Parenchymal border macrophages regulate the flow dynamics of the cerebrospinal fluid

Macrophages are important players in the maintenance of tissue homeostasis 1 . Perivascular and leptomeningeal macrophages reside near the central nervous system (CNS) parenchyma 2 , and their role in CNS physiology has not been sufficiently well studied. Given their continuous interaction with the cerebrospinal fluid (CSF) and strategic positioning, we refer to these cells as parenchymal border macrophages (PBMs). Here we demonstrate that PBMs regulate CSF flow dynamics. We identify a subpopulation of PBMs that express high levels of CD163 and LYVE1 (scavenger receptor proteins), closely associated with the brain arterial tree, and show that LYVE1 + PBMs regulate arterial motion that drives CSF flow. Pharmacological or genetic depletion of PBMs led to accumulation of extracellular matrix proteins, obstructing CSF access to perivascular spaces and impairing CNS perfusion and clearance. Ageing-associated alterations in PBMs and impairment of CSF dynamics were restored after intracisternal injection of macrophage colony-stimulating factor. Single-nucleus RNA sequencing data obtained from patients with Alzheimer’s disease (AD) and from non-AD individuals point to changes in phagocytosis, regulate CSF flow dynamics, an integral aspect of brain physiology, under homeostatic conditions. We identify arterial-associated PBMs that display a transcriptomic profile of scavenger cells and control extracellular matrix (ECM) remodelling, which affects arterial motion, a driving force of CSF flow dynamics. Depletion of PBMs results in the accumulation of ECM proteins and impairment of brain perfusion by the CSF. We demonstrate that normal ageing is associated with PBM dysfunction. Moreover, treatment of aged mice with macrophage colony-stimulating factor (M-CSF) improves CSF dynamics.

The meninges, a tripartite membranous covering of the brain parenchyma, are densely populated by immune cells and their derived cytokines can affect mouse behaviour [3][4][5] . Cytokines from the periphery and the meninges can be carried along in CSF, which is driven by arterial pulsation and vasomotion to circulate throughout the brain parenchyma [6][7][8] . This parenchymal perfusion by CSF not only provides meningeal immune molecules direct access to brain cell signalling but also performs 'glymphatic clearance' 9 by facilitating, through meningeal lymphatics, the removal and subsequent clearance of brain metabolites 10,11 . Antigens are transported along the same routes into the dura, where they are sampled by dural antigen-presenting cells and presented to patrolling T cells to ensure immune surveillance of the CNS 12 . Some of these brain-derived antigens subsequently also drain into the deep cervical lymph nodes (dCLNs) through meningeal lymphatic vessels 10 .
The CNS myeloid niche in the homeostatic brain comprises microglia and leptomeningeal and perivascular macrophages 13 (we collectively term these two populations of border macrophages as PBMs). Unlike microglia, which are located within the brain parenchyma, PBMs reside in the leptomeninges and perivascular spaces along the vasculature, and are therefore constantly in direct contact with the CSF [14][15][16] . Previous studies have suggested a detrimental role for such macrophages in chronic hypertension 17 , AD 18,19 , stroke and experimental autoimmune encephalomyelitis 20 . The functions of PBMs in brain homeostasis, however, are still largely unexplored.
Here we show that PBMs regulate CSF flow dynamics, an integral aspect of brain physiology, under homeostatic conditions. We identify arterial-associated PBMs that display a transcriptomic profile of scavenger cells and control extracellular matrix (ECM) remodelling, which affects arterial motion, a driving force of CSF flow dynamics. Depletion of PBMs results in the accumulation of ECM proteins and impairment of brain perfusion by the CSF. We demonstrate that normal ageing is associated with PBM dysfunction. Moreover, treatment of aged mice with macrophage colony-stimulating factor (M-CSF) improves CSF dynamics.

PBMs sample CSF and regulate its flow
PBMs are found in leptomeningeal and perivascular spaces in the brain, at the vicinity of larger blood vessels, and can be distinguished from microglia by their location and expression of the mannose receptor CD206 (ref. 16 ) (Fig. 1a, Extended Data Fig. 1a-c and Supplementary Video 1). We were able to distinguish two subtypes of PBMs on the basis of their expression of either LYVE1 or major histocompatibility complex II (MHCII) (Fig. 1b,c). Using flow cytometry, PBMs were distinguished using MHCII and CD38 as a substitute to LYVE1, as previously described 21 (Extended Data Fig. 1d).
Perivascular spaces are filled with CSF and constitute an interface between blood and the CNS parenchyma 22 . CSF flows along the perivascular space, can cross astrocytic endfeet and flow into the brain (a process termed glymphatic) 9,23 . We studied CSF flow dynamics by injecting fluorescent tracers into the mouse cisterna magna and then assessing tracer diffusion into the brain. Fluorescent ovalbumin (OVA) was delivered through intra-cisterna magna (i.c.m.) injection, and the mice were perfused after 1 h. The entire brain was extracted, fixed and imaged by light sheet microscopy (Extended Data Fig. 1e and Supplementary Video 2) or by stereomicroscopy ( Fig. 1d and Extended Data Fig. 1f). Using these methods, we observed that OVA was mostly located in the regions of the olfactory bulbs, the cerebellum and the middle cerebral artery (MCA) (Fig. 1d). Although most of the tracer accumulated at the perivascular space, 28.9% of tracer was sampled by cells ( Fig. 1d and Extended Data Fig. 1f).
Tracer penetration into the brain parenchyma is reportedly greater for small fluorescent tracers 24 . However, we found that tracers, independent of their size, accumulated in both perivascular and leptomeningeal spaces in CD206 + macrophages (or PBMs; Fig. 1e,f and Extended Data Fig. 1g). We also observed tracer uptake by PBMs when tracers were infused into the striatum (Extended Data Fig. 1h), which suggested that PBMs sample CSF content on its way into and out of the brain. Indeed, we observed double-positive PBMs 2 h after the tracers were co-injected into both the striatum and CSF, although CSF influx was reduced owing to concomitant intra-striatal injection, as previously described 23 (Extended Data Fig. 1i).
Given the close association of PBMs with CSF, we hypothesized that CSF flow dynamics may be partially controlled by PBMs. To test this hypothesis, liposomes containing clodronate (CLO) were administered through i.c.m. injections to deplete PBMs (about 75% depletion was achieved when brain tissue was examined 1 week later; Fig. 1g-i). We confirmed that both the number and the morphology of microglial cells were not affected by i.c.m. CLO liposome administration (Extended Data Fig. 2a). Immunohistochemistry (IHC) and single-cell RNA-sequencing (scRNA-seq) analyses also demonstrated that PBMs are the major population of brain border-associated cells that phagocytosed i.c.m.-injected DiI-labelled liposomes. By contrast, microglia and other stromal cells did not sample the tracer (Extended Data Fig. 2b-e) and hence could not be directly affected by the liposomes. Notably, we observed two major PBM subtypes on the basis of their gene expression  Fig. 2f-i). Specifically, CD206 + LYVE1 + MHCII lo/neg PBMs were highly phagocytic and endocytic cells, expressing high levels of scavenger receptors such as Cd163, Mrc1, Lyve1, Msr1 and Siglec1. Notably, these cells also upregulated genes involved in the interferon-γ (IFNγ) pathway such as Irf8, Ifitm2 and Ifitm6 (Extended Data Fig. 2h,i). Gene ontology pathway analyses highlighted important roles of LYVE1 + PBMs in metabolic processes and chemotaxis (Extended Data Fig. 2i). Conversely, CD206 + LYVE1 -MHCII + PBMs upregulated pathways that are involved in immune response, response to viruses, cytokine production, cell-cell adhesion and antigen presentation (Extended Data Fig. 2h,i). These results indicated that the two PBM subtypes have different functions.
To assess the role of PBMs in CSF dynamics, fluorescent OVA was given through i.c.m. injection 1 week after PBM depletion. After allowing the tracer to diffuse freely for 1 h, mice were perfused and the whole brain was extracted, fixed and imaged by stereomicroscopy (Fig. 1j). In PBM-depleted mice, the OVA coverage was significantly reduced (Fig. 1j). Brains were then sectioned to evaluate OVA coverage in coronal sections, a method commonly used to evaluate CSF influx 9,25 (Fig. 1k). OVA coverage of brain slices was also significantly reduced in PBM-depleted mice (Fig. 1k). Three days after CLO treatment, CSF flow was impaired to a lesser extent than 1 week after, which correlates with a lower level of (nevertheless, significant) PBM depletion at this time point (Extended Data Fig. 2j-m).
Our group recently showed that CSF flow is impaired after dural lymphatic ablation 25 . After 3 days of CLO treatment, the number of CD206 + dural macrophages located at the vicinity of the superior sagittal sinus was reduced (Extended Data Fig. 2n-o). Conversely, there was no effect of CLO treatment after 1 week on either dural lymphatic vessels or dural CD206 + macrophages, which indicates that the observed impairment of CSF flow in this study cannot be attributed to dural lymphatic ablation (Extended Data Fig. 2p-s). There was also no effect of CLO treatment on choroid plexus CD206 + macrophages (Extended Data Fig. 2t). However, a population of IBA1 + cells were absent in dCLNs after CLO treatment (Extended Data Fig. 2u), which are presumably sinus subcapsular macrophages that were depleted once CLO liposomes drained to dCLNs.
We repeated the CSF flow experiment using 4 kDa dextran conjugated with fluorescein isothiocyanate (FITC-dextran) or 3 kDa Texas Red, which are more diffusive owing to their small molecular weight. Similar to OVA, the influx of small tracers was impaired after PBM depletion (Extended Data Fig. 3a,b). Furthermore, we observed significant accumulation of tracers (both OVA and FITC-dextran) in the brain parenchyma 1 h after intra-striatal injection in PBM-depleted mice. This result suggests that both influx and efflux of CSF are impaired after PBM depletion (Extended Data Fig. 3c,d) or that CSF influx and efflux are interdependent, as previously described (for example, intra-striatal injection impairs influx of CSF tracers 23 ). To assess whether the impairment in CSF flow dynamics affects CSF protein content itself, we performed a proteomics analysis of CSF sampled from PBM-depleted mice and control mice (Extended Data Fig. 3e). Accumulation of synapse-related proteins in PBM-depleted mice, such as NRXN1, PTPRS, NRCAM and CDH2, was observed (Extended Data Fig. 3f-h). Moreover, there was accumulation of clusterin (CLU), apolipoprotein E (APOE) and amyloid precursor protein (APP), proteins that have been associated with AD, in the CSF of PBM-depleted mice (Extended Data Fig. 3i-k).
Next we used magnetic resonance imaging (MRI) to better evaluate the dynamics of CSF flow in vivo 26,27 (Fig. 2a). Diffusion of the contrast agent (Dotarem; 0.754 kDa) over time was reduced in PBM-depleted compared with control-treated mice (Fig. 2b,c and Supplementary Videos 3 and 4), without any notable effect of PBM depletion on ventricular size (Extended Data Fig. 3l-n). Intracranial pressure was also increased 1 week after PBM depletion, which normalized 3 weeks after depletion (Extended Data Fig. 3o), probably due to cells starting to repopulate the niche 17,28 . MRI analyses also revealed a reduction of tracer diffusion at the vicinity of the MCA (Extended Data Fig. 3p-r). IHC analyses showed that OVA coverage of dCLNs was reduced after PBM depletion (Extended Data Fig. 3s-v). However, there was no difference in OVA drainage to dCLNs, as assessed by in vivo imaging (Extended Data Fig. 3w-y). This is probably because of the rapid efflux of OVA from the dCLNs in PBM-depleted mice, as evidenced from live imaging of drainage (Extended Data Fig. 3y,z). This in turn is probably as a result of sinus subcapsular macrophage depletion by drained CLO liposomes (Extended Data Fig. 2u).
To better understand how PBM depletion affects CSF flow dynamics, we developed a new in vivo approach that enabled us to monitor fluorescent tracer movement over time. CSF tracers rapidly diffuse at the proximal part of the MCA perivascular space 6,27 . In addition, given the observed reduced tracer coverage around the MCA by MRI (Extended Data Fig. 3p-r), we decided to visualize fluorescent macromolecule movement through the intact lateral parietal bone after retraction of the right temporalis muscle (Fig. 2d,e). Immediately after i.c.m. injection of fluorescent OVA, mice were turned onto their side for in vivo stereomicroscopy imaging of the proximal part of the MCA through the intact skull ( Fig. 2d and Extended Data Fig. 4a-c). Using this method, we observed in vivo that OVA rapidly localized at the MCA perivascular space (Extended Data Fig. 4d) and was sampled by perivascular cells (Extended Data Fig. 4e). We validated this method by exposing mice to different anaesthetics that either enhance (ketamine and xylazine (KX) cocktail) or inhibit (isoflurane) the movement of CSF tracers 29,30 (Extended Data Fig. 4f-j). Using this approach, we observed that OVA coverage over time was strongly reduced in PBM-depleted mice at the MCA level. This was in accordance with ex vivo results showing that global tracer coverage was reduced (Fig. 2f,g and Supplementary Videos 5 and 6).
Collectively, these data provide evidence that PBMs are strategically located at the interface between blood and the brain parenchyma, they sample CSF and regulate its flow dynamics.

PBMs, extracellular matrix and arterial motion
After PBM depletion, OVA was barely able to reach the perivascular space of penetrating vessels (Extended Data Fig. 4k). This raised questions about the morphology of the perivascular space after PBM depletion. Aquaporin-4 (AQP4), a water channel present in the astrocytic endfeet that form the glia limitans (that is, the outer layer of the perivascular space), has been proposed as a mediator of CSF influx 23 . Using IHC, we were unable to detect any effect of PBM depletion on AQP4 coverage or AQP4 polarization (Extended Data Fig. 4l-n). By measuring the diameter of the perivascular space using the vascular marker CD31 in combination with AQP4 (Extended Data Fig. 4o), the space between the two markers (arguably representing the perivascular space) was substantially smaller in PBM-depleted mice than in control mice (Extended Data Fig. 4o-q). Using i.c.m.-injected fluorescent microbeads 6 , we were able to assess perivascular space in vivo (Extended Data Fig. 4r-t). However, this method did not allow assessment of the perivascular space in PBM-depleted mice because few beads could be detected (Extended Data Fig. 4s,t). To circumvent in vivo imaging, we perfused the mice with PBS without paraformaldehyde (PFA) to preserve the perivascular space, and bead accumulation was assessed at the vicinity of the MCA 6 (Extended Data Fig. 4u,v). First, we compared perivascular space measurements in PBS-treated mice using both methods. A slight decrease in ex vivo compared to in vivo measurements was observed (Extended Data Fig. 4u-w). Using this method, the bead coverage was strongly reduced in PBM-depleted mice (Extended Data Fig. 4x). Although the perivascular space and MCA diameter did not change (Extended Data Fig. 4y), the space filled by the beads was reduced in PBM-depleted mice (Extended Data Fig. 4z).
The changes in CSF flow dynamics were accompanied by mild behavioural alterations (Extended Data Fig. 5). PBM-depleted mice froze more on the first day of the cued-fear conditioning test (Extended Article Data Fig. 5a). However, PBM-depleted mice did not show any deficits in the following parameters: anxiety (assessed using the elevated plus maze test; Extended Data ). We also measured vital signs such as respiratory rate, heart rate and arterial pulsations and diameter, and did not find any difference between groups (Extended Data Fig. 5f).
To understand whether stromal cells (that is, endothelial cells and mural cells such as pericytes, vascular smooth muscle cells (VSMCs) and fibroblasts) are affected by PBM depletion, we sorted CD45 -CD13 + CD31 -(mural) cells and CD45 -CD13 -CD31 + (endothelial) cells and performed scRNA-seq 1 week after PBM depletion (Extended Data Fig. 6a). Fibroblasts (and pericytes, to a lesser extent) were the main producers of ECM, and ECM-related genes (Lgals3, Col1a2, Tgfb2, Lum, Col11a1 and Col8a2) were upregulated in PBM-depleted mice (Extended Data Fig. 6b). We also observed transcriptional changes in pericytes, mainly related to angiogenesis and DNA methylation (Extended Data Fig. 6c), as well as in capillaries (Extended Data Fig. 6d). However, there were no differences in terms of mural cell (Extended Data Fig. 6e), endothelial cell (Extended Data Fig. 6f) or VSMC coverage (Extended Data Fig. 6g).
Macrophages have been proposed to regulate ECM degradation in other organs through the production of matrix metalloproteinases (MMPs) such as MMP2 and MMP9, which degrade the ECM secreted by other cell types such as fibroblasts and VSMCs 26 . We proposed that PBM depletion could lead to changes in MMP activity, which in turn results in ECM accumulation that might interfere with CSF flow dynamics. Collagen type IV and laminin, both ECM proteins that are highly expressed in the brain, were accumulated around CD31 + vessels in PBM-depleted mice compared with vehicle-treated controls (Fig. 3a-c and Extended Data Fig. 6h-l). When analysed separately, we also observed an accumulation of ECM proteins at the vicinity of large surface and penetrating blood vessels stained positive for α-smooth muscle actin (αSMA + ) (Extended Data Fig. 6m,n). Furthermore, fluorescence spectrometry showed that MMP activity in the soluble brain fraction of PBM-depleted mice was significantly reduced (Fig. 3d,e). These results indicated that the functional perivascular space is regulated by PBMs through ECM remodelling.
Arterial pulsation 6,7 and vasomotion 8 are major drivers of CSF flow dynamics. It has been proposed that ECM deposition at the aorta level might affect its stiffness 31 . To test whether arterial motion might also be affected by PBM depletion, we created a small cranial window to observe vessel reactivity in vivo by photoacoustic microscopy (Fig. 3f). Head-restrained awake mice were placed on a moving platform and were able to move freely during image acquisition. At 1 min after the start of the imaging session, mice received 10% carbon dioxide (CO 2 ) in medical air through a nose cone to induce vessel dilation. Using this approach, we were able to record vessel diameter before and during the CO 2 challenge (Fig. 3f). Vasodilatory responses of arteries in PBM-depleted mice were impaired compared with control mice (Fig. 3g,h and Supplementary Videos 7 and 8). To confirm these findings, we assessed neurovascular coupling by recording vascular dilation through a thinned-skull cranial window at the barrel cortex level during contralateral stimulation of mouse whiskers 32 (Fig. 3i). Vessel dilation after whisker stimulation was impaired in PBM-depleted mice compared with control counterparts (Fig. 3j and Supplementary Videos 9 and 10). Systemic application of dobutamine, a β 1 -adrenergic agonist previously shown to enhance arterial pulsations 7 , rescued the impaired CSF dynamics in PBM-depleted mice, as determined by i.c.m. injection of OVA and assessment of its distribution through the perivascular spaces and its influx into the parenchyma (Extended Data Fig. 6o-q).
To better understand the biology of PBMs, we performed scRNA-seq from mouse brain samples (Fig. 3k, including Mrc1 and Ms4a7) and obtained 5 PBM clusters (Fig. 3k). To delineate crosstalk between PBMs and other cell types, we used the RNA Magnet algorithm, which predicts paired physical and signalling interactions between cell types 33 . Notably, we found that cluster 2, which exhibited high expression of scavenger markers such as Lyve1 and Cd163, interacted specifically with VSMCs, which are located at the arterial level (Fig. 3l). Spatial proximity between PBMs and VSMCs was confirmed by electron microscopy, which indicated that these two cell types may interact, although they are separated by the basal lamina (Extended Data Fig. 7b). These results confirmed recent findings that described important interactions between VSMCs and perivascular macrophages to allow their migration at the perivascular space 2 . Looking at differentially expressed genes between PBM clusters, cluster 2 was identified as potential professional scavenger cells (Cd163, Cd38, Lyve1, Msr1 and Cd36) when compared to the other clusters, which expressed genes related to antigen presentation (H2-Ab1, Cd74, Cd83, Cd14 and Nlrp3) (Extended Data Fig. 7c-e). Immunostaining confirmed that LYVE1 + PBMs were highly concentrated at the vicinity of αSMA + brain arteries and arterioles, whereas MHCII + PBMs were mostly localized to αSMAbrain blood vessels (Extended Data Fig. 7f,g). We then took advantage of LYVE1 expression by PBMs to genetically target these cells using  Fig. 7j-n). In support of the results from pharmacological ablation experiments, genetic ablation of LYVE1 + PBMs also affected ECM protein levels (Extended Data Fig. 7o,p) and CSF influx (Extended Data Fig. 7q-v). However, there were no changes in vessel coverage, MMP activity or intracranial pressure (Extended Data Fig. 7w-y), and CSF flow impairment was more subtle in these mice (Extended Data Fig. 7q-v). The differences may be because these mice have reduced PBM depletion compared with mice treated with CLO, or because a lifelong reduction in CSF flow resulted in adaptation or emergence of alternative pathways to regulate intracranial pressure. In summary, these observations suggest that PBMs regulate CSF dynamics along the perivascular space and its efflux and influx of the brain parenchyma. The mechanism underlying PBM regulation of CSF dynamics is based on their ability to regulate ECM remodelling, which in turn affects arterial stiffness.

PBMs in ageing and AD
CSF flow is impaired in old mice, and this impairment could be ameliorated in part through enhancement of meningeal lymphatic vessels 25 .

Article
We proposed that PBMs participate in the age-related deterioration of CSF dynamics. MRI analyses demonstrated that CSF flow is globally impaired in old mice. Fluorescent tracers also confirmed impairment at the MCA level, and different molecular weight fluorescent tracers confirmed impairment in brain coronal sections (Fig. 4a-c and Extended Data Fig. 8a-f). Comparison of the PBMs in young adult (3-month-old) and old (24-month-old) mice showed no difference in overall CD206 + cell numbers, but aged mice exhibited a significant reduction in LYVE1 + cells and an increase in MHCII + cells, which is in agreement with previous reports 21 ( Fig. 4d-g and Extended Data Fig. 8g-k). CD38 + PBMs in young mice were the major cell type that phagocytosed i.c.m. injected pHrodo particles (Extended Data Fig. 8l-o). The change in PBM phenotype observed in aged mice appeared to be associated with impaired pHrodo particle phagocytosis (Extended Data Fig. 8p-r), reduced functional perivascular space filled by the beads (Extended Data Fig. 8s-y) and accumulation of ECM proteins ( Fig. 4h-j and Extended Data Fig. 8z). M-CSF has been previously shown to improve pathophysiology of AD, presumably through the enhancement of amyloid-β (Aβ) phagocytosis by blood-derived monocytes 35 . We proposed that such acute activation of PBMs in old mice might enhance their ECM degradation capacity and improve CSF dynamics. Using scRNA-seq, we confirmed that brain cells expressing Csf1r were mainly PBMs, microglial cells and monocytes, although Csf1 was mostly expressed by microglial and mural cells (Extended Data Fig. 9a). A single i.c.m. injection of M-CSF (or artificial CSF (aCSF) as a control) was performed in old mice, and CSF dynamics was evaluated after i.c.m. injection of OVA tracer 6 h after M-CSF treatment ( Fig. 4k). At 1 h after OVA injection, brains were collected and analysed by stereomicroscopy. OVA distribution at the MCA level in M-CSF-treated old mice was significantly enhanced compared with control old mice (Fig. 4l). The same phenotypes were also found in brain coronal sections (Fig. 4m). Notably, acute M-CSF treatment in aged mice significantly increased MMP activity (Fig. 4n,o) and decreased ECM protein deposition ( Fig. 4p-r and Extended Data Fig. 9b, c). However, subacute M-CSF treatment (24 h) in aged mice was ineffective at enhancing CSF flow, although MMP activity was still significantly increased, but to a lesser extent than at 6 h. These results indicate that M-CSF treatment acutely drives MMP activity, which results in a short-term improvement in CSF flow (Extended Data Fig. 9d-g). Our scRNA-seq results revealed that VSMC-associated PBMs can be differentiated from other PBMs by their expression of scavenger receptors (Extended Data Fig. 7c-e). Pathway analysis confirmed that cluster 2 showed upregulated genes characteristic of receptor-mediated endocytosis and phagocytosis (Extended Data Fig. 7e). This analysis also showed that cluster 2 was linked to a cellular response to Aβ. Furthermore, it has been shown that depletion of perivascular macrophages worsens outcome in a mouse model of cerebral amyloid angiopathy (CAA) 36 , which suggests that PBMs are processing brain-derived Aβ. Finally, our CSF proteomics data indicated that PBM depletion leads to an accumulation of AD-associated risk factors such as CLU, APOE and APP (Extended Data Fig. 3i-k). We therefore proposed that PBMs might be involved in Aβ clearance. To test this hypothesis, 2-month-old 5×FAD mice (a mouse model of AD) each received a single i.c.m. injection of CLO liposomes or, as a control, PBS liposomes (Fig. 5a). CSF flow was impaired in PBM-depleted 5×FAD mice compared with their 5×FAD control littermates (Extended Data Fig. 10a,b). When their Aβ plaque loads were evaluated 1 month later, PBM-depleted 5×FAD mice exhibited significantly increased plaque load compared with their 5×FAD control littermates (Fig. 5b), specifically in the brain cortex and amygdala (Extended Data Fig. 10c). To better understand the role of PBMs in AD pathophysiology, we performed scRNA-seq on brains from 6-7-month-old 5×FAD and their wild-type (WT) littermates ( Fig. 5c and Extended Data Fig. 10d). We reclustered all macrophages and identified PBMs using the gene marker Mrc1 (Fig. 5d and Extended Data Fig. 10e). We also identified a damage-associated microglia (DAM) cluster specifically in 5×FAD mice, as previously described 37 (Extended Data Fig. 10f). PBMs from 5×FAD mice exhibited altered phagocytosis and endocytosis and response to IFNγ pathways ( Fig. 5e and Extended Data Fig. 10g). We confirmed using the RNA Magnet algorithm that PBMs interact with mural cells, notably VSMCs and fibroblasts (fibroblast-like cells) (Extended Data Fig. 10h). Of note, a human dataset that we had previously used to assess microglial function in AD 38 also contained a small population of PBMs (Fig. 5f). PBM populations from unaffected individuals and from patients with familial AD substantially differed, with 445 upregulated and 249 downregulated genes, respectively ( Fig. 5g and Extended Data Fig. 10i). Among the most notably dysregulated gene signatures in human PBMs from patients with familial AD were those involved in phagocytosis and endocytosis (CD163 expression) and IFNγ signalling ( Fig. 5h and Extended Data Fig. 10j), which recapitulated the findings from the AD mouse model. Notably, both the IFNγ receptor genes Ifngr1 and Ifngr2 were more highly expressed in brain immune cells (that is, microglia and PBMs) than in stromal cells in our mouse single-cell dataset, which indicated an important interaction between PBMs and IFNγ (Extended Data Fig. 10k). To test the possibility that excess IFNγ in CSF may cause dysfunction in CSF dynamics, we injected (i.c.m.) young adult WT mice with IFNγ. These mice exhibited impaired CSF flow compared with mice injected with PBS (Extended Data. Fig. 10l,m).
Collectively, these findings reveal that PBMs are pivotal players in CSF flow dynamics in ageing and in AD (Extended Data Fig. 10n). Therefore, PBMs should be further explored as potential new therapeutic targets  Article for AD and other age-associated diseases characterized by protein aggregation and CSF dysfunction.

Discussion
The results of this study demonstrated that perivascular and leptomeningeal macrophages express similar markers and are located around the CNS parenchyma, constantly interacting with CSF. Given their location, function and marker expression, we suggest referring to them as a single functional population, namely PBMs. PBMs are composed of two major subtypes: LYVE1 + MHCII lo/neg and LYVE1 lo/neg MHCII + . Although PBMs and microglia are both derived from early erythromyeloid progenitors in the yolk sac 39 , a recent study suggested that LYVE1 + macrophages predominantly originated from embryonic-derived progenitors and are maintained locally in the peritoneal mesothelium 34 . Our data demonstrated that LYVE1 + MHCII lo/neg PBMs regulate arterial motion and ECM remodelling (along large vessels and capillaries). Indeed, depletion or dysfunction of PBMs resulted in impaired arterial motion, accumulation of ECM and impairment of CSF flow. Previous studies that used CLO liposomes to deplete perivascular macrophages did not observe major changes in cerebral blood flow 17,19 . The apparent discrepancy may be a result of the use of cranial windows and dura mater removal used in those studies, which might have masked the effects we observed here with thinned-skull preparations. Spontaneous low-frequency oscillations of arterioles in brain parenchyma have been proposed as the driving force of CSF and interstitial fluid clearance 8 . Notably, this paravascular clearance was impaired in the context of CAA. Moreover, depletion of perivascular macrophages in CAA exacerbates the disease 36 , and perivascular macrophages can produce excessive reactive oxygen species, which then leads to neurovascular dysfunction in the context of hypertension and AD 17,19 . Together with our data, it is plausible that PBMs are at the interface between spontaneous arterial oscillations, ECM remodelling, neuronal activity and CSF flow.
Notably, aged mice exhibited an increase in LYVE1 lo/neg MHCII + PBMs, increased ECM deposition and impaired CSF dynamics. We showed here that treatment of aged mice with M-CSF increased MMP activity, reduced ECM accumulation and acutely restored impaired CSF flow. PBM depletion also resulted in increased accumulation of parenchymal plaques in the 5×FAD mouse model of amyloidosis. PBMs from patients with familial AD exhibited altered expression profiles compared with individuals without AD, including dysregulated pathways involving phagocytosis and endocytosis and IFNγ signalling on PBMs. Similar pathways were also among the most differentially regulated in PBMs of mice from WT compared with 5×FAD mice. Stimulation of PBMs in vivo in young mice through acute injection of IFNγ resulted in CSF flow impairment. Given the pleiotropic nature of IFNγ, however, its effect on CSF dynamics may not be solely as a result of its signalling on PBMs, although expression of both IFNγ receptors was higher in PBMs than other cell types in the brain. Future studies should analyse in further depth the role of IFNγ in CSF flow dynamics and developing intervention therapies for AD and other age-associated disorders based on IFNγ signalling or alternative modulations of PBMs.
Macrophages have previously been associated with cardiovascular diseases such as chronic hypertension 17,40 and with neurovascular coupling in both AD and homeostasis 41 . Although our appreciation of the part played by tissue-resident macrophages in tissue homeostasis has increased in recent years, the exact functions performed by the PBM population are still largely unknown. In this study, we unravelled an unexpected role for PBMs in CSF flow dynamics and demonstrated their potential therapeutic capacities in ageing and in AD. These findings may lead to the development of new therapeutic approaches for diseases associated with CSF dysfunction and with protein accumulation and aggregation, such as AD, Parkinson's disease, among others.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05397-3.

Treatments with i.c.m. injections
Mice were anaesthetized using an intraperitoneal injection of KX cocktail (100 mg kg -1 ketamine and 10 mg kg -1 xylazine) diluted in 0.9% Na (saline) solution. The fur of the neck was shaved and cleaned with 70% iodine. Then, mice were placed in a stereotaxic frame to maintain the head in a fixed position, and an ophthalmic solution was applied to prevent dry eyes. The skin from the neck was longitudinally incised, and muscles were retracted using hooks to expose the cisterna magna. The solutions, diluted in aCSF, were injected using a 33-gauge Hamilton syringe (1-5 μl; 2.5 μl min -1 ). The syringe was left in place for 1 min after injection to prevent backflow. For survival surgeries, the skin was sutured and mice were kept on a heating pad until fully awake. Mice received a subcutaneous injection of ketoprofen (2.5 mg kg -1 ) at the end of the surgery. To deplete PBMs, mice received an i.c.m. injection of CLO-loaded liposomes (5 μl; 5 mg ml -1 ; Fisher Scientific, CLD-8901). The control group consisted of mice that received an i.c.m. injection of PBS-loaded liposomes. The effect of PBM depletion was mostly studied 1 week after liposome injection to avoid any side effects from an inflammatory reaction due to the depletion. Moreover, at 1 week, we observed strong depletion (80-85% depletion), as previously described 28 . We chose to wait for 1 month instead of 1 week to evaluate the mid-to-long-term effect of PBM depletion on plaque accumulation in the 5×FAD mouse model of AD.
To evaluate the role of PBMs on CSF flow in old mice, 24-month-old mice received an i.c.m. injection of 5 μl aCSF or M-CSF (10 μg ml -1 diluted in aCSF; Sigma-Aldrich, M9170). Six hours or 24 h later, mice received an i.c.m. injection of OVA to evaluate CSF flow.
To evaluate PBM phagocytosis, mice received an i.c.m. injection of pHrodo particles (1 mg ml -1 in aCSF; Deep Red Escherichia coli bioparticles, Life Technologies, P35360). These particles emit 647 nm wavelength fluorescence only after being phagocytosed by cells (pH-dependent). Phagocytic activity was measured by pHrodo coverage by IHC, and pHrodo + cells was quantified by flow cytometry.

Intrastriatal injections
Anaesthetized mice (through KX cocktail) were shaved on the top of the head and placed in a stereotaxic frame. After skin incision, a small craniectomy was made using a drill, and the different solutions were injected using a glass capillary (1 μl; 0.2 μl min -1 ) (coordinates from the Bregma: anterior-posterior: +1.5 mm; medial-lateral: -1.5 mm; dorsal-ventral: +2.5 mm). The glass capillary was left in place for five additional minutes to prevent backflow. Mice were then sutured and placed on a heating pad until further experiments.

Dobutamine injection
For one experiment, mice received an intraperitoneal injection of dobutamine (40 μg kg -1 diluted in saline; Sigma Aldrich, D0676), a β 1 adrenergic agonist, or saline as a control, before CSF flow evaluation.
Proteomics analysis of CSF CSF collection. Mice were anaesthetized using a KX cocktail. The fur of the neck was shaved and cleaned with 70% iodine. Mice were then placed in a stereotaxic frame to maintain the head in a fixed position, and an ophthalmic solution was applied to prevent dry eyes. The skin from the neck was longitudinally incised and muscles were retracted using hooks to expose the cisterna magna. A glass capillary was inserted into the cisterna magna to collect CSF, which was transferred into 1.5 ml Eppendorf tubes for further analyses.
Peptide preparation. CSF (7-10 μl) samples from mice were dried in a speed-vac and solubilized with 30 μl of SDS buffer (4% (w/v), 100 mM Tris-HCl pH 8.0, and 0.2% dichloroacetone (DCA)). The protein disulfide bonds were reduced using 100 mM dithiothreitol with heating to 95 °C for 10 min. Peptides were prepared as previously described using a modification of the filter-aided sample preparation method 42 . The samples were mixed with 200 μl of 100 mM Tris-HCL buffer, pH 8.5, containing 8 M urea and 0.2% DCA (UA buffer). The samples were transferred to the top chamber of a 30,000 MWCO cut-off filtration unit (Millipore, part MRCF0R030) and spun in a microcentrifuge at 14,000 r.c.f. for 10 min. An additional 200 μl of UA buffer was added, and the filter unit was spun at 14,000 r.c.f. for 15-20 min. The cysteine residues were alkylated using 100 μl of 50 mM iodoacetamide (Pierce, A39271) in UA buffer. Iodoacetamide in UA buffer was added to the top chamber of the filtration unit.
The samples were gyrated at 550 r.p.m. for 30 min in the dark at room temperature using a thermomixer (Eppendorf). The filter was spun at 14,000 r.c.f. for 15 min, and the flow through was discarded. Unreacted iodoacetamide was washed through the filter with two sequential additions of 200 μl of 100 mM Tris-HCl buffer, pH 8.5 containing 8 M urea and 0.2% DCA, and the samples were centrifuged at 14,000 r.c.f. for 15-20 min after each buffer addition. The flow through was discarded after each buffer exchange-centrifugation cycle. The urea buffer was exchanged with digestion buffer (DB; 50 mM ammonium bicarbonate buffer, pH 8, containing 0.2% DCA). Two sequential additions of DB (200 μl) with centrifugation after each addition to the top chamber was performed. The top filter units were transferred to a new collection tube, and 100 μl DB containing 1 mAU of LysC (Wako Chemicals, 129-02541) was added and samples were incubated at 37 °C for 2 h. Trypsin (1 μg; Promega, V5113) was added and samples were incubated overnight at 37 °C. The filters were spun at 14,000 r.c.f. for 15 min to recover the peptides in the lower chamber. The filter was washed with 5 μl of 100 mM ABC buffer and the wash was combined with the peptides. Residual detergent was removed by ethyl acetate extraction 42 . After extraction, the peptides were dried in a speed-vac concentrator (Thermo Scientific, Savant DNA 120 speed-vac concentrator) for 15 min. The dried peptides were dissolved in 1% (v/v) trifluoroacetic acid (TFA) and desalted using stage tips (C18) as previously described 43 . The peptides were eluted with 60 μl of 60% (v/v) MeCN in 0.1% (v/v) formic acid (FA) and dried in a speed-vac (Thermo Scientific, Savant DNA 120 concentrator). The peptides were dissolved in 20 μl of 1% (v/v) MeCN in water. An aliquot (10%) was removed for quantification using a Pierce Quantitative Fluorometric Peptide Assay kit (Thermo Scientific, 23290).
The remaining peptides were transferred to autosampler vials (Sun-Sri, 200046), dried and stored at −80 °C.
Ultra performance liquid chromatography-Orbitrap mass spectrometry. The peptides were analysed using ultra performance liquid performance-Orbitrap mass spectrometry with the modifications described below. A volume of 2.5 μl of sample in 1% (v/v) FA (1%) was loaded onto a 75 μm i.d. × 50 cm Acclaim PepMap 100 C18 RSLC column (Thermo-Fisher Scientific) on an EASY nanoLC (Thermo Fisher Scientific) at a constant pressure of 700 bar at 100% buffer A (0.1% FA). Before sample loading, the column was equilibrated to 100% buffer A for a total of 11 μl at 700 bar pressure. Peptide chromatography was initiated with mobile phase A (1% FA) containing 2% buffer B (100% acetonitrile (ACN), 1% FA) for 5 min, then increased to 20% B over 100 min, to 32% B over 20 min, to 95% B over 1 min and held at 95% B for 19 min, with a flow rate of 250 nl min -1 . Data were acquired in data-dependent acquisition mode. The full-scan mass spectra were acquired with an Orbitrap mass analyzer with a scan range of m/z = 325-1,500 and a mass resolving power set to 70,000. Ten data-dependent high-energy collisional dissociations were performed with a mass resolving power set to 17,500, a fixed lower value of m/z of 110, an isolation width of 2 Da, and a normalized collision energy setting of 27. The maximum injection time was 60 ms for parent-ion analysis and product-ion analysis. The target ions that were selected for tandem mass spectrometry (MS/MS) were dynamically excluded for 30 s. The automatic gain control was set at a target value of 1 × 10 6 ions for full MS scans and 1 × 10 5 ions for MS2. Peptide ions with charge states of unassigned or one were excluded for higher energy collisional dissociation acquisition Identification of proteins. MS raw data were converted to peak lists using Proteome Discoverer (v.2.1.0.81, Thermo Fisher Scientific). MS/ MS spectra with charges greater than or equal to two were analysed using Mascot search engine (Matrix Science, v.2.7.0). Mascot was set up to search against a custom non-redundant UniProt database of mouse (version March 2021, 16,997 entries), assuming the digestion enzyme was trypsin with a maximum of 4 missed cleavages allowed. The searches were performed with a fragment ion mass tolerance of 0.02 Da and a parent ion tolerance of 20 ppm. Carbamidomethylation of cysteine was specified in Mascot as a fixed modification. Deamidation of asparagine, deamidation of glutamine, formation of pyro-glutamic acid from amino-terminal glutamine, acetylation of protein N terminus and oxidation of methionine were specified as variable modifications. Peptides and proteins were filtered at 1% false-discovery rate by searching against a reversed protein sequence database.

Intracranial pressure measurements
Mice were anaesthetized using a KX cocktail and placed in a stereotaxic frame. After the cisterna magna was exposed, a glass capillary containing a pressure sensor was inserted into the cisterna magna. A baseline measure was done in saline solution. The capillary was left in place in the cisterna magna for 3 min. Intracranial pressure was recorded using FISO Evolution software (v. 2.2.0.0). Respiratory rates were also measured using this method (breaths per min).

Western blots of isolated brain blood vessels
Brains were homogenized in a 2-ml tissue grinder in 1.5 ml of microvessel isolation buffer (MIB; 15 mM HEPES, 147 mM NaCl, 4 mM KCl, 3 mM CaCl 2 and 12 mM MgCl 2 ), containing a cocktail of protease inhibitors (complete, mini, EDTA-free protease inhibitor cocktail, Sigma-Aldrich). Samples were then transferred into a 15-ml conical tube and centrifuged at 1,000g for 10 min at 4 °C. The supernatant was removed and the pellet was resuspended in 5 ml of MIB containing 18% dextran (from Leuconostoc mesenteroides, MW 60,000-90,000; Sigma-Aldrich) and spun at 4000g for 20 min at 4 °C. The resulting supernatant was discarded, and the pellet was resuspended in 1 ml of MIB. The homogenate was then filtered through a 20-μm nylon filter (Pluristrainer 20 μm, Fisher Scientific). Microvessels were retained on the filter, whereas the parenchymal fraction was contained in the filtrate. The vascular fraction was then rinsed from the filter with PBS containing 1% BSA and pelleted at 4,000g for 10 min. The pellet was re-suspended in 1.5 ml PBS and transferred into an Eppendorf tube and pelleted again at 4,000g for 10 min.
Isolated vessels were lysed in RIPA buffer (Bioworld) containing protease inhibitors, and protein concentration was determined using a bicinchoninic acid assay kit (Pierce). Thirty micrograms of protein was separated by PAGE on 7.5% Mini-Protean TGX precast protein gels (Bio-Rad) and transferred onto nitrocellulose membranes. Ponceau S (Sigma-Aldrich) was used to confirm loading of equal amounts of protein and to monitor the transfer procedure. After blocking with blocking buffer (TBS (50 mM Tris, 150 mM NaCl, pH 7.6) containing 0.1% Tween-20 and 5% milk), the membranes were probed overnight (4 °C) with a primary antibody (collagen IV antibody 134001, Bio-Rad, 1:500) diluted in blocking buffer. Membranes were rinsed in TBS containing 0.1% Tween-20 and incubated with appropriate horseradish peroxidase (HRP)-conjugated secondary antibody (donkey anti-goat IgG (Abcam ab97120, 1:10,000) diluted in blocking buffer.

Behavioural tests
Behavioural tests were conducted at least 1 h after the dark cycle started. Behavioural tests were conducted in the following sequence: elevated plus maze > open-field test > three-chamber social-interaction test > forced-swim test (group 1) or cued-fear conditioning test (group 2). Each behavioural test was conducted with 50 dB white noise and at least 2 days apart to prevent stress. Experimental and social target mice were handled for 3 days before starting the first experiment. Before any behavioural tests, cages were located in a dark room with 50 dB white noise for 30 min.

Three-chamber social-interaction test.
The size of the three-chambered apparatus was 40 × 20 × 26 cm (width, height and depth, respectively), with a centre chamber that was 12 cm wide and side chambers that were 14 cm wide. Illuminance was kept at 50 lux. In the first session, the mouse could freely move around the entire three-chambered apparatus with two small containers in the left or right corner for 10 min (session 1). The mouse was then gently confined in the centre chamber while a new 'object' and a WT stranger mouse, 'stranger 1' (aged-matched C57BL/6J strain), was placed in one of the two plastic containers. The subject mouse was then allowed to freely explore all three chambers for 10 min (session 2). In the third session, the subject mouse was again gently guided to the centre chamber while the object was replaced with a WT 'stranger 2' mouse. The subject mouse again freely explored all three chambers for 10 min (session 3). Object and stranger exploration was defined by the nose of the subject mouse being oriented towards the target and coming within 2 cm of it as measured by EthoVision XT 15 (Noldus).
Open-field test. Mice were placed in an open-field box (35 × 35 × 35 cm) and recorded with a video camera for 60 min. The centre zone line was 9 cm apart from the edge. The testing room was illuminated at 0 lux. Mice movements were analysed using EthoVision XT 15 (Noldus).
Elevated plus maze. The elevated plus maze consisted of two open arms, two closed arms (for all arms, dimensions were 35 × 7 cm) and a centre zone, and was elevated to a height of 1 m above the floor. The illuminance of the closed arm was 80 lux, whereas the open arm was 120 lux. Mice were placed in the centre zone and allowed to explore the space for 8 min. Data were analysed using EthoVision XT 15 (Noldus).
Forced-swim test. Mice were placed into a 4-litre beaker that was three-quarter filled with tap water and recorded with a video camera on the top and side for 5 min. Water temperatures were kept at 20-21 °C, and the room was at 120 lux. Behaviour was manually analysed.
Cued-fear conditioning test. On the first day, mice were placed in the fear conditioning chamber (Ugo Basile). Before starting the experiment, 70% ethanol was sprayed once in the chamber. After 2 min of acclimation in white noise and the experimental condition, 1 kHz tone was applied for 20 s accompanied by 0.7 mA electric shock in the last 2 s. After 1 min of waiting time, tone-to-waiting was repeated three times in total. After repetition, mice were taken out of the conditioning chamber and returned to their home cage. On the second day (24 h after), the chamber was wiped with vanilla-flavoured oil, and the floor was exchanged with a grey-coloured plastic plate. Mice were placed in the chamber. After 2 min of acclimation to white noise, a 1 kHz tone was applied for 1 min, followed by 2 min of white noise. Mice were returned to the home cage. After 7 days of conditioning, the protocol for the second day was repeated. The data were analysed using activity parameters of EthoVision XT 15 (Noldus). Freezing time was measured during the waiting time (for the first day) and during the cue (for the second and eighth day).

Tissue collection and processing
Mice received a lethal intraperitoneal injection of euthasol (10% v/v in saline, 250 μl) and transcardially perfused with PBS containing 10 U ml -1 heparin. In some experiments, mice received an intravenous injection of lectin (30 μl; Dylight 649 labelled Lycopersicon Esculentum; Fisher Scientific, DL-1178) 5 min before perfusion. After removal of the skin, muscles and mandibles, the head was drop-fixed in 4% PFA for 24 h. Then, the skull caps (skull and attached dorsal dura mater) were detached and brains were kept in 4% PFA for an additional 24 h (48 h in total). When collected, the dCLNs were drop-fixed in 4% PFA for 12 h. After fixation, the tissues were cryoprotected with 30% sucrose solution and frozen in Tissue-Plus OCT compound (Thermo Fisher Scientific). Brains were sliced (100-μm-thick sections) with a cryostat and kept in 24-well plates filled with PBS at 4 °C. The dCLNs were sliced (30-μm-thick sections) and collected on gelatin-coated slides. In one experiment, whole brains were post-fixed with 4% PFA, then washed with PBS and were directly stained and imaged by stereomicroscopy.

Ex vivo stereomicroscopy imaging
Mice received a lethal intraperitoneal injection of euthasol (10% v/v in saline, 250 μl) and transcardially perfused with PBS containing 10 U ml -1 heparin. After removal of the skin, muscles and mandibles, the head was drop-fixed in 4% PFA for 24 h. Then, the skull caps (skull and attached dorsal dura mater) were detached and brains were kept in 4% PFA for an additional 24 h (48 h in total). The whole brains were then placed on a Petri dish and imaged by stereomicroscopy. For OVA measurements, whole brains were imaged using the following parameters: CY5 channel: zoom = 0.78, exposure time = 2 s. Quantification of OVA coverage at the MCA level was done using the following parameters: CY5 channel: zoom = 5, exposure time = 500 ms. For bead or pHrodo experiments, brains were placed on the side, and quantifications were done using the following parameters: GFP or CY5 channels: zoom = 5, exposure time = 250 ms or 500 ms, respectively.

Light sheet microscopy
For whole brain imaging, the vDISCO method was used to clear the brain 44 . In brief, mice received an i.c.m. injection of OVA (5 μl). One hour later, mice received an intravenous injection of lectin (30 μl) and were perfused 5 min later. Whole brain was post-fixed in 4% PFA for 24 h, then permeabilized and cleared using vDISCO protocol with passive tissue immersion. Brains were immersed in ethyl cinnamate (Sigma-Aldrich, W243019) and placed in chambered coverglass (Thermo Fisher, 155360) for light sheet imaging (LaVision BioTec).

In vivo fluorescent tracer dynamics evaluation
Mice were anaesthetized with a KX cocktail, the head was shaved and mice were placed on a stereotaxic frame. A lateral incision was made between the right eye and the right ear. The temporalis muscle was gently separated from the temporal bone. The surface of the skull was cleaned with a cotton bud. After i.c.m. injection of the fluorescent tracer, mice were positioned on a heating pad on their side to expose the right temporal bone under a stereomicroscope (Leica, M205 FA). The average time between the i.c.m. injection and the first image was 2 min. Mice were imaged over 1 h (240 frames in total: 4 frames per min). At the end of the imaging session, mice were euthanized using euthasol and tissues were collected for further analyses. For one experiment, mice were anaesthetized with KX or isoflurane (induction at 4.5% and continuously exposed at 0.75-2%). The fold increase was measured by calculating the ratio between x value divided by the minimal value.
For in vivo dCLN imaging, mice were anaesthetized with a KX cocktail, and the fur of the neck was shaved. After receiving an i.c.m. injection of OVA (5 μl), mice were placed on supine position, the skin from the neck was incised and retracted using hooks to expose the dCLNs. Live imaging was done under the stereomicroscope, as previously described 38 (average of 10 min between the i.c.m. injection and the beginning of the imaging).

MRI
Immediately after i.c.m. injection of Dotarem (a gadolinium-based MRI contrast agent; 0.754 kDa; 5 μl) under KX anaesthesia (average time between i.c.m. injection and start of the MRI acquisitions: 2 min), mice were placed in a prone position on the MRI device (9.4 Tesla MRI, Bruker Biospin). During the imaging session, 0.5-0.8% isoflurane was provided through a nose mask to prevent mice from waking up. Body temperature and respiratory rates were monitored during imaging. Isoflurane levels were adjusted with respiratory rates, and body temperature was controlled using a heating pad. A series of post-contrast T1 Fast Low Angle SHot (FLASH)-3D weighted images were taken through the head with the following parameters: repetition time = 30 ms; echo time = 8 ms; number of echo images = 1; number of averages = 1; number of repetitions = 12; scan time = 272,640 ms per sequence (4 m 54 s); flip angle = 20, field of view = 160 × 160 × 80 μm with a 128 × 128 × 64 matrix; spatial resolution = 125 × 125 × 125 μm (8 pixels per mm; voxel size = 0.125 mm 3 ), number of slices = 64; receiving coil 4 elements RF ARR 400 1H M. The total acquisition time was about 1 h per mouse (4 m 54 s ×12 sequences per mouse). To calculate the volume of ventricles, mice were placed on supine position on the MRI device (7 Tesla MRI, Bruker Biospin). T2-weighted sequences were taken through the mouse head using the following parameters: repetition time = 3,000 ms, time to echo = 139 ms, field of view = 26 × 20.5 mm, slice thickness = 0.13 mm, number of slices = 160 and number of excitations = 3 (total acquisition = 16 min per mouse).

In vivo photoacoustic microscopy
Mice were anaesthetized using a bolus of 4% isoflurane and medical air and maintained in 1-2% isoflurane during the surgical procedure. After fixing the head in a stereotaxic frame, a longitudinal incision was made to expose the skull. A small cranial window was made to be able to image brain blood vessels. Mice were sutured and allowed to recover after surgery in a clean cage. On the same day, mice were anaesthetized with 4% isoflurane and were transferred to the photoacoustic microscope, restrained using a nut and medical air was used for inhalation. The imaging session started 20 min after the mouse woke up. Thirty seconds after the acquisition started (average of 5,000 frames per mouse, frequency = 12 Hz), mice received a mixture of 10% CO 2 in medical air for 5 min. The vessel diameter fold-increase was measured by calculating the ratio between the measured values and the average of the first thousand values (before CO 2 challenge).

In vivo whisker stimulation
Mice were anaesthetized using a KX cocktail and placed in a stereotaxic frame. The body temperature was constantly monitored and adjusted using a heating pad. Whiskers on the right side of the mice were cut to about 0.5 cm. A longitudinal incision was made to expose the skull, and a thinned-skull window was made at the left barrel cortex to enable imaging of the distal part of the MCA. Mice were then placed under a stereomicroscope for imaging. Mechanical whisker stimulation was performed for 10 s, 6 s after the beginning of the imaging session. An average of ten videos were made per mouse (five with stimulations and five without stimulation), using the following parameters: brightfield channel; ×16 zoom; exposure time: 200 ms (5 frames per s). Heart rates (beats per min) and basal arterial pulsations (pulsatile amplitude per vessel diameter) were measured using the videos without whisker stimulations.

Electron microscopy
Mice received a lethal dose of euthasol by intraperitoneal injection and were transcardially perfused with warm Ringer's solution followed by perfusion with warm fixative containing 2.5% glutaraldehyde and 2% paraformaldehyde in 0.15 M cacodylate buffer (2 mM CaCl 2 at pH 7.4). Brains were transferred to a fixative solution and allowed to fix overnight at 4 °C. Brains were rinsed in cacodylate buffer 3 times for 10 min and fixed in 1% osmium tetroxide and 1.5% potassium ferrocyanide in cacodylate buffer for 1 h. Brains were then washed in ultrapure water 3 times for 10 min and stained in 1% thiocarbohydrazide solution for 1 h, followed by 1 h staining in aqueous 2% osmium tetroxide. Brains were rinsed in ultrapure water 3 times for 10 min and stained overnight in 1% uranyl acetate at 4 °C. The brains were then washed in ultrapure water 3 times for 10 min and stained with 20 mM lead aspartate at 60 °C for 30 min. After staining was complete, samples were washed in ultrapure water, dehydrated in a graded acetone series (50%, 70%, 90%, 100% three times) for 10 min in each step, and infiltrated with microwave assistance (Pelco BioWave Pro) into Durcupan resin. Samples were flat embedded and cured in an oven at 60 °C for 48 h. After resin curing, 70-nm-thick sections from brain cortex were prepared on copper grids, post-stained with uranyl acetate and Reynold's lead and imaged on a scanning electron microscope (Zeiss Merlin FE-SEM) using the followed parameters: voltage = 5.00 kV; probe current = 3.0 nA; and WD = 6.9 mm.

MMP activity assay
The left hemisphere of the brain was homogenized in 6 volumes of PBS using a mini bead beater (Sigma Aldrich) and 2.3 mm diameter zirconia/silica beads (Biospec). The homogenate was then centrifuged in a microcentrifuge (Eppendorf) at 14,000 r.c.f. for 20 min at 4 °C. Protein (100 μg) from the supernatant (representing the soluble fraction of the brain) was incubated with 25 μm of the quenched fluorescent MMP substrate BML-P128-0001 (Enzo Life Sciences) in PBS and incubated at 37 °C for 15 min. Afterwards, the fluorescence of the cleaved fluorescent product was measured using a BioTek Synergy H1 plate reader (excitation/emission = 340 nm/440 nm).

RNA-seq
Mus musculus DiI-liposome leptomeningeal scRNA-seq. Sample preparation. Fifteen mice (3-month-old males; C57BL6/J) received an i.c.m. injection of 5 μl (23 mg ml -1 ) of DiI-liposomes (Liposoma, I-005). Mice were perfused with heparinized PBS and tissues were collected the next day. Subdural meninges were gently collected from the brains and were dissected in FACS buffer and digested in a digestion solution containing 1:50 collagenase VIII, 1:500 DNase, 1:50 FBS in FACS buffer for 15 min. Tissues were then mashed through 70-μm strainers in 50-ml tubes containing FACS buffer and 10% FBS to stop enzymatic digestion. After centrifugation, the supernatant was removed and cells were resuspended in FACS buffer and transferred to a V-bottom plate. Cells were sorted, and only live Dil + cells were used for sequencing. Single-cell data pre-processing. Reads were aligned to the mm10 genome using the Cellranger software pipeline (v.3.0.2) provided by 10x genomics. The resulting filtered gene-by-cell matrix of unique molecular identifier (UMI) counts was read in R using the read10x-Counts function from the Droplet Utils package. Filtering was applied to remove low-quality cells by excluding cells expressing fewer than 200 or greater than 5,000 unique genes, having fewer than 1,000 or greater than 30,000 UMI counts, as well as cells with greater than 20% mitochondrial gene expression. Expression values for the remaining cells were then normalized using the scran and scater packages. The resulting log 2 values were transformed to the natural log scale for compatibility with the Seurat (v.3) pipeline 45-47 . Dimensionality reduction and clustering. The filtered and normalized matrix was used as input to the Seurat pipeline, and cells were scaled across each gene before the selection of the top 2,000 most highly variable genes using variance stabilizing transformation. Principal components analysis was conducted, and an elbow plot was used to select the first ten principal components (PCs) for t-distributed stochastic neighbour embedding (tSNE) analysis and clustering. Shared nearest neighbour (SNN) clustering optimized with the Louvain algorithm, as implemented by the Seurat FindClusters function, was performed before manual annotation of clusters based on the expression of canonical gene markers. Macrophages were then subset out, rescaled and clustered as above with the first 21 PCs and a resolution of 0.3.

M. musculus CLO depletion scRNA-seq. Sample preparation.
To deplete PBMs, mice received an i.c.m. injection of CLO-loaded liposomes (5 μl; 5 mg ml -1 ; Fisher Scientific; CLD-8901). The control group consisted of mice that received an i.c.m. injection of PBS-loaded liposomes. One week later, mice received a lethal intraperitoneal injection of euthasol (10% v/v in saline, 250 ml) and transcardially perfused with PBS containing 10 U ml -1 heparin. Lateral choroid plexuses were gently removed, then whole brains were digested, myelin was removed and brain were stained with CD13 (to stain for mural cells), CD31 (endothelial cells) and CD45 (immune cells). Stromal cells (that is, endothelial plus mural cells) were then sorted and prepared for single-cell sequencing. Single-cell data pre-processing. Reads were aligned to the mm10 genome using the Cellranger software pipeline (v.6.0.0) provided by 10x genomics. The resulting filtered gene-by-cell matrix of UMI counts was read in R using the read10xCounts function from the Droplet Utils package. Filtering was applied to remove low-quality cells by excluding cells expressing fewer than 200 or greater than 6,000 unique genes, having fewer than 1,000 or greater than 30,000 UMI counts, as well as cells with greater than 10% mitochondrial gene expression. Expression values for the remaining cells were then normalized using the scran and scater packages. The resulting log 2 values were transformed to the natural log scale for compatibility with the Seurat (v.3) pipeline 45-47 . Dimensionality reduction and clustering. The filtered and normalized matrix was used as input to the Seurat pipeline, and cells were scaled across each gene before the selection of the top 2,000 most highly variable genes using variance stabilizing transformation. Principal components analysis was conducted, and an elbow plot was used to select the first 30 PCs for tSNE analysis and clustering. SNN clustering optimized with the Louvain algorithm, as implemented by the Seurat FindClusters function, was performed before manual annotation of clusters based on the expression of canonical gene markers.
M. musculus brain scRNA-seq. Sample preparation. Four mice (4-month-old males; Aldh1l1Cre-ER T2 /RosaCAG-tdTomato (C57BL6/J background)) received 100 μl of tamoxifen solution (20 mg ml -1 ) on three consecutive days to induce expression of tdTomato in astrocytes. A sham surgery of ligation of the lymphatic vessels afferent to the dCLNs consisting of skin incision and sternocleidomastoid muscle retractation was also performed. Afterwards, the skin was sutured and mice were allowed to recover on a heating pad until fully awake. Five weeks after surgery, mice were perfused with heparinized PBS. Brains were gently dissected using scissors and incubated in digestion solution 3 times for 20 min each (brains were also mechanically ground using descending diameter plastic pipettes between incubations). Tissues were then mashed through 70-μm strainers in 50-ml tubes containing FACS buffer and 10% FBS to stop enzymatic digestion. Myelin was removed by transferring samples into 3 ml FACS buffer containing 22% BSA and centrifuged (1,000g, 9 accelerations, 2 decelerations for 10 min at 4 °C). The remaining supernatant and the myelin layer were carefully removed, the pellet resuspended in FACS buffer and transferred to a V-bottom plate. FcBlock solution (1:50; 50 μl) was added to the wells and after 20 min, and antibodies (CD13 (BD Biosciences, 558744), CD31 (BioLegend, 102516) and CD45 (eBioscience, 550994); 1:200) were added (total volume = 200 μl). Cells were enriched in CD11b and negative for Ly6G and sorted in four different categories: CD45 + CD13 -CD31 -(immune), CD45 -CD13 + CD31 -(mural), CD45 -CD13 -CD31 + (endothelial) and CD45 -dTomato + astrocytes. Overall, the recovered yield of astrocytes was low, and the maximum number of sorted astrocytes was used for sequencing. For the remaining cell populations, equal cell numbers were pooled and used for subsequent analyses. Single-cell data pre-processing. Reads were aligned to the mm10 genome using the Cellranger software pipeline (v.6.0.0) provided by 10x genomics. The resulting filtered gene-by-cell matrix of UMI counts was read in R using the read10xCounts function from the Droplet Utils package. Filtering was applied to remove low-quality cells by excluding cells expressing fewer than 200 or greater than 7,000 unique genes, having fewer than 1,000 or greater than 50,000 UMI counts, as well as cells with greater than 15% mitochondrial gene expression. Expression values for the remaining cells were then normalized using the scran and scater packages. The resulting log 2 values were transformed to the natural log scale for compatibility with the Seurat (v.3) pipeline 45-47 . Dimensionality reduction and clustering. The filtered and normalized matrix was used as input to the Seurat pipeline, and cells were scaled across each gene before the selection of the top 2,000 most highly variable genes using variance stabilizing transformation. Principal components analysis was conducted, and an elbow plot was used to select the first 30 PCs for tSNE analysis and clustering. SNN clustering optimized with the Louvain algorithm, as implemented by the Seurat FindClusters function, was performed before manual annotation of clusters based on the expression of canonical gene markers. Cell-cell interaction analysis. To evaluate potential cell-cell or ligand-receptor interactions in an unbiased way, the RNA Magnet package 33 was utilized with membrane, ECM, as well as both, ligandreceptor pairs queried and all vascular network cell types (arterial, capillary and venous endothelial cells, pericytes and VSMCs) included as anchors for RNAMagnetAnchors. Signalling interactions were also investigated with RNAMagnetSignaling, and the top signalling pair molecules were examined for both endothelial cells and mural cells with each other cell type present in the dataset. Macrophages were then subset out, rescaled and clustered as above with the first seven PCAs and a resolution of 0.3.
M. musculus 5×FAD scRNA-seq. Sample preparation. Mice received a lethal intraperitoneal injection of euthasol (10% v/v in saline, 250 ml) and transcardially perfused with PBS containing 10 U ml -1 heparin. Cortexes were gently dissected in PBS, digested and stained after myelin removal with CD13 (to stain for mural cells), CD31 (endothelial cells) and CD45 (immune cells). Stromal cells (that is, endothelial and mural) and immune cells were then sorted and prepared for single-cell sequencing.
Single-cell data pre-processing. Reads were aligned to the mm10 genome using the Cellranger software pipeline (v.6.0.0) provided by 10x genomics. The resulting filtered gene-by-cell matrix of UMI counts was read in R using the read10xCounts function from the Droplet Utils package. Filtering was applied to remove low-quality cells by excluding cells expressing fewer than 200 or greater than 7,500 unique genes, having fewer than 1,000 or greater than 40,000 UMI counts, as well as cells with greater than 25% mitochondrial gene expression. Expression values for the remaining cells were then normalized using the scran and scater packages. The resulting log 2 values were transformed to the natural log scale for compatibility with the Seurat (v.3) pipeline 45-47 . Dimensionality reduction and clustering. The filtered and normalized matrix was used as input to the Seurat pipeline, and cells were scaled across each gene before the selection of the top 2,000 most highly variable genes using variance stabilizing transformation. Principal components analysis was conducted, and an elbow plot was used to select the first 22 PCs for tSNE analysis and clustering. SNN clustering optimized with the Louvain algorithm, as implemented by the Seurat FindClusters function, was performed before manual annotation of clusters based on the expression of canonical gene markers. Cell-cell interaction analysis. To evaluate potential cell-cell or ligand-receptor interactions in an unbiased way, the RNA Magnet package 33 was utilized with membrane, ECM, as well as both, ligand-receptor pairs queried and all vascular, stromal and mural cell types (arterial, capillary and venous endothelial cells, ependymal cells, fibroblasts, pericytes and VSMCs) included as anchors for RNAMagnetAnchors. Macrophages were then subset out, rescaled and clustered as above with the first 14 PCs and a resolution of 0.6.
Homo sapiens single-nucleus RNA-seq. Sample preparation. The Neuropathology Core of the Knight Alzheimer's Disease Research Center and the Dominantly Inherited Alzheimer Network (DIAN) provided the parietal lobe tissue of postmortem brains for each sample. These samples were obtained with informed consent for research use and were approved by the review board of Washington University in St Louis. AD neuropathological changes were assessed according to the criteria of the National Institute on Aging-Alzheimer's Association (NIA-AA). From the 60 frozen human parietal lobes, approximately 500 mg of tissue was cut and weighed on dry ice using sterile disposable scalpels. The parietal tissue was homogenized in ice-cold homogenization buffer (0.25 M sucrose, 150 mM KCl, 5 mM MgCl 2 , 20 mM tricine-KOH pH 7.8, 0.15 mM spermine, 0.5 mM spermidine, EDTA-free protease inhibitor and recombinant RNase inhibitors) with a dounce homogenizer. Homogenates were centrifuged for 5 min at 500g, at 4 °C, to pellet the nuclear fraction. The nuclear fraction was mixed with an equal volume of 50% iodixanol and added on top of a 35% iodixanol solution for 30 min at 10,000g, at 4 °C. After myelin removal, the nuclei were collected at the 30-35% iodixanol interface. Nuclei were resuspended in nuclei wash and resuspension buffer (1% BSA and recombinant RNase inhibitors in PBS) and pelleted for 5 min at 500g and 4 °C. Nuclei were passed through a 40-μm cell strainer to remove cell debris and large clumps. Nuclei concentration was manually determined using DAPI counterstaining and a haemocytometer. Nuclei concentration was adjusted to 1,200 nuclei per μl and processed immediately following the 10x Genomics Single Cell Protocol instructions. We generated single-nucleus RNA-seq libraries using a 10x Chromium single cell reagent Kit v3 for 10,000 cells per sample and sequenced 50,000 reads per cell from 31 frozen human parietal lobes. Single-cell data pre-processing. We prepared a pre-mRNA reference according to the steps detailed by 10x Genomics based on the GRCH38 (3.0.0) reference, and reads were aligned to the using the Cellranger software pipeline (v.3.0.2). The resulting filtered gene-by-cell matrices of UMI counts for each sample were read in R using the read10x-Counts function from the Droplet Utils package. Filtering was applied to remove low-quality cells by excluding cells expressing fewer than 500 or greater than 10,000 unique genes, having fewer than 2,000 or greater than 100,000 UMI counts, as well as cells with greater than 25% mitochondrial gene expression. Samples were then randomly assigned to one of five cohorts and individually processed to screen for the possible presence of perivascular macrophages. Expression values for the remaining cells in each cohort were merged by gene symbol into one dataframe and normalized using the scran and scater packages. The resulting log 2 values were transformed to the natural log scale for compatibility with the Seurat (v.3) pipeline 47 . Each was then scaled across each gene before the selection of the top 2,000 most highly variable genes using variance stabilizing transformation. Principal components analysis was conducted, and an elbow plot was used to select PCs for tSNE analysis and clustering. SNN clustering optimized with the Louvain algorithm, as implemented by the Seurat FindClusters function, was performed before manual annotation of clusters based on the expression of canonical gene markers. After removal of cells identified as neurons, oligodendrocytes, oligodendrocyte precursor cells and astrocytes, the remaining cells were split by original sample identity. Integration, dimensionality reduction and clustering. Reference samples were chosen as those with more than 500 cells per sample and were prepped for integration utilizing the SCT normalization provided by Seurat with functions SelectIntegrationFeatures and PrepSCTIntegration. Sample integration was then performed with FindIntegrationAnchors specifying a k.filter = 100 and the reference samples determined above, and IntegrateData. Principal components analysis was conducted, and an elbow plot was used to select the first 30 PCs for tSNE analysis and clustering. SNN clustering optimized with the Louvain algorithm, as implemented by the Seurat FindClusters function, was performed before manual annotation of clusters based on the expression of canonical gene markers. Differential expression. For analysis of differentially expressed genes between conditions, each cluster was filtered to include genes that had at least 5 transcripts in at least 5 cells, then the top 2,000 highly variable genes were determined and included for further analysis using the SingleCellExperiment modelGeneVar and getTopHVGs functions. After filtering, observational weights for each gene were calculated using the ZINB-WaVE zinbFit and zinbwave functions 48 . These were then included in the edgeR model, which was created with the glmFit function, by using the glmWeightedF function 49,50 . Results were then filtered using a Benjamini-Hochberg-adjusted P value threshold of less than 0.05 as statistically significant. Pathway enrichment. Over-representation enrichment analysis with Fisher's exact test was used to determine significantly enriched Gene ontology terms (adjusted P < 0.05) for the sets of significantly differentially expressed genes. For each gene set, genes were separated into upregulated and downregulated, and separately, the enrichGO function from the clusterProfiler package was used with a gene set size set between 10 and 500 genes and P values adjusted using Benjamini-Hochberg correction.

Statistical analyses and reproducibility
All data are presented as the mean ± s.e.m. All the experiments (except single-cell and single-nuclei RNA-seq data) were repeated independently at least two times (biological replicates). Statistical significance was determined using two-tailed unpaired Welch's t-test (nonparametric) when comparing two independent groups or by paired t-test when comparing values from the same group. For comparisons of multiple factors, one-way or two-way analysis of variance (ANOVA) with appropriate multiple-comparisons tests were used. Statistical analyses were performed using Prism 9.0 (GraphPad software). Exact P values are all provided in figures. We provide all raw data and the statistical analyses as a PRISM file in Supplementary Data 1. All experiments were done blinded, and groups were revealed only after all the analyses were performed.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Mouse single-cell mRNA sequencing data are available at the Gene Expression Omnibus under the accession number GSE188285. The human single-nucleus data from the Knight ADRC is publicly available by request from the National Institute on Aging Genetics of Alzheimer's Disease Data Storage Site with accession number NG00108.v1 (https:// www.niagads.org/datasets/ng00108). To access the data from the DIAN brain bank, special request must be made using this URL: https://dian. wustl.edu/our-research/for-investigators/.

Code availability
Custom code used to analyse the mRNA sequencing data is freely available at https://doi.org/10.5281/zenodo.7047054.  Fig. 3 | CSF flow after PBM depletion. a, One week after PBM depletion, mice received an i.c.m. injection of FITC-Dextran (FITCDex; 4 kDa; 5 μl), brains were harvested one hour later and FITCdex (green) coverage was measured on coronal sections co-stained for DAPI. Representative images and corresponding quantifications are shown. Scale bar, 2 mm. n = 5 mice/group; two-tailed unpaired Welch's t-test. b, One week after PBM depletion, mice received an i.c.m. injection of Texas Red (3 kDa; 5 μl), brains were harvested one hour later and Texas Red (red) coverage was measured on coronal sections co-stained for DAPI. Representative images and corresponding quantifications are shown. Scale bar, 2 mm. n = 4 mice treated with PBS, and 5 mice treated with CLO; two-tailed unpaired Welch's t-test. c, One week after PBM depletion, mice received an intrastriatal injection of OVA (45 kDa; 1 μl) and brains were harvested one hour later. Representative images and corresponding quantifications are shown. Scale bar, 2 mm. n = 4 mice treated with PBS, and 5 mice treated with CLO; two-tailed unpaired Welch's t-test. d, One week after PBM depletion, mice received an intrastriatal (i.s.) injection of FITC-Dextran (FITCdex; 4kDa; 1 μl) and brains were harvested one hour later. Representative images and corresponding quantifications are shown. Scale bar, 2 mm. n = 4 mice/group; Two-tailed unpaired Welch's t-test. e, One week after CLO or PBS liposome injection, mice were anesthetized, and a glass capillary was inserted i.c.m. to collect CSF for proteomic analyses. f, Volcano plot corresponding to downand up-regulated proteins in CSF comparing PBM-depleted and control mice. F-test with adjusted degrees of freedom based on weights calculated per gene with a zero-inflation model and Benjamini-Hochberg adjusted P values. g, Corresponding GO Pathway analysis showing down-and up-regulated pathways in PBM-depleted and control mice. Over-representation test. h, Sunburst plot representing the location of the upregulated CSF-derived neuronal/synaptic-related proteins after PBM depletion. i-k, Quantification of relative spectral counts for i, Clusterin (CLU); j, Apolipoprotein E (APOE) and k, Amyloid Precursor Peptide (APP). For e-k: n = 4 mice treated with PBS, and 5 mice treated with CLO; two-tailed unpaired Welch's t-test. l, MRI based T2-weighted anatomical sequences were performed before and one week after PBM depletion. m, Representative T2 images showing lateral ventricles (in hypersignal) before and after PBM depletion. Scale bar, 2 mm. n, Quantification of ventricle volume in mm 3 . n = 5 mice/group; one-way ANOVA with Tukey multiple comparisons test. o, Intracranial pressure was measured one-(7d) and three (21d) weeks after PBM depletion. n = 5 mice treated with PBS, 7 mice treated with CLO at 7d; 6 mice treated with PBS, and 7 mice treated with CLO at 21 d; two-way ANOVA with Sidak's multiple comparisons test. . c, Open field test: quantification of the distance moved over an hour, the total distance moved, the time spent in the center of the box over an hour and the total time spent in the center of the box. d, Forced swim test: quantification of the total floating time (left) and the latency to float (right). For a-d: n = 10 mice/group; two-tailed unpaired Welch's t-test. e, Three-chamber test: mice were first exposed to a mouse (S1) or an object (O), and then to a previously-exposed mouse (S1) or a new mouse (S2). Quantifications of the total sniffing time and the total time spent in the chamber for the two tests. n = 17 mice treated with PBS, and 13 mice treated with CLO; two-tailed unpaired Welch's t-test. f, One week after PBM depletion, respiratory rate, heart rate, arterial pulsation and diameter were monitored. n = 5 mice/group; two-tailed unpaired Welch's t-test. All data are presented as mean values +/− SEM. Fig. 6 | See next page for caption.