the first 3-5 days after stroke, CX3CR1 GFP/+ Mo/MΦ number strikingly increased at 14 and 28 days. CCR2 + and CX3CR1 + Mo/MΦ accumulated in the ischemic brain with a distinctive biphasic profile. At day 3, we observed a drastic increase of CCR2 + Mo/MΦ, whereas CX3CR1 + Mo/MΦ only moderately increased. Nearly all CCR2 + Mo/MΦ were located in the infarct area, consistent with an active involvement in the inflammatory processes occurring in the territory most affected by ischemia. In contrast, the CX3CR1 + Mo/MΦ were mostly found at the infarct's periphery, both rostrally and caudally. By 7 days, CX3CR1 + Mo/MΦ outnumbered CCR2 + Mo/MΦ in the infarct's periphery and continued to increase reaching a maximum at 14 days. At 28 days, numbers of CX3CR1 + Mo/MΦ were still elevated, suggesting that Mo/MΦ participate in inflammatory processes also in the chronic phase of the injury.
Our data define two distinct regions within the chronic ischemic brain lesion: a cavitated core filled with CCR2 + and CX3CR1 + Mo/MΦ and a surrounding area of extensive neuronal loss heavily populated by CX3CR1 + but not CCR2 + Mo/MΦ.

Phenotypic differentiation of CX3CR1 + Mo/MΦ occurs surrounding the ischemic lesion
While CCR2 + Mo/MΦ retained a round/amoeboid phenotype, commonly associated with phagocytic activity [38] throughout the time period studied, CX3CR1 + Mo/MΦ displayed three distinct phenotypes: amoeboid, ramified and elongated perivascular. Like CCR2 + Mo/MΦ, amoeboid CX3CR1 + Mo/MΦ were predominantly located in the center of the lesion. Interestingly, we observed an increased numbers of double-positive cells in this area, possibly indicating that the conversion of CCR2 + into CX3CR1 + Mo/MΦ occurs in the lesion center. The fact that we did not observe CCR2 + Mo/MΦ with any other morphology than amoeboid suggests that the morphological polarization of Mo/MΦ is linked to CCR2 down-regulation and concomitant up-regulation of CX3CR1. Indeed, in human monocytes, CCR2 is down-regulated as the monocytes differentiate into macrophages [39]. With the exception of the necrotic center, ramified CX3CR1 + Mo/MΦ were evenly distributed throughout the infarct without any apparent regional preference. Due to their high Iba1 expression and cytoplasmic arborization, these cells were essentially indistinguishable from activated microglia. Many of CX3CR1 + Mo/ MΦ with elongated cell bodies were associated with cerebral blood vessels. These CX3CR1 + Mo/MΦ were strictly located on the abluminal side of the vessel and were also abluminal to pericytes and smooth muscle cells lining the vessel. Interestingly, CX3CR1 + Mo/MΦ did not cover all vessels uniformly but were associated with blood vessel that showed disruption of integrity as manifested by uneven or discontinuous smooth muscle cell coverage (Fig. 4c). While we can only speculate on their function, it is possible that vessel-associated CX3CR1 + Mo/MΦ support vascular integrity by physically strengthening the wall of vessels damaged by the ischemic injury. The recent findings that monocyte depleted mice showed a higher rate of hemorrhagic transformation after ischemic stroke support this view [8]. On the other hand, vessel-associated Mo/MΦ may participate in the removal of damaged vessels to allow vascular replacement by angiogenesis. While this possibility cannot be excluded, we did not see any intracellular inclusions positive for vascular markers (e.g. GLUT1, α-actin) that would indicate a phagocytic activity of these cells.
Infiltration of inflammatory CCR2 + Ly6C hi may give rise to CX3CR1 + Ly6C lo in the ischemic brain Several perivascular CX3CR1 + Ly6C lo Mo/MΦ were immunopositive for Ki67 3 days after MCAo but only few proliferative CX3CR1 + Ly6C lo Mo/MΦ, exhibiting mostly a ramified phenotype, were observed at later periods. Low rates of Mo/MΦ proliferation have also been reported in a mouse model of permanent cerebral ischemia [19]. Considering these low proliferation rates, it is more likely that the increased numbers of brain Mo/MΦ are the result of continuous recruitment of hematogenous monocytes than from cell proliferation in situ. Kokovay et al. [40] also reported that recruited bone marrow-derived cells into the brain expressing microglia markers were only occasionally proliferative. Interestingly, this study described a high number of Ki67 + cells associated with blood vessels 3-14 days after ischemia and between 10 and 20 % of the Ki67 + cells were identified as BM-derived pericytes. Contrary to the Kokovay study, however, we found that these perivascular cells expressed markers of the myeloid lineage but not pericytes, a discrepancy possibly due to the use of different pericyte markers (PDGFRβ in our study vs. desmin).
On the other hand, the increase in CX3CR1 GFP/ + CCR2 RFP/+ double-positive cells in the ischemic lesion before the rise of CX3CR1 GFP/+ single positive cells suggests that CX3CR1 + Ly6C lo derive from infiltrating CCR2 + Ly6C hi Mo/MΦ and not from a hematogenous source. This premise is supported by the observation that mice deficient in CCR2 did not exhibit significant numbers of both Ly6C hi and Ly6C lo Mo/MΦ in the ischemic brain, despite a normal number of circulating Ly6C lo monocytes. Furthermore, in chimeric mice lacking the transcription factor NR4A1 in BM-derived cells, Ly6C lo monocytes were drastically reduced in the circulation, but Ly6C lo Mo/MΦ were still observed in the brain 14 days after stroke. Therefore, the Ly6C lo Mo/MΦ present in the post-ischemic brain could not have originated from circulating Ly6C lo monocytes. These observations raise the possibility that CX3CR1 + Ly6C lo Mo/MΦ differentiate from CCR2 + Ly6C hi Mo/MΦ in situ, rather than originate from infiltrating CX3CR1 + Ly6C lo blood monocytes. Our data are consistent with the observations by Gliem et al. [8] of reduced accumulation of differentiated macrophages (identified as Ly6c lo F4/80 hi ) in CCR2 −/− mice 6 days after ischemia. Thus, CCR2 + Ly6C hi Mo/MΦ could be precursors for CX3CR1 + Ly6C lo Mo/MΦ from which they develop through environmental cues provided by the ischemic brain tissue, as has been recently suggested [19]. The signals and their target receptors responsible for this phenotypic shift remain to be determined.
Because CX3CR1 + Ly6C lo Mo/MΦ have been linked to post-ischemic repair in several organs including the brain [8,[41][42][43], our findings have important implications for therapies targeting the migration of inflammatory leukocytes into the ischemic brain. Indeed, depletion of circulating CCR2 monocytes impairs long-term spontaneous behavioral recovery after stroke [37], while Ly6C lo monocyte deficiency does not affect functional recovery and severity of injury after hypoxia/ischemia in mice [18]. All these findings suggest that inhibiting the recruitment of presumably proinflammatory CCR2 + Ly6C hi monocytes would also prevent the development of potentially protective CX3CR1 + Ly6C lo Mo/MΦ, which might affect repair processes during the subacute and chronic phase of ischemic brain injury. Therefore, therapies directed at interfering with monocyte recruitment into the ischemic brain must take into account the stage of evolution of the damage and the phenotype of the cells involved.

Conclusions
In conclusion, we have investigated the time course and phenotypic characteristics of monocytes recruitment into the post-ischemic brain. We found that "inflammatory" monocytes (CCR2 + Ly6C hi ) are recruited into the brain in the acute phase of the damage resulting in early accumulation of CCR2 + Ly6C hi Mo/MΦ. A secondary accumulation of CX3CR1 + Ly6C lo Mo/MΦ ensues weeks later, which is most likely the result of a phenotypic switch of CCR2 + Ly6C hi Mo/MΦ rather than recruitment of blood borne "patrolling" monocytes (CX3CR1 + Ly6C lo ). The morphological diversity of CCR2 + Ly6C hi and CX3CR1 + Ly6C lo Mo/MΦ suggests a diversity of roles in the development and resolution of the damage. Although the molecular cues driving the phenotypic switch and its impact on tissue outcome remain to be defined, the data provide new insight into the complexities of monocyte trafficking into the brain with significant implications for immune-based approaches to ameliorate ischemic brain injury by modulating the cellular bases of post-ischemic inflammation.