Aquaporin-4 Gene Disruption in Mice Reduces Brain Swelling and Mortality in Pneumococcal Meningitis*

The astroglial water channel aquaporin-4 (AQP4) facilitates water movement into and out of brain parenchyma. To investigate the role of AQP4 in meningitis-induced brain edema, Streptococcus pneumoniae was injected into cerebrospinal fluid (CSF) in wild type and AQP4 null mice. AQP4-deficient mice had remarkably lower intracranial pressure (9 ± 1 versus 25 ± 5 cm H2O) and brain water accumulation (2 ± 1 versus 9 ± 1 μl) at 30 h, and improved survival (80 versus 0% survival) at 60 h, through comparable CSF bacterial and white cell counts. Meningitis produced marked astrocyte foot process swelling in wild type but not AQP4 null mice, and slowed diffusion of an inert macromolecule in brain extracellular space. AQP4 protein was strongly up-regulated in meningitis, resulting in a ∼5-fold higher water permeability (Pf) across the blood-brain barrier compared with non-infected wild type mice. Mathematical modeling using measured Pf and CSF dynamics accurately simulated the elevated lower intracranial pressure and brain water produced by meningitis and predicted a beneficial effect of prevention of AQP4 upregulation. Our findings provide a novel molecular mechanism for the pathogenesis of brain edema in acute bacterial meningitis, and suggest that inhibition of AQP4 function or up-regulation may dramatically improve clinical outcome.

The astroglial water channel aquaporin-4 (AQP4) facilitates water movement into and out of brain parenchyma. To investigate the role of AQP4 in meningitisinduced brain edema, Streptococcus pneumoniae was injected into cerebrospinal fluid (CSF) in wild type and AQP4 null mice. AQP4-deficient mice had remarkably lower intracranial pressure (9 ؎ 1 versus 25 ؎ 5 cm H 2 O) and brain water accumulation (2 ؎ 1 versus 9 ؎ 1 l) at 30 h, and improved survival (80 versus 0% survival) at 60 h, through comparable CSF bacterial and white cell counts. Meningitis produced marked astrocyte foot process swelling in wild type but not AQP4 null mice, and slowed diffusion of an inert macromolecule in brain extracellular space. AQP4 protein was strongly up-regulated in meningitis, resulting in a ϳ5-fold higher water permeability (P f ) across the blood-brain barrier compared with non-infected wild type mice. Mathematical modeling using measured P f and CSF dynamics accurately simulated the elevated lower intracranial pressure and brain water produced by meningitis and predicted a beneficial effect of prevention of AQP4 upregulation. Our findings provide a novel molecular mechanism for the pathogenesis of brain edema in acute bacterial meningitis, and suggest that inhibition of AQP4 function or up-regulation may dramatically improve clinical outcome.
Streptococcus pneumoniae (pneumococcus) is the most common and aggressive meningeal pathogen (1)(2)(3). The overall incidence of pneumococcal meningitis is rising (4,5) and the emergence of penicillin-resistant S. pneumoniae makes treatment harder (6,7). Even with effective antibiotic treatment, mortality from pneumococcal meningitis is 10 -30%, with 30 -50% of patients acquiring permanent neurological deficits (1)(2)(3). A major complication associated with unfavorable outcome in bacterial meningitis is brain edema, which causes a rise in intracranial pressure potentially leading to brain ischemia, herniation, and death (8,9). The molecular mechanisms responsible for the formation and absorption of excess fluid in brain edema associated with meningitis are poorly understood.
The two main types of brain edema are cytotoxic and vaso-genic (10). Cytotoxic edema, as occurs in ischemic stroke, refers to cell swelling that primarily affects astroglial cells. Vasogenic edema, as occurs in brain tumors, involves accumulation of excess fluid in the extracellular space of the brain parenchyma because of a leaky blood-brain barrier (BBB). 1 Although both types of brain edema are thought to co-exist in meningitis (8,9), their relative contributions to brain swelling are not known.
Recently, the bidirectional water channel aquaporin-4 (AQP4) has been found to play an important role in brain-water homeostasis (11)(12)(13)(14)(15)(16). AQP4 protein is expressed strongly in astroglia at the BBB and CSF-brain interfaces (17,18), suggesting involvement in water movement between fluid compartments (blood and CSF) and brain parenchyma. AQP4 deletion markedly reduced brain swelling in mouse models of cytotoxic brain edema, including water intoxication and focal cerebral ischemia (16). In contrast, AQP4 deletion in mice significantly worsened outcome in mouse models of vasogenic brain edema, including intraparenchymal fluid infusion, focal cortical freeze injury, and brain tumor (19). Thus, AQP4 appears to facilitate water movement into brain astroglia in cytotoxic edema, and water movement out of the brain in vasogenic edema.
To investigate the functional role of AQP4 in meningitis, we created a mouse model of meningitis involving injection of live S. pneumoniae into the CSF. Although the severity of infection was similar in wild type and AQP4 null mice as assessed by CSF bacterial and white cell counts, the AQP4 null mice had remarkably lower intracranial pressure, brain water accumulation, astrocyte foot process swelling, and mortality. Based on experimental data and computational modeling, we conclude that reduced permeability of the BBB in AQP4 deficiency accounted for the improved outcome in AQP4 null mice. Our results have specific implications regarding aquaporin-based therapies of bacterial meningitis.

MATERIALS AND METHODS
Mice-AQP4 null mice were generated by targeted gene disruption as described previously (20). These mice have normal growth, appearance, survival, gross cerebrovascular anatomy, brain water content, intracranial pressure, bulk intracranial compliance, and serum chemistries, except for a mild reduction in maximal urinary concentrating ability after prolonged water deprivation (16,19,20) and sensorineural hearing loss (21). Experiments were performed on weight-and sexmatched wild type and AQP4 null mice in an outbred (CD1) genetic background. Investigators were blinded to genotype information in all experiments. Protocols were approved by the University of California San Francisco Committee on Animal Research.
Bacterial Culture-S. pneumoniae serotype 14 (clinical isolate) was purchased from ATCC (catalog number 700676) and grown at 37°C on trypticase soy agar-enriched with defribrinated sheep blood (Remel, Lenexa, KS). Sensitivity to cefotaxime was confirmed using antibiotic resistance discs (Fisher Scientific, Los Angeles, CA). Bacteria were periodically subcultured from infected mouse CSF to maintain virulence. About 8 h prior to injection, two bacterial colonies were suspended in brain heart infusion broth (Fisher Scientific, Los Angeles, CA) supplemented with 5% fetal bovine serum and incubated at 37°C in 5% CO 2 . S. pneumoniae was grown in brain heart infusion broth to mid-log phase, estimated using optical absorbance at 570 nm. The bacterial suspension was then centrifuged at 6,000 ϫ g for 10 min. The pellet was resuspended in sterile, 0.9% saline to an absorbance of 0.20 at 570 nm, equivalent to 3-4 ϫ 10 6 colony forming units/50 l, which was confirmed for each set of experiments by quantitative bacterial cultures.
Pneumococcal Meningitis Model-Mice were anesthetized with 2.5% 2,2,2-tribromoethanol (intraperitoneal 125 mg/kg, Sigma). The head was secured in a stereotactic frame (Benchmark, Neurolab, St. Louis, MO) with the mouthpiece set at 5 mm and the ear bars at 10 mm. After incising the scalp, a 1-mm diameter burr hole was made 2-mm lateral and 0-mm posterior to the bregma, using a micromotor drill (Foredom, Bethel, CT). A blunted 27-gauge needle attached to a gas-tight glass syringe (Hamilton, Reno, NV) was stereotactically introduced through the burr hole until the needle tip made contact with the skull base (Fig.  1A). After 2 min, 50 l of S. pneumoniae suspended in 0.9% saline was infused over 5 min and 2 min later the needle was gradually removed. In control studies only saline was infused. About 5% mice died during the injection, most likely because of intracerebral hemorrhage and/or rapid increase in intracranial pressure. Mice that died prior to recovery from anesthesia were excluded from the study. After recovery mice were kept at room temperature and provided with standard chow and water ad libitum. Mice were given subcutaneous injections of 0.5 ml of normal saline at 10, 20, and 30 h after bacterial infusion to prevent dehydration.
Immunodetection of AQP4 -For immunocytochemistry, mouse brains were fixed in 10% formalin overnight, processed through graded concentrations of ethanol, immersed in Citrisolv (Fisher Scientific, Los Angeles, CA), and embedded in paraffin. Tissue sections (7 m thick) were deparaffinized in Citrisolv and rehydrated with graded ethanols. After blocking with goat serum, samples were incubated with a rabbit anti-AQP4 polyclonal antibody (1:100, Chemicon, Temecula, CA) and washed in phosphate-buffered saline. Bound antibody was detected using the Vectastain Elite ABC kit (Vector Laboratories, Burlingame, CA) according to the manufacturer's instructions.
For immunoblot analysis, the left cerebral cortex (contralateral to infusion site) was homogenized in 250 mM sucrose, 1 mM EDTA, 10 mM Tris-HCl, containing 1 g/ml aprotinin, 1 g/ml pepstatin A, and 1 mM phenylmethylsulfonyl fluoride, pH 7.4, and centrifuged at 5,000 ϫ g for 10 min. The supernatant (crude membrane extract) was loaded on a 4 -12% SDS-polyacrylamide gel (20 g of protein/lane) and transferred to a polyvinylidene difluoride membrane. The membrane was incubated with rabbit anti-AQP4 antibody (1:1000, Chemicon, Temecula, CA) followed by anti-rabbit IgG horseradish peroxidase-linked antibody, and visualized using enhanced chemiluminescence (Roche Molecular Biochemicals, Indianapolis, IN). AQP4 protein was quantified by scanning densitometry using serial dilutions of brains from meningitis mice as standards.
Intracranial Pressure-Intracranial pressure (ICP) was recorded from the cisterna magna as described (19). After positioning the mouse in the stereotactic frame, a midline incision was made at the back of the neck and the underlying muscles were dissected to reveal the cisterna magna. A 27-gauge needle attached using short, non-compliant tubing to a pressure transducer (TSD104A; Biopac Systems, Santa Barbara, CA) was inserted into the cisterna magna. Intracranial pressure was recorded at 50 Hz. Rectal temperature was maintained at 37-38°C using a heating lamp. ICP values were averaged over 2-5 min.
Brain Water Content-Rectal temperature was maintained between 37 and 38°C using a heating lamp for 1 h prior to mouse sacrifice. Mice were anesthetized, sacrificed by cervical dislocation, and the brain was immediately removed and divided into right and left hemispheres and cerebellum. Brain fragments were weighed immediately and then dried in a vacuum oven for 12 h at 105°C. The dried brain was re-weighed and % brain water content calculated as (wet weight Ϫ dry weight) ϫ 100/wet weight. The net change in brain water in meningitis was computed from % brain water after normalization to a constant brain weight of 500 mg.
Neurological Score-Global neurological deterioration, which occurs when intracranial pressure rises, was quantified using an established neurological scale (22): 1 ϭ dead, 2 ϭ no righting reflex, 3 ϭ ataxia, 4 ϭ loss of scatter reflex, and 5 ϭ normal. Mice were scored prior to bacterial injection and every 10 h thereafter.
Physiological Parameters-CSF white cell count was determined by counting cells in CSF smears after Gram stain (Remel, Lenexa, KS). The number of live bacteria in the CSF was quantified by culturing serial dilutions of CSF on trypticase soy broth agar with defibrinated sheep blood. Glucose concentrations were measured with a glucometer (OneTouch Ultra, LifeScan, Milpitas, CA) using 1-l samples of CSF and serum. Arterial blood pressure was recorded from the carotid artery and central venous pressure from the internal jugular vein using the Biopac system. Blood, obtained from the left cardiac ventricle, was analyzed using a blood gas machine (model 248, Chiron Corp., Norwood, MA) and at the Moffitt Hospital clinical laboratory.
Evans Blue Extravasation-In some experiments, Evans blue dye (Sigma) (4% in saline, 160 mg/kg) was injected into a femoral or jugular vein at 30 h after bacterial infusion. Fifteen minutes later, the left cardiac ventricle was perfused with 20 ml of phosphate-buffered saline. The brain was removed and divided into right and left hemispheres and cerebellum. Each part of the brain was immersed in 2 ml of formamide at 55°C overnight, and the extracted dye was quantified by optical absorbance at 610 nm against Evans blue/formamide standards. For comparison, Evans blue dye extravasation was also measured 1 h following focal cortical freeze injury, performed as previously described (19).
Cortical Fluorescence Recovery after Photobleaching-Photobleaching measurements were done as described previously (23). Briefly, after anesthesia the head was immobilized in a stereotactic frame and a craniotomy was fashioned to reveal the intact dura and underlying brain surface. FITC-dextran (70 kDa, 30 mg/ml, Sigma) dissolved in artificial cerebrospinal fluid was applied to cover the exposed dura for 2 h. After loading, the dural surface was washed with dye-free artificial CSF. In some experiments, the dura was then opened and the cortical surface was exposed for 10 min to CSF that had been collected from control mice or mice with meningitis. The stereotactic frame was then transferred to the stage of an upright epifluorescence microscope. For cortical fluorescence recovery after photobleaching measurements, the first order beam of an argon ion laser (488 nm) diffracted by an acoustooptic modulator was focused onto the surface of the mouse brain through a dichroic mirror (510 nm) and objective lens (Nikon ϫ50 air). Emitted fluorescence was detected by a photomultiplier. Bleaching was accomplished by increasing laser illumination intensity ϳ4000-fold for 1-2 ms. Apparent diffusion coefficients in brain versus saline (D/D 0 ) were computed from fluorescence recovery curves.
Electron Microscopy-Brain ultrastructure was evaluated at 30 h after bacterial infusion. Mice were killed and the non-injected hemisphere was immersed overnight in 3% (w/v) Karnovsky fixative. Samples were postfixed in 1% buffered osmium tetroxide, dehydrated in ethanol, and embedded in epoxy resin. Ultrathin sections were stained with aqueous saturated uranyl acetate and Reynold's lead citrate and examined in a 1200 EX JEOL electron microscope. The astrocytic foot process cross-sectional area was measured in randomly selected transmission electron micrographs by image analysis software as described (16).
Osmotic Water Permeability of the Blood-brain Barrier-The product of apparent osmotic water permeability of the BBB, P f , and BBB surface area, S, was estimated experimentally from the kinetics of brain water accumulation in response to intraperitoneal injection of distilled water (200 ml/kg), according to the relation, where v w is the partial molar volume of water (0.018 cm 3 /mmol), brain is brain osmolality (ϳ310 mOsm, (24)), and plasma is plasma osmolality after intraperitoneal water infusion (ϳ290 mOsm at 3-6 min) measured using a micro-osmometer (Micro-Osmette, Precision Systems Inc., Natick, MA). Water flux into the brain J v was estimated by wet/dry brain weight measurements at 3 and 6 min after water injection. Statistical Analysis-Data are presented as the mean Ϯ S.E. Statistical analysis was performed using Student's t test, analysis of variance, and Log Rank tests. Analyses were done using WinSTAT (version 2001.1, A-Prompt, Lehigh Valley, PA) or XLStat-Pro (version 7.5, Addinsoft, Brooklyn, NY) software.

RESULTS
Mouse Model of Meningitis-As evidenced by using Evans blue dye, fluid injected at the base of the skull rapidly distributed throughout the CSF (Fig. 1B). Histological examination of the needle tract in two wild type and two AQP4 null mice did not reveal brain abscess or significant hemorrhage (not shown).
No focal histological abnormality was noted after hematoxylin/ eosin staining of brain sections from these mice.
The pneumococcal infection in mice showed multiple similarities to human meningitis. There was marked leukocyte infiltration of the leptomeninges in infected, but not in salineinjected mice (Fig. 1C). Inflammation was prominent in the basal meninges, but was absent from meninges overlying the hemispheric convexities, and was qualitatively similar in wild type and AQP4 null mice. Gram stain of CSF smears revealed that Ͼ90% of leukocytes were polymorphonuclear and associated with Gram-positive streptococci (Fig. 1D).
The severity of pneumococcal infection in the CSF and the systemic reaction were evaluated ( Fig. 1D and Table I). At 30 h after bacterial infusion, CSF leukocyte and bacterial counts were comparably elevated in wild type and AQP4-deficient mice. There was no significant difference in serum sodium concentration to account for the differential brain swelling between wild type and AQP4 null mice with meningitis (presented below). The CSF:serum glucose ratio was low in all mice with meningitis compared with saline-injected control mice. Pneumococcus-infected wild type and AQP4 null mice became comparably hypothermic, which is a manifestation of systemic sepsis, whereas three wild type and three AQP4 null mice injected with sterile saline remained normothermic. Because hypothermia can attenuate brain swelling, measurements of ICP and brain water content were performed after re-establishing normothermia using a heating lamp for 1 h prior to measurements. Arterial blood gas analysis at 30 h after bacterial infusion revealed metabolic acidosis with compensatory respiratory alkalosis. Hemodynamic parameters that could independently influence ICP, including central venous and arterial blood pressures, were also comparable in wild type and AQP4 null mice.
AQP4 Expression-Because AQP4 expression in brain is reported to be up-regulated in brain edema from multiple causes (12,25), AQP4 protein expression was determined in normal mice and mice with pneumococcal meningitis. Immunoblot analysis showed AQP4 protein in brain homogenates from wild type mice as bands at ϳ30 and 34 kDa representing the two AQP4 protein isoforms expressed in brain, which were absent in homogenates from AQP4 null mice ( Fig. 2A). Quantitative immunoblot analysis of brain homogenates taken at 30 h after bacterial infusion indicated ϳ7-fold increased AQP4 protein expression in wild type mice compared with control wild type mice. Immunohistochemistry showed AQP4 expression in wild type mice in pericapillary astrocyte foot processes, external glial limiting membrane, and ependyma (Fig. 2B). Increased AQP4 immunolabeling was found in the external glial limiting membrane, pericapillary regions, and in a fibrillary pattern throughout the brain parenchyma in meningitis (Fig. 2C). AQP4 protein was absent from the brain of an AQP4 null mouse with meningitis (Fig. 2D).
Outcome Measures-After injection of pneumococcus into the CSF and recovery from anesthesia, we noted progressive global neurological deterioration of wild type and AQP4 null mice (Fig. 3A). At 30 h post-infection, AQP4 null mice showed less neurological impairment than wild type mice, as quantified by neurological score. AQP4 null mice also had significantly lower mortality than wild type mice assessed over a one-week period after bacterial infusion (Fig. 3B). Because human subjects with meningitis are likely to receive antibiotics, in the survival studies mice were treated with intraperitoneal cefotaxime (115 mg/kg twice daily for 5 days) starting at 24 h after bacterial infusion. Cefotaxime is the first line treatment for pneumococcal meningitis in humans and has good CSF penetration in meningitis (26).  Because the severity of CSF infection and the systemic response were similar in wild type and AQP4-deficient mice over the first 30 h, we investigated whether the improved outcome in AQP4 null mice was because of reduced brain swelling. Brain swelling was quantified using ICP and brain water content was determined from measurement of wet and dry brain weights. In wild type mice at 30 h after bacterial infusion, ICP was elevated to ϳ3 times normal (Fig. 3C), and associated with a substantial increase in total brain water content (Fig. 3D). Remarkably, ICP did not rise above baseline in AQP4 null mice with meningitis, and the increase in brain water was much reduced (Fig. 3D). In control studies, there was no rise in ICP or brain water content at 30 h after injecting sterile saline into the CSF.

Cytotoxic Versus Vasogenic Brain Edema in Meningitis-
Because AQP4 has opposing roles in cytotoxic (16) versus vasogenic (19) edema, we sought to determine the type(s) of edema that occur(s) in the bacterial meningitis model used here. Electron microscopy of cerebral cortex from mice with meningitis revealed swelling of pericapillary astrocyte foot processes, a marker of cytotoxic edema (Fig. 4A). Averaged pericapillary foot process area was ϳ6-fold greater in brain cortex from wild type compared with AQP4 null mice, which had about the same area as non-infected control mice. No tight junction opening (the hallmark of vasogenic brain edema) was found in 20 randomly chosen capillary endothelial tight junctions from wild type and AQP4 null mice (Fig. 4B). Fig. 4C shows that meningitis produced little increase in extravasation of Evans blue into brain compared with cortical freeze injury, an established model of vasogenic brain edema (10).
Cortical fluorescence recovery after photobleaching was used as an independent method to assess cytotoxic versus vasogenic brain edema. We showed recently that diffusion of FITC-dextran in brain extracellular space is remarkably slowed in cytotoxic brain edema (23), whereas diffusion is accelerated in vasogenic edema (27). The differential response occurs because the extracellular space contracts in cytotoxic (cell swelling) brain edema, but expands in vasogenic brain edema. As shown in Fig. 5A, the diffusion of 70 kDa FITC-dextran was markedly impaired at 30 h after bacterial infusion. Consistent with previous findings (23), the baseline apparent diffusion coefficient was slightly higher in AQP4 null mice. Meningitis produced greater slowing of FITC-dextran diffusion in wild type than AQP4 null mice (67 versus 39% of baseline).
Mechanism by Which AQP4 Deficiency Limits Brain Edema-Two possible mechanisms were investigated to explain the reduced cytotoxic brain edema in meningitis in wild type versus AQP4 null mice: (a) greater intrinsic brain cell swelling in response to meningitis in wild type versus AQP4 null mice, and (b) reduced BBB water permeability in AQP4 null mice. To investigate mechanism a, we measured FITC-dextran diffusion in the extracellular space in brains of wild type versus AQP4 null mice after exposure to meningitic CSF. Meningitic CSF applied directly to brain cortex after removal of the dura produced marked cell swelling within 10 min (Fig. 5B). However, the magnitude of slowing of FITC-dextran diffusion (60 -64% reduction in D/D 0 compared with baseline) was independent of the genotype of mice receiving the meningitic CSF. Therefore, mechanism a is unlikely to explain the reduced cytotoxic brain edema in AQP4 null mice.
Mechanism b predicts reduced BBB osmotic water permeability in AQP4 null mice compared with wild type mice with meningitis. BBB water permeability was estimated by intraperitoneal injection of a water bolus (0.2 ml/g body weight), which reduced serum osmolality from 310 Ϯ 5 to 290 Ϯ 5 mOsm at 3-6 min after infusion. Fig. 5C shows a ϳ9-fold greater rate of water entry into the brain in control (non-meningitis) wild type mice compared with AQP4 null mice. After 30 h of meningitis, water entry into the brain was greatly increased in wild type mice (ϳ5-fold compared with control wild type mice), which is consistent with the up-regulation of AQP4 protein in meningitis shown in Fig. 2.
Mathematical Model of Brain Swelling in Meningitis-A quantitative model of brain swelling was developed to examine whether the reduced water permeability of the BBB in AQP4 null mice could account for their reduced ICP and brain water gain in meningitis. As diagrammed in Fig. 6A, model parameters include the rates of CSF secretion (CSF in ) and outflow (CSF out ), brain compliance (C br ), osmotic water permeability of the BBB (P f (t)), and a time-dependent term describing menin-FIG. 2. AQP4 protein up-regulation in brain in meningitis. A: top, representative immunoblot (20 g of protein/lane); and bottom, data summary (6 mice per group, *, p Ͻ 0.05) of total AQP4 protein expression. B, control wild type mouse brain showing AQP4 expression (brown) around capillaries (black arrowheads) and in glia limitans (white arrowheads). C, brain of a wild type mouse 30 h after bacterial infusion. Increased AQP4 immunoreactivity seen around capillaries (black arrowheads), throughout the brain parenchyma, and in glial limiting membrane (white arrowheads). D, brain of AQP4 null mouse.
Intracranial pressure (ICP(t)) is related to V brain (t) from the compliance relation as follows.
As measured in mice by intraventricular perfusion and dye dilution, CSF in ϭ 0.4 l/min and CSF out (t) ϭ 0.14(ICP(t) Ϫ 0.87) l/min (28). From data in Fig. 5C, the P f ⅐S product for wild type (P f ϩ/ϩ (t)⅐S) and AQP4 null (P f Ϫ/Ϫ (t)⅐S) control mice were 1.7 ϫ 10 Ϫ2 and 0.18 ϫ 10 Ϫ2 cm 3 /s, respectively, and 9.3 ϫ 10 Ϫ2 and 1.3 ϫ 10 Ϫ2 cm 3 /s at 30 h after bacterial infusion. For computations we assumed constant P f (t)⅐S for the first 240 min followed by a linear increase thereafter (Fig. 6A). The functional form of ⍀ in was chosen to reflect bacterial growth in CSF: where ⍀ in (0) ϭ 0.04 l/cm 3 and ␣ ϭ 1.4 ϫ 10 Ϫ3 were specified to give the measured ICP of ϳ29 cm H 2 O at 30 h after bacterial infusion in wild type mice. Fig. 6B shows the predicted V brain (t) and ICP(t) from numerical integration of Equations 2 and 3. ICP and brain water increased progressively in the wild type mice. As found experimentally (Fig. 3, C and D), ICP and brain water accumulation in AQP4 null mice were remarkably reduced, which was simulated by replacing P f ϩ/ϩ (t) by P f Ϫ/Ϫ (t). A third simulation, labeled "constant P f S, " was done in which (P f ϩ/ϩ (t)⅐S) was fixed at (P f ϩ/ϩ (0)⅐S) as measured in control wild type mice, rather than allowed to increase as a consequence of meningitis-induced regulation. Interestingly, maintaining a constant P f ϩ/ϩ (0)⅐S produced little increase in ICP and brain water over 30 h. These computations support the idea that reduced water permeability of the BBB in AQP4 deficiency accounts for the improved outcome in meningitis, and predict substantially reduced brain swelling in meningitis after either inhibition of AQP4 water permeability or its up-regulation. DISCUSSION Our data implicate AQP4 as the major determinant of BBB water permeability. In wild type and AQP4 null mice with meningitis, inflammatory and bacterial products in the CSF induced water movement from the vascular compartment into the brain. About 80% of the excess water in meningitis entered the brain parenchyma through AQP4 water channels. AQP4 null mice with meningitis had a much better outcome than wild type mice, with remarkably lower intracranial pressure and brain water accumulation, and improved neurological status and survival. FIG. 4. Cytotoxic brain edema in mouse model of meningitis. A: left, electron micrographs showing marked swelling of pericapillary astrocyte foot processes (arrowheads) in AQP4 ϩ/ϩ , but not AQP4 Ϫ/Ϫ mice. L, capillary lumen; N, endothelial cell nucleus. Right, astrocyte foot process areas in control mice (5 micrographs per mouse, 1 mouse each) and mice with meningitis (10 micrographs per mouse, 2 mice each), *, p Ͻ 0.05. B, electron micrographs of capillary endothelial tight junctions (arrowheads) from brains of AQP4 ϩ/ϩ and AQP4 Ϫ/Ϫ mice with meningitis. Each tight junction is representative of 20 examined. BM, endothelial cell basement membrane; L, capillary lumen. C, Evans blue dye extravasation into brain parenchyma (g of dye per g dry hemisphere weight) of control mice, mice with meningitis, and mice after cortical freeze injury (5 mice per group). There was no significant difference between AQP4 ϩ/ϩ versus AQP4 Ϫ/Ϫ mice.
FIG. 5. FITC-dextran diffusion in brain extracellular space measured by cortical surface photobleaching. A, representative fluorescence recovery curves (left), and calculated apparent diffusion coefficient (D/D 0 ) (right), for 70 kDa FITC-dextran in brain extracellular space of normal mice (4 per group) and mice with meningitis (5 per group, p Ͻ 0.01 comparing AQP4 ϩ/ϩ versus AQP4 Ϫ/Ϫ meningitis mice). B, FITC-dextran diffusion in brain extracellular space after exposing brains of AQP4 ϩ/ϩ and AQP4 Ϫ/Ϫ mice to CSF collected from control mice (control CSF) or mice after 30 h of meningitis (meningitis CSF) (5 mice per group, p Ͻ 0.001 comparing normal versus meningitis CSF). C, water accumulation in brains of control mice and mice with meningitis induced by intraperitoneal injection of water (200 ml/kg).
Pneumococcal infection caused a 7-fold up-regulation of AQP4 protein expression in brain, which further increased BBB water permeability and hence the accumulation of excess brain water. It is not impossible to study experimentally whether the increase in AQP4 expression is key to meningitisinduced brain edema because inhibitors of AQP4 up-regulation are not available. To assess the role of AQP4 up-regulation, we constructed a mathematical model taking into account the different pathways for water brain water entry/exit. The model predicted that the majority of the excess water in meningitis entered the brain through up-regulated AQP4 water channels. AQP4 up-regulation in meningitis is thus a maladaptive response that accelerates brain swelling. AQP4 up-regulation has also been found in the brain of a human patient with brain edema secondary to acute bacterial meningitis (29), suggesting that similar mechanisms might occur in humans. Increased astrocyte AQP4 expression is not restricted to meningitis, but occurs in several pathologies associated with cytotoxic brain edema, including traumatic brain injury in humans (29) and rats (30), cerebral ischemia in humans (31) and rats (32), and hyponatremia in rats (33). Direct measurement of BBB water permeability (P f S), as was done here for meningitis, will be required to determine whether AQP4 up-regulation also accelerates brain water accumulation in these conditions. In general, brain edema is associated with increased AQP4 transcript expression in astroglia (25,30,32,34), suggesting that upregulation of AQP4 protein involves increased mRNA synthesis or decreased degradation. Altered AQP4 expression in astroglia might be mediated by changes in osmolality through a p38 mitogen-activated protein kinase-dependent pathway (34). It is not known whether cytokines or bacterial products can also influence AQP4 expression.
Although it has been suggested that both cytotoxic and vasogenic edema types might co-exist in acute bacterial meningitis, several studies have shown that astroglia are primarily damaged, suggesting that cytotoxic edema predominates. In vitro, lipopolysaccharides cause ultrastractural changes (35) and selectively depolarize (36) rodent astroglia, whereas pneumococcal cell walls selectively kill human astroglia (37). In rodents in vivo, the invading neutrophils become intimately associated with astroglia (38) and Haemophilus influenzae causes massive astroglial swelling (39). Our data also suggest that excess brain water in meningitis primarily accumulates in the intracellular compartment (cytotoxic brain edema). This is supported by morphological evidence of diffuse astroglial foot process swelling and biophysical evidence of reduced macromolecular diffusion in brain extracellular space. Pneumococcal meningitis produced only little opening of the BBB as demonstrated by electron microscopy of brain capillary endothelial tight junctions and measurement of Evans blue dye extravasation, suggesting little vasogenic brain edema. This agrees with the observation that in humans (40), dogs (41), and rats (42) non-complicated bacterial meningitis does not on magnetic resonance scans cause parenchymal enhancement (which would suggest disruption of the BBB), but only meningeal enhancement.
We hypothesize that, during meningitis, astroglia become unable to regulate ionic gradients across the plasma membrane. This causes accumulation of intracellular sodium (with accompanying chloride and water), resulting in cell swelling. The mechanisms of cytotoxic brain edema have been extensively investigated in early focal cerebral ischema in cats (43), rats (44 -46), and gerbils (47), which manifest increased salt and water accumulation from the blood into the brain. Many factors present in CSF of human subjects or animals with acute bacterial meningitis have been proposed to impair the ability of cells to regulate their ionic gradients, including cytokines, free radicals, excitatory amino acids, interleukin-1␤, tumor necrosis factor-␣, and H 2 O 2 (48 -54).
When the BBB is intact, water flows from the blood into the brain through AQP4, which are abundant at the capillaryfacing plasma membranes of astroglial foot processes (Fig. 6C). The excess water enters astroglia and increases intracellular space volume at the expense of extracellular space volume, as found in Fig. 5. Not all parts of the astroglial cell swell equally: foot processes swell more because they are adjacent to compressible capillaries (Fig. 4), but cell bodies expand less because they are next to poorly compressible swollen brain cells.
For these studies, a mouse model of acute bacterial meningitis was developed that reproduced many features of acute bacterial meningitis in humans including leptomeningeal inflammation, CSF leukocytosis, low CSF glucose, impaired neurological function, and fulminant course. Several approaches were evaluated to introduce bacteria into the CSF, including cisternal, intraparenchymal, and transparenchymal basal infusions at different rates and using different needle types. We FIG. 6. Mathematical model of brain water accumulation in meningitis. A, the rate of water entry into the brain/CSF compartment is the sum of the rates of CSF production (CSF in ) and edema fluid formation (P f S⅐⍀ in ); the rate of edema fluid elimination is governed by CSF absorption (CSF out ) (left). See "Results " for definition of symbols. ⍀ in (t) (right top) and P f (t)⅐S (right bottom) after bacterial infusion used for computations. B, predicted effect of meningitis on ICP (left) and brain water (right) in wild type mice (AQP4 ϩ/ϩ ), AQP4 null mice (AQP4 Ϫ/Ϫ ), and wild type mice in which AQP4 up-regulation is inhibited (constant P f S). C, edema fluid is produced by water moving from the capillary lumen across endothelial cells (arrows), through AQP4 channels (open circles) into astroglial foot processes.
found that slow infusion of a bacterial suspension through brain parenchyma into the basal CSF space with a blunted 27-gauge needle avoided problems of fluid leakage and abscess formation, and gave a highly reproducible course of meningitis. The needle did not cause significant brain injury as assessed by histological examination of the needle tract, and normal intracranial pressure and neurological function in saline-injected control mice. We chose S. pneumoniae, because it is a major cause of meningitis in humans whose incidence is increasing compared with Neisseria meningitidis and H. influenzae (1)(2)(3)(4)(5). The presence of pneumococcus in the CSF produced marked brain swelling with increased intracranial pressure and brain water content, resulting in impaired neurological function and mouse survival.
From the present findings and previous reports (14,16), blockers of AQP4 water permeability are predicted to reduce brain swelling in meningitis, early stroke, hyponatremia, and other pathologies where cytotoxic edema is prominent. However, AQP4 channel blockers may impair hearing (21) and vision (55), and increase brain swelling where vasogenic brain edema predominates as in brain tumor or late stroke (11). Because increased AQP4 expression accounts for the majority of the excess brain water in meningitis, we propose that inhibitors of AQP4 up-regulation may be better drugs than AQP4 channel blockers. Inhibitors of AQP4 up-regulation would efficiently reduce brain edema in meningitis without the potential side effects of inhibition of AQP4 water permeability.