Structural and functional conservation of non-lumenized lymphatic endothelial cells in the mammalian leptomeninges

Founder C57BL/6j mice were purchased from Charles River and male offspring were bred and maintained at UCL on a 12-h light/12-h dark cycle with ad libitum supply of food (Envigo 2018 Teklad global 18% protein rodent diet) and water. Mice were deeply anaesthetised with Euthatal then transcardially perfused with heparinised (5000 U/l) physiological saline, followed by 10% formalin solution (4% PFA in PBS, prepared fresh on day of perfusion). Freshly dissected brain hemispheres of 5 month-old mice were immersed in 4% PFA/PBS (all PBS was DEPC treated) solution, incubated for 24 h at 4 °C, washed twice with PBS and then immersed in 30% sucrose/PBS for 1–2 days at 4 °C. Fixed brain tissue was then washed twice in PBS, frozen in pre-cooled isopentane (2-methylbutane, Sigma-Aldrich) on dry ice and stored at − 80 °C. Frozen brain hemispheres were embedded in OCT and 30 μm coronal sections were cut on a Leica CM1850 cryostat.

Digoxigenin and fluorescein-labelled probes were made via standard protocols, amplified by PCR from cDNA using primer pairs found on Allen Mouse Brain Atlas [34] as described below.

The in situ hybridization protocol was adapted from a previous published protocol [3] with the following modifications:

30 µm-thick sections were transferred to a 12-well falcon plate into PBS/Tween 20 (PBST) and washed 3 × 10 min to remove sucrose. Brain sections were pre-hybridised overnight in hybridization solution at 60 °C, in a hybridization oven gently rocking until the sections sank. The following day riboprobes were pre-heated in blocks to 80 °C for 10 min prior to being mixed in hybridization buffer, at a final concentration of 100–300 ng/µL.

Brain sections were hybridised overnight at 60 °C in a hybridization oven with gentle rocking. Probe removal was done with a series of hot washes in preheated wash solution (50% Formamide, 1 × SSC, 0.1%Tween-20) at 65 °C for 3 × 1 h. Sections were subsequently washed in Maleic Acid Buffer (100 mM Maleic Acid, pH 7.5) and blocked for at least 1 h at RT in blocking solution (2% Boehringer Blocking Reagent in Maleic Acid Buffer; 11096176001, Roche). Anti-Fluorescein-POD antibody (11426346910, Sigma) was applied at 1:500 dilution in blocking solution, and incubated overnight at 4 °C. Unbound antibody was subsequently removed in a series of PBST washes at RT. Detection of fluorescein-labelled riboprobes was done using as substrates the bench-made fluorescein tyramide (green) in some wells and Cy3 tyramide (red) in others, as previously described [33]. Brain sections were incubated for 20 min in either fluorescein tyramide (1:500) or cy3 (1:100) diluted in PBST, at RT and with gentle shaking, protect from light. 2 µL of 0.5% H₂O₂ in PBST was added to 1 mL of the tyramide solutions for a final concentration of 0.001%. The tyramide reaction ran for 40 min protected from light and without agitation. Samples were subsequently rinsed and washed 3× 15 min in PBST. Samples were then washed for 10 min each in 30%, 50%, 75% and 100% methanol in PBST.

Sections were then incubated in a solution of 1% H₂O₂ in methanol for 30 min to inactivate the first peroxidase. Samples were then rehydrated in the wells in a reverse order for 10 min each at RT then washed 2× in PBST for the removal of all traces of methanol. Brain slices were incubated in blocking buffer for 1 h at RT. Anti-DIG POD (11207733910, Sigma) was added into the appropriate wells with the block at 1:1000 O/N at 4 °C. The following morning samples were washed 4 × 30 min in PBST at RT. Samples were washed several times in PBST and then mounted onto glass slides with coverslips using an anti-fade mounting media (CitiFluor, UK) and kept at 4 °C in the dark until imaging.

Brains from single and double fluorescent ISH were imaged on a Leica SP8 LSCM with 25× water of 63× oil immersion objectives.

Post-image processing using raw data, including image thresholding and maximum intensity z-stack projections (adjusted for printing), were performed using ImageJ 7.7.2 software. Iso surface rendered images were processed via Imaris 8.4.1 (Bitplane A.G.). Images and figures were assembled using Adobe Photoshop and Adobe Illustrator. Statistical analysis and graphs were prepared using Microsoft Excel, GraphPad PRISM version 7.05 and the R bootstrap library using Q3.R. Bootstrapping was run without an assumption of normality, with 100 k iterations and α = 0.05.

In zebrafish, BLECs are unambiguously identified as vegfr3 expressing cells that are in close association with meningeal blood vessels (Fig. 1a, b). Immunohistochemistry (IHC) on the cortical leptomeninges from a 17-week-old Tie2-GFP;NG2-DsRed double reporter mouse [25] using antibodies against mouse VEGFR3 also labelled cells that resided in close proximity to Tyrosine Protein Kinase Receptor 2 (Tie-2)-positive blood vessels (Fig. 1c, d). As in zebrafish, these cells did not associate with vessels that had penetrated into the brain. These cells did not correspond to Neural/Glial Antigen 2 (NG2)-positive pericytes or smooth muscle cells. Similar results were obtained with alternative VEGFR3 antibodies on paraffin-embedded tissue (Supplementary Fig. 1a, b) as well as by in situ hybridization against Vegfr3 mRNA (Supplementary Fig. 1d, e), ruling out antibody staining artefacts. To confirm the identity of these VEGFR3-positive cells as the mammalian BLEC homologue, we also examined mouse leptomeninges for the co-expression of VEGFR3, MRC1, and LYVE1, which are BLEC-associated markers in zebrafish. Although leptomeningeal cells expressed a heterogeneous combination of markers, numerous cells co-expressed all three tested BLEC markers (Fig. 1e–e’’’). Cell counts from independent brains found that VEGFR3 co-localized with LYVE1 95% (93–97%, n = 2 brains) of the time, while MRC1 co-localized with LYVE1 72% (70–77%, n = 3) of the time (Fig. 1f, g). Because MRC1 and LYVE1 are expressed in subsets of macrophages and dendritic cells, and VEGFR3 has been reported in peripheral, but not meningeal, macrophages [30, 71] we examined the leptomeninges for expression of the additional lymphatic endothelial marker, PROX1, which along with VEGFR3 has not been described in CNS border macrophages [18, 29]. IHC on paraffin-embedded sections with antibodies against PROX1 labelled leptomeningeal cells (Supplementary Fig. 1c). Double in situ hybridization confirmed that these Prox1-positive cells also expressed Vegfr3 mRNA (Supplementary Fig. 1f, h). Finally, we tried antibodies against the widely-used LEC marker, PODOPLANIN (PDPN), but, similar to a previous report [61], found in mouse tissue a nearly ubiquitous expression in the pia that extended into the glia limitans (Supplementary Fig. 2). Thus, the use of PDPN to identify individual cells in the meninges was not possible.

These data demonstrate that mouse leptomeninges contain a cell type that co-expresses at least three and likely four zebrafish BLEC markers that have not been described as co-expressed in other known leptomeningeal cell types. However, Mato/Fluorescent Granule Perithelial (FGP) cells, a phagocytic cell type of the mammalian meninges with auto-fluorescent inclusions, have been proposed to be the mammalian equivalent of BLECs [66]. We therefore tested whether zebrafish and mammalian BLECs auto-fluoresced under an epifluorescence microscope as was applied by Mato and colleagues [42] as well as by Venero Galanternik and colleagues [66]. No auto-fluorescent inclusions were detected with standard GFP, RFP and UV band passes. We also applied a 500–588 nm bandpass with high-intensity laser exposure under confocal microscopy. We detected auto-fluorescence from the blood vasculature but failed to observe auto-fluorescent inclusions in the meninges of either zebrafish (Supplementary Fig. 3) or mouse (Supplementary Video 1), suggesting that auto-fluorescence is not a consistent feature of this cell type.

Based on their expression profile similarity to zebrafish BLECs, we named these mammalian cells Leptomeningeal Lymphatic Endothelial Cells (LLECs) and sought to further characterize whether this cell type shared other features of BLECs.

As previously described [11, 36], transmission electron microscopy (TEM) of zebrafish meninges identified cells outside of blood vessel basal laminae that possessed distinctively large, homogenous, electron-dense spherical inclusions (Fig. 2a). TEM of adult mouse cortical leptomeninges also identified cells with large, homogenous spherical inclusions (Fig. 2b). Although we cannot rule out their existence, we have not been able to identify other cell types containing such large, homogenous inclusions. To confirm unambiguously that these cells corresponded to BLECs, we first performed immunogold EM for YFP on the meninges of zebrafish carrying a transgenic BLEC reporter [Tg (flt4:mCitrine)] and found the cells with large, round inclusions were labelled (Fig. 2c, asterisks). In mouse leptomeninges, immunogold EM with antibodies against LYVE1 labelled cells containing large, homogeneous inclusions that are features of BLECs (Fig. 2d, asterisks). LYVE1 is also expressed in meningeal macrophages, and immunogold EM additionally labelled an ultrastructurally distinct population of rounded cells that possess numerous, irregularly shaped, heterogenous inclusions (Fig. 2f, g) that are characteristic of macrophages. Consistent with this interpretation, a similar macrophage-like cell type was observed in the meninges of Tg(flt4:mcitrine) zebrafish, but these cells were never labelled by immuno-EM against YFP, indicating that these cells lack canonical BLEC marker expression. Together, these results show that the ultrastructural distinctiveness of zebrafish BLECs is conserved in mouse LLECs and that these cells are a distinct population from CNS macrophages in both species.

To further explore whether mouse LLECs are distinct from border macrophages, we investigated the developmental origins of these cells. In zebrafish and mice, meningeal macrophages are derived from a monocytic lineage and therefore require the transcription factor PU.1 to differentiate [2, 18, 43]. In contrast, zebrafish BLEC development does not require pU.1, consistent with their distinct ontological origin that is more similar to peripheral lymphatic vessels [36, 66]. We therefore examined whether cells that express LLEC markers were still present in the meninges of PU.1 knockout mice, which lack all macrophages [43].

Because PU.1 knockout mice are perinatal lethal, we first confirmed the presence of LLECs in the developing meninges (E12.5–E18.5) using IHC against LYVE1, PROX1, and MRC1 and found cells expressing these three markers as early as E12.5 (Fig. 3a–l). Similarly, in PU.1 knockout mice, which lack macrophages and other differentiated myeloid cells [18, 43], cells were still detected that co-expressed at least three BLEC markers (Fig. 4e–h), although MRC1 expression appeared reduced compared to wild-type controls (Fig. 4c, g). We confirmed these PU.1-independent cells were expressed in the developing meninges by double labelling these cells with antibodies to LYVE1 and LAMININ, which delineates the meninges from the parenchymal surface and surrounding tissue (Supplementary Fig. 4a, c) [10]. We conclude that murine LLEC development is not dependent on the PU.1 transcription factor, which is similar to zebrafish BLECs and unlike neighbouring macrophages [36, 66].

Zebrafish BLECs have the ability to internalize macromolecules up to 500 kD in size [36]. However, since not all molecules below this size limit (e.g., Evan’s Blue, molecular weight, 961 Da) can be internalized by zebrafish BLECs, we decided to examine whether disease-relevant molecules such as Amyloid β1-40 (Aβ; molecular weight, 5kD) can also be endocytosed by BLECs. We specifically chose Aβ1-40 over Aβ1-42 because Aβ1-40 is a major contributor to cerebral amyloid angiopathy (CAA, [60]) and experimentally aggregates less readily, making experiments much easier to control. Fluorescently-tagged Aβ was injected into a pocket of CSF dorsolateral to the posterior hindbrain ventricle of 7 days post fertilization (dpf) transgenic Tg(kdr-l:mCherry); Tg(flt4:mCitrine);casper zebrafish larvae, in which the vasculature is labelled with mCherry (red) and BLECs with mCitrine (green), and the subsequent distribution of Aβ was followed by live confocal imaging (Fig. 5a, c). After 30 min, Aβ accumulated in flt4/vegfr3-positive BLECs but not in the kdr-l-positive vasculature (Fig. 5c, d; Supplementary Video 2). To observe whether meningeal macrophages also internalized Aβ, we injected Aβ into a 7 dpf transgenic line, Tg(mpeg1:Gal4-FF)^gl25;Tg(UAS-E1b:nfsB.mCherry)^c264;Tg(flt4:mCitrine);nacre^−/−, in which BLECs express mCitrine (green) and macrophages/microglia express mCherry (red). After 45 min, Aβ again accumulated in BLECs, but no Aβ was detectable in any of the nearby mpeg1-positive macrophages (Fig. 5e, e′). We conclude that zebrafish BLECs outperform macrophages in the acute internalization of Aβ.

To test whether mouse LLECs internalize macromolecules, both 10kD cascade blue dextran and fluorescently-tagged Aβ (HiLyte-647) were slowly co-infused (1 µL/minute) into the cisterna magna of 2-month old mice (Fig. 6a, arrow) that were sacrificed after 60 min and processed for IHC. The 10kD dextran was detected within PROX1 and LYVE1 co-expressing cells (Supplementary Fig. 5), demonstrating that LLECs can internalize macromolecules. In addition to the walls of meningeal blood vessels (e.g. Fig. 6f′′), Aβ was also internalized by cells that co-expressed pairwise combinations of the BLEC associated markers MRC1, LYVE1, VEGFR3, and PROX1 (Fig. 6d–f′′′ and Supplementary Fig. 5). Although technical constraints precluded the simultaneous imaging of fluorescently-tagged Aβ and three LLEC markers, the high rate of LLEC marker co-localization (often > 90% e.g. see Fig. 1 and Supplementary Fig. 1) in the leptomeninges means the uptake rates observed in these pairwise combinations is a strong indication that LLECs expressing at least four markers are internalizing Aβ.

Since Aβ is preferentially internalized by BLECs and not by macrophages in zebrafish, we quantified the rates of Aβ internalization by these populations. On average, 95% of mouse LLECs, which express both VEGFR3 and LYVE1, had internalized Aβ after 60 min (Fig. 6c). In contrast, 72% of MRC1, LYVE1 double positive cells, which identifies both putative BLECs and macrophages, internalized Aβ, a significantly lower rate compared to LLECs (p < 0.05, bootstrap). The rates of Aβ internalization were even lower for cells expressing only MRC1, with only 30% of these cells internalizing Aβ (Fig. 6c). Together with the zebrafish data, these results suggest that a greater capacity to internalize Aβ is a hallmark feature of BLECs and LLECs that functionally distinguishes them from meningeal macrophages.

To determine whether LLECs are also present in human meningeal tissue, we performed immunostaining with antibodies against LLEC proteins on sections from post-mortem human brains that lacked obvious neuropathology (Table 1). As in zebrafish and in mice, VEGFR3, LYVE1 and MRC1 were all found to be expressed within a heterogeneous population of cells within the pia and subarachnoid space of the leptomeninges of these samples (Fig. 7a–c). We then examined post-mortem meninges obtained from elderly human samples (aged 88–92) showing various degrees of neuropathology (Table 1) and again found VEGFR3, LYVE1, and MRC1 expression (Fig. 7d–f, and Supplementary Fig. 6). VEGFR3 was also strongly expressed in the lining of leptomeningeal vascular walls (Supplementary Fig. 7e, black arrows), in a tight layer of interdigitated cells comprising the arachnoid barrier (Supplementary Fig. 7d, orange arrows) [63, 70], and in clusters of cells corresponding to the arachnoid trabeculae (Supplementary Fig. 7b), including cells with pale nuclei with small prominent nucleoli typical of leptomeningeal and meningioma cells (Supplementary Fig. 7c) [12, 62]. This helps to identify these cells as phenotypically leptomeningeal cells. These observations raise the possibility that human leptomeninges contain LLECs, although the expression of some markers such as VEGFR3 is more widespread in other meningeal tissue than observed in fish and rodents.

To determine if the cells in the pia and subarachnoid space that express lymphatic markers represent LLECs, we performed triple-labelling fluorescent IHC on post-mortem human brain tissue for MRC1, LYVE1, and VEGFR3. We found several examples of these three LLEC markers co-expressed in the same leptomeningeal cells (Fig. 7g–p and Supplementary Video 3). To emphasize the spatial locations of LLECs relative to vessels and each other and to more clearly make out their morphological features, we used the Imaris imaging software to make 3-D reconstructions from the serial confocal images of the leptomeningeal human post-mortem tissue (Supplemental Fig. 5e–h). Consistent with LLECs in zebrafish and mouse, a subset of triple-positive cells were closely associated with vessels within the leptomeningeal layer, while their cell bodies remained outside of the vascular basement membrane. Morphologically, these cells were similar to zebrafish and mouse BLECs, with long, thin processes near blood vessels. Several of these processes formed unusual loop-like structures of unknown functional significance that were also a common morphological feature observed in zebrafish BLECs (Supplementary Fig. 8).

Taken together, the co-expression of three lymphatic markers in subarachnoid cells with a distinctive morphological character supports the existence of LLECs in human leptomeninges.

We provide evidence that zebrafish BLECs have an equivalent cell type in both mouse and humans that we have named LLECs. Do LLECs represent a unique cell type? Although cells expressing the lymphatic markers of BLECs have not to our knowledge been previously described in the leptomeninges, there are three other meningeal cells types that share at least a few LLEC features, including Mato Cells/FGPs, meningeal macrophages, and telocytes [8, 16, 40]. We consider whether LLECs are equivalent to, or a subpopulation of, each of these cell types in turn.

The ability of BLECs to readily endocytose macromolecules up to 500 kD in an Mrc1-dependent manner [36] and for mammalian LLECs to internalize dyes of at least 10 kD suggests that they participate in the surveillance and clearance of CSF solutes. Mrc1 is a pattern recognition receptor that is involved in the identification, binding, and endocytosis of pathogens, including bacteria involved in tuberculosis (Mycobacterium tuberculosis) and meningitis (Streptococcus pneumoniae) and fungi such as Candida albicans [6, 17, 49, 54, 56, 57]. In other cellular contexts, MRC1 can be exploited by pathogens. For example, viruses such as Influenza type A can enter and infect MΦs via MRC1 [50], and Mycobacterium tuberculosis can co-opt lymphatic endothelial cells in lymph nodes to serve as a replicative niche following MRC-dependent endocytosis [35]. Whether LLECs also participate in the uptake of pathogens or can serve as a reservoir for infection remains to be determined. MRC1 also participates in immune organization, for example by trafficking lymphocytes along lymphatic endothelium to draining lymph nodes, a process also exploited by tumours to metastasize to nodes [41]. Whether the densely interconnected network of LLECs that reside in the meninges perform a similar immune organizational function, for example by providing signalling cues or sites of attachment for lymphocytes [53], is an interesting question for future investigation.

While BLECs and LLECs can internalize a diversity of macromolecules, the selective ability of BLECs/LLECs to internalize Aβ more readily than meningeal macrophages suggests that these cells may especially participate in the clearance of Aβ, which accumulates within the wall of leptomeningeal arteries in CAA [68]. Because CAA may commence with Aβ accumulation in leptomeningeal vessels during aging as well as in Alzheimer’s Disease patients [19, 55], and cases of CAA solely within the leptomeninges have been reported [58], LLECs are well positioned to participate in the progression of this disease. Interestingly, chemical ablation of MRC1-expressing perivascular macrophages (pvMφ) using mannosylated clodronate containing liposomes exacerbates the development of CAA in a mouse model [20]. Since MRC1-expressing LLECs would likely also be eliminated by this methodology, this result should be revisited to examine whether LLECs participate in playing a protective role against Aβ-related disease. If so, whether LLECs have been lost or have diminished function in patients with CAA should also be examined.

More generally, the discovery of LLECs expands the diversity of “brain” lymphatics beyond the lymphatic vessels of the dura mater to also include cells within the leptomeninges. Although the role of dural brain lymphatics in the clearance of the brain and CSF remains controversial, there are several possibilities by which LLECs may interact with dural lymphatics to participate in the drainage and macromolecular clearance of the CSF that warrant further exploration. First, although under normal conditions the LLECs are not lumenized like the dural lymphatics, zebrafish BLECs can transiently form tubes that penetrate the brain parenchyma upon brain vascular injury [13]. Whether mammalian LLECs can also form tubes in certain disease or injury contexts to directly participate in lymphatic drainage of the brain should be examined in future studies. Second, the loosely interconnected meshwork of BLECs/LLECs could physically interact with dural lymphatics to participate in solute removal. For example, fluorescent tracer imaging experiments have revealed lymphatic articulations from the dura into the subarachnoid space near tracer “hot spots” at several points along the skull [37], and a recent paper has characterized a set of dural lymphatic vessels in the base of the mouse skull that have valves and other structures of classical lymphatics, lie close to the subarachnoid space, and reside in a thinner part of the dura mater [1]. Whether the distribution of LLECs within the leptomeninges shows a similar intracranial variation is unknown, but holocranohistochemistry to preserve anatomical architecture [60] could facilitate mapping the positions of LLECs relative to dural lymphatic articulations and other important structures of the meninges.

The ability of LLECs to participate CSF macromolecular clearance (at least under experimental conditions) also makes them potentially interesting targets for clinical applications. For example, enhancing dural lymphatic development using soluble VEGF improved macromolecule drainage and cognitive impairment in an AD mouse model [44]. How the VEGFR3-expressing LLECs respond to VEGF signalling is not known, but LLECs may also have been affected alongside the dural lymphatics when manipulated with VEGFR signalling [1, 44]. Whether LLEC number or endocytic function can be enhanced, and whether this has a protective effect in animal models of CAA, AD, or other pathological states would not only demonstrate a critical role for LLECs in disease progression but would also highlight LLECs as a promising target for therapeutic intervention.