CNS border-associated macrophages in the homeostatic and ischaemic brain

CNS border-associated macrophages (BAMs) are a small population of specialised macrophages localised in the choroidplexus,meningealandperivascularspaces.Untilrecently,thefunctionofthiselusivecelltypewaspoorly understood and largely overlooked, especially in comparison to microglia, the primary brain resident immune cell. However, the recent single cell immunophenotyping or transcriptomic analysis of BAM subsets in the homeostatic brain, coupled with the rapid emergence of new studies exploring BAM functions in various cerebral pathologies, including Alzheimer ’ s disease, hypertension-induced neurovascular and cognitive dysfunction, and ischaemic stroke, has unveiled previously unrecognised heterogeneity and spatial-temporal complexity in BAM populations as well as their contributions to brain homeostasis and disease. In this review, we discuss the implications of this new-found knowledge on our current understanding of BAM function in ischaemic stroke. We ﬁ rstprovideacomprehensiveoverviewanddiscussion ofthecell-surfaceexpressionpro ﬁ les,transcriptional signaturesandpotentialfunctionalphenotypesofhomeostaticBAMsubsetsdescribedinrecentstudies.Evidence for their putative physiological roles is examined, including their involvement in immunological surveillance, waste clearance, and vascular permeability. We discuss the evidence supporting the accumulation and genetic transformation of BAMs in response to ischaemia and appraise the experimental evidence that BAM function might be deleterious in the acute phase of stroke, while considering the mechanisms by which BAMs may in ﬂ u-ence stroke outcomes in the longer term. Finally, we review the therapeutic potential of immunomodulatory strategiesasanapproachtostrokemanagement,highlightingcurrentchallengesinthe ﬁ eldandkeyissuesrelat-ing to BAMs, and how BAMs could be harnessed experimentally to support future translational research.


Introduction
The consequences of cerebrovascular disease are among the leading health issues.Ischaemic stroke is perhaps the best-known consequence (Campbell et al., 2019;Hu et al., 2017); however, despite the identification of drugs that target the mechanisms of ischaemic brain injury in experimental stroke models, clinical trials have thus far failed to present the same positive results yielded from animal studies (Lourbopoulos et al., 2021).This highlights the importance of developing novel therapeutics for stroke intervention, which will require the further identification of targets implicated in stroke pathogenesis and brain repair.
Microglia constitute the primary resident immune cell in the brain and their contribution to the immune response after ischaemic stroke is rapidly being uncovered (Ma et al., 2017).By contrast, our understanding of the roles of non-parenchymal central nervous system (CNS) border-associated macrophages (BAMs) is only beginning to emerge.Relative to parenchymal microglia, BAMs are a small immune cell population comprising only ~10% of total leukocytes in the homeostatic murine brain (Mrdjen et al., 2018).There are three main types of BAMs named after their localisation at anatomical borders between the periphery and CNS -perivascular macrophages, meningeal macrophages, and choroid plexus macrophages.However, recent evidence indicates that BAMs are in fact a highly heterogenous population of immune cells that comprise of multiple subtypes with distinct transcriptional signatures (Van Hove et al., 2019).Importantly, the positioning of BAMs at CNS borders rather than in the parenchyma suggests that this immune cell population contributes to brain homeostasis and disease in distinct ways to that of microglia.
Although our understanding of the roles of BAMs still lags behind that of microglia, there has been a recent and rapid emergence of studies exploring the functions of this enigmatic immune cell population in various diseases affecting the brain including Alzheimer's disease and cerebral amyloid angiopathy (CAA) (Hawkes & McLaurin, 2009;Park et al., 2017), hypertension-induced neurovascular and cognitive dysfunction (Faraco et al., 2016a;Iyonaga et al., 2020;Kerkhofs et al., 2020;Santisteban et al., 2020), and ischaemic and haemorrhagic stroke (Pedragosa et al., 2018;Rajan et al., 2020;Wan et al., 2021;Zheng et al., 2021).In this review we will discuss the implications of our emerging knowledge of the specialised roles of BAMs in stroke.More specifically, we will: (i) Discuss our current understanding of the genetic signature, potential functional phenotypes, and proposed physiological roles for BAMs subtypes in the homeostatic brain; (ii) Examine the responses of BAMs to ischaemic stroke and discuss emerging clinical and experimental evidence for roles for BAMs in stroke outcomes including a coverage of the potential putative mechanisms involved; (iii) Finally, we will present an appraisal of the therapeutic potential and challenges of targeting BAMs for therapeutic benefit.

Integral role of inflammation in ischaemic stroke
Although the consequences of inflammation in the ischaemic brain manifest in the days and weeks following stroke onset (Iadecola & Anrather, 2011;Planas, 2018), the immune response occurs immediately following artery occlusion.Indeed, the inflammatory process begins in the intravascular compartment where hypoxia, changes in shear stress due to stagnant blood, and the production of reactive oxygen species (ROS) triggers the activation of platelets and cerebral endothelial cells (Iadecola & Anrather, 2011).This is followed by the rapid deployment of adhesion molecules including P-and E-selectin, intercellular adhesion molecule-1 (ICAM-1), and integrins on platelets, endothelial cells, and leukocytes (Yilmaz & Granger, 2010).The release of pro-inflammatory mediators and excess ROS production by injured cells triggers blood-brain barrier (BBB) disruption which together with the expression of adhesion molecules facilitates the infiltration of circulating leukocytes into the ischaemic brain.Within the ischaemic parenchyma itself, the brain's innate immune system is also rapidly activated by danger-associated molecular patterns (DAMPs) released from injured and dying brain cells (Kono & Rock, 2008) (see Section 3).DAMPs can enter the systemic circulation via the disrupted BBB or meningeal lymphatic vessels where they activate the adaptive immune system in primary and secondary lymphoid organs (Liesz et al., 2015).In addition, CNS antigens are similarly released into the circulation where they prime the adaptive immune system (Becker, 2009;Urra et al., 2014).Collectively, these events lead to a spatiotemporaldependent infiltration and accumulation of peripheral-derived immune cells in the ischaemic brain, which include granulocytes, dendritic cells, bone marrow (BM)-derived monocytes/macrophages, and T and B lymphocytes (Brait et al., 2012;. Chu et al., 2014;Gelderblom et al., 2009;Kleinschnitz et al., 2010).Importantly, although inflammation contributes to ischaemic brain injury it is also considered essential for repair which can persist for weeks and months (Iadecola & Anrather, 2011).Also, activation of the immune system is followed by a state of systemic immunodepression and resultant increased susceptibility to infections, which leads to considerable mortality after stroke (Corrado et al., 2006;Kimura et al., 2005;Shim & Wong, 2018).The roles of the various immune cell populations in ischaemic brain injury and repair have been comprehensively covered by excellent review articles (Iadecola & Anrather, 2011;Planas, 2018).However, here we will provide a briefly overview of the roles of BM-derived monocytes/macrophages.
In mice, circulating BM-derived monocytes exist as two populations characterised by their expression of antigen complex Ly6C.Monocytes expressing Ly6C hi and the chemokine receptor CCR2 are regarded as 'inflammatory', whereas Ly6C lo CX3CR1 + monocytes are 'patrolling or anti-inflammatory' (Geissmann et al., 2003).Although now recognised as an oversimplification, macrophages were historically classified into two main types -M1 ('pro-inflammatory') and M2 ('alternative antiinflammatory') macrophages.M1 macrophages are typically identified by their expression of inducible nitric oxide synthase (iNOS), CD32, CD16, major histocompatibility complex class II (MHCII) and costimulatory molecules CD80 (B7-2) and CD86, and the production of pro-inflammatory cytokines TNF-α, IL-1β, and IL-6, whereas M2 are characterised by the expression of the pathogen pattern recognition receptor CD206 (also known as mannose receptor C type 1 [MRC1]), arginase-1 (ARG-1), and Ym1/2, and the production of antiinflammatory cytokines IL-10 and TGF-β (Mantovani et al., 2005).Most studies investigating the contribution of Ly6C hi monocytes in ischaemic brain injury have exploited their dependence on the CCR2/CCL2 axis for infiltration into the brain (Kuziel et al., 1997;Lu et al., 1998;Tsou et al., 2007).Early studies showed that global CCR2 genetic deficiency reduced infarct size in mice after ischaemia-reperfusion (Dimitrijevic et al., 2007).However, more recent studies using pharmacological inhibitors or anti-CCR2 antibodies show that global CCR2 inhibition exacerbates acute ischaemia-reperfusion brain injury and haemorrhagic transformation, impairs long-term functional recovery, and increases pro-inflammatory cytokine expression in mice after ischaemic stroke ( Chu et al., 2015;Gliem et al., 2012;Wattananit et al., 2016).Also, a more recent study using mice with CCR2-deficient myeloid cells demonstrated that although CCR2 expressing monocytes contribute to the acute inflammatory response following ischaemiareperfusion, they may also play roles in brain repair (Pedragosa et al., 2020).Notably, these different functions coincide with an initial prevalence of Ly6C hi CCR2 + CX3CR1 low/− monocytes in the acute phase followed by a shift towards a LyGC lo 'anti-inflammatory' phenotype with different expression levels of CCR2 and CX3CR1.The protective or reparative roles of monocytes may occur, at least in part, by promoting M2 macrophage polarisation ( Hu et al., 2012).For example, in the acute phase (24 h after ischaemia-reperfusion), genes indicative of the M2 phenotype are reduced in monocyte CCR2-deficient mice (Pedragosa et al., 2020), and global CCR2 inhibition prevents the typical upregulation of M2 markers (Chu et al., 2015).Whole-genome RNA sequencing data of the murine ischaemic brain also supports the view that infiltrating monocytes/macrophages undergo genomic reprogramming in the subacute phase after permanent ischaemia acquiring a transcriptome favouring inflammation resolution and neurovascular repair (Wang et al., 2020).BM-derived monocytes/macrophages cooperate with other immune cells in the ischaemic brain including microglia (Perego et al., 2016), which have protective or reparative roles.Whether monocytes/macrophages similarly interact and/or influence the phenotype of BAMs in the post-stroke milieu is unknown.Notably, however, there is evidence that monocytes may give rise to or acquire features of perivascular macrophages (pvMΦ) in models of neuroinflammation (Audoy-Rémus et al., 2008), suggesting interactions between the two immune populations.Similarly, as will be discussed in Section 5.1, CX3CR1 + monocytes may infiltrate perivascular spaces and acquire features of pvMΦ in the ischaemic brain (Rajan et al., 2020).

Microglia
Microglia are derived from early erythromyeloid precursors in the yolk sac, colonise the CNS during early development, and can be replenished throughout life by self-renewal (Ginhoux et al., 2010;Utz et al., 2020).Microglia display a highly ramified morphology characterised by a small soma and fine processes.Historically, they were termed as 'resting' in the physiological state; however, although they are non-motile, they are highly dynamic and constantly survey the brain parenchyma via their fine processes (Ginhoux et al., 2010;Nimmerjahn et al., 2005).Extensive evidence indicates that microglia can have both beneficial and deleterious roles following ischaemic stroke (Ma et al., 2017;Qin et al., 2019).In brief, although microglia are one of the first immune cells to respond to ischaemic injury their activation can occur in all phases of an ischaemic stroke (Rupalla et al., 1998;Schilling et al., 2003).Immediately following artery occlusion, microglia detect changes in ionic gradients, neuronal activity, and the release of DAMPs from injured or dying cells in the boundary zone of the infarct via their processes and Toll-like receptor family of receptors.Activation of microglia by DAMPs occurs via the NF-κβ pathway, which can also trigger their proliferation and migration to the ischaemic lesion (Davalos et al., 2005;Zhao et al., 2018).Also, microglia rapidly release pro-inflammatory mediators, which facilitate the recruitment of peripheral-derived immune cells via activation of the endothelium (Uno et al., 1995).Early efforts classified the activated microglia into two major phenotypes that are akin to the classical phenotypes of macrophages: 'pro-inflammatory/neurotoxic M1 microglia' and 'anti-inflammatory/reparative M2 microglia'.However, it is now understood that their polarisation states are much more diverse than previously considered.Indeed, microglia express diverse transcription profiles in response to pathological conditions, suggesting a variety of intermediate microglia phenotypes beyond the M1-M2 paradigm (Deng et al., 2020;Guo et al., 2021;Zheng et al., 2021).It is broadly accepted, however, that the transitions of these 'M1/M2' phenotypes are closely related to brain injury and repair (Hu et al., 2012).For example, 'M1 microglia' secrete pro-inflammatory cytokines such as TNF-α, IL-1β, and IFN-γ, as well as cytotoxic substances including ROS and excess levels of NO, which contribute to inflammation and cell injury after stroke.M2 microglia on the other hand express surface antigens such as CD206, which is involved in pathogen recognition and receptormediated endocytosis, CD164, and ARG-1, and are involved in phagocytic clearance of cell fragments and dead cells.Furthermore, they produce anti-inflammatory cytokines such as IL-4, IL-10, and TGF-β which reduce inflammation and promote neuronal survival.

CNS border-associated macrophages (BAMs)
The brain contains diverse border regions which form barriers and interfaces between the brain and the periphery, namely the perivascular space, meninges (comprising of the dura, arachnoid, and pia maters), and choroid plexus.Each of the borders have distinct roles in regulating CNS homeostasis and exhibit their own degree of 'immune privilege' (Engelhardt et al., 2017).They contain a diverse array of resident or patrolling immune cells, including BAMs, BM-derived macrophages, T and B lymphocytes, and dendritic cells, thereby representing the first line of defence (Korin et al., 2017;Mrdjen et al., 2018;Van Hove et al., 2019).Indeed, it is proposed that most of the diversity of the brain's immune compartment is restricted to its border regions (Van Hove et al., 2019).
There are three classical types of BAMs aptly named after their border localisation (Faraco et al., 2017).Specialised perivascular macrophages (pvMΦ) primarily reside in the perivascular space surrounding penetrating and parenchymal cerebral vessels larger than 20 μm but can also be found around pial and subpial vessels, and venules in the peri-venule space (Faraco et al., 2016b;Ookawara et al., 1996;Yang et al., 2019).In comparison to highly ramified microglia, pvMΦ are elongated cells with a relatively simpler morphology, allowing them to wrap around vessel walls (Nayak et al., 2012).Like microglia they appear to be non-motile but continuously extend and retract their protrusions along the perivascular space (Barkauskas et al., 2013;Goldmann et al., 2016).Meningeal macrophages (mMΦ) populate the pia and dura mater of the meninges border, where they sit alongside pial and subpial vessels.Like pvMΦ, mMΦ in the pia mater are elongated with a simple morphology but higher motility compared to both microglia and pvMΦ, whereas dura mMΦ have a bipolar morphology with more dendrites (Kierdorf et al., 2019;McMenamin et al., 2003) and align with blood vessels in the dura mater but are separated from the vessels by pericytes and fibroblast (Sato et al., 2021).Similarly to pvMΦs, mMΦ use their processes to constantly survey the meningeal space (Nayak et al., 2012) but only a third of all mMΦ display motility in the uninflamed brain (Goldmann et al., 2016).PvMΦ and mMΦ are often co-localised on the pia mater, therefore it is challenging to distinguish between them based on morphology alone.Lastly, choroid plexus macrophages (cpMΦ) reside in the choroid plexus and are characterised by a more compact shape with less elongations, and are best explored through sagittal brain sections (Goldmann et al., 2016).
The origin of BAMs is discussed extensively in recent reviews (Prinz et al., 2017) and will only be mentioned briefly here.BAMs were originally thought to be populated postnatally and constantly replaced by BM-derived monocytes.More recently, the employment of genetic fate-mapping approaches and single-cell transcriptomics has revealed that like microglia, BAMs originate from yolk sac-derived erythromyeloid progenitors (Goldmann et al., 2016;Prinz et al., 2017;Utz et al., 2020).For example, to examine whether embryonic progenitors contribute to the pool of mature BAMs, Goldmann et al. used the Cx3CR1Cre ER Rosa26 YFP fate-map system (Yona et al., 2013).This seminal study showed that following a single injection of tamoxifen at embryonic day 9, YFP labelling persisted in microglia, pvMΦ, and mMΦ of young (6-week-old) mice, whereas YFP labelling was progressively lost in cpMΦ.Thus, pvMΦ and mMΦ populations in the steady state CNS are primarily maintained throughout life by self-renewal with minimum replenishment by haematopoietic stem cell (HSC)-derived progenitors (Goldmann et al., 2016).However, they require macrophage colony-stimulating factor-1 receptor (CSF1R) signalling for their survival, proliferation, and differentiation (Ginhoux et al., 2010).Interestingly, deletion of the highly conserved super enhancer fms-intronic regulatory element (FIRE) found on the Csf1r locus depletes microglia without affecting pvMΦ/mMΦ populations, suggesting that transcriptional and epigenetic mechanisms differentially regulate CSF1R in microglia and pvMΦ/mMΦ (Rojo et al., 2019).Recent evidence suggests that not all mMΦ are self-renewing.Indeed, pulse-chase labelling in Cx3cr1 CreER :R26-YFP mice to assess input from HSC-derived progenitors, or Flt3 Cre :R26-YFP mice to distinguish between embryonically and HSCderived progenitors, showed that although dura mMΦ are derived from yolk sac progenitors, they are gradually replaced by BM-derived precursors (Van Hove et al., 2019).Similarly, despite their embryonic origins, findings by Goldman et al., and others indicate that a proportion of cpMΦ are replaced via CCR2 dependent recruitment of HSC-derived monocytes, most likely Ly6C hi CCR2 + monocytes, rather than selfreplacement (Chinnery et al., 2010;Van Hove et al., 2019).Thus, tissue accessibility is a likely major determinant for the replenishment of BAMs.Indeed, both the dura mater and choroid plexus contain fenestrated blood vessels and, thus, are more permeable than the pia mater and perivascular space.

Surface protein expression patterns and genetic signature
Until very recently, BAMs were characterised and discriminated from microglia by their cell surface marker expression, localisation, and/or morphology using the classical FACS based approach or immunohistochemistry methods.They were often identified by their high expression (relative to CD45 lo microglia) of CD45, major histocompatibility complex class II (MHCII) and co-stimulatory molecules (CD86 and CD40), expression of surface markers shared by all CNS myeloid cells -including the fractalkine receptor CX3CR1, CD11b, the glycoprotein F4/80, and tyrosine-protein kinase Mer (MeTK) -as well as low expression of Iba-1 relative to microglia.Furthermore, BAMs have been identified by their expression of the pathogen pattern recognition receptors CD206 and C-type lectins dendritic cell-specific ICAM-3 grabbing nonintegrin (DCSIGN), the phagocytic scavenger receptors CD36, CD163 and CD169, and the lymphatic vessel endothelial hyaluronan receptor 1 (LYVE-1) (Faraco et al., 2016b;Park et al., 2017;Pedragosa et al., 2018;Rajan et al., 2020).For immunohistochemistry methods, the distinct patterns of BAM localisation, their relatively simple morphology, and phagocytic activity have also been used as a surrogate for their discrimination from microglia (Schwabenland et al., 2021;Swanson et al., 2020).These properties, together with their cell surface epitope expression, led to the identification of the three main BAM subsets (pvMΦ, mMΦ and cpMΦ), as well as the beginnings of research to better understand their functional roles.However, as discussed below, the use of multiparameter single cell immunophenotyping or transcriptomic analysis, albeit in rodent rather than human tissue, has revealed that BAMs are in fact much more heterogenous than previously considered.Also, BAMs exhibit surface protein expression patterns and core gene signatures that are distinct from microglia.Furthermore, whilst more research is clearly needed, there is evidence of clear border-specific gene signatures suggesting distinct functional specialization tailored to the needs of environment.
Cell surface phenotyping has shown that although BAMs have an expression profile that closely resembles microglia; CD45 + CD11b lo F4/80hi CD64 hi MeTK + CX3CR1 + , they are discriminated from microglia by their relatively higher expression of CD45, lower relative expression of Iba-1, higher levels of CD38, the absence of the classical microglia marker SIGLEC-H, and the expression of CD206, CD169, and LYVE-1 (Mrdjen et al., 2018) (Fig. 1).Notably, some BAMs express lower levels of CD45, thus, making it challenging to distinguish these BAMs from microglia using the CD45 hi versus CD45 lo/int FACS gating strategy.The same study has also uncovered that at least four distinct subsets reside in the steady-state CNS, distinguished by their differential cell surface expression of CD38 or LYVE-1, MHCII, and CCR2.The largest BAM subset (~75% of all BAMs) is positive for CD38 or LYVE-1, but negative for MHCII, indicating a large proportion of BAMs do not constitutively express MHCII.This is followed by a second smaller subset that is identified by the expression of CD38 or LYVE-1, and MHCII.Notably, both subsets lack CCR2 suggesting they primarily comprise of pvMΦ and mMΦ (except for dura mMΦ), which undergo self-renewal with minimum replenishment by CCR2 expressing progenitors.The final two BAM subsets are almost identical (CD38 − LYVE-1 − MHCII + CCR2 + ) apart from differing expression levels of CCR2 and CD44, the latter of which was previously reported to be selectively expressed on infiltrating immune cells but absent from resident myeloid cells in the naïve murine brain (Korin et al., 2017).Lastly, all subsets express macrophage markers CD206, CD169, and CD64; however, levels of CD206 are highest in the largest subset, which lacks MHCII, CCR2, and CD44.Interestingly, some BAM subsets appear to be enriched in distinct CNS border locations.Indeed, whereas single positive CD38 or LYVE-1 BAMs are found in all border regions, those that were double positive for LYVE-1 (or CD38) and MHCII are enriched in the pia mater and perivascular space.Also, the dura mater is enriched with relatively more singlepositive MHCII BAMs but very few double positive BAMs (LYVE-1 + MHCII + ).Notably, only LYVE-1 − MHCII + BAMs, enriched in the dura mater and choroid plexus, are positive for CCR2, consistent with the mixed renewal of cpMΦ and dura mMΦ discussed in Section 3.2.Lastly, evidence indicates that a large proportion of pial BAMs are often CX3CR1 negative or low, suggesting pial BAMs may not be efficiently targeted by the commonly used Cx3CR1 CreER Rosa26 fate-map system (Yona et al., 2013), mentioned in Section 3.2.
In addition to the identification of a BAM core gene signature, sets of genes have been highlighted that are either shared between the various BAM subsets or are unique for a specific subset (Van Hove et al., 2019).For example, subtypes with low expression of MHCII (sd-BAM, cp lo -BAM, and d lo -BAM) exhibit differentially expressed genes compared to the MHCII hi subsets (cp hi -BAM and d hi -BAM), and clear border-specific signatures exist for BAMs in the subdural meninges, dura matter, and choroid plexus.Furthermore, consistent with the MHCII + CCR + BAM subset identified by cell surface phenotyping (Mrdjen et al., 2018), MHCII expressing BAMs found predominately in the dura and choroid plexus express transcriptional gradients of monocyte signature genes (e.g., Ccr2) that correspond with their transition from a monocyte to a mature BAM.

Proposed physiological roles
There has been an emergence of studies exploring the physiological roles of BAMs (Fig. 2), however, the breadth of their roles remains to be fully elucidated.Studies thus far have focussed on the roles of pvMΦ and mMΦ, whereas very little is known about cpMΦ.Nevertheless, as discussed below, their position at CNS borders together with their phagocytic activity (Graeber et al, 1989;Kida et al., 1993;Santisteban et al., 2020), and ability to extend and move their processes along border spaces, collectively infer BAMs have roles in monitoring and filtering cerebrospinal fluid (CSF), and in the scavenging of CNS-derived metabolites and antigens, and invading microbes (Mato et al., 1996;Polfliet et al., 2001;Rua & McGavern, 2016).Thus, it is likely that BAMs represent an early warning system for cellular and pathogenic components, which may be particularly true for the dura mater of the meninges and choroid plexus, which are more permeable borders.Furthermore, given BAMs are located at CNS borders containing a diverse array of cell types including immune and vascular cells, it appears likely that they mediate their physiological roles through cooperative interactions with other brain cells.

Perivascular drainage
Two principal clearance pathways have been proposed in the brain: the intramural periarterial drainage (IPAD) pathway and the glymphatic system (Carare et al., 2008;Gouveia-Freitas & Bastos-Leite, 2021) .Tracers cleared by the IPAD pathway are simultaneously taken up by pvMΦ, suggesting that a fraction of macromolecules in the interstitial fluid may move radially through the arterial wall to be cleared by pvMΦ (Carare et al., 2008), which might help to maintain flux through the IPAD pathway.Moreover, considering that recent mathematical modelling and experimental data firmly supports vascular smooth muscle (VSM) contraction as a mechanism driving IPAD flux (Aldea et al., 2019;van Veluw et al., 2020), and that arterial resident macrophages positively modulate contractility in healthy peripheral vessels, it is conceivable that brain pvMΦ also regulate bulk flow through the IPAD pathway via the regulation of VSM tone (Lim et al., 2018).Also, loss of function studies in the normal mouse aorta have suggested that pvMΦ protect against arterial wall stiffness (Lim et al., 2018), a vascular Fig. 1.Summary of the cell surface protein expression profile and core gene signature of BAMs in the homeostatic and ischaemic rodent brain.In the homeostatic brain, BAMs have a similar cell surface profile to microglia but can be distinguished from microglia by the absence of SIGLEC-H, low expression of Iba, and the expression of cell surface markers including CD38, LYVE1, and pathogen pattern recognition receptors such as CD206.Several studies have proposed a BAM core signature gene which include Lyve-1, Mrc1 (coding for CD206), Lyz2, and Pf4 (as known as chemokine Cxcl4).As discussed in Section 4.1, some choroid plexus BAMs (cpMΦ) express MHCII protein and high levels of MHCII related genes, and high levels of Ccr2, whereas some perivascular (pvMΦ) and meningeal (mMΦ) BAMs lack MHCII protein/genes or CCR2/Ccr2.After ischaemic stroke the BAM core gene signature is largely unchanged during the hyperacute and acute phases apart from an increased Cd38 and Cd44 expression in the hyperacute phase.There is evidence however that some BAMs may transition to a 'M1 like' phenotype in the acute phase as evidenced by increased expression of iNOS and decreased ARG-1.BAMs undergo changes in their gene expression in the hyperacute phase which include increased expression of genes related to antigen presentation, leukocyte chemotaxis, and vascular leakage.In the acute phase, genes related to antigen presentation are decreased, but genes related to cell proliferation and degradation of the extracellular matrix (Mmps) are increased.This figure was created using BioRender.com.
Fig. 2. Proposed mechanisms of BAM function in the homeostatic and ischaemic brain.CNS border-associated macrophages (pvMΦ, cpMΦ and mMΦ) have been implicated in immune surveillance in the healthy brain via roles in antigen presentation, scavenging function, and leukocyte interactions.Other homeostatic roles, particularly relevant to pvMΦ may include waste clearance, nutrient uptake, preservation of vascular permeability, regulation of VSMC contractility and activation of the HPA axis.Evidence suggests BAMs may be harmful in the acute phase of ischaemic stroke.They may contribute to the post-stroke response via roles in extracellular remodelling, granulocyte chemotaxis, vascular inflammation, and vascular leakage.Effects on angiogenesis via the HIF-1α pathway, and release of ROS may also be important.ROS reactive oxygen species, HPA hypothalamic-pituitary-adrenal axis, PVS perivascular space, VEGF vascular endothelial growth factor, VSMC vascular smooth muscle cell, IPAD intramural periarterial drainage pathway.This figure was created using BioRender.com.
property deemed to negatively impact IPAD (Bakker et al., 2016;Weller et al., 2008).In the glymphatic system, subarachnoid CSF passes through the perivascular space and enters the interstitial space and parenchyma via astrocytic end feet (Iliff et al., 2012).The localisation of pvMΦ in the perivascular space means these macrophages have direct access to the glymphatic system, and, thus, it is conceivable that pvM Φ contribute to waste clearance by preventing the accumulation of large particles in the perivascular space which would otherwise impair glymphatic flux.Indeed, reports that BAMs can rapidly take up substances (e.g., dextran, ovalbumin, ferritin, and horseradish peroxidase) administered either into the brain or systemically, further substantiate their role in filtering CSF (Carare et al., 2008;Mato et al., 1996;Santisteban et al., 2020).The importance of waste clearance for brain homeostasis is emphasised by the fact that impaired perivascular fluid drainage is associated with CAA and Alzheimer's disease (Peng et al., 2016;Roher et al., 2003;Silva et al., 2021).Of note, pvMΦ have a role within this context, with loss and gain of pvMΦ function studies demonstrating that pvMΦ contribute to the clearance of vascular amyloid deposition in CAA (Hawkes & McLaurin, 2009).

Antigen recognition, uptake, and presentation
Early studies showing the expression of MHCII, co-stimulatory molecules (e.g., CD86 and CD40), and pathogen recognition receptors (e.g., CD206 and DCSIGN) on BAMs in the naïve brain, as well as their positioning in close vicinity to T lymphocytes, initially supported a crucial role in constitutive antigen recognition, uptake, and presentation (Fabriek et al., 2005;Hickey et al., 1992;Polfliet et al., 2001).However, as discussed, recent work indicates that the majority of BAMs do not express MHCII in the steady-state CNS (Van Hove et al., 2019), highlighting that not all BAMs are likely to be involved in antigen presentation to T lymphocytes (Mrdjen et al., 2018).Notably, as previously mentioned, MHCII + BAMs are enriched in the choroid plexus and dura mater, suggesting the strategic localisation of antigen presentation to T lymphocytes in immunologically vulnerable CNS borders.Indeed, the dura mater encloses the glymphatic system which is patrolled by a vast array of immune cells including T lymphocytes.Also, as discussed in Section 5, there is evidence that BAMs upregulate MHC related molecules in response to neuroinflammation, but the extent to which they contribute to T lymphocyte activation is less clear (Goddery et al., 2021;Jordão et al., 2019).Thus, questions remain over the extent BAMs contribute to antigen presentation in the CNS, both in the steady-state and neuroinflammatory conditions.

Physiological functions involving interactions with other brain cells
Other physiological roles for BAMs are beginning to emerge including the regulation of BBB permeability and metabolic processes in the CNS.These are likely to involve rich connections with other cell types occupying the CNS border regions including endothelial and vascular smooth muscle cells, and pericytes of the cerebrovascular system.In vitro, noncontact co-culture of primary brain capillary endothelial cells with blood-derived macrophages decreases paracellular permeability, suggesting that pvMΦ may contribute to vascular barrier function (Zenker at al., 2003).Indeed, in the area postrema, a circumventricular organ which lacks a typical BBB and has a fenestrated endothelium, pvMΦ limit the influx of tracers larger than 10 kDa into the parenchyma (Willis et al., 2007).This aligns with findings in peripheral vessels, including in the dermis and peritoneum, where pvMΦ reduce vascular leakage under baseline conditions and under stimulation with permeability mediator VEGF, potentially via direct interaction with endothelial cells and the regulation of endothelial VE-cadherin phosphorylation (He et al., 2016).Notably, in the presence of vascular endothelial growth factor (VEGF), macrophages dissociate from endothelial cells resulting in reduced vessel coverage by macrophages which suggests dynamic regulation of pvMΦ-vessel interactions according to environmental cues.Moreover, pvMΦ can themselves produce VEGF (Jais et al., 2016) and thus may additionally modulate barrier function via paracrine effects on other perivascular cells and endothelial cells (Pedragosa et al., 2018), although this is currently poorly understood.Work from rodent models of hypertension and ischaemic stroke also supports the involvement of pvMΦ in vascular permeability regulation, although notably under these pathological conditions, BAM function appears harmful and exacerbates BBB dysfunction and vascular leakage (Pedragosa et al., 2018;Santisteban et al., 2020).There is also evidence that arterial resident macrophages modulate VSM tone in the systemic circulation and are a source of vascular ROS (Park et al., 2017), which may have roles as physiological vasodilators in the cerebral circulation (Miller et al., 2007;Miller et al., 2005;Miller et al., 2006).Notably, however, liposomeencapsulated clodronate depletion of pvMΦ and mMΦ has no effect on resting cerebral blood flow (CBF), neurovascular coupling, or endothelium-dependent dilator responses, suggesting they are unlikely to play significant roles in regulating cerebrovascular tone and CBF under homeostatic conditions.Although, emerging data in rodent models suggests that pvMΦ contribute to neurovascular deficits observed in Alzheimer's disease and hypertension (Faraco et al., 2016a;Park et al., 2017).
Others have suggested that BAMs may influence cellular metabolism by regulating nutrient uptake in the brain.Exposure of mice to a high fat diet leads to acute downregulation of cerebral endothelial GLUT1 expression, which is subsequently restored by pvMΦ via the production of VEGF, allowing brain glucose uptake to be maintained and cognitive function preserved (Jais et al., 2016).Also, Mato et al. described the incorporation of circulating lipids into the cytoplasm of pvMΦ in rats fed a high fat diet ( Mato et al., 1982), suggesting pvMΦ may facilitate the transportation of circulating lipids into the CNS.In support, bulk RNA sequencing of macrophage populations in the brain identified upregulated expression of genes involved in lipid metabolism and cholesterol storage in pvMΦs relative to microglia (Van Hove et al., 2019).BAMs may also contribute to iron uptake, as a subset of pvMΦs and cpMΦs express Slc40a1, which encodes for the iron transporter ferroportin-1 (Dani et al., 2021).Lastly, there is evidence that pvMΦ regulate activation of the hypothalamic-pituitary-adrenal axis (HPA) in response to acute emotional and immune/inflammatory stress (Sayd et al., 2020), although their role in this process is complex; influenced by the nature of the stimulus and, most likely, by dynamic interactions with endothelial cells (Serrats et al., 2010).

Proliferation and migration
Emerging experimental (Pedragosa et al., 2018;Rajan et al., 2020) and clinical evidence (Holfelder et al., 2011) indicates that BAMs accumulate in the early phases following ischaemic stroke.For example, whole-brain scRNA-seq analysis reveals a marked increase in the proportion of BAMs 24 h after ischaemia-reperfusion in mice (Zheng et al., 2021).Also, using CD163 + as a marker for rat BAMs, a separate study found that although numbers of immunosorted CD163 + cells are unchanged in the first 24 h after ischaemia-reperfusion in rats, by day 3 CD163 + cells dramatically accumulate, particularly within the meninges and perivascular spaces of the ischaemic hemisphere (Rajan et al., 2020).CD163 + cells are also found to accumulate around cerebral blood vessels in post-mortem tissue samples of patients with ischaemic stroke (Holfelder et al., 2011) and there was a progressive accumulation of CD163 + cells in the meninges proximal to the ischaemic lesion in tissue from patients 6 to 140 days after stroke onset (Rajan et al., 2020), suggesting the accumulation of BAMs might persist into the chronic phase after stroke.Data from a recent review article also indicates that pvMΦ accumulate in response to chronic cerebral hypoperfusion in mice induced by two-vessel carotid occlusion ( Yang et al., 2019).After 8 weeks of hypoperfusion, the numbers of CD206 + pvMΦ increased in sub-cortical regions with no change in cortical regions.Also, this increase primarily related to a greater number and proportion of CD206 + pvMΦ co-expressing the classical M1 markers CD32 and CD16, whereas numbers of CD206 + pvMΦ co-expressing LYVE-1 decreased, suggesting the emergence of altered BAM phenotypes and/or subtypes in response to ischaemia (see Section 5.1.2).One major caveat of all these findings is that these approaches clearly rely on brain excavation, which inadvertently destructs the dura mater and potentially the arachnoid mater, thereby excluding analyses of macrophages in these locations.The application of a refinement method for brain excavation that preserves the meninges (Louveau et al., 2018) may provide an opportunity to fully examine the responses of all BAM subsets to ischaemia.Interestingly, there is evidence that cells expressing BAM markers also accumulate in the ischaemic parenchyma.For example, although there was no evidence for the accumulation of CD163 + cells in the ischaemic parenchyma during the early phase (24 h) in rats (Rajan et al., 2020), CD169 + cells (negative for CD206) are found in the ischaemic parenchyma of mice on day 4 after ischaemiareperfusion. Similarly, in post-mortem tissue of stroke patients there is a progressive appearance of CD163 + cells within the infarcted parenchyma (Holfelder et al., 2011;Rajan et al., 2020) and in the meninges proximal to the infarct core (Rajan et al., 2020).These findings may indicate the emergence of a distinct BAM subset that migrates from the perivascular space to the ischaemic parenchyma, however, definitive evidence is lacking.Indeed, it is conceivable that this observation is due to upregulation of BAM makers on resident microglia or BMderived macrophages (Henning et al., 2009;Holfelder et al., 2011;Rajan et al., 2020).
Genes related to cell-cycle and cell division (e.g., marker of proliferation Ki67; cyclin-dependent kinases) are increased in CD163 + cells in the ischaemic hemisphere of rats.Consistent with this, immunofluorescence for Ki67 is detected in CD163 + cells in perivascular and meningeal spaces of the ischaemic hemisphere of rats on day 3 after ischaemia-reperfusion, and this occurs in parallel with increased gene expression of Csf1 (Rajan et al., 2020).Thus, like microglia, their accumulation following ischaemic stroke appears to primarily stem from the self-proliferation of resident BAMs.Notably, however, a proportion of BAMs in the ischaemic brain is replaced by infiltrating BM-derived monocytes.Using chimeric mice transplanted with BM from transgenic CX3CR1 GFP CCR2 GFP mice, ~30% of CD169 + cells originate from BM-derived CX3CR1 + CCR2 + patrolling monocytes after ischaemic stroke (Rajan et al., 2020).Furthermore, in perivascular spaces most of the CD169 + cells are CX3CR1 + suggesting monocytes infiltrate the perivascular space and acquire features of pvMΦ.The mechanisms driving the recruitment of BM-derived monocytes and their maturation into BAMs have not been characterised but may parallel findings from a rodent model of endotoxemia, where classical inflammatory signals involving a TNF/Angiopoietin 2-dependent pathway promote the adhesion of monocytes to the brain capillary endothelium, followed by their proliferation, transmigration across the BBB and subsequent adoption of a pvMΦ phenotype (including an elongated morphology and expression of CD11b, CD163, and stabilin-1) (Audoy-Rémus et al., 2008).Consistent with this, in a model of experimental autoimmune encephalomyelitis (EAE) the proliferation of BAMs occurs alongside continuous monocytic infiltration up to the peak of disease severity (Jordão et al., 2019).These monocytes acquire phenotypic markers often associated with BAMs, and in the chronic phase of disease are transiently integrated into the CNS as BAMs undergo apoptosis.Also, in a mouse model of Alzheimer's disease there is evidence that circulating monocytes are recruited to cerebral vessels in a CCR2-dependent manner and contribute to the pool of pvMΦ (Mildner et al., 2011).Notably, the recent discovery of communicating channels between the skull BM and the meninges raises the possibility that BM-derived CX3CR1 + monocytes may migrate from the skull BM directly into the perivascular spaces following ischaemia (Herisson et al., 2018).

Genetic transformations
Studies employing scRNA-seq analyses or single-cell immunophenotyping of cells from CNS borders or whole-brain indicate that BAMs rapidly undergo changes in response to neuroinflammation suggesting phenotypic alterations.For example, during EAE, BAMs strongly upregulate MHCII and related molecules including H2-Aa, H2-Ab1, H2-Eb1, and Cd74, suggesting a role in antigen presentation in the inflamed brain (Jordão et al., 2019).However, functional studies indicate that T lymphocyte activation is largely driven by peripheral CCR + myeloid cells in EAE, which show long-lasting interactions with T lymphocytes compared to BAMs (Jordão et al., 2019).In contrast to other immune cells, BAM numbers might decrease during EAE-induced neuroinflammation, supporting the concept that BAM numbers decline as BMderived monocytes infiltrate the inflamed brain (Mrdjen et al., 2018).Also, there is evidence that BAMs might downregulate most of their core signature genes and lose their heterogenicity, with almost all BAMs co-expressing CD38 and MHCII in EAE-induced neuroinflammation (Mrdjen et al., 2018).Similar findings are also found in the aged murine brain, where there is an increased frequency of CD38 + and MHCII + (CCR2 − ) BAM subsets (Mrdjen et al., 2018).Interestingly, however, in contrast to microglia, which shift towards an activated phenotype in murine models of Alzheimer's disease (e.g., APP/PS1 5xFAD mice), the distribution and phenotype of BAM subsets are unaltered (Keren-Shaul et al., 2017).
Studies have also begun to uncover that BAMs undergo rapid changes in their gene expression profile in the early phases after ischaemic stroke (Fig. 1).However, these studies have been restricted to analyses of whole-brain or CD163 + immunosorted cells rather than analyses of BAMs subsets from specific CNS borders and have focussed only on the acute phase after ischaemia-reperfusion.In the rodent ischaemic brain, BAMs express genes consistent with the core gene signature described in the naïve brain (e.g., Cd38, Cd206, Cd163, Cd44, and Lyve-1) (Pedragosa et al., 2018;Rajan et al., 2020).Most of these genes are unchanged in the hyperacute (24 h) phase, with the exception of Cd38 and Cd44 (Pedragosa et al., 2018), which show increased expression.Consistent with the presence of distinct BAM subsets in the naïve murine brain, six BAM subsets have been identified 24 h after ischaemia-reperfusion (Zheng et al., 2021).Notably, the proportion of cells in some of these BAM subsets increases further beyond 24 h, supporting the view that BAMs accumulate in response to ischaemia.Some of the subsets seemed to undergo genetic changes, which include higher levels of genes for antigen presentation molecules such as H2-Aa and H2-Ab, as well as higher genes involved in chemotaxis (e.g., Ccl2, Cxc11, and Cxcl2), suggesting their involvement in antigen presentation and early leukocyte recruitment to the ischaemic brain (Zheng et al., 2021) (Fig. 1).Importantly, however, data showing a functional role of BAMs in these processes after stroke is lacking.It is conceivable that, similarly to models of EAE, BAMs may play a minor role in T lymphocyte activation in the post-stroke milieu (Jordão et al., 2019).Functional analyses of genes over-expressed in rat CD163 + BAMs in the hyperacute (16 h) after ischaemia does, however, reveal an overrepresentation of genes related to leukocyte chemotaxis and neutrophil recruitment (e.g., IL-6, L-selectin, Cxcl14, Ccl1, and Ccl22).Also, at this early time point after stroke, the classical M2 marker Arg-1 is one of the most highly up-regulated genes, whereas the classical M1 marker Nos2 (encoding iNOS) is not overrepresented, suggesting the acquisition of a distinct metabolic and inflammatory signature akin to the 'M2-Like' macrophage phenotype.Also, hypoxia inducible factor-1 (HIF-1) signalling is one of the most highly upregulated pathways in rat CD163 + BAMs, further illustrated by the upregulation of Hif-1α and its target VEGF-A (Vegfa), as well as genes related to angiogenesis, blood vessel remodelling, and extracellular remodelling.Consistent with this, VEGF is detected around some cerebral vessels and CD163 + pvMΦ at the periphery of the infarct core in post-mortem human brain tissue of patients who died 24 h after stroke onset.Collectively, these findings suggest that ischaemia changes the homeostatic functions of BAMs in the very early phase after stroke onset (Fig. 2).
Rat CD163 + BAMs undergo further changes in their gene expression at later stages of the acute phase after ischaemia-reperfusion (day 3); however, notably their transcriptomes become more similar to microglia and BM-macrophages, suggesting they acquire a similar functional phenotype in response to ischaemia (Rajan et al., 2020).In contrast to BAMs in the hyperacute phase (Pedragosa et al., 2018), CD163 + BAMs also appear to acquire a pro-inflammatory phenotype at this later time point as evidenced by the expression of numerous inflammatory genes.Also, immunohistochemistry methods show that the proinflammatory cytokine TNF-α is co-expressed in CD163 + cells in the perivascular, meningeal, and choroid plexus CNS borders on day 7 after stroke onset in the spontaneously hypertensive rat (Henning et al., 2009).Interestingly, however, CD163 + BAMs in the perivascular and meningeal spaces do not express iNOS, whereas CD163 + cells in the ischaemic parenchyma have strong expression of iNOS but undetectable ARG-1 immunoreactivity (Fig. 1).Thus, similar to microglia, CD163 + cells infiltrating the brain parenchyma from perivascular spaces might acquire a pro-inflammatory phenotype, which also supports the view that distinct BAM subsets with potentially unique functions emerge in response to ischaemic injury (Zheng et al., 2021).Consistent with this, there is a marked increase in genes for MMPs in BAMs in parallel with decreased expression of collagen, suggesting roles in extracellular degradation which may facilitate their migration into the ischaemic parenchyma (Rajan et al., 2020) (Fig. 1).Indeed, gene ontology (GO) analyses showed BAMs upregulate genes characteristic of a migratory phenotype.Lastly, genes involved in antigen presentation such as Mrc1 and H2-Aa are also downregulated at day 3 after ischaemic stroke in the rat, indicating that BAMs may acquire a phenotype of impaired antigen presentation capacity as the ischaemic lesion develops (Rajan et al., 2020).By contrast, immunohistochemistry methods show that MHCII protein is co-expressed in rat CD163 + BAMs at all CNS borders at the later time point of day 7 after stroke onset (Henning et al., 2009).

BAMs and stroke outcomes
One of the major challenges of investigating the roles of BAMs in stroke outcomes is that the immune landscape changes dramatically following ischaemia, making it difficult to unambiguously discriminate and target BAM populations from peripheral-derived immune cells.Furthermore, as discussed, BAMs become transcriptionally similar to microglia and BM-derived macrophages in the acute phase after stroke and may lose their heterogenicity.As discussed in Section 4.2.3,BAMs appear to closely interact with other immune and vascular cells to mediate their physiological functions.Also, although their interactions with other brain cells in the ischaemic brain remain to be experimentally determined, gene expression data and clodronate depletion experiments (discussed below) infer that BAMs may interact with other brain cells to modulate the ischaemic milieu.Indeed, as discussed, BAMs upregulate genes involved in leukocyte chemotaxis, extracellular modelling, and angiogenesis suggesting they may interact with immune and vascular cells within the CNS borders.
To begin to unravel the functions of BAMs in stroke outcomes, studies thus far have depleted BAMs by injection of liposome-encapsulated clodronate into the cerebral ventricles, a method developed over 20 years ago (Polfliet et al., 2001) that has been recently used to elucidate the roles of BAMs, primarily pvMΦ, in models of neuroinflammation, myocardial infarction, Alzheimer's disease, CAA, or hypertension (Faraco et al., 2016a;Iyonaga et al., 2020;Kerkhofs et al., 2020;Santisteban et al., 2020;Yu et al., 2010).Although this approach leads a robust depletion of BAMs (~70%) without affecting microglia, BMderived macrophages, or circulating leukocytes (Faraco et al., 2016a;Pedragosa et al., 2018), it is transient, and re-population by BMderived monocytes/macrophages may occur.Most studies thus far have used clodronate to study the functions of pvMΦ; however, this approach also depletes mMΦ making it challenging to define their unique spatial-temporal functions or the functions of cpMΦ.Also, as discussed in Section 4, BAMs are now regarded as a heterogenous immune cell population comprising of multiple subsets beyond these classical types.Thus, this depletion strategy does not allow the discrimination of BAM subsets.Lastly, although i.c.v.injection of clodronate effectively depletes BAMs, evidence suggests that it also provokes a series of neuroinflammatory responses per se, highlighting another potential limitation of using this approach to study BAMs in ischaemic stroke (Drieu et al., 2020).
These latter points notwithstanding, evidence thus far indicates that the collective pvMΦ and mMΦ population may make a deleterious contribution to stroke pathophysiology, at least within the very acute phase after stroke onset in young and otherwise healthy rodents.Depletion of BAMs with liposomal clodronate injection (i.c.v.) prior to induction of ischaemia-reperfusion in rats led to small, yet significant, improvements in neurological function with no change in lesion size at 24 h after ischaemia (Pedragosa et al., 2018).Notably, however, at this early time point the infarct is not fully developed in rodents.Thus, more research is needed to examine the impact of BAM depletion on final infarct volumes.Compatible with data showing an overrepresentation of genes related to leukocyte chemotaxis and neutrophil recruitment in rat CD163 + BAMs after ischaemia-reperfusion (See Section 5.1.2),depletion of BAMs reduces the number of infiltrating granulocytes in cortical regions of the rat ischaemic brain, without altering numbers of T lymphocytes, NK cells, or myeloid mononuclear cells (Pedragosa et al., 2018).Furthermore, BAM depletion reduces the expression Vegfa and VEGF protein in the cerebral cortex and reduces the permeability of pial and cortical cerebral vessels, which is consistent with the finding that genes related to HIF1 signalling are upregulated in rat CD163 + cells.Collectively this work suggests that BAMs may have roles in attracting granulocytes to the perivascular and meningeal spaces in response to ischaemia and may also facilitate their infiltration into the brain through VEGF-dependent vascular leakage (Schoch et al., 2002).Indeed, it is known that neutrophil extravasation from meningeal vessels and neutrophil accumulation in perivascular spaces both occur in response to cerebral ischaemia.Similarly, peripheral CD163 + macrophages, expressing elevated levels of HIF1 and VEGF, are implicated in vascular leakage and inflammatory cell recruitment in atherosclerosis (Guo et al., 2018).A recent study by Drieu et al. also showed that clodronate BAM depletion has no effect on lesion size at 24 h after stroke onset but modulated ischaemia-indued inflammatory responses including a decrease in leukocyte rolling/adhesion on cerebral vessels, and microglia activation (Drieu et al., 2020).Interestingly, the main purpose of this study was to investigate the roles of BAMs in the aggravating effects of chronic alcohol drinking on ischaemic stroke outcome.The study showed that in the absence of ischaemic stroke, chronic alcohol consumption by itself provokes a neurovascular inflammatory priming which includes an increased number of pvMΦ.Also, using BAMdepleted mice, this study provides evidence that 'priming' of pvMΦ contributes to the aggravating effect of alcohol on ischaemic lesion size.There is growing evidence for alterations in BAM numbers and phenotype in cerebrovascular risk factor such as (Faraco et al., 2016a).Thus, the functions and impact of BAMs on stroke outcomes may be altered in the setting of cerebrovascular risk factors.
The complexity and temporal evolution of pvMΦ and mMΦ function in response to haemorrhagic stroke has recently been highlighted in a murine model of subarachnoid haemorrhage (SAH).Depletion of BAMs led to decreased perivascular inflammation and reduced microthrombus formation regardless of whether clodronate was administered prior to or after experimental SAH (Wan et al., 2021).However, whilst pre-depletion of BAMs hindered the clearance of erythrocytes from the subarachnoid space and had no impact on neuronal cell death or neurological status, depletion of BAMs 3 h after SAH had no effect on erythrocyte clearance but reduced neuronal cell death and improved neurological outcomes.The different effects on stroke outcomes between the two groups may reflect a dual role for BAMs in haemorrhagic stroke: immediately following SAH erythrocyte phagocytosis by BAMs may be advantageous by preventing the breakdown of the blood clot within the subarachnoid or perivascular spaces, masking any aggravating effects of BAMs on perivascular inflammation, however, once the erythrocytes have been cleared, the beneficial effects of BAM function may be reduced, revealing harmful effects on neuronal survival and neurological status.
As discussed, data thus far points towards a prominent functional role for BAMs in promoting vascular leakage, and leucocyte attraction and infiltration into the ischaemic brain.Furthermore, BAMs appear to acquire a pro-inflammatory phenotype as the ischaemic lesion develops suggesting they modulate stroke-induced inflammatory responses.However, drawing from the roles in other diseases affecting the brain it is conceivable that their roles are much broader.For example, in the angiotensin II-induced model of hypertension and chronically hypertensive BPH/2 J mice, pvMΦ contribute to BBB disruption and neurovascular dysfunction via ROS production by Nox2 oxidase and cooperative interaction with cerebral endothelial cells (Faraco et al., 2016a;Santisteban et al., 2020).In spontaneously hypertensive rats (SHR), pvMΦ contribute to endothelial dysfunction and vascular remodelling of the middle cerebral artery (Pires et al., 2014).The roles Nox2 oxidase-derived ROS in promoting cerebrovascular dysfunction and ischaemic brain injury are well documented (De Silva et al., 2011;Jackman et al., 2009;Kahles & Brandes, 2013).Thus, it is conceivable that pvMΦ might similarly contribute to deficits in cerebrovascular function and cerebral perfusion after ischaemic stroke.Currently it is unknown if and how cpMΦ respond to ischaemia or whether they contribute to ischaemic processes.Notably, the choroid plexus is a route of entry for BM-derived monocytes/macrophages (Ge et al., 2017), raising the possibility that cpMΦ may interact with immune cell populations entering the ischaemic brain.
The influence of BAMs on outcomes beyond the acute phase of stroke is unknown and clearly warrants further investigation.In a small number of ischaemic stroke patients, Kösel et al. reported the appearance of MHCII + perivascular cells containing lipid droplets in the lateral-cortical spinal tract in areas affected by Wallerian degeneration up to 10 years after stroke (Kösel et al., 1997).The identity of these cells was not defined; however, it is conceivable that pvMΦ may participate long-term demyelination after stroke.Indeed, the persistence of elevated numbers of cells expressing BAM markers in the perivascular, meningeal, and parenchymal spaces in the weeks to months following stroke implies that their contribution may extend beyond the acute phase of stroke.Putative brain reparative roles for BAMs should also be considered.Indeed, as discussed in Section 2, evidence suggests that infiltrating monocytes/macrophages primarily acquire a phenotype favouring inflammation resolution and neurovascular repair.Furthermore, there is evidence for upregulation of genes related with angiogenesis and extracellular remodelling in rat CD163 + BAMs.Investigating the long-term roles of BAMs is of course hampered by the relatively rapid re-population of BAMs after clodronate depletion, and this approach does not allow for selective depletion of the various BAM subsets.Thus, studies using different experimental approaches are needed to specifically define the roles of BAM subsets for the evolution of the ischaemic lesion as well as their involvement in repair processes.

Therapeutic potential and challenges of targeting BAMs
The research field is perhaps too young for us to propose that BAMs are a therapeutic target for ischaemic stroke.Indeed, although recent work provides evidence the BAMs have important roles in the homeostatic and diseased brain, BAMs are likely to be far outnumbered by their distinct cousins (microglia) and infiltrating leukocytes in the ischaemic brain.As with other immune cells involved in stroke pathogenesis the roles of BAMs are likely to be complex and multiphasic, which will pose a major challenge for the development of BAM modulatory therapies.Furthermore, it is important to highlight that approaches that prevent deleterious roles of immune cells have so far failed to translate into the clinic for the treatment of acute ischaemic stroke (Veltkamp & Gill, 2016).A major challenge of any inflammatory treatment is that should counteract deleterious post-ischaemic inflammation but leave the reparative roles of the immune system intact.The therapeutic potential of BAMs is clearly very speculative; however, below we have provided an appraisal of potential immunomodulatory strategies that could be exploited to target BAMs.

Multi-target immunomodulatory strategies
Several anti-inflammatory drugs including glucocorticoids, cyclooxygenase (COX) 2 inhibitors, and minocycline have shown promise in experimental models of stroke (Murata et al., 2008;Sugimoto & Iadecola, 2003) but have failed improving stroke outcomes when tested in clinical trials (Kohler et al., 2013;Sandercock & Soane, 2011).However, nonselective COX inhibitors, such as aspirin, have shown clear benefit for secondary stroke prevention (Del Giovane et al., 2021) and can prevent macrophage accumulation in the stroke-prone SHR (Ishizuka et al., 2008).Nevertheless, the pursuit of identifying and developing novel treatments that possess anti-inflammatory properties is ongoing.Several experimental studies have shown the therapeutic potential of the IL-1β receptor antagonist (IL-1Ra), (Maysami et al., 2016); however, there has been no major benefit in clinical trials (Smith et al., 2018).The sphingosine 1-phosphate receptor agonist fingolimod (or FTY720) has been proposed by some to hold the most therapeutic potential among new immunomodulatory compounds in adult cerebral ischaemia.Fingolimod exerts anti-inflammatory effects (among other effects) and has been approved by the FDA as a treatment for multiple sclerosis (Gasperini & Ruggieri, 2012).Its anti-inflammatory effects primarily relate to its ability to reversibly promote homing of T and B lymphocytes to secondary lymphoid organs, thereby causing lymphopenia (Cohen & Chun, 2011).In addition, evidence suggests that fingolimod may modulate the inflammatory response by shifting macrophages towards a M2 anti-inflammatory phenotype ( Qin et al., 2017).It has been extensively tested in experimental stroke models, where it has shown beneficial effects which extend beyond its immunomodulatory properties (Czech et al., 2009;Hasegawa et al., 2010;Wei et al., 2011).Interestingly, however, in contrast to adult stroke, studies of neonatal hypoxia-induced brain injury have found that fingolimod either has no effect or exacerbates the extent of brain injury (Herz et al., 2018;Yang et al., 2014).Furthermore, the study showing that fingolimod worsens neonatal brain injury provided evidence that this relates to sustained depletion of circulating T lymphocytes, and a concomitant increase in infiltration of neutrophils and macrophages (Herz et al., 2018).These findings highlight the challenges of translating findings of immunomodulatory agents from adult to neonatal stroke (Mallard & Vexler, 2015).This latter point notwithstanding, positive results for fingolimod have been reported from the small clinical trials of adult cerebral ischaemia conducted thus far including in patients receiving thrombolysis (Tian et al., 2018).Whilst research is still in its infancy, an elegant study recently described the promise of human amnion epithelial cells (hAECs) for ischaemic stroke, showing that intravenous delivery of hAECs to mice up to 3 days after ischaemic stroke onset markedly reduces infarct size and functional deficits in young and aged mice of both sexes, and limits infarct development in the nonhuman primate (Evans et al., 2018).This study provided evidence that hAECs mediate their protective and reparative effects by modulating the inflammatory response, which likely includes inhibition of neutrophil and macrophage migration, the promotion of macrophages to a M2 antiinflammatory phenotype, and the modulation of T and B lymphocyte activation and phenotype.Although there is no current evidence that these agents also modulate BAMs in the stroke setting, these examples demonstrate that multi-target immunomodulatory strategies may offer more promise than approaches that selectively target one immune cell population (e.g., BAMs) and/or inflammatory mediators.

Anti-inflammatory strategies related to macrophage and microglia M1/M2 phenotype
Not surprisingly, the research field is more advanced in the pursuit of BM-macrophage/microglia-specific strategies for ischaemic stroke.As discussed in this review, BAMs exhibit surface protein expression patterns and core gene signatures in the homeostatic brain that are distinct from microglia.However, the cell-surface expression profile and transcriptomes of BAMs may become more like microglia (and BMderived macrophages) in the inflamed brain (Mrdjen et al., 2018;Rajan et al., 2020).More research is needed to define the specific roles of BAM subtypes (relative to parenchymal microglia) in each specific CNS border and to determine if and how BAMs undergo similar M1/ M2-like phenotypic shifts in the ischaemic brain.However, if, as proposed, BAMs acquire a similar phenotype to microglia and BM-derived macrophages during ischaemia, it is conceivable that strategies designed for these immune cell populations may have the added benefit of targeting BAMs.
Given the importance of the balance of microglia M1/M2 polarisation states in ischaemic stroke and other brain diseases, studies have examined whether pharmacological agents can modulate the M1/M2 balance for therapeutic gain.The transcription factor peroxisome proliferator-activated receptor-γ (PPARγ) is proposed to be a master regulator of the transcriptome of monocytes/macrophages and thus might shape their reparative phenotype in the post-stroke brain (Wang et al., 2019;Zhang, 2019).Similarly, PPARγ can regulate the phenotype of microglia (Hasegawa-Moriyama et al., 2013); however, its roles in regulating the phenotype of BAMs are unknown.Genetic or pharmacological approaches that target this master regulator may be advantageous for shifting macrophages/microglia towards the M2 reparative phenotype.Indeed, several studies have shown that PPARγ agonists efficiently protect against ischaemia in experimental models, which involves reduced infiltration of macrophages and expression of inflammatory cytokines (Culman et al., 2012;Cuzzocrea et al., 2003).Also, PPARγ agonists can inhibit microglia activation and promote a phenotypic shift towards the M2 reparative phenotype in diseases models associated with neuroinflammation (Hasegawa-Moriyama et al., 2013).Collectively, this data suggests that PPARγ might be novel therapeutic target in determining reparative macrophage/microglia phenotypes following ischaemic stroke (Ding et al., 2020).Furthermore, PPARγ agonists are likely to have neuroprotective and reparative effects beyond immunomodulation (Ding et al., 2020).For example, PPARγ exerts protective effects on the cerebral vasculature when activated pharmacologically which includes the suppression of vascular oxidative stress, inflammation, and endothelial cell senescence (De Silva et al.,2018).Members of the signal transducer and activator of transcription (STAT) family are also involved in macrophage/microglia phenotypic switching.For example, together with PPARγ, STAT6 is a potential upstream regulator shaping the inflammation-resolving transcriptome of macrophages in the ischaemic brain (Zhang et al., 2019).Also, some miRNAs could be exploited to modulate macrophage/microglia polarisation in the ischaemic brain.For example, miR-146a, miR-511-3p, miR-223, and let-7c have all been linked with M2 macrophage polarisation likely by targeting PPARγ, STAT6, and other transcription factors (Curtale et al., 2019).
Other than the modulation of macrophage/polarisation, regulation of depletion and repopulation may offer another strategy to modulate inflammation in the ischaemic brain.As discussed in Section 3.2, similarly to microglia, pvMΦ and mMΦ populations in the homeostatic brain require CSF1R signalling for their survival and proliferation (Ginhoux et al., 2010).Also, CD163 + cells display increased gene expression of Csf1 in the ischaemic brain (Rajan et al., 2020), suggesting some BAM populations require this signalling pathway for self-proliferation.Several studies have demonstrated the potential of using agonists for the CSF1R to deplete microglia in the rodent brain.For example, small molecule CSF1 inhibitors such as PLX3397 and PLX5662 (which is more effective in crossing the BBB), can rapidly deplete microglia in the rodent brain; however, a major limitation of this strategy is that repopulation is also rapid once treatment is withdrawn (Najafi et al., 2018;Rice et al., 2017).Also, this strategy would have to be precisely timed to counteract the proposed deleterious effects of BAMs/microglia but allow the reparative actions to remain intact.Other depletion strategies such as liposomal clodronate and Mac-1-sapori are likely to have the same challenges.

Theoretical strategies for selectively targeting BAMs
Studies thus far have begun to reveal the potential roles of BAMs in ischaemic stroke, and the future application of modern techniques should expedite our understanding of the phenotypic transformations of BAM subtypes in response to ischaemia, their spatial heterogenicity, and most importantly their temporal roles in brain injury and/or repair.Of course, it is important to highlight that most of the data obtained on BAMs in the ischaemic stroke has arisen from experimental stroke models, supported by limited data from post-mortem tissue of stroke patients.Also, even if we identify approaches that can functionally alter or deplete BAM subsets, it is conceivable that the BAM populations will be replaced by either their counterparts or monocytes/macrophages.Thus, as it is often the case in an emerging field of research, we are left with more questions about BAMs than we started with.
Presuming BAMs do indeed have clear-cut, non-redundant functions in the ischaemic brain, this knowledge may allow the explicit distinction of BAMs from microglia and BM-derived macrophages, and ultimately the identification and testing of strategies that can selectively target BAMs.Indeed, their localisation at CNS borders might infer that this immune cell population contributes to ischaemic brain injury and repair in distinct ways to that of microglia.If so, it might be strategic to target BAMs without modulating microglia function.The CNS borders should be more accessible to pharmacological agents than the BBB 'protected' parenchyma following ischaemia, which may facilitate selective targeting of BAMs over microglia.However, CNS borders contain a diverse array of resident or patrolling immune cells including monocytes and BM-derived macrophages.Thus, the impact of any BAM strategy on these immune populations will need to be careful considered.Perhaps the most obvious avenue for specifically targeting BAMs is the generation of pharmacological agents that modulate cell surface receptor signalling pathways and/or genes unique to BAMs.For example, designer peptide sequences have been successfully used in targeting cell surface receptors (e.g., CD206) on tumour cells (Scodeller et al., 2017).The recent advent of several modern techniques has so far uncovered the cell surface protein profile on murine BAMs and a consistent BAM core signature (Jordão et al., 2019;Mrdjen et al., 2018;Pedragosa et al., 2018;Rajan et al., 2020;Van Hove et al., 2019), however, the specific roles of these cell surface proteins (and associated downstream signalling) and genes in regulating BAM function in health and disease are unknown.Also, there is evidence, albeit primarily from models of EAE which produce profound neuroinflammation, that BAMs may lose their heterogenicity in the inflamed brain (Mrdjen et al., 2018).Nevertheless, there is a strong rationale to exploit these modern approaches to provide a more in-depth understanding of the genetic and potential phenotypic transformations of BAM subtypes after stroke both in the early and later stages.

Summary
This review highlights that BAMs are a small but heterogenous immune cell population in the homeostatic CNS, which exhibit a surface protein expression profile and core gene signatures that are distinct from their distant cousins, microglia.It also highlights emerging evidence for the existence of multiple BAM subtypes residing in CNS border regions, suggesting functional specialization tailored to the specific needs of each border environment.Although the breadth of their physiological roles remains to be fully defined, pvMΦ and mMΦ are likely to have roles in immunological surveillance, scavenging of waste, CNS antigens, and pathogens, and the regulation of BBB and vascular permeability.Furthermore, these functions are likely to involve close interactions with other cells within the perivascular and meningeal spaces.As discussed, there is growing interest in defining the roles of CNS borders and BAMs in neuroinflammatory and neurodegenerative brain diseases, including ischaemic stroke.However, we are only at the very beginnings of understanding if and how BAMs undergo ischaemia-specific transformations, if they initiate and/or resolve ischaemic brain injury, and if they can be therapeutically targeted.The future application of recently developed techniques will go some way to answer these questions by facilitating the selective purification and/or targeting of BAM subtypes.Such knowledge will allow us to appraise if they indeed hold promise as a treatment for stroke.