Brain perivascular macrophages: Recent advances and implications in health and diseases

Abstract Brain perivascular macrophages (PVMs) belong to a distinct population of brain‐resident myeloid cells located within the perivascular space surrounding arterioles and venules. Their characterization depends on the combination of anatomical localization, phagocytic capacity, and molecular markers. Under physiological status, they provide structural and functional support for maintaining brain homeostasis, including facilitation of blood‐brain barrier integrity and lymphatic drainage, and exertion of immune functions such as phagocytosis and antigen presentation. Increasing evidence also implicates their specific roles in diseased brain, ranging from cerebrovascular diseases, Aβ pathologies, infections, and autoimmunity. Collectively, PVMs are key components of the brain‐resident immune system, actively participate in a broad‐spectrum of processes in normal and diseased status. Details of the processes are largely underexplored. Targeting PVMs would lead to new insights and be a promising strategy for a broad array of human diseases.


| INTRODUC TI ON
Increasing evidence suggests a significant role of brain-resident immune cells in physiological and pathophysiological settings. 1 Aside from parenchymal microglia that are extensively studied, border-associated macrophages (BAMs), including perivascular macrophages (PVMs), meningeal macrophages (MGMs), and choroid plexus macrophages have implicated in variable processes such as development, homeostasis, and diseases. 2 As their name suggests, PVMs are brain-resident macrophages located in the perivascular space, also known as Virchow-Robin space (VRS). Unless peripheral monocytes/macrophages that originate from the mesoderm, PVMs originate from yolk sac and fetal liver embryonically. 2 Nevertheless, their characterization and their roles in normal and disease status are largely underexplored. In this review, we discuss their characterization focusing on clarifying some confusing anatomical concepts and providing practical suggestions for future research. We also summarize evidence supporting their significant roles in both physiological and pathophysiological states.

| PVM CHAR AC TERIZ ATION
It has been about four decades since the discovery of elongated cells in the perivascular space, which possessed autofluorescent large intracellular granules and can uptake exogenous trypan blue and horseradish peroxidase. 3 Since then, scientists gain more and more knowledge on their identity. Details on the history of PVM discovery and identification are comprehensively reviewed elsewhere. 4,5 Nowadays, people have come to a consensus that brain PVMs belong to a group of distinct myeloid cells located in the VRS between the vascular basement membrane (BM) on the abluminal side and glial limitans of the brain parenchyma ( Figure 1A), and modern PVM characterization relies on a combination of location, function, and molecular markers. Herein we would like to discuss the practical, sometimes confusing issues on these aspects.

| Anatomical localization of PVMs
There is controversy about the location of PVMs in relation to BM.
Some report that PVMs are positioned outside the BM, 3,6 while others report the opposite. 2,5 After careful review of these controversial results, we believe this inconsistency comes from the presence of two layers of BM in the brain vasculature. The inner BM, namely endothelial BM, underlies the endothelium, and the outer BM, namely vascular BM or parenchymal BM, forms the vascular abluminal barrier against glial limitans. [7][8][9][10] The two BM layers are clearly visible under electron microscopy (EM). 11 However, immunostaining can hardly illustrate the delicate structure in the current scenario. Both BM layers are positive for laminin 5 unless different laminin subtypes are stained. 12 In addition, smooth muscle staining patterns are sometimes irregular 13 ( Figure 1B, field a). Therefore, application of EM is much more accurate and trustable, and EM images consistently reveal that PVMs locate between the vascular/parenchymal BM and the glial limitans/brain parenchyma 3,5,6 ( Figure 1A).
Multiple studies have reported the presence of PVMs surrounding musculated vessels, known as arterioles 10-35 μm in diameter. 4,14,15 Whether PVMs exist in nonmusculated vessels has not been reported. Although Ookawara et al did mention the possible existence of PVMs surrounding venules, they only identified PVMs surrounding vessels processing continuous or discontinuous smooth muscle layer, excluding the possibility of capillaries. 15 As an important complement to the field, our group proves the presence of PVMs in both peri-arteriole space and peri-venule space ( Figure 1B) by showing PVMs surrounding both musculated arterioles and nonmusculated venules. Subsequent cell counting revealed more peri-arteriole PVMs than peri-venule PVMs in both cortical and subcortical regions in mouse brain, but no difference if normalized to vessel length ( Figure 1C). In addition, we did not observe any difference in PVM distribution with regard to vessel diameter as others reported ( Figure 1C). 4 A reasonable explanation for this discrepancy is that we did not include any pial vessels which are larger in size and harbor much more CD206 + cells than intracerebral vessels. 4

| Functional characterization of PVMs
The phagocytotic function of PVMs is utilized for PVM identification. For example, by intravenous (IV) injecting fluorescence labeled dextran, PVMs can be visualized due to their phagocytosis of the fluorescence signals. 6
CD163 is highly expressed in PVMs but not in microglia. 19,20 Another celltype that expresses CD163 is monocyte. 21 CD206 is a mannose receptor and another widely accepted marker for PVM. CD206 is not present in monocytes, making it more specific to PVMs than CD163. 22,23 A very small portion of microglia may also express weak CD206 under naïve state ( Figure 1D'', field a), and after brain injury such as stroke and brain trauma, a subpopulation of microglia and infiltrating macrophages transiently express high level of CD206. 24,25 Lyve-1 is a receptor for hyaluronan and is expressed on the lymphatic vessels and PVMs, but not monocytes or microglia, 26,27 making it a more specific PVM marker than CD163 and CD206 ( Figure 1D'', field b). However, Lyve-1 is not as sensitive as CD206 to identify PVMs In summary, there is no single marker defining PVMs with good sensitivity and specificity by far. Selection of PVM markers is dependent on which assay is used and which cell type is the most crucial to differentiate. For example, CD11b + CD45 high CD206 + is a good way for PVM isolation with flow cytometry. In immunostaining, CD163 is good to tell PVMs from microglia but not monocytes. CD206 is good to tell PVM from monocyte, but worse than Lyve-1 in distinguishing PVM from microglia. A most recent research using mass cytometry followed by single-cell RNA sequencing revealed CD38 combined with major histocompatibility complex (MHC)-II to be reliable molecular signatures of BAMs, 28 but needs further validation. In addition, a combination of anatomical localization and phagocytotic function is more reliable in PVM identification.

| Relation of PVMs with MGMs
BAMs consist of PVMs, MGMs, and choroid plexus macrophages. 2 Same as microglia, BAMs are derived from yolk sac and fetal liver embryonically. 2  F I G U R E 1 PVM characterization and distribution. A, Anatomical localization of MGMs and PVMs. PVMs are located in the Virchow-Robin space surrounding pial arterioles and penetrating arterioles. Arrow indicates the direction of blood flow. B, Peri-arteriole macrophages (arrows) and peri-venule macrophages (arrowheads). Field a (i-iv) shows an arteriole with a smooth muscle layer, and field b (i-iv) shows a venule lacking a smooth muscle layer. *Vessel lumen. Scale bar = 20 μm. C, Perivascular CD206 + cell numbers in cortical and subcortical regions of mouse brains (upper panel). Cell numbers normalized to vessel length in vessels of different sizes (lower panel). Data are mean ± SD, n = 3. *P < .05 vs peri-arteriole by unpaired t test. D, Lyve-1 is more specific for PVM than CD206. D' shows individual channels of an arteriole. Fields a (i-ii) and b (i-ii) in D'' indicate microglia (arrow, weak CD206, negative for Lyve-1) and PVM (arrowhead, strong CD206 and Lyve-1), respectively. *Vessel lumen. Scale bar = 20 μm. CSF, cerebrospinal fluid; MGM, meningeal macrophage; PVM, perivascular macrophage; BM, basement membrane; α-SMA, α-smooth muscle actin To our knowledge, by far, none of the PVM characterizing approaches discussed above could definitively distinguish PVM from MGM. First, both of them have phagocytotic activities and are visualized after injection of fluorescence labeled dextran. 13 Second, depletion methods such as CLO liposomes or Cx3cr1 knockout nonselectively deplete both of them, 30 and repopulation from bone marrow monocytes happens to both of them. 6,13,29 Third, they share the same set of markers, such as CD206, CD163, Lyve-1, and MHC-II. 4,29 As a result, anatomical localization seems to be the last resort to differentiate them, but only to a limited degree. When focusing on macrophages in VRS surrounding penetrating vessels, one can comfortably say PVM. Problem emerges when it comes to the pia, which holds up pial arterioles and MGMs parallelly. Since MGMs tend to evenly distribute on the pia surface, some MGMs may happen to overlap with pial arterioles while others may not. 4,31 As for those overlapping with pial arterioles, it is hard to define if they are MGMs or PVMs. Out of this concern, all the data in the present article exclude the pial vessels and only focus on penetrating vessels.
Single-cell RNA sequencing may be a promising tool to distinguish PVM from MGM, although existing studies fail to do so, 27,32 as they applied a large scope focusing on whole brain myeloid cells. Narrowing down the scope by focusing only on BAMs may provide useful information.

| PVM FUN C TI ON S IN THE NORMAL B R AIN
Outlined below are some of the functions carried out by PVMs in healthy adult brain. Many of them are also implicated in disease settings which will be discussed in detail in Section 3.

| BBB integrity
Under physiological conditions, PVMs contribute to BBB integrity.
proteins from the blood, and along with laminin layer, play a critical role in restricting macromolecules over 10 kDa into the brain. 33 Similarly, PVMs in peripheral organs restrict vascular permeability, such as mesentery arteries 34 and cochlear. 35 Comparable finding is also reported in an in vitro vestibular BBB co-culture system consisting of endothelial cells (ECs), PVMs, and pericytes. 36 Interestingly, PVMs may lead to BBB disruption in diseased status. Details are discussed under Section 3.1.

| Phagocytosis
The existence of CD163 + "perivascular microglia" has long been observed in both rodents 11,37 and humans. 19 Nowadays, the phagocytic ability of PVMs is broadly accepted. 4 Interestingly, this ability can be restored in PVMs that are repopulated from bone marrow replenishment after experimental PVM depletion. In a bone marrow chimera in which irradiated recipient mice received bone marrow transplantation of green fluorescent protein (GFP)-labeled monocytes, perivascular GFP + donor cells could be observed as early as 2 weeks after bone marrow transplantation. 40 In addition, phagocytosis-dependent labeling of donor PVMs was observed 4 hours after ICV injection of biotin and rhodamine-labeled dextran (10 kD). 40

| Antigen presentation
Expression of MHC class II is a key feature of antigen presenting cells (APCs). 41 Back to 1988, MHC-II-expressing "perivascular microglia" have been found to present antigen to lymphocytes in an experimental allergic encephalomyelitis (EAE) model. 42 MHC-II-expressing PMVs have also been observed in mouse pineal gland at naïve status, 43 which are more evident in diseased status, such as mouse retina treated by interferon γ and CD40, 44 rat brain after transient middle cerebral artery occlusion (MCAO), 45 mouse EAE brain, 42 and autopsy from multiple sclerosis (MS) human brain. 22

| Lymphatic clearance
Four lymphatic clearance pathways have been identified in the brain, namely olfactory pathway, meningeal pathway, glymphatic pathway, and intramural perivascular pathway. 46 Given their close relationship to vessels, PVMs may facilitate the latter two pathways.
First reported by Iliff et al, 47 the glymphatic pathway involves the "paravascular space," which drains interstitial fluid and cerebrospinal fluid (CSF) from the para-arterial space via glial parenchyma to the para-venous space, finally into the internal cerebral vein. As a lowresistant drainage pathway, pulsatile paravascular flow generated by the cardiac cycle was observed. 48 Of note, the so-called "paravascular space" referred to by Iliff is located between the glial limitans and the vascular BM, which is actually VRS where PVMs reside.
Carare et al 49 reported the existence of another pathway that involves intracerebral arteries, termed the intramural perivascular drainage (IPAD) pathway. As the term "intramural" suggests, IPAD involves a route within the vessel wall, to be specific, the tunica media which consists of vascular smooth muscle cells (VSMCs). They find that tracers injected into the caudate putamen could gain access into the arterial wall, and travel along the intercellular spaces among VSMCs. PVMs facilitate IPAD clearance by taking up particles ranging from 2 nm to 1 μm. 49 Interestingly, the continuous flow of IPAD pathway relies on VSMC contraction rather than the arterial pulsation from the heart. 50 PVMs play an important role in tuning the vascular tone, as ablation of PVMs result in increased regional blood flow, probably through VSMC relaxation. 6,13 Therefore, PVMs may help control the velocity of IPAD through regulating VSMC constriction and relaxation. Of note, the velocity of IPAD is much slower in aged mice. 51 In summary, PVMs may facilitate lymphatic drainage by two means. First, as they directly access to lymphatic routes, such as paravascular space and intramural perivascular space, they can phagocytose large particles. Second, PVMs may facilitate IPAD indirectly through regulation of VSMC tone and subsequently regulate IPAD velocity. The exact role of PVMs in lymphatic clearance and their alteration with aging need to be addressed in future research.
Taken together, brain PVMs are located at the brain-peripheral interface with direct contact to the CSF and the brain parenchyma.
They express SRs and exert phagocytosis, acting as immune surveillant and APCs, providing structural and function support to the BBB and lymphatic clearance. These roles are important for brain homeostasis.

| PVM ALTER ATI ON S IN D IS E A S ED B R AIN
In human brains, increased CD163 + perivascular cells were found increased in autopsies from traumatic brain injury, intracerebral bleeding, ischemic stroke, and cerebral hypoxia patients, 52 indicating that PVMs are active players in disease processes. In this section, we discuss the most extensively studied pathological conditions, where PVMs seem to be able to modify the disease progress (Table 1).

| BBB disruption
Several studies described PVM accumulation after BBB injury in various disorders. For example, seizure patients were found to have more CD163 + PVMs than controls. And in rats with experimental seizure with BBB leakage, the severity of BBB leakage is positively correlated with CD163 + PVM cells. 53 In a mouse BBB injury model induced by transgenic expression of HIV-1 Tat 1-86 protein, phagocytic PVMs were increased by 5-fold in caudate/putamen. 54 In a rat traumatic injury model, CD163 + PMVs were increased for at least 4 days after. 55 In retina, PMV migration to the lesion site is evident after osmotic injury to the blood-retina barrier. 56 The role of PVMs in BBB permeability is complicated. It is possible that PVMs facilitate BBB integrity under physiological conditions, while participate in BBB disruption under diseased status. 57 Some evidence is discussed under Section 2.1 in this regard. Another possibility is that PVM may sense BBB leakage and help to seal the hole, since microglia and microphages were reported to do so. 58,59 Further studies should focus to solve this dilemma.   (Figure 2A,B), suggesting the existence of PVM subpopulation. 2VS led to increased parenchymal myeloid cells ( Figure 2B), along with increased PVM numbers in subcortex but not cortex ( Figure 2B,C), probably responsible for more severe white matter injury than gray matter in dementia. CD16/32 + PVMs increased in number and percentage after 2VS, while Lyve-1 + PVMs decreased, suggesting a subpopulation shifting ( Figure 2B,D,E).

| Cerebrovascular diseases and risk factors
Among CD206, CD16/32, and Lyve-1, CD206 is the most sensitive, and Lyve-1 is the most specific, which does not label parenchymal microglia/infiltrating microphages as CD206 and CD16/32 do even after brain injury ( Figure 2B, field a). In summary, we report the PVMs also play a role in vascular biology. In developing retina, PVMs could inhibit excessive angiogenesis by inducing EC apoptosis. 65,66 Interestingly after retinal vein occlusion, monocyte-derived PVMs accumulated to the lesion site and were protective against EC apoptosis. 67 In multiple tumors, PVMs were reported to support angiogenesis and maintain tumor microenvironments, partially via secreting microvesicles. [68][69][70] Although the role of PVMs on vascular biology in cerebrovascular diseases remains underexplored, it is implicated that PMVs are actively engaged in vascular biology in both disease pathogenesis and tissue repair. Future study should address its role specifically in cerebrovascular diseases and related risk factors.

| Aβ pathology
PVM's phagocytic function is critical in mitigating Aβ pathology, such as Alzheimer's disease and cerebral amyloid angiopathy (CAA). In one study using TgCRND8 mice (overexpression of human Aβ could also lead to blunted cerebrovascular responses. 71,72 PMVs were reported to participate in Aβ-induced neurovascular dysfunction through CD36 mediated oxidative stress. Park et al 13 applied three models of Aβ pathology, Aβ topical perfusion onto brain cortex, Aβ intravenous administration, and Aβ overexpression in Tg2576 mice (mice carrying Swedish mutation of APP). All three models were associated with neurovascular dysfunction as detected with whisker stimulation and acetylcholine superfusion.
Additionally, selected deletion of CD36 and NOX2 in PVMs by bone marrow chimera could abolish Aβ-mediated neurovascular dysfunction, and wild-type PVMs could re-establish neurovascular dysfunction in CD36 −/− mice. 13 However, clinical outcome of the mice, such as cognitive function and neurobehavioral performance were not evaluated in this study. Collectively, these findings suggest a complicated, bidirectional roles of PVMs in Aβ pathology. SRs, such as CD36 and SR-B1, may be crucial in Aβ clearance by PVMs.

| Multiple sclerosis
In human MS brain specimens, both CD163 + PVMs and MGMs were increased in numbers, especially in the center of acute lesions and the edge of chronic lesions. These cells were HLA-DR + suggesting a possible role in antigen presentation. 82

| FUTURE D IREC TI ON S
Current difficulties of PVM research include lack of specific markers and inability to separate PVMs from MGMs. Application of RNA sequencing, protein sequencing, and mass cytometry may provide more clues in the future. In addition, intramural perivascular space and VRS collapse after death, bringing extra difficulty studying PVMs and lymphatic routes in dead tissue. 86 Live imaging such as two-photon and multi-photon microscopy can be a powerful tool in this regard.
Although much progress has been achieved during past decades, is any interaction between them and brain parenchymal cells. Other features such as regional distribution pattern, renewal, gender difference, and aging remain unexplored. Under most diseased status, it seems that they act as double-edged swords. Considering the growing body of evidence showing their significant roles in both healthy and diseased status, in-depth studies on PVMs could lead to new insights in our understanding of CNS biology and human diseases.

ACK N OWLED G M ENTS
This work was supported in part by a grant from the National Institutes of Health (NS103810). We thank Pat Strickler for the administrative support.

CO N FLI C T O F I NTE R E S T
The author declare no conflict of interest.