Discussion
This study provides the first evidence for involvement of CSF–brain and blood–CSF interfaces at the pial glia limitans, the ependymal surface, and the choroid plexus in the histopathology of NMO (Table
2). At the pial glia limitans, AQP4 loss was colocalized with dystrophic astrocytes, microglial reactivity, and C9neo deposition. AQP4 loss at the pial surface also colocalized with meningeal infiltration of eosinophils and neutrophils. Likewise, the ependymal lining of ventricles in NMO tissue exhibited loss of ependymocytes and structural disorganization associated with the presence of subependymal reactive astrocytes, loss of AQP4 expression at both the ependymal surface and in the parenchyma, microglial reactivity, C9neo deposition, and granulocytic infiltrates at the CSF face of the ependymal layer. Finally, the choroid plexus in NMO tissues exhibited profound loss of AQP4 expression, an increase in number of stromal macrophages, and deposition of immunoglobulins and complement activation factors on choroid epithelial cell membranes. These changes were never observed in control tissues. The specificity of AQP4 loss in the choroid plexus in NMO cases was confirmed by the coincident retention of normal levels of AQP1 immunoreactivity. While MS cases did exhibit microglial reactivity and some evidence of stromal complement deposition, the pattern and extent of pathology were markedly different from NMO, indicating a unique pathology involving CSF interfaces in NMO that mirrors the pathology at blood–brain interfaces. Notably, while the glia limitans superficialis and the glia limitans perivascularis share a primary involvement of AQP4-expressing astrocytic foot processes, the ependymal and choroid plexus barriers are comprised of non-astrocytic cells expressing AQP4. Thus, while NMO is certainly a primary astrocytopathy, the disease also has an important component of an “AQP4-opathy” that involves antibody-mediated injury to other CNS cell types that express the water channel. As with our previous report demonstrating that sarcolemmal AQP4 is a target of IgG in patients with NMO [
19], these observations further expand the cellular repertoire of NMO IgG targets beyond the archetypal astrocyte.
Ependymoglial lineage cells exhibit apical-basal polarization and the expression of basolateral membrane specializations that facilitate contact with mesenchymal borders [
10]. These specializations anchor the basolateral membrane of ependymoglial cells to the basal lamina of mesenchymal cells via interactions between laminin, α-dystroglycan, and β-dystroglycan [
73]. Thus, astrocytic endfeet interact with capillary endothelial cell and meningeal fibroblast basement membranes to form the blood–brain barrier and the CSF–brain barrier, respectively [
69], and choroid plexus epithelial cells interact with capillary endothelial cell basement membranes to form the blood–CSF barrier [
28,
37]. These physical mesenchymal–ependymoglial barriers are vital to the maintenance of the unique CNS interstitial fluid composition required for normal neuronal function. Of note, AQP4 is a component of the dystrophin-associated protein complex and is, therefore, polarized to the basolateral membrane of ependymoglial cells, bringing the water channel into proximity with the mesenchymal borders [
50]. Indeed, this localization is fundamental to water, ion, and protein exchange between the blood, the brain parenchyma, and the CSF [
1].
The blood–brain barrier is formed at tight junctions between capillary endothelial cells within the brain parenchyma, preventing diffusion of blood-borne ions and proteins into the perivascular stroma. The endfeet of perivascular astrocytes, along with accessory cells, such as pericytes, provides a second layer of control over solute flow from blood to parenchyma, resulting in a two-step process of facilitated movement across the endothelial cell barrier and then across the astrocyte barrier [
69]. In contrast, the blood–CSF barrier at the choroid plexus is physically manifested by the apical tight junctions found between choroid plexus epithelial cells. The endothelial cells at this interface are fenestrated and lack tight junctions, permitting diffusion of ions and proteins into the choroid stroma [
74]. Pathogenically, this difference has a profound consequence for accessibility of NMO IgG to AQP4. At the blood–brain barrier, circulating NMO IgG has limited access to AQP4 on perivascular astrocytic endfeet. However, at the blood–CSF barrier, NMO IgG could readily enter the perivascular space and encounter AQP4 on the basolateral surface of the choroid epithelial cells.
Robust AQP4 expression on the basolateral surface of choroid plexus epithelial cells in normal tissue coupled to the presence of immunoglobulin immunoreactivity and the nearly complete loss of AQP4 expression in these cells in NMO tissue indicates that the blood–CSF barrier may be a primary site for entry of NMO IgG into the CNS. In fact, immunoglobulins have been detected within the choroid plexus stroma in healthy normal controls [
52], as well as in both neurological and non-neurological disease conditions, such as Alzheimer disease, systemic lupus erythematosus, and others [
3,
4,
9,
13,
17,
46,
55,
56,
66,
67,
81]. Of note, we also detected robust immunoreactivity for C9neo at the blood–CSF barrier in NMO tissues, with a pattern that suggested deposition along the choroid plexus epithelial cell membrane. Complement factors, ranging from C1q through to C9neo, are also observed in the choroid plexus in multiple conditions, including MS and Alzheimer disease [
22,
48,
54‐
56,
66]. However, in MS, as recently reported by Moore et al. [
48], deposition of C9neo is observed predominantly in stromal concretions and at the subepithelial basement membrane, with only infrequent staining on choroid epithelial cells. Our study corroborates this finding: in Fig.
9j, we show that C9neo is almost entirely restricted to subepithelial basement membrane and stromal concretions in MS tissue. What sets NMO apart from the normal, non-pathogenic deposition of complement within the choroid stroma is the presence of AQP4 on the basolateral membrane of the epithelial cells adjacent to the basement membrane and stroma. Rather than accumulate in concretions, some of the circulating NMO IgG that crosses into the choroid stroma may find antigenic target and trigger complement deposition on the epithelial cell membranes. This is supported by the staining in Fig.
9d, in which C9neo is clearly present on the epithelial cell membrane. Thus, while immunoglobulin and complement deposition in the choroid plexus are not unique to NMO, per se, the distribution of these factors beyond the stroma and basement membrane is uniquely driven by the availability of the AQP4 target antigen and the presence of anti-AQP4 antibody in NMO patients.
The presence of C9neo on choroid epithelial cells raises the possibility of complement-mediated damage to the choroid as a pathogenic trigger in NMO. However, paradoxically, while we observed ependymal disorganization and discontinuity coincident with C9neo deposition and subpial vacuolations associated with C9neo deposition at the pial glia limitans, we did not find evidence for overt choroid plexus damage. In this context, it is noteworthy that AQP4 is also expressed on the basolateral membrane of kidney epithelial cells in the distal tubule [
72] and is accessible to NMO IgG across the peritubular capillary endothelial cells in the same manner that it is accessible in the choroid plexus [
36]. However, kidney pathology has not been widely reported in NMO patients, with only two studies suggesting possible injury to AQP4-expressing tubular epithelial cells, both of which occurred in unusual circumstances [
25,
49]. This may be explained by the robust expression of complement regulatory factors on the basolateral membrane of distal tubule and collecting duct cells in the kidney [
63]. Choroid plexus epithelial cells also express complement regulatory proteins and upregulate expression of factors, such as CD46 and CD59 under pathogenic conditions [
5,
68]. Moreover, several studies indicate that C9neo deposition, despite being equated with obligate lytic destruction in much of the literature, can be detected in the absence of damage due to regulation of the terminal membrane attack complex by molecules, such as vitronectin and clusterin [
7,
21]. The expression of these molecules in the human choroid plexus is currently unknown, but the maintenance of physical integrity at the blood–CSF barrier despite heightened C9neo deposition suggests that something is preventing lytic destruction. Likewise, while we speculate that the blood–CSF barrier may serve as a unique entry point for NMO IgG into the CNS, it is not clear whether such transmission requires a physical compromise of the barrier or whether changes in AQP4 expression, microglial activation, and complement cascade engagement initiate a functional compromise or modulation that permits IgG introduction from the blood into the CSF.
If circulating NMO IgG enters the CSF at the choroid plexus, it may initiate a cascade of events that eventually leads to parenchymal astrocyte injury and subsequent secondary loss of myelin [
39]. Our observation of AQP4 expression on the basolateral surface of ependymal cells in normal tissue and the loss of ependymal AQP4 expression in NMO tissue, as well as evidence of ependymal discontinuity and C9neo deposition on ependymal cells, suggests that CSF NMO IgG gained access to this CSF–brain barrier and elicited damage. Likewise, astrocyte reactivity, loss of AQP4 expression, and deposition of C9neo at the pial glia limitans, associated with subpial vacuolated tissue, suggest that this CSF–brain barrier was also targeted by NMO IgG. While ependymal cells along the ventricle are connected by intermediate junctions [
51] and pial fibroblasts are joined by desmosomes [
18], neither of these structures are as restrictive to IgG diffusion as tight junctions [
44]. At these sites, NMO IgG would have access to AQP4 expressed on the basolateral surface of ependymal cells as well as to AQP4 expressed on astrocytic endfeet. Thus, serum NMO IgG may gain access to brain parenchyma by transitioning through the blood–CSF barrier at the choroid plexus, circulating through the CSF, and then transitioning through the CSF–brain barriers at the ependyma and the pial glia limitans. Ultimately, this blood-to-CSF-to-brain transition of NMO IgG may result in an inside-out antibody-mediated compromise of the blood–brain barrier, providing direct access of NMO IgG from blood-to-brain. In addition, the CSF route for NMO IgG entry into the brain parenchyma could generate reactive astrocyte responses that drive granulocytic infiltration even across an intact blood–brain barrier [
23,
77], inducing inflammatory-mediated, leukocyte-dependent pathology within the parenchyma [
40].
Our findings, therefore, suggest a possible pathogenic role for NMO IgG in the CSF. NMO IgG is detected in CSF and levels are predicted by recent relapse and high serum NMO IgG titer [
26,
43]. Increased total CSF protein and elevated CSF lactate and albumin levels also correlate with NMO disease severity and acute relapse, suggesting a relationship between dysfunction at the blood–CSF barrier and disability [
26,
31,
82]. Conversely, in a study of ten NMO patients with CSF NMO IgG measured at relapse, follow-up measurements after treatment revealed that a reduction in NMO IgG in CSF correlated with clinical improvement, though serum levels were not correlated with remission [
11]. The concept of distinct pathogenic roles for CSF and serum NMO IgG is further supported by evidence that plasmapheresis showed delayed efficacy in some NMO patients, suggesting that the initial washout was insufficient to improve function [
30,
38]. Together, these findings support the CSF compartment as an important source of pathogenic NMO IgG. The cascades elicited by CSF NMO IgG likely include the direct induction of complement-mediated injury to CSF–brain interfaces and parenchymal astrocytes, as well as the initiation of self-amplifying astrocytic stress responses, as discussed above. However, an equally important mechanism of injury triggered by NMO IgG in the CSF may be disruption of normal water homeostasis at the blood–CSF and CSF–brain barriers.
Pathologic changes in brain water content manifest in two distinct but inter-related ways: edema, associated with increased water in the brain parenchyma [
71], and hydrocephalus, associated with increased water in the CSF [
8]. Edema is generally categorized as cytotoxic, in which reduced water transport function in parenchymal cells leads to interstitial shrinkage and intracellular swelling in the absence of blood–brain disruption, or vasogenic, in which blood–brain barrier damage induces interstitial accumulation of water. Brain edema is cleared by interstitial or cellular water flow across the glia limitans superficialis into CSF and then into the blood, by water flow across the ependyma into the CSF and then into the blood, and by flow across the glia limitans perivascularis directly into the blood [
70]. AQP4 is clearly involved in edema [
71], such that genetic deletion of AQP4 results in reduced cytotoxic edema by limiting the rate of water accumulation in the parenchyma, but increased vasogenic edema by reducing water clearance at the CSF–brain and blood–CSF barriers [
62]. In NMO, this suggests that loss of AQP4-expressing cells or loss of AQP4 expression at the CSF–brain barrier in the ependyma or pial glia limitans will reduce the capacity to remove water from the parenchyma, facilitating edema. Indeed, evidence of extensive T2-weighted MRI hyperintensities in subcortical white matter coupled to an absence of gadolinium enhancement by T1 MRI and an increase in the apparent diffusion coefficient indicates the presence of pseudo-vasogenic interstitial edema in NMO [
64]. Similar observations of increased apparent diffusion coefficient in normal-appearing white matter in the absence of gadolinium enhancement [
27] further argue for a pseudo-vasogenic edema in NMO in which interstitial water is increased without blood–brain barrier dysfunction due to failure to transport water at CSF–brain interfaces [
12]. This model is consistent with the loss of AQP4 expression and vacuolated underlying tissue observed at these barrier sites in NMO, and with our previous report of intramyelinic edema [
20]. This model also highlights the role of AQP4 loss at CSF–brain interfaces in the transient edema we previously described in NMO patients with posterior reversible encephalopathy syndrome [
42].
We recently reported that obstructive hydrocephalus is found in NMO patients at a higher frequency than observed in the general adult population [
6]. On the basis of these cases, we predicted a primary involvement of ependyma in NMO and our current findings support this concept. However, the pathogenic mechanisms underlying hydrocephalus in NMO are not straightforward. While evidence from mice lacking AQP4 expression indicates about a 10% incidence of hydrocephalus triggered by complete obstruction of the cerebral aqueduct [
14], it is not clear why such obstructions occur. AQP4-null mice exhibit an intact blood–brain barrier but a disorganized ependymal cell layer, even in animals without hydrocephalus [
14], supporting a model in which loss of AQP4, whether genetic or antibody-mediated, induces defects in CSF homeostasis. However, loss of AQP4 at the blood–CSF barrier in the choroid plexus may have counteracting effects on CSF production, resulting in little change in overall CSF volume. This model argues against CSF hyper-production as the cause of hydrocephalus in NMO, though it remains possible that hypo-production of CSF due to loss of choroid plexus epithelial cells could impact ventricular flow in a manner that increases the likelihood of aqueductal blockage. Such blockage and subsequent obstructive hydrocephalus may also arise as a result of ependymal cell loss induced by the NMO IgG-mediated complement deposition we observed in this study. Humans and mice with primary ciliary dyskinesia exhibit hydrocephalus [
32], and genetic deletion or mutation of multiple specific ciliary proteins results in ependymal cilia beat defects and consequent hydrocephalus induced by reduced CSF flow [
15,
33,
45,
79]. Additional evidence indicates that laminar CSF propulsion mediated by ependymal cilia is required to maintain aqueductal patency, and defects in such propulsion result in aqueduct stenosis and triventricular hydrocephalus [
24]. Ultimately, short of catastrophic loss of ependymal cells and/or choroid plexus epithelial cells or the initiation of an amplification loop that results in a rapid rise in intracerebral pressure and consequent collapse of the aqueduct, it is likely that overt hydrocephalus in NMO is at one end of a spectrum that more frequently manifests as transient pressure increases and disorganized CSF flow. Regardless, these events may contribute to pathologic events in the brain parenchyma that run in parallel to direct NMO IgG-induced astrocyte pathology. Indeed, it is possible that preventing damage at the blood–CSF and CSF–brain interfaces may substantially reduce or delay direct parenchymal damage by the NMO IgG. In light of this, it is noteworthy that we observed robust complement deposition at the choroid plexus and ependymal interfaces, suggesting that these sites may be directly protected by complement cascade blocking therapies, such as eculizumab [
57]. It would be of interest in future studies to determine if eculizumab therapy alters the imaging hallmarks of pseudo-vasogenic edema in NMO patients as a secondary outcome measure.
The pathology findings we describe also provide mechanistic evidence for our previous report of periependymal and periventricular MRI abnormalities [
58]. These included fluid-attenuated inversion recovery (FLAIR) and T2 signal abnormalities in periependymal regions of the lateral ventricles and along the walls of the third and fourth ventricles and the aqueduct of Sylvius. In meningitis, FLAIR hyperintensity is caused by elevated CSF protein content related to leptomeningeal inflammation [
29], which extends the effective echo time beyond the inversion time used to suppress bulk CSF [
47]. This same effect results in T2 hyperintensity. Therefore, our previously reported MRI findings are consistent with pathological changes at the ependymal CSF–brain and choroid plexus blood–CSF barriers that may involve both inflammation (Fig.
6k, l) and decreased local CSF flow due to cell injury or loss and barrier breach (Fig.
6i, j). Likewise, the reports of the so-called “pencil thin” ependymal hyperintensities with contrast-enhanced FLAIR imaging [
2] support a dynamic alteration or disruption at the CSF–brain barrier. Notably, in common with infection-induced changes in the ependyma and choroid plexus [
16], intraventricular exudate (“ventricular debris”) may be an expected MRI observation during acute attacks in NMO patients and may contribute to a loss of aqueduct patency that triggers hydrocephalus [
60].
In conclusion, we provide evidence that the interfaces between blood and CSF and between CSF and brain are key sites for the initiation of NMO pathophysiology. Furthermore, the pathology at these sites may have important implications for disease evolution in NMO patients, including serving as the possible point at which serum NMO IgG enters the CNS and gains broad access to the brain parenchyma. Our findings also indicate that NMO is an “AQP4-opathy” with pathological targets beyond the astrocyte and provide insight into recent reports of edema [
12,
20,
27,
42,
64] and hydrocephalus [
6] in NMO patients that may suggest a unique model for CSF flow-dependent pathogenic events in this disease. Finally, the abundance of complement activation products in the choroid plexus, pial glia limitans, and ependymal layer suggests that these sites may be key targets for complement blocking therapies [
57]. As such, imaging of these interfaces and other measures of CSF homeostasis may offer quantifiable tests of efficacy in future trials using therapies that target the complement pathway in NMO.