Targeting Aquaporin-4 Subcellular Localization to Treat Central Nervous System Edema

Abstract

Swelling of the brain or spinal cord (CNS edema) affects millions of people every year. All potential pharmacological interventions have failed in clinical trials, meaning that symptom management is the only treatment option. The water channel protein aquaporin-4 (AQP4) is expressed in astrocytes and mediates water flux across the blood-brain and blood-spinal cord barriers. Here we show that AQP4 cell-surface abundance increases in response to hypoxia-induced cell swelling in a calmodulin-dependent manner. Calmodulin directly binds the AQP4 carboxyl terminus, causing a specific conformational change and driving AQP4 cell-surface localization. Inhibition of calmodulin in a rat spinal cord injury model with the licensed drug trifluoperazine inhibited AQP4 localization to the blood-spinal cord barrier, ablated CNS edema, and led to accelerated functional recovery compared with untreated animals. We propose that targeting the mechanism of calmodulin-mediated cell-surface localization of AQP4 is a viable strategy for development of CNS edema therapies.

Keywords: AQP4; Aquaporin; TRPV4; astrocyte; calmodulin; edema; oedema; protein kinase A; spinal cord injury; traumatic brain injury; trifluoperazine.

Graphical abstract

Figure 1

Hypoxia Induces AQP4 Subcellular Relocalization in Primary Cortical Astrocytes (A) Mean fold change in AQP4 surface expression (±SEM), measured by cell-surface biotinylation in primary cortical astrocytes. Cells were treated with 5% oxygen for 6 h (hypoxia) or 85 mOsm/kg H₂O (hypotonicity) compared with untreated normoxic astrocytes (control). The CaM inhibitor (CaM_i) was 127 μM trifluoperazine (TFP). The TRPV4 inhibitor (TRPV4_i) was 4.8 μM HC-067047, and the intracellular Ca²⁺ chelator was 5 μM EGTA-AM. The TRPV4 channel agonist (TRPV4_a) was 2.1 μM GSK1016790A. Kruskal-Wallis with Conover-Inman post hoc tests were used to identify significant differences between samples. ^∗p < 0.05; ns represents p > 0.05 compared with the untreated control (Table S2; n = 4). (B) Mean fold change in AQP4 surface expression (±SEM) with time under hypoxia. Rat primary cortical astrocytes were exposed to 5% oxygen, and AQP4 surface expression was measured by cell-surface biotinylation after 1, 3, and 6 h and compared with untreated normoxic astrocytes (normoxia). Cells were returned to normoxic conditions (21% oxygen), and AQP4 surface expression was measured at 1, 3, and 6 h. ^∗p < 0.05 by ANOVA followed by t test with Bonferroni correction (Table S2; n = 3). (C) Calcein fluorescence quenching in response to elevation of extracellular osmolality to 600 mOsm with mannitol was used to quantify astrocyte plasma membrane water permeability following the same hypoxia and inhibitor regimen described in (A). ^∗p < 0.05; ns indicates p > 0.05 by ANOVA followed by t test with Bonferroni correction (Table S2; n = 3). (D) Normalized membrane water permeability of hypoxic rat primary cortical astrocytes over 6 h. Cells were returned to normoxic conditions (21% oxygen), and permeability was measured over the subsequent 6 h. ^∗p < 0.05 compared with the t = 0 normoxic control (Table S2; n = 3). (E) Representative calcein fluorescence quenching traces for hypoxia and normoxia. (F) Intracellular cAMP accumulation in rat primary cortical astrocytes (±SEM) in response to forskolin, hypotonicity (with or without extracellular Ca²⁺ [the intracellular Ca²⁺ chelator was 5 μM EGTA-AM] or the CaM_i W-7 or TFP), and the CaM_a peptide CALP-3. PKA activity in lysates made from rat astrocytes subjected to the same treatments. Activity is normalized to the average of the untreated control. ^∗p < 0.05. ns represents p > 0.05 by ANOVA followed by t test with Bonferroni correction (Table S2; n = 3). (G) Phosphorylation of glutathione S-transferase (GST)-recAQP4ct, analyzed by Phos-tag SDS-PAGE. The phosphorylation stoichiometry was judged to be complete (∼1 mol phosphate/mol GST-recAQP4ct, 1P) after 2 h.

Figure S1

Knockdown of AQP4 Suppresses Acute Cytotoxic Edema and Improves Functional Recovery in Rats at 3 dpi but Not at 28 dpi following DC Crush Injury a, ELISA for AQP4 showing no increase in total AQP4 protein in primary astrocytes after 6 h under hypoxia (5% O₂; n = 6); b, Representative immunoblot and c, densitometry to confirm 60% and 75% knockdown of AQP4 using in vivo JetPEI-delivered shRNA to AQP4 (shAQP4) at 3 dpi and 28 dpi following DC crush injury (n = 3 independent repeats from 3 pooled rat spinal cords/experiment (total n = 9 rats/condition)); d, Spinal cord water content was significantly suppressed at 3 dpi, but at 28 dpi the water content in DC + shAQP4-treated animals was significantly higher than DC + Vehicle-treated controls, despite 75% AQP4 protein knockdown (n = 3-4 rats/condition, 3 independent repeats (total n = 10 rats/condition)); e, Knockdown of AQP4 improved tape sensing and removal time up to 1 week after DC crush injury, but the sensing and removal time gradually increased to above DC + vehicle-treated controls at 4 weeks after DC crush injury (n = 3-4 rats/condition, 3 independent repeats (total n = 10 rats/condition)); f, The early improvement in ladder crossing ability of rats gradually worsened 1 week after and at 28dpi is higher than DC+vehicle-treated controls (n = 3-4 rats/condition, 3 independent repeats (total n = 10 rats/condition)). ^∗ represents p < 0.05, ns represents p > 0.05 (see Table S2 for p values). Animals were euthanized at 28dpi due to episodes of vomiting, gait problems, lethargy and convulsions, possibly reflecting their inability to regulate water in the CNS. Related to Figure 2.

Figure 2

Inhibition of CaM or PKA Reduces Spinal Cord Water Content, AQP4 Translocation to the BSCB, and AQP4 Expression after DC Crush Injury In Vivo (A) Water content of the thoracic spinal cord 3 days after dorsal column (DC) crush and treatment with CaM or PKA inhibitors (CaM_i or PKA_i, respectively). Sham, laminectomy only; DC+vehicle, T8 DC crush +PBS; DC+CaM_i, DC crush + intra-lesion injection of 41 mM TFP; DC+PKA_i, DC crush + intra-lesion injection of 10 μM H89; n = 4 rats per treatment group, except DC+vehicle 3 dpi (n = 8) and DC+vehicle 28 dpi (n = 3). (B) Water content of the thoracic spinal cord 3 days after DC crush to investigate the mode of action of TFP. DC+CaM_i, DC crush + intra-lesion injection of 41 mM TFP or 164 mM W-7; DC+A1R_i, DC crush + intra-lesion injection of 53 mM α1 adrenergic receptor antagonist terazosin; DC+D2Ri_i, DC crush + intra-lesion injection of 6.6 mM dopamine D₂ receptor antagonist L-741,626; DC+PKC_i, DC crush + intra-lesion injection of 9.94 μM PKC inhibitor Gö 6983. n = 3–12 rats per treatment group, normalized to sham controls across multiple experiments. (C–F) Representative images used to quantify data in (G) and (H) (scale bar, 50 μm). Sham tissue (C) was compared with tissue from animals 3 days after DC crush (3 dpi; D), for which an increase in total AQP4 expression and translocation to the blood-spinal cord barrier (BSCB; white arrows identify astrocyte endfeet) is observed, which is ablated by treatment with TFP (E) or treatment with H89 (F). Two representative magnified images are shown for each panel. The contrast in the insets was manually adjusted to aid the eye (scale bars, 10 μm). (G) Quantification of at least 24 images (a minimum of 4 images per animal, n = 6) shows changes in AQP4 BSCB localization. (H) Quantification of at least 18 images (a minimum of 3 images per animal, n = 6) shows changes in total AQP4 expression. For all comparisons, ANOVA was followed by post hoc Bonferroni-corrected t tests. ^∗p < 0.05, ns represents p > 0.05 (Table S2). See also Figure S1, Figure S2, Figure S3.

Figure S2

Targeted Inhibition of CaM or PKA Reduces Acute Brain Water Content after an In Vivo Brain Stab Injury Model of Cytotoxic Edema; Increases in AQP4 Expression and AQP4 Translocation to the Blood-Spinal-Cord Barrier are Localized to the Injury Site In Vivo. a, Water content of the ipsilateral and contralateral brain cortex 3 days after stab injury and treatment with CaM, PKA or PKC inhibitors (CaM_i or PKA_i). Sham = craniotomy only; Stab injury+Vehicle = 3 mm cortical stab injury + PBS; Stab injury+CaM_i = 3 mm cortical stab injury + intra-lesion injection of 41 mM trifluoperazine (TFP); Stab injury + PKAi = 3 mm cortical stab injury + intra-lesion injection of 10 mM H89; Stab injury+PKCi = 3 mm cortical stab injury + intra-lesion injection of 9.94 μM Gö 6983; n = 18 rats per treatment group, ^∗ represents p < 0.05, ns represents p > 0.05 (see Table S2 for p values). Central bars represent median, crosses represent the mean, outer bars represent upper and lower quartiles. Outliers (data points more than 1.5 × IQR from the median) are shown as separate points; b, Increases in AQP4 expression and AQP4 translocation to the blood-spinal-cord barrier are localized to the injury site in vivo. Representative images showing AQP4 expression (green) with endothelial cells labeled with RECA1 (red) in Sham animals; c, 3 days after DC crush (3 dpi), an increase in total AQP4 expression and translocation to the blood-spinal-cord barrier (BSCB) is observed at the injury site; d, At a minimum of 3 mm away from the lesion, total AQP4 expression and translocation to the BSCB are indistinguishable from that of sham animals; e, Normalized relative perivascular fluorescence (A.U. ± SD 3 days after DC crush injury; sham = laminectomy only; DC+vehicle = T8 DC crush + PBS at the injury site; DC+vehicle (away) = T8 DC crush + PBS at sites at a minimum of 3 mm away from the lesion site; f, Normalized total AQP4 fluorescence (A.U. ± SD) following the same experimental conditions described in panel e. ^∗ represents p < 0.05, ns represents p > 0.05 (see Table S2 for p values). Related to Figure 2.

Figure S3

Subcellular Fractionation of Primary Rat Astrocytes Reveals Foxo3a Nuclear Translocation a-d, Immunofluorescence micrographs of rat spinal cord tissue stained for Foxo3a (green) and DNA (DAPI, blue) 3 days after dorsal column (DC) crush and treatment with PBS (DC + vehicle), TFP (DC + CaM_i) or H89 (DC + PKA_i) injected into the lesion site; zoomed-in images are shown for each panel; e, Relative protein kinase activity in cultured primary rat astrocytes subjected to PKA activation with forskolin (10⁻⁵ M) or 10 minutes extracellular hypotonicity (85 mOsm) in the absence or presence of TFP. Data (n = 3) are normalized to untreated controls. ^∗ represents p < 0.05, ns represents p > 0.05 compared to untreated control by ANOVA followed by t test with Bonferroni correction, see Table S2 for p values); f, Immunoblot of fractionated primary rat astrocytes showing the abundance of Foxo3a, which is higher in the nuclear fraction (only degradation products are seen the cytoplasmic fraction as shown in panel g) following PKA activation with forskolin or hypotonicity. This is consistent with the image in panel b showing localization of Foxo3a to nuclei in injured spinal cord tissue. C = cytoplasmic fraction, N = nuclear fraction; g, Primary rat astrocytes were subjected to subcellular fractionation following activation of PKA by forskolin or hypotonicity. Intact Foxo3a was detected only in the nucleus (predicted molecular weight 71 kDa, black arrowhead). Degraded Foxo3a was detected in the cytoplasm (white arrowhead). h, Densitometry of nuclear Foxo3a signals shown in panel g and normalized to nuclear Lamin B. ^∗ represents p < 0.05, ns represents p > 0.05 (see Table S2 for p values). Related to Figure 2.

Figure S4

The Predicted AQP4 CBD and Its Comparison with the CBD in the AQP0 Crystal Structure a, Crystal structure of the human AQP4 tetramer viewed from the side of the membrane and from the extracellular side. The carboxyl terminus, for which there is no structural information, is shown as beads. The sequence of the predicted CaM-binding domain (green beads) is shown in the box with hydrophobic residues highlighted in green. The phosphorylation site at Ser 276 is highlighted with a red circle; b, Crystal structure of bovine AQP0 (PDB code 1YMG) showing the carboxy-terminal helix (black box) which harbors the CaM-binding domain of AQP0; c, Zoom-in on the CaM binding domain with residues involved in binding shown in stick representation; d, Helix wheel representation of the AQP0 carboxy-terminal helix showing its amphipathic character. Colors indicate residue types as follows: hydrophobic-green, basic-blue, acidic-red and polar-yellow; e, Top scoring structural model of the predicted AQP4 CaM-binding site generated by PEP-FOLD3. Hydrophobic and charged/polar residues on either side of the predicted helix are shown in stick representation; f, Helical wheel representation of the predicted helix in panel e showing its amphipathic character. Colors indicate residue types as follows: aromatic-purple, hydrophobic-green, basic-blue, acidic-red and polar-yellow; g, Hypotonicity-induced translocation of AQP4 in HEK293 cells is abrogated by a triple F258/262/266A mutation, but not by the corresponding single point mutations. ^∗p < 0.05 by ANOVA followed by Bonferroni-corrected t test, ns denotes p > 0.05 (see Table S2 for p values). Related to Figure 3.

Figure S5

MST and CAP Experimental Controls a, Typical MST traces. The relative fluorescence is plotted against the experiment time. Each trace corresponds to a sample with a different concentration of AQP4 whereas calmodulin (CaM) concentration remained constant, except for the trifluoperazine (TFP) titration experiment (bottom right) in which TFP was varied and both AQP4 and CaM were kept constant. The difference in relative fluorescence before (blue column) and after (red column) heating is used to calculate ΔF_norm. The position of the red column was determined by the M.O. Affinity Analysis software (Nanotemper) as the time interval that gave the best signal to noise ratio; b, Recombinant AQP4 is a functional water channel. Fluorescence traces from a proteoliposome shrinking assay showing that liposomes containing purified AQP4 (blue) have significantly higher water permeability than empty liposomes (gray). The increase in fluorescence corresponds to the fluorophore ((5)6-carboxyfluorescein) present on the inside of the liposomes becoming more fluorescent as the liposomes shrinks when mixed with hyperosmotic solution. The data were fitted to a double exponential function (solid blue and black lines for AQP4-containing and empty liposomes, respectively) and the rate constant (k) was used to calculate the osmotic water permeability (P_f). For AQP4-containing liposomes P_f = 5.9 ± 0.5 × 10⁻² cm/s and for control liposomes P_f = 1.2 ± 0.2 × 10⁻² cm/s, corresponding to an AQP4 single channel permeability of 1.1 ± 0.1 × 10⁻¹³ cm³/s. Analysis by t test showed a statistically-significant difference between AQP4-containing liposomes and empty liposomes (p = 0.0015; Table S2). Related to Figure 3; c, Representative Spike 2 software processed CAP traces. A normal short-latency negative-positive CAP trace was observed in sham animals which was ablated in DC+vehicle- and DC+PKCi-treated rats, but partially restored in DC + CaMi and DC + PKAi-treated rats. Following a dorsal hemi-section in the same animals at the end of the experiment, CAP traces were ablated in all animals, indicating DC axon regeneration. Related to Figure 4.

Figure 3

Binding of CaM to the AQP4 Carboxyl Terminus (A) Microscale thermophoresis (MST) data showing that full-length AQP4 interacts directly with CaM. The binding curve, obtained by plotting ΔF_norm against [AQP4], could be fitted to a one-to-one binding model (estimated K_d of 29 ± 5.6 μM). Addition of 5 mM EDTA demonstrated that binding was Ca²⁺ dependent. The interaction was also inhibited by addition of 1 mM TFP. Truncation of the AQP4 carboxyl terminus before the predicted CBD (AQP4-Δ256) resulted in a construct that did not interact with CaM. AQP4 F258/262/266A did not bind CaM. The phospho-mimetic mutant AQP4-S276E bound CaM with approximately 2-fold higher affinity (K_d = 17 ± 3.1 μM) than WT AQP4 (p = 0.031), suggesting that phosphorylation of S276 affects the interaction with CaM. (B) Response curve showing that TFP inhibits the interaction between AQP4 and CaM in a concentration-dependent manner. The concentration of TFP needed for 50% inhibition (IC₅₀) was estimated to be 790 ± 2 μM. (C) Left: ¹H, ¹⁵N-HSQC NMR data showing the interaction of the recombinant carboxyl terminus of AQP4 (recAQP4ct; residues 254–323) with CaM at 30°C. Chemical shift perturbations (CSPs) were observed by titrating CaM into 0.5 mM ¹³C, ¹⁵N-labeled recAQP4ct with 0 (black) and 2 (green) molar equivalents of CaM in the presence of 6 mM Ca²⁺. Top right: CSPs induced by CaM binding to recAQP4ct were plotted as a function of the residue number (X indicates that data are not available). Bottom right: chemical shift index (CSI) values of the Cα and C′ atoms of recAQPct. The sequence of the CBD is underlined. The region with consecutive positive CSI values (red) represents an α-helical conformation. (D) Anti-CaM immunoblotting following nickel affinity chromatography (IMAC) from 1% Triton X-100 lysates (input) of HEK293 cells transfected with AQP4-His₆ (AQP4) or AQP4 F258/262/266A-His₆ (AQP4 CBD_mut) demonstrates that the F258A/F262A/F266A mutation abrogates AQP4-CaM binding. (E) Cell-surface biotinylation followed by neutravidin/anti-AQP4 ELISA demonstrates a reduced rate of AQP4 plasma membrane accumulation in AQP4-transfected HEK293 cells upon F258A/F262A/F266A mutation (AQP4 CBD_mut) compared with the wild-type control (AQP-WT). Normalized data were fitted to functions of the form 1−e^−kt, and t_1/2 was calculated as −ln(0.5)/k. See also Figures S4 and S5.

Figure 4

Inhibition of AQP4 Expression and Subcellular Relocalization after Cytotoxic Edema In Vivo Improves Electrophysiological, Sensory and Locomotor Functional Recovery, BSCB Breakdown, and Spinal Cord Cavitation (A) Superimposed representative spinal cord compound action potential (CAP) traces demonstrating the significant functional improvement in CaM_i-treated (TFP) and PKA_i-treated (H89) rats compared with vehicle-treated and PKC_i-treated (Gö 6983) rats following DC crush. CaM_i treatment was 41 mM TFP. PKA_i treatment was 10 mM H89. PKC_i treatment was 9.94 μM Gö 6983. In all cases, these were injected directly into the lesion site at a volume of 2.5 μL. (B and C) CaM_i and PKA_i significantly improved the mean CAP area (B) and the mean CAP amplitude (C). (D) CaM_i and PKA_i treatment significantly improved the hindpaw tape sensing and removal time, and within 2 weeks, CaM_i-treated and PKA_i-treated rats recovered completely and were indistinguishable from sham controls. (E) The mean ratio of slips/total steps in a ladder-crossing test was significantly improved after CaM_i and PKA_i treatment, returning to sham control levels by 2 weeks. For the tape sensing/removal and ladder crossing tests, significant deficits remained in vehicle-treated and PKC_i-treated rats. #p < 0.05, linear mixed model; ##p < 0.05, generalized linear mixed model; calculated as described previously (Fagoe et al., 2016; Table S2; n = 6 rats/group; 3 independent repeats; total n = 18 rats/group). (F) Treatment with 41 mM TFP significantly reduced BSCB breakdown 7 days after DC crush (right panel), as determined by albumin staining (green) around the lesion site (representative images are shown in the left panel). ^∗p < 0.05 (Table S2) by ANOVA followed by t test with Bonferroni correction. (G) TFP also suppressed the normal process of cavitation that occurs at lesion sites after DC crush, significantly reducing the cavity area 6 weeks after DC crush at all depths through the DC (right panel; black-filled dots represent DC+vehicle; white dots represent DC+CaM_i), as determined by laminin staining (brown) (representative images are shown in the left panel). n = 6 rats/group; 2 independent repeats (total n = 12 rats/group); p = 4.0 × 10⁻¹⁰, comparing lesion volume in DC+vehicle with DC+CaM_i by t test (Table S2). See also Figure S5 and Table S1.

Figure 5

AQP4 Subcellular Relocalization Drives Cytotoxic Edema: The Proposed Roles of CaM and PKA For a Figure360 author presentation of this figure, see https://doi.org/10.1016/j.cell.2020.03.037. Following hypoxic insult, failure in Na⁺, K⁺, and Cl⁻ pumps in the plasma membrane leads to osmotic dysregulation. The mechanosensitive TRPV4 channel facilitates an influx of Ca²⁺ ions into astrocytes, which activates CaM. CaM interacts with an adenylyl cyclase, activating cyclic AMP (cAMP)-dependent PKA, which phosphorylates AQP4 at Ser276, causing it to relocalize to the plasma membrane. CaM interacts directly with AQP4; this regulatory interaction drives AQP4 subcellular relocalization (created with https://biorender.com).