Advances and controversies in meningeal biology

Abstract

The dura, arachnoid, and pia mater, as the constituent layers of the meninges, along with cerebrospinal fluid (CSF) in the subarachnoid space and ventricles, are essential protectors of the brain and spinal cord. Complemented by immune cells, blood vessels, lymphatic vessels, and nerves, these connective tissue layers have held many secrets that only recently have begun to be revealed. Each meningeal layer is now known to have molecularly distinct types of fibroblasts. CSF clearance through peripheral lymphatics and lymph nodes is well documented, but the routes and flow dynamics are debated. Advances made in meningeal immune functions are also debated. This review considers the cellular and molecular structure and function of the dura, arachnoid, and pia in the context of conventional views, recent progress, and what is uncertain or unknown. The hallmarks of meningeal pathophysiology are identified toward developing a more complete understanding of the meninges in health and disease.

Introduction

The three connective tissue layers of the meninges covering the brain and spinal cord have been known since ancient times¹. The dura (tough), arachnoid (spider web-like), and pia (tender) mater—the latter two collectively referred to as leptomeninges (thin meninges), along with the enclosed cerebrospinal fluid (CSF) and overlying skull and vertebral column, are recognized as essential protectors of the central nervous system (CNS) from mechanical forces, infections, and other forms of injury (Fig. 1). The essential contribution of meningeal membranes to brain development and their involvement in diverse disease processes are also well documented. Bacterial and viral infections (meningitis), intrinsic tumors (meningiomas), and metastases from malignancies in other organs can be life-threatening, as can hematomas formed by hemorrhage from blood vessels located above or within the meninges. These and other aspects of meningeal pathophysiology are discussed in many reviews^2–5.

Fig. 1 |. Hallmarks of meningeal pathophysiology.

Illustration of the diverse and distinctive functions of the meninges as they relate to health and disease. The meninges, consisting of the outermost dura, central arachnoid, and innermost pia, are connective tissue layers that enclose cerebrospinal fluid (CSF) in the subarachnoid space (SAS) and ventricles and serve as mechanical protectors of the CNS. The dura contains lymphatic vessels and is innervated by nociceptive sensory axons, derived mainly from neurons of the trigeminal ganglion, that contribute to the symptoms and pathophysiology of migraine. The arachnoid barrier layer of the meninges has tight intercellular junctions that create a restrictive permeability barrier. Other regions are sites for regulated immune surveillance by a diverse repertoire of immune cells. Bidirectional communication for immune cell migration between skull bone marrow and dura through veins in bone channels has been proposed and is currently being studied. The SAS serves as a reservoir for CSF circulation and clearance through mechanisms involving dural lymphatics that drain to lymph nodes in the neck and elsewhere. The SAS also contains arteries that supply the brain and veins that drain into dural venous sinuses. These diverse meningeal functions can be compromised by traumatic head injury, hemorrhage, meningitis, or meningioma, and can also be impaired in neuroinflammatory and neurogenerative conditions and by metastases from tumors in other organs.

The purpose of this review is to consider the advances made in understanding the structure and function of the dura, arachnoid, and pia in the context of conventional views, recent progress, and what is still uncertain or unknown. With this perspective, controversial issues (Box 1) and open questions (Boxes 2–3) are identified and discussed where evidence is insufficient or conflicting. The strengths and limitations of previous and current data, interpretations, and approaches are considered in the setting of rapidly growing evidence with the goal of ensuring a solid foundation for defining the hallmarks of meningeal pathophysiology (Fig. 1) and developing a more complete understanding of meningeal functions in health and disease (Table 1).

BOX 1. Controversies in meningeal biology.

1. Do Prox1 expressing cells form another barrier in the meninges? Recent reports argue that the meninges consists of four cell layers, not three^{25, 28}. The putative 4^th meningeal layer is reported to consist of cells that express Prox1 and form a permeability barrier, dividing the SAS into functionally different inner and outer compartments, and are separate from arachnoid barrier cells²⁵. These claims have generated considerable debate^29–34. As evidence against the claims, subsequent studies reported that Prox1 mRNA and Prox1 promoter-driven fluorescent reporter mark a subset of VE-cadherin⁺ fibroblasts located in the inner arachnoid adjacent and adherent to arachnoid barrier cells^{6, 7}. This subsequent work also revealed that Prox1⁺/VE-cadherin⁺ arachnoid fibroblasts form a discontinuous cell layer lacking expression of the tight junction proteins claudin-11 and angulins, unlike arachnoid barrier cells⁷. Attribution of a barrier function to Prox1⁺ cells²⁵ could therefore be a misinterpretation of the barrier function of the adjacent arachnoid barrier cells^{6, 7}.

2. What is the functional significance of Prox1 expression in arachnoid fibroblasts? Prox1 is a master regulator for lymphatic endothelial cell differentiation from venous precursors during embryogenesis. Expression of Prox1 in venous endothelial cells downregulates blood endothelial cell-specific genes and upregulates lymphatic endothelial cell genes¹³³. However, Prox1 is also expressed in many non-lymphatic cells, including neurons, oligodendrocytes, retinal cells, neuroendocrine cells, lens cells, cardiac myocytes, liver and gut epithelial cells, and platelets³⁰. The contribution of Prox1 to defining the gene expression profile of a subset of fibroblasts in the inner arachnoid, but not in the arachnoid barrier or pia, merits further investigation, but it does not merit consideration as a resemblance to lymphatics, as proposed by some investigators^{25, 28}. Prox1 is reported to promote epithelial-mesenchymal transition in colon cancer cells by inhibiting E-cadherin via miR-9¹³⁴, raising the possibility that Prox1⁺ cells in the inner arachnoid are precursors to arachnoid barrier cells that upregulate E-cadherin after Prox1 is downregulated. Questions about why the arachnoid consists of multiple layers of transcriptionally distinct fibroblasts, with Prox1⁺/Cdh5⁺ cells and Slc22a2⁺/Cdh5⁺ cells next to the SAS and Cldn11⁺/Cdh1⁺ arachnoid barrier cells remain to be answered.

3. Do pial fibroblasts form a continuous layer or a reticulum? Contrary to previous evidence^{50, 54}, a recent report claims, based on TEM imaging and ERTR7 antibody staining, that the pia is a multi-layered reticular mesh of cells that loosely cover the brain surface¹³⁵. However, some cells in the TEM images in that report (Figs. 1 and S1¹³⁵) resemble arachnoid cells more than pial cells. As the leptomeninges usually remains attached to the brain after the calvarium is removed, perhaps the reported multi-layered pia reflects instead the pia and arachnoid together. Also, the ERTR7 antibody used¹³⁵ recognizes type 6 collagen¹³⁶, which is expressed and secreted by arachnoid fibroblasts as well as by pial cells⁷. Staining of type 6 collagen in the ECM by this antibody would be expected to show mainly a reticular mesh of extracellular collagen fibrils, unlike the largely continuous pial cell membrane (Fig. 3d–e). The ERTR7 antibody is, therefore, a useful tool for studying the ECM but not cell continuity in meningeal layers. More functional studies are needed to understand the role of ECM composition and extracellular arrangement in the arachnoid, pia, and subpial and perivascular components of the meninges.

4. Is the glymphatic hypothesis consistent with meningeal structure? Division of the SAS into two compartments by a Prox1⁺ putative barrier-forming 4th meningeal layer has been used to explain how CSF can selectively enter periarteriolar spaces at the origin of a unidirectional waste clearance system for brain macromolecules²⁵, referred to as the glymphatic system⁸⁷. Although frequently cited as established CSF physiology^{26–28, 36, 135}, many aspects of the glymphatic hypothesis⁸⁷ remain contested and unproven^137–139. Still debated is whether nutrients and metabolites in interstitial fluid of the brain parenchyma move by diffusion down a concentration gradient or by bulk (convective) flow driven by a hydrostatic pressure gradient. The glymphatic hypothesis supposes that pressure-driven bulk influx of CSF along periarterial spaces clears waste from the brain parenchyma by bulk efflux of interstitial fluid mixed with CSF through perivenous spaces by a process dependent on astrocyte aquaporin-4 (AQP4) water channels⁸⁷. However, a hydrostatic pressure gradient driving such convective influx and efflux has not been detected. It is unclear how meningeal layers could be arranged to enable CSF influx from the SAS into periarteriolar spaces while simultaneously enabling fluid efflux through perivenous spaces into the SAS en route to lymphatic vessels in the dura and around cranial and spinal nerves. Evidence that Prox1⁺ cells in the meninges are part of the inner arachnoid and do not form a barrier^{6, 7} weighs against these cells partitioning the SAS into two compartments. There is need for more studies using contemporary and newly developed methods to determine whether any meningeal layers functionally compartmentalize the SAS to direct CSF flow in ways that differ from the conventional view.

5. What is the functional significance of arachnoid villi and granulations? At the present time there is no accepted model for CSF transport through arachnoid granulations into dural venous sinuses⁵. Arachnoid granulations are small, infrequent, or absent in dural venous sinuses of mice and other rodents and in human neonates⁹⁶. They are also absent in some adult humans⁹⁶. Furthermore, the cells lining arachnoid villi and dural venous sinuses do not have structural features that would enable bulk outflow of CSF into blood by any known mechanism⁵. One possibility is that arachnoid granulations, when present, serve as a pathway for CSF outflow only under conditions of increased intracranial pressure¹⁴⁰. An alternative stems from a study of human arachnoid granulations visualized by MRI and examined postmortem histologically and by immunohistochemistry, which confirmed age-related differences and revealed additional variations in size, location, interior structure, and surface lining⁴⁷. The latter findings were interpreted as evidence that arachnoid granulations have channels that serve as passageways for CSF to reach the tissue around dural venous sinuses⁴⁷. Dynamic studies of CSF movement are needed to test this concept.

BOX 2. Open questions about meningeal structure.

1. What is the most meaningful terminology for the principal cells of meningeal layers? Terms used in the historical literature include endothelial cells⁵³, mesothelial cells^{93, 129}, neurothelial cells⁶⁶, and more recently epithelial-like cells⁴. Use of the term epithelial cells implies a junction-connected cell layer with accompanying basement membrane, similar to the cells lining the airways, gut, urinary tract, and other organs with an epithelial lining; whereas endothelial cells implies similarity to the cells lining blood vessels and lymphatics, and the term mesothelial cells implies similarity to the cells lining the peritoneal, pleural, and joint spaces. Both pial cells and arachnoid barrier cells are associated with basement membranes, yet both also make multiple collagens and other ECM proteins typical of fibroblasts. Referring to all meningeal cells as fibroblasts¹⁴¹ has been questioned, because arachnoid barrier cells make E-cadherin - a classical epithelial protein - and also have tight junctions, similar to junctions underlying the barrier function of epithelial cells and endothelial cells. However, scRNA-seq analysis has shown greater similarity of all meningeal cell types to fibroblasts than to other classes of cells⁷. Moreover, the transcriptional diversity of meningeal cells fits the observed diversity of fibroblasts in other organs⁵². The classification of fibroblasts and other mesenchymal cells with mixed characteristics, including mesothelial cells, synoviocytes, and myoepithelial cells, is however unfinished, and further work is needed to characterize the diversity of cell types in the dura, arachnoid, and pia and to determine the terms that best fit the molecular and functional features of these cells.

2. Is the pial layer continuous with perivascular fibroblasts? Electron microscopic imaging has shown that pial fibroblasts, variably referred to as pia mater cells¹³¹, pial cells¹³², or leptomeningeal cells^{51, 76} (Table 2) are continuous with the sleeve of fibroblasts in the adventitia of pial blood vessels^{76, 130}. Although the two fibroblast populations are connected by adherens junctions⁷, debate continues on the exact configuration and location of the convergence of perivascular sleeves with the pia and the depth it extends from the brain surface^{141, 142}. What was initially considered non-convergence, with the separation resulting in a continuous pathway between the SAS and perivascular (Virchow-Robin) spaces in the brain^{53, 93}, was later recognized as pial invaginations around penetrating blood vessels^{66, 131, 132}. Accordingly, Virchow-Robin spaces are continuous with the subpial space but are separated by the pia from the SAS (Fig. 2)¹⁴³. In SEM images, the convergence of pial fibroblasts and perivascular fibroblasts varies with vessel size and type from tent-like protrusions into the SAS to depressions in the brain parenchyma. Whether different cellular arrangements exist in different anatomical locations or around blood vessels of different sizes remains to be determined. Genetic mouse lines with fluorescently labelled pial and perivascular fibroblast cells (e.g., Col1a1 or Pdgfra driven reporters) could help address this issue.

3. Which brain compartments are continuous with the subarachnoid space? Although CSF in cerebral ventricles clearly communicates with CSF in the SAS via the foramina of Luschka and Magendie, a further connection is hinted by the apparent continuity of the basement membranes of the glia limitans, tela choroidea, and choroid plexus epithelial cells in the third and fourth ventricles⁵¹. Such continuity of a thin region of interstitium - the basement membrane - between the choroid plexus epithelium and the subpial space¹⁴⁴ could allow direct passage of extravasated plasma solutes into the SAS. If such a route indeed exists, questions regarding its capacity and functional significance should be addressed.

4. How do subarachnoid space dimensions vary among species? The reported depth of the SAS varies across species. Values of 570–676 μm in humans over the cerebral cortex and nearly twice that in the Sylvian sulcus have been obtained by optical coherence tomographic intravital imaging¹⁴⁵. This contrasts with estimates in mice of less than 10 μm over the cerebral cortex and up to 50 μm over the spinal cord near the dorsal vein⁶. As differences in SAS depth can influence CSF flow and immune cell trafficking, the differences limit extrapolation of CSF and cellular dynamics in mouse SAS to corresponding values in humans. Development of animal models in rats, pigs, or other species with larger SAS dimensions for intravital imaging of fluorescently labeled immune cells moving within CSF and across meningeal layers could overcome this limitation.

BOX 3. Open questions about meningeal function.

1. Is the arachnoid barrier functionally dynamic? The barrier properties of the leptomeninges have been localized and elucidated by following the dynamics of fluorescent or electron dense tracers injected into the CSF-filled cisterna magna or cerebral ventricles. Administration of fluorescent tracers into the cisterna magna of living VE-cadherin-GFP reporter mice confirmed that the arachnoid forms a barrier even to small molecular tracers⁶. However, other studies report that tracers or pathogens in CSF can accumulate along the dural sinuses in mice^{12, 42}. Tracer accumulation in the dura along the superior sagittal sinus was also observed in humans by MRI after intrathecal injection of the contrast agent gadobutrol¹⁴⁶. These observations were initially interpreted as evidence that molecules as small as gadobutrol (molecular weight 604.7) can pass from CSF across the arachnoid into the dura. However, if that interpretation is valid, the tracer would be expected to spread into the dura over the entire brain surface. Drainage of small molecules from CSF into the systemic circulation can be detected as soon as 20 minutes after injection in mice and within 2 hours in humans^{88, 147}. An alternative interpretation is that the contrast agent enters the systemic circulation and then reaches the dura near the sagittal sinus from the bloodstream¹⁴⁷. Technical advances in labelling and tracking native proteins in CSF are needed to overcome the limitations of studies that involve injections of contrast agents or tracers into the CSF that are prone to artifacts and misinterpretation of results.

2. What is the functional relevance of regional differences in meningeal structure? Anatomical studies across higher vertebrate species report similar layers in the cranial meninges and also in spinal cord meninges. However, regional differences in meningeal structure include gaps in the arachnoid barrier near the cribriform plate^{10, 102}. Openings in specialized regions of the arachnoid around veins that bridge from the SAS to the dura have also been reported⁸¹. In addition, regional differences in gene expression have been reported for the developing meninges²¹. The functional relevance of region-specific differences in meningeal layer structure and gene expression warrants further investigation.

3. Where can CSF access lymphatics through routes that bypass the arachnoid barrier? Although CSF egress near cranial and spinal nerves is well documented^{10, 11, 48, 98, 101}, further work is needed to identify the cellular and junctional features of the openings and connections at these sites. In particular, the CSF transit route along olfactory nerves across the cribriform plate to nasal lymphatics has been documented as a site where the arachnoid barrier is lacking^{10, 98}. Openings in the arachnoid barrier for CSF transit along veins that bridge the arachnoid to merge with dural venous sinuses have also been reported⁸¹. However, an earlier report examining the potential anatomical connections of bridging veins proposed the opposite hypothesis based on evidence that perivascular meningeal cells around these veins function as a barrier between the dura and CSF¹⁴⁸. Still needed are functional assessments of the relative magnitude of CSF egress, cell transit, and lymphatic uptake through such routes. As interpretations are currently based on findings after acute intracranial injection of tracers^{81, 148}, further studies are needed to determine the functionality and fraction of CSF drainage through such putative openings under physiologic conditions. Also to be considered is how the openings are normally regulated and change under pathological conditions. Additional questions to be addressed include whether CSF that exits through such openings can access the functionally specialized button junction, fluid-entry region of lymphatics in the dura. Intravital imaging of cell-specific reporter mice and 3-dimensional immunofluorescence imaging should help answer these questions.

4. What is the status of proposed pathways between the skull bone marrow and brain? Studies that have combined scRNA-seq, light sheet microscopy, ex vivo X-ray computed tomography, and other contemporary methods have questioned the existence of CNS immune privilege and proposed that immune cells can enter the CNS directly from bone marrow in the skull and vertebrae and cross the meninges without entering the bloodstream^{24, 110, 113, 149, 150}. The barrier function of meningeal layers that limit immune cell and/or protein exchange between the dura and CNS has not yet been tested with sufficient rigor in studies that have reported the existence of these potential pathways^{12, 42, 112}. More work is needed to identify and functionally test routes that could cross or circumvent meningeal barriers under normal conditions or in disease.

5. Do meningeal fibroblasts contribute to drug uptake and transport? Drug transporting and metabolizing mechanisms in dural cells could help protect the brain and spinal cord. Like arachnoid barrier cells, dural border cells and dural fibroblasts strongly express the multidrug resistance transporters Abcb1a and Abcg2. These cells also express multidrug and toxin extrusion proteins-1 (MATE1, encoded by the dural border cell marker Slc47a1) and −2 (MATE2, encoded by Slc47a2)⁷. Still to be determined are the relative contributions of arachnoid barrier cells, dural border cells, dural cells, brain endothelial cells, and choroid plexus epithelial cells to protecting the brain from low molecular weight drugs and xenobiotic substances through expression of drug efflux pumps and metabolizing enzymes, in addition to the physical barrier properties. Conditional deletion of transporters from arachnoid barrier cells by using newly generated Cre drivers⁸¹ could be useful for determining the contribution of the arachnoid barrier to drug uptake and transport.

6. How are the barrier properties of pial blood vessels induced and maintained? Although pial blood vessels in the subarachnoid space are surrounded by fibroblasts and not glial cells, they provide an endothelial barrier that prevents small molecules from passing from the blood into CSF comparable to the blood-brain barrier². The barrier properties of brain endothelial cells are dynamically reinforced by neuronal and astrocyte-derived Wnt signaling and pericyte contacts¹⁵¹. This feature raises the question of whether meningeal fibroblasts play a yet unexplored role in maintaining the barrier properties of endothelial cells of pial blood vessels. Data from scRNA-seq analysis show that pial perivascular fibroblasts express mRNA for Wnt4, Wnt 5a, Wnt 6, Wnt 9a, and other Wnt ligands, raising the possibility that Wnt signaling activation promotes the barrier properties of neighboring pial vascular endothelial cells^{7, 21}.

Table 1 |.

Functions of the three meningeal layers

Function	Dura mater	Arachnoid mater	Pia mater
Mechanical protection & support	Dampens effects of mechanical trauma Falx cerebri and tentorium cerebelli restrict brain displacement Dural connective tissue and border cells can separate to create a subdural space	Encloses CSF which acts as a shock absorber and provides buoyancy to the CNS tissues	Viscoelastic properties allow expansion of brain or spinal cord with changes in brain volume Denticulate ligaments provide stability to the spinal cord in the vertebral column
Permeability barrier	Permeable to low molecular weight tracers in blood (40kDa HRP, 10kDa dextran, 3kDa dextran)	Arachnoid barrier layer: Size-restrictive physical barrier for small molecules and proteins (40kDa HRP, 40kDa dextran, 66kDa BSA-biocytin-TMR, 10 kDa dextran, 3kDa dextran) Metabolic barrier: Expression of solute transporters and efflux pumps Reported gaps near cribriform plate, arachnoid granulations, along dural venous sinuses– permissive to efflux-influx of CSF	Permeable to molecular tracers (3, 10, and 40 kDa dextran, 55 kDa BSA)
Arterial supply & venous return	Dural arterial supply is separate from the brain and spinal cord Houses venous sinuses, drainage site for brain, skull and scalp Dural capillaries have endothelial fenestrations and are more permeable than CNS capillaries with the blood-brain barrier	Bridging veins penetrate the arachnoid as they connect pial veins to dural sinuses. Arachnoid trabeculae attach to pial vasculature and potentially provide structural support	Houses pial vasculature (arterial and venous) that supplies blood to the brain and spinal cord Attachment to blood vessels through convergence of pial and perivascular fibroblast layers as the vessels enter and exit the brain Pial/parenchymal perivascular fibroblast surround blood vessels (adventitia), potential structural support
Immune barrier & surveillance	Contains a unique subset of dural macrophages in addition to B cells, B-cell precursors - concentrated along the sinus in mice Reported to receive cells directly from skull and spinal bone marrow in mice	Subarachnoid space harbors tissue resident border associated macrophages Upper border of CSF-filled subarachnoid space allowing for CNS immune surveillance	Supports immune cell interactions, including T-cell crawling Lower border of CSF-filled subarachnoid space allowing for CNS immune surveillance Could influence immune cell movement
CSF reservoir, circulation & clearance	Dural venous sinuses receive CSF outflow through arachnoid granulations (currently debated) Dural lymphatic vessels drain some portion of CSF	Arachnoid barrier layer confines CSF within the subarachnoid space Permissive for fluid clearance at cribriform plate, along cranial nerves, and potentially other locations Possible CSF outflow through arachnoid granulations (currently debated)	Allows diffusive transport between CSF and brain interstitial fluid Allows fluid communication to/from the perivascular spaces along penetrating blood vessels
Sensory innervation	Nociceptive afferent axons that mediate pain of migraine Release CGRP, substance P, and other vasodilating neuropeptides Mechanosensory functions	No known innervation	Innervation by sympathetic and parasympathetic autonomic nerves contributes to vasomotor activities of pial arteries

Advances in meningeal biology: Technical progress brings deeper understanding and new controversies

As experimental approaches and methods have advanced, so has the understanding of many aspects of meningeal structure and function. Among these are the distinctive features of individual cells that comprise the dura, arachnoid, and pia and their relation to adjacent bone, neural tissue, and enclosed CSF^{6, 7}. Also receiving increasing attention are meningeal lymphatic vessels as a route for CSF clearance^{5, 8–10}. Lymphatics have long been known to be present in the dura (here referred to as dural lymphatics) but not in other meningeal layers¹¹. Immune cell involvement in neuroinflammation and infection^{2, 12} and the contribution of dural sensory nerves to migraine^{13, 14} are other areas of active investigation. Progress is also being made in understanding the structure and function of the physical and transport barrier in the arachnoid that isolates CSF from interstitial fluid in the dura and beyond^{3, 4}. Related work is seeking to elucidate barrier properties that result in subdural hematomas forming specifically between the dura and arachnoid¹⁵, with the objective of understanding the cells and extracellular matrix (ECM) at the separation site^{7, 16}. This broadening interest in meningeal biology is building on the solid historical foundation and increasing knowledge of the protective functions of the dura, arachnoid, arachnoid trabeculae, CSF, and pia against mechanical injury of the brain and spinal cord^{17, 18}.

Technological advances

Advances in experimental approaches include 2-photon intravital imaging of meningeal layers in genetically engineered reporter mice, which has revolutionized the understanding of structure-function relationships^{6, 19}. Similarly, 2-photon and near-infrared imaging of fluorescent tracers and magnetic resonance imaging (MRI) of contrast agents infused into the subarachnoid space (SAS) are providing insights into the flow dynamics and clearance routes of CSF in living mice⁵. By building on earlier studies¹¹, these new approaches have elucidated the contribution of lymphatics to CSF clearance and neuroimmune responses^{5, 8–10}. Characterization of the meninges has also been advanced through single-cell RNA sequence analysis (scRNA-seq), which has revealed that meningeal membranes consist of multiple types of fibroblasts distributed in anatomically distinct layers^{7, 20, 21}. Additional work has highlighted the diversity of macrophages and other immune cells that normally reside in the meninges²² and of immune cells that migrate into or out of the blood vasculature, skull bone marrow, and brain parenchyma in response to injury^{4, 12, 23, 24}.

Current debates

Technological advances have also led to provocative claims about the meninges that are disputed by conflicting evidence and alternative interpretations (Box 1). The claim that a previously unrecognized fourth layer of meninges forms a barrier that divides the SAS into inner and outer compartments is a recent example²⁵ (Box 1a). This claim has been repeated in ensuing reviews and reports^26–28, but the validity of the original evidence has been questioned^29–34. Data from subsequent studies using different approaches support the conventional view of one subarachnoid compartment for CSF and three meningeal layers^{6, 7}, creating a controversy that deserves further investigation. This and other controversies in meningeal biology are considered in more detail in the following sections and in Box 1.

Unanswered questions

Many important questions about meningeal biology remain unanswered (Boxes 2–3). For some, the existing evidence is insufficient or too ambiguous for a definitive understanding. After the arachnoid barrier layer was found to have tight junctions that restrict permeability¹⁵, additional questions arose, including the barrier properties of other meningeal layers to macromolecules and cells^{3, 35}. Also unclear are the regions of the arachnoid barrier permissive to CSF drainage into dural venous sinuses or lymphatics en route to lymph nodes^{5, 8–10}. Despite solid evidence for CSF outflow through lymphatics associated with cranial or spinal nerves and some regions of dura^{5, 8, 10, 11}, structural connections between the SAS and these lymphatics have proven difficult to identify (Box 3c).

Similarly challenging has been determining the fraction of CSF that drains through lymphatics in specific regions of the head, neck, and spine^{5, 10}. Arachnoid granulations in dural venous sinuses have long been considered CSF exit routes in humans, but convincing evidence is still lacking⁵ (Box 1e). Claims that expansion of dural lymphatics by vascular endothelial growth factor-C can improve CSF clearance³⁶ have been brought into question by subsequent evidence to the contrary^37–39. Extensive work has implicated the meningeal innervation by trigeminal afferent nerve fibers in vasomotor changes, inflammation, and migraine, but the underlying mechanisms and adaptive advantages of such neural pathways remain incompletely understood^{13, 14}. This and other unanswered questions are considered in more detail in the following sections and in Boxes 2–3.

Refining meningeal layer identity

Since the first description in ancient medical texts, many histological and transmission and scanning electron microscopic (TEM and SEM) studies have characterized the cellular features and organization of the dura, arachnoid, and pia (Table 2). This work has also yielded numerous creative but potentially confusing names for the meningeal layers (Table 2) and the cells that form them (Box 2a).

Table 2 |.

Meningeal layer nomenclature, molecular and cellular characteristics, and species studied

Meningeal layers	Alternate names	Gene expression enrichment in mouse *potential **mouse and human	Documented cellular features	Documented species
Dura mater
Outer layer of dura	Inner periosteal layer126, outer dural border layer127	Col1a1, Matn4*, Npp4*7, 41	Adjacent to osteoblast layer of calvarium True periosteum in brain but not in spinal cord	Primates127, rodents41, 126
Dural connective tissue	Fibrous dura, middle fibrous layer of dura127	Col1a1high, Fxyd5, Slc47a1, Foxp2, Col14a17, 21	Extensive collagen fibrils with embedded dural fibroblasts	Primates15, 127, rodents15, 126, fish128
Dural border cell layer	Mesothelial cells66, 129, neurothelium66, inner dura layer40, subdural neurothel of the pachymeninx126, subdural mesothelium49, subdural compartment65	Col1a1 low , Fxyd5, Slc47a1, Crapb2, Foxp2 7	Fibroblasts with overlapping cell processes, connected to each other and to arachnoid barrier cells by cell-cell junctions	Primates15, 49, 127, rodents15, fish128
Arachnoid mater
Arachnoid barrier cell layer	Neurothelium40, neurothel130	Cdh1**47, Cldn11**48, Cdh5, Lsr, Klf5, Vim**47, Slc4a4, Dpp47, 21	Cells with long, overlapping processes, connected by tight junctions between two or three cells (tricellular junctions), no collagen fibrils	Primates15, 49, 127, rodents15, fish128
Inner arachnoid cell layer	Mesothelial cells (of the arachnoid)129, arachnoid reticular layer65, outer arachnoid layer40, 127, 130, inner arachnoid layer49	Slc4a4, Crapb2, Cdh5, Aldh1a2 Subtype specific expression: Ppp1r1a, Prox1, Slc22a27	Fibroblasts with long, overlapping cell processes, interconnected by cell-cell junctions (gap and adherens junctions, desmosomes) with pockets of collagen fibrils. Inner arachnoid cells are connected to arachnoid barrier cells by cell-cell junctions.	Primates15, 49, 127, rodents15 Fish* - loose arrangement of cells called the ‘inner layer’, no anatomically distinct pia128
Subarachnoid space
Trabeculae		Cdh5, Crabp2**	Collagen fibril core surrounded by fibroblasts interconnected by gap and adherens junctions, desmosomes, connected to inner arachnoid cells and pial cells by cell-cell junctions	Primates49, dogs50, rats78, mouse6
Pia mater
Pial cell layer	Inner arachnoid layer40, 130, inner and outer pial layers40, 130	Lama1, Lum, Col1a1, Cdh5, Lamc1, Col4a1, Ngfr**7	Fibroblasts with cell processes connected by adherens junctions, intercellular stomata.	Primates49, 131, dogs50, rodents6, 15
Pial perivascular fibroblasts	Adventitia76, 132, pial cells131, leptomeningeal cells76	Pdgfra 7	Fibroblasts that surround blood vessels in pial layer, interconnected by adherens junctions

Proposals have been made to subdivide the meninges into four²⁵ or even seven layers⁴⁰. However, the conventional terminology for the layers and cells links new findings to earlier literature and facilitates communication and understanding of the components. With this in mind, the terms used in this review reflect the historical names validated by contemporary molecular methodologies (Fig. 2, Table 2)^{7, 21, 41, 42}.

Fig. 2 |. Composition of meningeal membranes, barriers, extracellular matrix, and sites of hemorrhage.

The arachnoid and pial membranes consist of five transcriptionally distinct types of fibroblasts: dural border cells (brown); arachnoid barrier cells (pink); two types of inner arachnoid cells (blue and turquoise); and pial and perivascular fibroblasts (green). Also shown are dural fibroblasts (grey), dural immune cells (blue), dural nerves (brown), collagen fibrils (black), basement membrane (violet), blood vessels and lymphatics (yellow). Distinctive features of the meninges include: a. Dural fibroblasts are dispersed among dense bundles of collagen fibrils that form most of the dural volume. b. Dural border cell layers have little extracellular matrix and are sites of potential separation and subdural space formation. c. Arachnoid barrier cells in two overlapping layers are connected by adherens junctions and sealed by tight junctions and tricellular junctions. d. Inner arachnoid cells form two discontinuous layers, one Prox1⁺ and the other Sidt1⁺, that are structurally supported by basement membrane, collagen fibrils, and other extracellular matrix components. e. The subarachnoid space is spanned by trabeculae from the inner arachnoid to the pia that vary in size, density, and collagen content in different CNS regions and in different species. f. The pial membrane is a largely continuous cell monolayer with scattered intercellular openings (stomata). g. Pial fibroblasts are transcriptionally closely related to and merge with fibroblasts around blood vessels in the subarachnoid space and penetrating the brain parenchyma. h. The subpial space contains a meshwork of collagen fibrils that provide structural support to the pial membrane. i. The subpial space is continuous with Virchow-Robin spaces around brain blood vessels. j. Astrocyte cell bodies form the glia limitans over the brain surface, whereas astrocyte foot processes form the glia limitans around blood vessels. k. All blood vessels shown have a tight endothelial barrier, except for the dural sinus and fenestrated capillaries in the dura. l. Vascular smooth muscle cells (red-brown) surround arteries and arterioles. Mural cells on veins and capillaries are not shown. m. Meningeal hemorrhage includes epidural hematoma after injury to meningeal arteries at the dura-skull interface, subdural hematoma from damaged bridging veins that creates a space between layers of loosely attached dural border cells, and subarachnoid hemorrhage from aneurysms or other bleeding into the subarachnoid space.

Dura mater

Outer dura

The thick, leathery outermost meningeal layer, the dura, consists of multiple layers of dural fibroblasts embedded in a voluminous and dense extracellular matrix, consisting primarily of type 1 collagen (Fig. 2, Table 2). The dura also contains blood vessels, lymphatic vessels, sensory and autonomic nerves, and diverse populations of immune cells, which are considered in separate sections below. The outer part of the dura over the brain merges with periosteal fibroblasts and osteoblasts of the inner surface of the calvarium.

The molecular and spatial heterogeneity of dural fibroblasts reflects the three molecular and spatially distinct layers of the mouse fetal dura, where periosteal dural cells differ from other dural cells⁴¹. Similarly, scRNA-seq analysis has revealed multiple molecularly distinct fibroblast populations in the adult dura⁷, but their locations remain to be identified. Regions of dura around venous sinuses also exhibit immunological specialization^12,42. Further exploration of the heterogeneity of dural cells over the brain and spinal cord is likely to reveal additional regional molecular and functional specializations.

Dural border cells

The inner part of the dura is lined by so-called dural border cells, which are connected to each other and to arachnoid barrier cells by adherens junctions and gap junctions, with no intervening ECM. Because of these features, some dural border cells remain attached to the arachnoid after the calvarium is removed. In mice, dural border cells and dural fibroblasts are transcriptionally similar, but are molecularly distinguishable from pia, arachnoid, and perivascular fibroblasts by enriched expression of Slc47a1, Fxyd5 and Foxp2, among other genes (Table 2)⁷. However, dural border cells have lower expression of Col1a1 and several other fibrillar collagen genes than dural fibroblasts. Both dural border cells and dural fibroblasts have enriched expression of Foxp2, but only dural fibroblasts express Col14a1 (Table 2)⁴³. Although these examples of mRNA signatures apply to the meninges, many of the markers are also enriched in fibroblast populations in other organs and in some cases in other cell types (https://www.genecards.org).

Arachnoid mater

Diversity of arachnoid fibroblasts

The arachnoid, located between the SAS and dura, is composed of at least four different fibroblast-like cell types organized in distinct layers (Fig. 2, Table 2). As described in numerous light and electron microscopic studies, the inner part of the arachnoid, which forms the roof of the SAS, consists of cells with overlapping cell processes that are joined to each other and to the outer part of the arachnoid by adherens junctions and gap junctions and are separated by bundles of collagen fibrils and other ECM components^{1, 7, 15, 44, 45}. The outer part of the arachnoid consists of arachnoid barrier cells and dural border cells joined by tight junctions, tricellular junctions, adherens junctions, and gap junctions with little or no ECM between the cells^{15, 44} (Figs. 2, 3a–c). Of practical consideration, when the calvarium is removed in adult mice, the arachnoid remains adherent to the brain surface. Because most dural border cells remain attached to the arachnoid, they are treated as a constituent of the arachnoid but also part of the dura where some remain attached. The plane of mechanical splitting of dura from arachnoid is considered further in the context of subdural hematomas (see Mechanical Protection and Support and Arterial Supply and Venous Return sections).

Fig. 3 |. Intercellular junctions in arachnoid and pia.

a. Region of meningeal layers in Fig. 2 showing junctions marked by lines at cell borders. All cells shown are interconnected by adherens junctions. Arachnoid barrier cells also have tight junctions and tricellular junctions that contribute to the permeability barrier of the layer^{7, 43}. b. Diagram illustrating the location and components of tricellular junctions between arachnoid barrier cells (from⁷). Conventional bicellular tight junctions and adherens junctions are also present but are not shown for simplicity. c. Arachnoid barrier cells. Junctions in two overlapping layers are shown by staining of tight junction protein claudin-11, adherens junction protein cadherin-1/E-cadherin, and tricellular junction protein Lsr/angulin-1. Tricellular junctions (arrows) have strong staining for all three proteins. Some other junctions (arrowheads) have strong claudin-11 and E-cadherin but weak or no Lsr. Confocal microscopic images of adult mouse cerebral meningeal whole mount (from⁷). d. Pia. Historical drawing showing the continuous layer of pial cells interconnected by junctions (black). Silver nitrate staining of dog pia (from⁵³). e. Pia. Intravital 2-photon image of mouse spinal cord showing VE-cadherin-GFP (green) at pial cell borders and blood vessels (orange) marked by fluorescent albumin (from⁶). f. Pia. SEM image of stoma through which the subpial space is visible. Stoma (F). Cell nuclei (N). Macrophage (M). Dog cranial pia (from⁵⁰).

Fibroblasts of the inner arachnoid

The inner arachnoid has two of the four arachnoid fibroblast-like cells types. These are transcriptionally similar but distinguishable from each other and from other leptomeningeal cells by enriched expression of double-stranded RNA transporter Sidt1 among other genes in one type, and of protein phosphatase 1, regulatory subunit 1A (Ppp1r1a) and prospero-related homeobox 1 (Prox1) in the other. Neither forms a fully continuous layer (Fig. 2). Fluorescently labeled markers show that Ppp1r1a⁺ and Prox1⁺ cells are immediately next to but separate from arachnoid barrier cells⁷.

Fibroblasts of the outer arachnoid

The outer arachnoid consists of arachnoid barrier cells and dural border cells. Arachnoid barrier cells are distinguished from other arachnoid cells by expression of genes for adherens junction protein E-cadherin (Cdh1) (Fig. 3a–c) and dipeptidyl peptidase 4 (encoded by Dpp4) and by having ultrastructural features of tight junctions in freeze-fracture replicas¹⁵. They also express transcription factor Klf5 and have comparatively low expression of ECM genes^{7, 21, 46}. Unlike the inner arachnoid, the outer arachnoid is continuous and forms the main physical barrier of the meninges (see Permeability barrier section). The barrier in adult mice consists of two layers of overlapping arachnoid barrier cells covered on the outside by layers of dural border cells⁷ (Figs. 2, 3a).

The outside of the arachnoid barrier cell layer is anchored to dural border cells that are connected to each other and to arachnoid barrier cells by adherens junctions. Dural border cells express tight junction protein claudin-11 (Cldn11) and tricellular junction protein Lsr/angulin-1 similar to arachnoid barrier cells, but they lack expression of E-cadherin and instead express the solute transporter-encoding transcript Slc47a1 and other markers in common with dural fibroblasts⁷.

Arachnoid barrier cells have been considered epithelial cells because of their expression of E-cadherin and barrier forming properties (Box 2a), but advances in meningeal development revealed that arachnoid barrier cells arise from the same mesenchymal precursor as other meningeal fibroblasts⁴⁶ and have an overall transcriptome more similar to fibroblasts than epithelial cells⁷. Arachnoid barrier cells also express genes encoding mesenchymal markers, such as cadherin-11 (Cdh11) and vimentin (Vim). Similarly, human arachnoid cells have enriched expression of E-cadherin and vimentin^{47, 48}. Based on clustering of scRNA-seq data, arachnoid barrier cells are more similar to dural border cells than to inner arachnoid cells. Clustering also reveals that the extent of resemblance fits the anatomical order from outside-to-inside: dural border cells > arachnoid barrier cells > Ppp1r1a⁺/Prox1⁺ inner arachnoid cells > Sidt1⁺ inner arachnoid cells > pial cells⁷. The functional significance and implications of this heterogeneity in gene expression deserve further study.

Arachnoid fibroblasts that cover trabeculae

Delicate trabeculae, consisting of arachnoid-covered connective tissue (Fig. 2), span the SAS in all mammalian species studied (Table 2)¹⁷. In mice, trabeculae are present in the SAS around the spinal cord but have not been found over the cerebral cortex⁶, where cells or cell processes span the narrow SAS in place of typical trabeculae^{17, 44, 45}. Studies of trabeculae in humans show the collagen core is coated by fibroblasts connected by adherens junctions and gap junctions to each other and to arachnoid cells of the SAS roof and to pial cells of the SAS floor⁴⁹. Some trabeculae contact perivascular fibroblasts, which are transcriptionally similar to pial cells^{7, 50}. Most fibroblasts coating trabeculae in the mouse spinal cord are VE-cadherin⁺ but lack expression of Prox1⁶. In the human fetus, fibroblasts covering trabeculae are CRABP2⁺, which is also enriched in mouse arachnoid and dural fibroblasts but not pial or perivascular fibroblasts^{7, 21}. Although this marker profile for trabeculae fits the Ppp1r1a⁺ type of inner arachnoid fibroblasts, more studies of molecular identity are needed to determine whether trabeculae-associated fibroblasts represent a separate population that produces ECM components specialized for formation and maintenance of trabeculae.

Pia mater and perivascular fibroblasts

The pia is composed of flattened pial fibroblasts that cover the surface of the brain and spinal cord (Fig. 2, Table 2). The pia is a monolayer of cells (Box 1c) but is thicker in humans than in mice⁵¹. Pial cells are closely related and connected by junctions to perivascular adventitial fibroblasts around pial blood vessels on the brain surface. The two types of fibroblasts are transcriptionally indistinguishable⁷, even though pial fibroblasts neighbor astrocytic end-feet and perivascular adventitial fibroblasts neighbor vascular mural cells. Both of these cell populations are identified as fibroblasts by their high expression of ECM genes Col1a1, Col3a1, Dcn, Lum, among others^{7, 21}, which are widely accepted albeit imperfect markers of fibroblasts in other organs⁵².

Pial fibroblast barrier properties and stomata

Pial fibroblasts are interconnected by adherens junctions around much of their circumference, based on TEM imaging^{7, 15, 44, 45}. Pial cells joined by adherens junctions form a nearly continuous layer over the floor of the SAS and surrounding pial blood vessels^{6, 7}, as initially described in the 19^th century by Key and Retzius⁵³ (Fig. 3d, Boxes 1c, 2b). TEM evidence of adherens junctions fits with VE-cadherin (Cdh5) expression identified in pial cells by scRNA-seq analysis and by VE-cadherin-GFP fluorescence at pial cell borders in transgenic reporter mice^{6, 7} (Fig. 3e). The pial cell layer has scattered circular intercellular openings, called stomata, but is otherwise largely continuous^{50, 54} (Figs. 2, 3d–f).

The subpial space, located between pial fibroblasts and astrocytic end-feet of the glia limitans (Fig. 2, Box 2b–c), contains types 4 and 6 collagen, laminin, nidogen-1, and other ECM components^{44, 45}, consistent with the high expression of ECM genes in pial fibroblasts (Table 2)⁷. Collagen fibrils are also found in the narrow spaces between fibroblasts in the adventitia of pial blood vessels.

Pial fibroblast relationship to blood vessels

Perivascular fibroblasts surround not only blood vessels on the brain surface but also arterioles and venules within the brain and spinal cord parenchyma (Box 2b). Parenchymal perivascular fibroblasts have been visualized live or after fixation in mouse transgenic lines that express Col1a1-GFP^{55, 56}, Col1a2-CreERT¹⁹, Pdgfra-H2B-GFP²⁰, or Lum-CreERT⁷. Perivascular fibroblasts extend to 4^th order arterioles, although the coverage is incomplete¹⁹. Pdgfra-H2B-GFP⁺ fibroblasts surrounded by LAMA1⁺ basement membrane are found around arteries, arterioles, venules, and veins but not capillaries in mouse brains²⁰.

Parenchymal perivascular fibroblasts in mice are transcriptionally closely related to pial perivascular fibroblasts but can be distinguished by scRNA-seq and in situ mRNA hybridization. Parenchymal perivascular fibroblasts have higher expression of Col15a1, Col12a1, Spp1 and several Col4 subunits, whereas pial perivascular fibroblasts have higher expression of Mgp, Serpine2 and Clec3b⁷. These differences could reflect different developmental programs⁵⁶. The question of whether pial fibroblasts and perivascular fibroblasts form a continuous sleeve is debated (Box 2b).

Mechanical protection and support

One of the essential functions of the meninges is providing mechanical protection to the brain and spinal cord. The dense fibrous dura is considerably stiffer than the underlying neural tissue and dampens mechanical forces exerted on the skull and spinal column. The leptomeninges, supported by arachnoid trabeculae composed of fibroblasts around a collagen fibril core¹⁷ together with CSF in the subarachnoid space, act as a shock absorber and provide buoyancy to the brain (Table 1, Box 2d). The human brain, despite weighing on average 1.2–1.4 kg, has essentially neutral buoyancy with a weight of only 4% of the actual mass⁵⁷. This buoyancy is essential for maintenance of blood flow and neuronal function. Specialized extensions of the dura (falx cerebri, tentorium cerebelli, dentate ligaments) provide further mechanical support. The falx cerebri, located between the cerebral hemispheres, limits brain lateral displacement and rotation within the skull. The tentorium cerebelli provides additional support between the cerebral hemispheres and cerebellum¹⁸. Similarly, the thick collagenous bundles of the dentate ligaments stabilize the spinal cord within the vertebral column through rigid fixation to the dura. The heterogeneity of mechanical properties of different regions of meninges are now being elucidated by atomic force microscopy, second harmonic generation imaging of collagen, and other methods^{58, 59}.

Changes in CSF volume normally compensate for increases or decreases in cerebral blood flow, as stipulated by the Monro-Kellie doctrine, whereby the sum of volumes of the brain, CSF, and intracranial blood remains constant⁶⁰. Accordingly, intracranial pressure is normally maintained constant. Mechanical testing of the dura has revealed nonlinear stress-strain curves similar to other collagenous tissues, affording tissue compliance during variations in CSF volume⁶¹. Similarly, the pia has viscoelastic properties due to the presence of collagen fibrils in the subpial space, which allow for stretch to accommodate changes in brain and spinal cord volume in health and after injury^{62, 63}.

Mechanical forces from trauma, surgery, or postmortem manipulation can cause a subdural fluid-filled space to form at the dura-arachnoid interface due to separation of dural border cells (Fig. 2). The process of dural border cells separation and formation of a subdural space has the adaptive value of contributing to the absorption of mechanical forces between the brain and skull but can also rupture bridging veins and result in a subdural hematoma^{64, 65}.

Permeability barrier

The barrier function of the arachnoid protects the brain and spinal cord from the changing milieu outside the CNS by limiting free entry of molecules from the dura and beyond (Fig. 2, Table 1). Dural capillaries have endothelial fenestrations, are leakier than brain capillaries, and make the dura accessible to plasma components¹⁵. However, extravasated plasma is blocked from entering the CNS by the blood-CSF barrier formed by arachnoid barrier cells, which are the only meningeal cells that express E-cadherin and are uniformly connected by tight junctions that prevent intercellular transit of solutes^{15, 66}.

Barrier properties of arachnoid barrier cells

Arachnoid barrier cells are interconnected by tight junctions that are visible by TEM imaging of thin epoxy sections and freeze-fracture replicas¹⁵. Claudin-11 is one of the tight junction proteins that join arachnoid barrier cells⁴⁸. Claudin-11 is also found in oligodendrocytes, where it forms junctions between myelin sheaths and in testicular Sertoli cells where it contributes to the blood-testis barrier⁶⁷. Arachnoid barrier cells are also joined by adherens junctions that contain E-cadherin, and less so VE-cadherin^{6, 7, 15, 46}. Adherens junctions are required for maturation of tight junctional complexes^{6, 7, 46}.

Arachnoid barrier cells are also joined by tricellular junctions, as demonstrated by immunohistochemistry, ultrastructural analysis, and scRNA-seq identification of Lsr/angulin-1 and angulin-3 expression in the mouse meninges⁷. Tricellular junctions are tight junction complexes formed by angulin family proteins and the transmembrane protein tricellulin at contacts where three cells meet (Fig. 3a–c). Angulins and tricellulin are also located in tricellular junctions of brain endothelial cells⁶⁸, where expression of Lsr/angulin-1 is required for blood-brain barrier maturation⁶⁹.

Arachnoid barrier cells not only form a physical barrier but also a metabolic barrier mediated by expression of drug resistance/efflux transporters of the ATP-binding cassette (ABC) and solute carrier (SLC) families (Box 3a, 3e). Such transporters expressed by arachnoid barrier cells include P-glycoprotein (PGP, encoded by Abcb1a), breast cancer resistance protein (BCRP, encoded by Abcg2), and organic anion transporters-1 (OAT1, encoded by Slc22a6) and −3 (OAT3, encoded by Slc22a8)^{7, 70–72}, as well as the drug metabolizing enzyme DPP4. These molecules in arachnoid barrier cells resemble the blood-brain barrier, where endothelial cells strongly express Abcb1a, Abcg2 and Slc22a8 and many other molecules that facilitate transport of nutrients, remove metabolites²⁰, and restrict drug transport into the CNS (Box 3e).

Absence of barrier properties of inner arachnoid cells

Inner arachnoid fibroblasts, located between arachnoid barrier cells and the SAS, are not continuous and are not joined by tight junctions but are connected by adherens junctions and gap junctions^{15, 66}. These adherens junctions contain VE-cadherin, a protein typically associated with endothelial cells of blood vessels and lymphatics. In blood vessels, VE-cadherin plays a central role in mechanosensitive processes that regulate cytoskeletal remodeling and endothelial integrity in response to shear forces⁷³. VE-cadherin in inner arachnoid cells is also likely to contribute to mechanosensing of CSF shear stress and regulating cellular stiffening, ECM production, migration, and other responses. VE-cadherin in pial fibroblasts could have a similar function^{6, 7}.

Claims that Prox1⁺ cells form a fourth cell layer of the meninges^{25, 28}, separate from the dura, arachnoid, and pia, have generated considerable debate (Box 1a). However, subsequent studies revealed that Prox1⁺ cells are part of the inner arachnoid, where they form a discontinuous layer distinct from the arachnoid barrier layer (see Fibroblasts of the inner arachnoid section)⁷. The significance of meningeal cell Prox1 expression, which is a feature of lymphatic endothelial cells and neural and oligodendrocyte precursors, has proven to be another controversial topic (Box 1b).

Pial cell continuity supported by adherens junctions

The pia is a largely continuous layer of cells connected by VE-cadherin-containing adherens junctions punctuated by rounded intercellular stomata overlaying the reticular collagen network in the subpial space^{6, 7, 50, 54} (Fig. 3d–e, Box 1c). However, unlike the arachnoid barrier, the pia is not a barrier to macromolecules, which can readily pass from the SAS into the subpial space^{6, 74, 75}. This leakiness is consistent with the absence of tight junctions and presence of intercellular stomata between pia cells (Figs. 2, 3f)^{7, 50, 54}. However, erythrocytes in the SAS after subarachnoid hemorrhage do not enter the brain parenchyma without pial breakdown⁷⁶.

Beneath the pia, the glia limitans (Fig. 2), formed by astrocytes and basement membrane, permits passage of fluid and low molecular weight solutes from CSF into the CNS parenchyma but limits immune cell entry^{2, 6}.

Arterial supply and venous return

The meningeal vasculature is important not only for nutritional and clearance functions but also for features that can lead to epidural, subdural, and subarachnoid hemorrhage and as a route for treating some of these conditions (Fig. 2, Table 1). The arterial blood supply of the dura - mainly from the middle meningeal artery - is separate from that of the brain, which comes from the internal carotid and vertebral arteries. The middle meningeal artery arises from the maxillary branch of the external carotid artery and enters the skull through the foramen spinosum⁵³. The dura receives additional blood from branches of the ophthalmic, ethmoidal, occipital, and vertebral arteries.

Dural arteries, capillaries, and venous sinuses

Arterial blood flow to the dura is almost as great as to the cerebrum but is regulated independently from that of the brain. The dura has abundant fenestrated capillaries that permit extravasation of some plasma constituents². However, the arachnoid barrier protects the brain and spinal cord from the changing milieu of the overlying dura. Vasoactive neuropeptides released from nociceptive afferent nerves by electrical stimulation can increase plasma leakage from dural blood vessels (see Innervation section)^{13, 14}.

Dural veins drain into venous sinuses between the outer and inner layers of dural connective tissue¹². Cranial dural venous sinuses include the superior sagittal sinus and transverse sinuses, which also drain venous blood from the brain, skull, and scalp. Dural venous sinuses have higher mechanical stiffness than the remainder of the dura due to the high collagen content of the ECM⁷⁷. This rigidity prevents dural venous sinuses from easily collapsing, and, because of the absence of valves, ensures continuous blood flow.

Arteries and veins of the subarachnoid space and pia

Unlike the dura, the arachnoid and pia are thin and do not have capillaries, but they are accompanied by large vessels located between the brain and arachnoid barrier that carry blood to and from the brain (Fig. 2). These blood vessels have barrier properties similar to the blood-brain barrier (Box 3f). In humans and other mammals with a large brain, some blood vessels traversing the SAS are anchored to arachnoid trabeculae^{50, 76, 78}, which could protect against mechanically induced rupture and subarachnoid hemorrhage. This may not apply to mice where trabeculae have been found in the spinal cord but not on the dorsal aspect of the brain⁶.

The carotid and vertebral arteries join at the circle of Willis in the SAS at the base of the brain. Successive branches of these arteries give rise to pial arteries that supply penetrating arteries/arterioles to all parts of the brain parenchyma. Similarly distributed pial veins receive venous blood from the brain. Pial veins typically drain into the nearest dural venous sinus. To reach a dural sinus, veins cross the SAS and the arachnoid as bridging veins (Fig. 2), which are considered clinically important sources of bleeding in subdural hematoma^79–81. Insight into the cellular organization of the meningeal layers at the bridging sites comes from evidence that sleeves of arachnoid accompany bridging veins into the dura⁸¹ (Box 3b, 3c).

Spinal cord vasculature

The spinal cord venous system in humans is divided into diverse venous plexuses⁸². Longitudinal veins follow a route across the SAS, arachnoid, and dura to the external venous plexus. While in humans the architecture of veins on the dorsal aspect of the spinal cord is highly variable, the spinal cord in mice is drained by one large midline vein located in the SAS^{2, 83}.

Meningeal hemorrhage

Meningeal blood vessels are sources of intracranial hemorrhage that can be life-threatening from the rapid increase in intracranial pressure and subsequent events (Fig. 2). Cranial epidural hematomas most commonly occur laterally after traumatic rupture of a meningeal artery or medially from a dural venous sinus or vein⁸⁴. Subdural hematomas typically arise from ruptured bridging veins or dural capillaries after head injury⁷⁹, where extravasated blood creates and is confined by a subdural space that forms between dural border cells and adjacent connective tissue^{64, 65}. In subdural hematomas, blood usually does not cross the arachnoid barrier into the SAS. In contrast, subarachnoid hemorrhage, caused by a ruptured aneurysm, arteriovenous malformation, or trauma, results in blood accumulation in the SAS^{85, 86}. Although the pia prevents erythrocytes from crossing⁷⁶, hemoglobin and breakdown products of extravasated blood can enter the brain with potentially lethal consequences, including cerebral vasospasm, ischemia, and neuronal toxicity^{85, 86}.

CSF circulation and clearance

Established and uncertain CSF circulation routes

The arachnoid and pia confine CSF to the SAS. CSF produced by the choroid plexuses flows from the ventricles to the SAS through the foramina of Luschka and Magendie in the brain stem. CSF circulates in multiple directions through all regions of the SAS, including large and small cisternae and perivascular spaces near the brain and spinal cord surface and within the parenchyma (Table 1). It is debated whether CSF enters the brain by bulk flow driven by a hydrostatic pressure gradient (convective flow, Box 1d) through spaces around penetrating blood vessels and in the interstitium⁸⁷ or instead is largely confined to the SAS⁸⁸. Pressure gradients needed to drive CSF by convective flow from the SAS into or out of the brain parenchyma have never been convincingly demonstrated⁸⁹. While CSF in the SAS could be driven by a hydrostatic pressure gradient across the pia and subpial space into brain interstitial spaces, there is evidence that this occurs only after sudden loss of blood pressure, hyperosmotic plasma challenge, or other special conditions^{88, 90}. To the extent that CSF serves as a sink for interstitial fluid to preserve the internal milieu of the neural parenchyma, metabolites and toxic substances would have to diffuse and/or move by convective flow through brain interstitium and perivascular (Virchow-Robin) spaces or vascular adventitia to reach the SAS^{91, 92} (Box 1d).

Arachnoid granulations do not explain CSF clearance

Regardless of the routes of CSF circulation, pathways across the arachnoid barrier must exist for CSF clearance, totaling about 500 mL per day in humans⁵. Historically, the major route for CSF efflux was thought to be through arachnoid granulations/villi in dural venous sinuses⁹³. Fluid channels lined by collagen fibrils within arachnoid granulations reportedly lead to an apex covered by arachnoid cap cells⁹⁴. A pressure gradient between CSF and blood in dural sinuses was thought to provide the driving force for efflux through a valve-like mechanism⁹⁵. Yet, in vivo physiological evidence for this mechanism of efflux is lacking⁵ (Box 1e). Instead, compelling evidence is accumulating against an essential role for arachnoid granulations in CSF clearance. An MRI study of the number, size, and distribution of arachnoid granulations in 120 humans between 0 and 80 years of age with no known CSF pathology reported substantial variability in number and a surprisingly large proportion of individuals with no granulations in the dural sinuses (85% at age 2 years, 15% at age 60 years, and 35% at age 80 years)⁹⁶. Arachnoid granulations were also infrequent in neonates⁹⁶. At the present time there is no accepted model for CSF transport through arachnoid granulations into dural venous sinuses⁵ (Box 1e).

CSF clearance through lymphatics

Tracer studies in multiple species performed over many years have consistently demonstrated multiple CSF clearance routes to lymphatic vessels¹¹ (Box 3c). Indeed, in mice, pathways leading to lymphatics appear to be the predominant CSF efflux route⁹⁷. Specific routes identified by tracer studies include lymphatics traversing the cribriform plate to lymphatic plexuses in the nasal mucosa and nasopharynx^{10, 11, 98}, in the meningeal/epineurial sheath around other cranial and spinal nerves^{97, 99}, and in the dura over the brain and spinal cord^{8, 100}. E-cadherin or claudin-11 staining in arachnoid barrier cells appears discontinuous above the cribriform plate^{48, 51, 98, 101}. Lymphatic vessels that traverse foramina in the cribriform bone with olfactory nerve bundles provide routes for egress of particles as large as 1 μm^{10, 98}, suggesting that CSF drainage to lymphatics can occur at this site through focal openings in the arachnoid barrier (Box 3c). Further research is needed to identify the cellular and junctional details of the connections of meningeal layers lining the SAS and lymphatic endothelial cells in this region and in the dura/epineurium around other cranial and spinal nerves.

Immune barrier and surveillance

Neuronal signaling requires a constant specialized environment that is incompatible with conventional humoral and cellular mechanisms of immune surveillance (Table 1). However, the CNS has a unique relationship with the immune system, referred to as CNS immune privilege¹⁰². In this context, brain barriers establish compartments in the CNS that differ with respect to accessibility of immune cells and immune mediators. One view is that brain barriers evolved for protection and defense of the CNS parenchyma analogous to the architecture of medieval castles with a two-walled moat¹⁰². In this concept, the CNS parenchyma is protected by the outer tight junction-fortified arachnoid barrier, the CSF-filled moat, and the inner pia and glia limitans. The SAS is patrolled by CD206-positive tissue-resident macrophages, referred to as border-associated macrophages or CNS-associated macrophages, that are transcriptionally distinct from microglial cells¹⁰³. During embryonic development, both border-associated macrophages and microglia enter the CNS from the yolk sac and occupy and maintain their respective niches throughout life by self-renewal^{22, 103}. Border-associated macrophages appear in the developing meninges at embryonic day 10.5 and enter the brain parenchyma along penetrating blood vessels together with perivascular fibroblasts around postnatal day 5⁵⁶, consistent with molecular interactions of the two cells types^{7, 104}.

Dynamics of meningeal T cell and macrophage populations

Circulating activated CD4 and CD8 T cells can migrate from pial blood vessels into the SAS, independent of antigen specificity or inflammatory stimuli, as documented by live-cell imaging in vivo^{6, 83}. T cells in the SAS can move with CSF flow but are highly motile and can crawl over pial cells⁶. Border-associated macrophages in the SAS and perivascular spaces are strategically positioned to internalize antigens in CSF for presentation to T cells. Recognition of cognate antigens by major histocompatibility complexes (MHC) on macrophages in the SAS triggers local T-cell reactivation, which can be visualized by shuttling of fluorescently tagged nuclear factor of activated T cells 5 (NFAT5) into the nucleus¹⁰⁵. Only T cells that have been reactivated in the SAS cross the glia limitans into the CNS parenchyma, which in experimental animal models of multiple sclerosis is associated with development of clinical signs of disease¹⁰⁶. T-cell migration across pial blood vessels and the glia limitans requires molecular mechanisms that are distinct from those required for T-cell migration across the blood-brain barrier^{106, 107}.

ScRNA-seq and functional studies have revealed the heterogeneity of border-associated macrophages and emphasized their contribution to maintenance of tissue integrity under normal conditions and potential involvement in Alzheimer’s disease, cerebral amyloid angiopathy, subarachnoid hemorrhage, and other neurological pathologies^{108, 109}.

Issues raised by dural macrophage contribution to brain immunity

The dura serves as an immune hub that hosts macrophages, neutrophils, T cells, B cells, and B-cell precursors in mice^{110, 111}. Dural macrophages are distinct from border-associated macrophages in the SAS, as demonstrated by scRNA-seq analysis and immunophenotyping^{12, 22, 23}, consistent with meningeal layer-specific functions of these macrophages.

Some investigators¹¹⁰, but not others¹¹¹, propose that dural antigen-presenting macrophages and B cells are continuously populated from the neighboring skull and vertebral bone marrow rather than from the bloodstream (Box 3d). This feature deserves further study, as does the relationship of the arachnoid barrier to potential routes for immune cell migration from skull bone marrow to the dura and into the brain parenchyma¹¹². MHC class II^hi macrophages, a special subset of dural macrophages, are concentrated at sites where venules from the CNS join dural venous sinuses¹². As these are reported to be sites where CSF can access the dura, the region of dura around venous sinuses could be specialized for antigen monitoring^{12, 42}. Accordingly, MHC class II^hi macrophages around dural venous sinuses could serve as an immune barrier against systemic pathogens that is continuously refreshed by myeloid cells from the systemic circulation¹².

Issues raised by reports of skull bone marrow-brain connections

Chemotactic signals from CSF could drive immune cell recruitment from the bone marrow to the dura and potentially into the SAS and CNS parenchyma¹¹³. However, as few studies have directly tested the restrictive properties of the arachnoid barrier to such signals, it remains to be determined how they would cross or circumvent the arachnoid barrier. Equally important is consideration of how disease-associated alterations in the arachnoid barrier, reported in mouse models of neuroinflammatory disease¹¹⁴ and elevated meningeal reactive oxygen species¹¹⁵, enable entry of dura-derived cytokines or immune cells into the CSF. Investigators studying the contribution of interleukin-17 (IL17) and brain macrophages to cognitive impairment accompanying salt-sensitive hypertension implicated the involvement of defective arachnoid barrier function¹¹⁶, but further research is needed to define how IL17 reaches the brain parenchyma under these conditions (Box 3d).

Innervation

The innervation of the meninges contributes to sensing mechanical forces imposed on the brain and spinal cord, regulating vasomotor activities, and mediating the pain of migraine (Table 1)^{13, 14}. The dura is richly innervated by neuropeptide-releasing nociceptive somatosensory afferent axons and by adrenergic sympathetic and cholinergic parasympathetic autonomic axons^{13, 14}. Unlike the dura, the arachnoid is not innervated, and the pia is innervated largely by sympathetic axons in the adventitia of pial arteries, as first described by Thomas Willis in his eponymous Circle of Willis¹¹⁷. The sympathetic innervation of pial blood vessels does not extend to penetrating arterioles of the CNS parenchyma.

Autonomic innervation and migraine

Sympathetic axons from the superior cervical sympathetic ganglion that innervate meningeal blood vessels promote vasoconstriction by releasing norepinephrine. Vasoconstricting agents (dihydroergotamine) have been used in the treatment of migraine to counter the vasodilating action of sensory neuropeptides¹⁴. Parasympathetic axons from the sphenopalatine ganglion of the facial nerve (cranial nerve VII) that innervate meningeal arteries contribute to migraine by promoting vasodilatation. Facial nerve axons of sphenopalatine ganglion cells release pituitary adenylate cyclase-activating polypeptide (PACAP), which can evoke vasodilatation and migraine symptoms. Pharmacological blockade of sphenopalatine ganglion cells has been used to treat some cases of migraine^{118, 119}.

Sensory innervation and migraine

Nociceptive afferent axons in the dura above the tentorium are thought to be responsible for much of migraine pain^{13, 14}. These axons travel in the three divisions of the trigeminal nerve and have neuronal cell bodies in the trigeminal Gasserian/semilunar ganglion. The axons project to nociceptive regions of the spinal trigeminal tract and nucleus^{13, 14}. Sensory axons in the dura below the tentorium—considered responsible for occipital migraine pain—travel to structures in the posterior fossa along blood vessels from cell bodies in dorsal root ganglia at cervical levels C1-C3 of the spinal cord¹²⁰. The spinal dura has much less innervation than the cranial dura¹²¹.

Therapeutic relevance of meningeal innervation

When activated, some sensory nerve endings in the dura release calcitonin gene-related peptide (CGRP), substance P, PACAP, and other neuropeptides that have potent vasodilating activity implicated in the symptoms of migraine¹¹⁹. Substance P not only causes vasodilatation but also increases vascular permeability (neurogenic inflammation). CGRP inhibitors are widely used in the treatment of migraine and are effective in about half of the cases^{118, 119, 122}.

Recent studies have revealed functions of sensory nerves in the dura beyond their contribution to migraine. A neuroimmune mechanism has been proposed to be involved in bacterial meningitis¹²³. Bacterial toxins can induce the release from nociceptive nerve endings of CGRP that binds to receptors on macrophages for receptor activity modifying protein 1 (RAMP1), leading to suppression of immune responses¹²³. Calcium-signaling in meningeal afferents can be detected by 2-photon imaging during normal locomotion of awake mice, indicating that these axons have mechanosensing capabilities¹²⁴. Indeed, Piezo2 expression has been found in trigeminal ganglion neurons, consistent with mechanosensory functions of this population of meningeal afferent axons¹²⁵.

Future outlook

The recognition long ago of the dura, arachnoid, and pia mater as outer connective tissue protectors of the brain and spinal cord has evolved into the understanding of these layers as complex structures with multiple distinctive features and great clinical importance. Advances in meningeal biology have improved the understanding of epidural, subdural, and subarachnoid hemorrhage, meningioma, meningitis, and other primary conditions of the meninges. Perhaps even more important has been the benefit to understanding migraine and neuropathic pain, neuroinflammatory diseases such as multiple sclerosis, and neuronal degeneration accompanying Alzheimer’s disease and aging.

Advances made through the use of contemporary genetic, molecular, physiological, and imaging methods have raised many questions that must still be answered. These include the role of the meninges in overall adult brain physiology, especially the functions as a barrier and communication interface with the immune system, blood vasculature, lymphatics, sensory innervation, and other organs. Also included are questions about the contribution of the meninges to brain development and topics not covered by this review. Molecular profiling endeavors have revealed many details of the diversity of meningeal fibroblasts, expression of molecular transporters in specific cell types of individual meningeal layers, and complexity and heterogeneity of resident immune cells. The physiological and pathophysiological roles of these cells can now be elucidated through new cell-type restricted genetic tools and other technological advances. As reflected by the hallmarks of meningeal pathophysiology (Fig. 1), the contribution of the meninges to many normal CNS functions and clinical conditions of public health importance worldwide underscores the value of continuing to advance the understanding of the dura, arachnoid, pia, and CSF as multifunctional protectors of the brain and spinal cord. These advances should also lead to novel diagnostics and approaches for drug delivery to the CNS.