Adult, newborn and embryonic mice of the C57/BL6J strain (Jackson Laboratories, Boulder, CO, United States) were used in this study. The mice were allowed ad libitum access to food and drinking water. Experimental protocols were approved by the Norwegian Animal Research Authority (NARA) with license number FOTS 8572 and carried out in accordance with the European Directive 2010/63/EU. The derivation of mouse embryonic stem cells (mESC) was approved by the ethical committee for the use of laboratory animals at Lund University and the Swedish Work Environment Authority with permit number M43-15 and carried out in accordance with European and Swedish regulations.

Wildtype (C57BL/6) mouse embryonic stem cell (mESC) line was derived as previously described (Chumarina et al., 2017). mESCs (lines CSC 2.3), at passages 14 and 17, were cultured on a monolayer of CF1 irradiated MEFs in DMEM medium containing 15% fetal bovine serum (FBS), penicillin/streptomycin (P/S; 100 Units/mL and 100 mg/mL, respectively), 2 mM L-glutamine (GIBCO), 100 mM non-essential amino acids (NEAA, GIBCO), 1% nucleosides (Millipore), 10% FBS (Millipore), β-mercaptoethanol (110 μM), and supplemented with leukemia inhibitor factor (LIF, 10,000 units/mL; Millipore, ESG1106). Cells were maintained at 37 °C, 5% CO₂, and passaged using 0.5% trypsin (Millipore) every 4–5 days.

mESC-derived astrocytes were generated as previously described (Roybon et al., 2013; Savchenko et al., 2019). Briefly, mESCs were harvested by trypsinization, transferred to low-attachment culture flasks and cultured as free-floating spheres in 50% of DMEM and 50% of Neurobasal media containing penicillin/streptomycin (P/S; 100 Units/mL and 100 mg/mL, respectively), 2 mM L-glutamine (GIBCO), β-mercaptoethanol (110 μM) and 10% KnockOut Serum Replacement (Invitrogen). The medium was supplemented with 1 μM Retinoic acid (RA) (Sigma-Aldrich, R2525) and 0.5 μM Smoothened Agonist (SAG) (Millipore, 566660) and changed at days 2 and 5. On day 7, the free-floating spheres were mechanically dissociated and transferred to neurosphere medium composed of DMEM/F12 (Thermo Fisher Scientific), 2% B27 supplement, 2 mM L-glutamine, 100 mM non-essential amino acids, P/S (100 Units/mL and 100 mg/mL, respectively) and 2 mg/mL heparin (Sigma-Aldrich), supplemented with FGF2 (20 ng/mL) and EGF (20 ng/mL) for 2 weeks. Additionally, spheres were passaged on day 14 to generate secondary neurospheres. On day 21, spheroids were dissociated into single cells and plated onto polyornithine (100 μg/mL)/laminin (2 μg/mL) coated plates in neurosphere medium supplemented with 10% Fetal Bovine Serum (FBS) (Thermo Fisher Scientific, 10270106). The medium was changed twice per week. On day 36, the cells were used for the experiments.

Murine astrocyte cultures were prepared from C57/BL6J embryonic E16-18 or P1-P2 mouse cortices by a protocol modified from de Vellis and Cole (2012); Pacifici and Peruzzi (2012), and Schildge et al. (2013). The skin was removed, and skull opened with separate sterile scissors to expose the brain, and the meninges removed under a stereomicroscope with sterile filter paper. Approximately 1 mm x 1 mm pieces of cortex were extracted with a sterile scalpel from each brain hemisphere and collected in cold complete medium containing 10% FBS (BioWest, Nuaillé, France), 90% DMEM High glucose (BioWest) and 1% Penstrep (Thermo Fisher Scientific, Waltham, MA, United States). Cortex pieces were washed with cold complete medium two times before enzymatic digestion with warm 1–2 mL TrypLE Express (Thermo Fisher Scientific) for 10 min in a 37^°C water bath with occasional shaking. Complete medium was added, and the tubes were centrifuged for 5 min at 1,000 rpm. The pellet was triturated in complete medium using first a 10 mL then 5 mL stripette. The cell suspension was filtered through a 70 μm cell strainer to remove debris. The filtered cells were plated at a density of 3–5 × 10⁶ cells per flask. Care was taken to complete the process within 1–2 h. Medium was changed on day 2 after plating, and every 3 days thereafter. After 7–8 days, astrocytes were confluent and were shaken in warm serum free medium for several minutes to remove contaminating microglia and oligodendrocytes, which grow on top of the astrocyte monolayer. The serum free media containing contaminating cells was removed, the astrocyte monolayer washed twice, trypsinized, and centrifuged at 180 g for 5 min. The cells were resuspended at a 1:2 ratio into T75 flasks and cultured in medium with serum. To ensure matured astrocytes, experiments were performed 2–3 weeks after the first split, and passages were limited to three (Schildge et al., 2013). For all experiments, astrocytes were grown on high molecular weight (Mol W > 300,000) poly-D-lysine (Merck Millipore, Darmstadt, Germany) coated flasks and plates and cultured in a 5% CO₂ 37^°C 95% O₂ humidity incubator. For BMP treatments, primary astrocytes were grown to 90% confluence on 96-well glass plates (for Plate Runner assays), 24-well plates (for RNA extraction) or 6-well plates (for protein extraction). The cells were washed with PBS and kept in serum free media overnight. The next day, cells were treated with 50 ng/mL BMPs in serum free media for 6 days, 72 or 12 h.

We assessed BMPs –2, –3, –4, –5, –6, –7, and –10 for their effect on AQP4 and GFAP protein expression in mouse embryonic cell derived astrocytes (mESC) and primary astrocytes through a high throughput immunofluorescent assay (Plate Runner HD, Trophos), as previously described (Savchenko et al., 2019). Astrocytes were seeded in 96-well plates in serum free medium and treated with 50 ng/μL recombinant human BMP1 (1927-ZN, R&D Systems, MN, United States), BMP2 (120-02C, Peprotech), BMP3 (120-24B, Peprotech), BMP4 (PHC9531, Thermo Fisher Scientific), BMP5 (120-39, Peprotech), BMP6 (120-06, Peprotech), BMP7 (120-03P, Peprotech), BMP10 (120-40 Peprotech), FGF2 as a negative control (100-18B, Peprotech) or in BMP-free medium (condition used as a negative control for 6 days, before fixation in 4% PFA in PBS followed by a 1 h blocking with 5% BSA in PBS. The fixed cells were immunolabeled overnight (Table 1), washed with PBS and incubated with secondary antibodies 1:500 anti-chicken Alexa Fluor 488 (Thermo Fisher Scientific Cat# A-11039, RRID:AB_2534096) and anti-rabbit Alexa Fluor 555 (Thermo Fisher Scientific Cat# A27039, RRID:AB_2536100) for 1 h at room temperature. DAPI (Thermo Fisher Scientific) diluted 1:50,000 in 1× PBS was used to stain nuclei. Duplicate wells from three separate experiments with astrocyte culture originating from two separate batches of embryonic cortices isolated on the same day were used. After imaging on Plate Runner HD (Trophos), images were analyzed by the Metamorph (Cell Scoring Application Module; Molecular Devices) analysis software. Data is presented as mean average cell intensities of staining.

Adult (4-month-old) male C57/BL6J mice (n = 4) were anesthetized with isofluorane (Baxter, Deerfield, IL, United States) and decapitated before regional dissection of frontal cortex, striatum, parietal cortex, hippocampus, midbrain and cerebellum was performed on ice and the regions flash frozen in liquid nitrogen. Regions from each hemisphere were stored at –80C and used for RNA or protein extraction, respectively.

Gene expression measurement was performed as previously described (Rao et al., 2019) with some modifications. For primary astrocyte samples, RLT lysis buffer was added directly to plate and scraped. Total RNA was extracted using the RNeasy^® Plus Mini Kit (QIAGEN) and RNA concentrations were quantified by NanoDrop (Thermo Fisher Scientific). 50 ng RNA was reverse transcribed into cDNA by Oligo (dT)15 primers using the GoScript Reverse Transcription System (Promega) in a reaction volume of 20 μL. For adult mouse brain regions, regions were homogenized in RLT buffer and total RNA extracted using the RNeasy^® Plus Mini Kit and RNA concentrations were quantified by NanoDrop. 500 ng total RNA was reverse transcribed to cDNA as described above. The cDNA was diluted in 10 mM Tris-HCl (pH 8.0). Real time PCR was performed usingSYBR^® Green PCR Master Mix (Applied Biosystems) with specific primers at 200 nM concentration. Real time assays were carried out in 96 well plates using the StepOnePlus™ Real Time PCR System (v2.3, Applied Biosystems). Thermal cycling was performed on the StepOnePlus system (Applied Biosystems) under the following conditions: 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 60 s. 1.25–5 ng input was used for RT-qPCR depending on gene expression. Copy numbers were calculated according to an absolute standard for quantification. Copy numbers were determined by calculating copy numbers per ng RNA input into the cDNA synthesis, using the formula: Copy number/ng = [A]/Mw, where [A] is the Avogadro’s number (6.02 × 10²³/mol) and Mw is the molecular weight of the amplicon (in ng/mol). Oligocalc (Kibbe, 2007) was used to calculate the molecular weight of the amplicon. Thereafter, a standard curve was prepared for each primer with 10-fold serial dilutions. NormFinder (Andersen et al., 2004) was used to identify the best housekeeping gene or combination of housekeeping genes. For primary astrocyte samples Tbp, Gapdh, Actb, Hprt1, Ubc, Ppia, and H2afz were analyzed, and the best stability value of 0.063 was achieved with a combination of Ppia and H2afz. These genes were used for double normalization of the data. For adult mouse brain regions Housekeeping genes Tbp, Ppia, H2afz, Hprt1, and Actb were evaluated by NormFinder. Ppia alone was found to have the best stability value of 0.069 and was used for normalization. Non-reverse transcriptase (NRT) and non-template (NTC) controls were included on each plate. One-way ANOVA post hoc LSD test was used for statistical analysis. Data are presented as mean values ± SEM. P < 0.05 was considered as significant. Primers used are listed in Table 2.

Experiments were performed as previously described with some modifications (Katoozi et al., 2020). Total protein extraction was performed by homogenization in RIPA buffer with freshly added SigmaFAST protease inhibitor (Sigma-Aldrich, St. Louis, MO, United States) and PhosSTOP phosphatase inhibitor (Roche Life Science, Basel, Switzerland). For primary astrocyte cultures, lysis buffer was added directly in the wells and the cells collected by scraping. For adult mouse brain regions, samples were homogenized in Eppendorf tubes. The homogenates were incubated on ice for 30 min with occasional vortexing before centrifugation at 22,000 g at 4°C for 15 min. The supernatant was collected, and protein concentrations measured using a Pierce™ BCA protein assay kit (Thermo Fisher Scientific, Waltham, MA, United States). Briefly, samples containing 10 μg protein (primary astrocytes) or 40 μg protein (brain samples) were heated in 1 × Laemmli sample buffer at 37°C for 15 min and separated on Criterion™ 18- or 26-well gels (BioRad, Hercules, CA, United States) by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) using the Criterion™ (BioRad) Tris-glycine system at 160 V for 1 h 15 min. Proteins were transferred to 0.2 μm Immun-Blot PVDF membranes by wet blotting at 100 V for 45 min at 4°C (BioRad). Uniform transfer of proteins was verified by reversible Ponceau S staining (0.1%w/v, 1% acetic acid, Sigma-Aldrich). Membranes were blocked for 1 h at RT in 3% BSA in 1 × Tris-buffered saline (TBS) (BioRad), washed in 1 × TBS with 1% Tween-20 (Sigma-Aldrich) and cut before separate overnight incubation at 4°C with primary antibodies (Table 1). Subsequently, membranes were incubated with anti-igG Horseradish Peroxidase conjugated antibodies; rabbit (GE Healthcare Cat# NA934, RRID:AB_772206) or mouse (GE Healthcare Cat# NA931, RRID:AB_772210) secondary antibodies at 1: 20,000 dilution for 1 h at room temperature, washed in TBST 3 × 10 min and immunoreactive bands detected by SuperSignal™ West Pico Chemiluminescent Substrate (Thermo Fisher Scientific) on a ChemiDoc™ Touch (BioRad) imaging system and bands quantified as arbitrary background-subtracted density units in Image Studio Lite (Ver 5.2, Licor Biosciences, Nebraska, United States). For blots with unspecific background labeling, background correction was performed by automated computation of the median intensity of pixels in an area adjacent to individual bands and subtracting it from the band intensities. Normalization was performed by dividing intensities of protein bands of interest with the normalizing control band intensity for their respective laneRabbit anti-α-tubulin (Abcam; Cat# ab4074, RRID:AB_2288001) and GAPDH (Abcam Cat# ab9484, RRID:AB_307274) were used as normalizing controls for primary astrocyte and brain regions, respectively. The obtained values were transferred to SPSS Version 25 (SPSS, Chicago, IL, United States) and compared using ANOVA post hoc LSD. Data are presented as mean values ± SEM. P < 0.05 was considered as significant.

For light microscopic immunofluorescence experiments of BMP2 and BMP4 expression, adult (3–4 months) C57/BL6J male (n = 4) and female (n = 1) mouse brains were harvested and fresh frozen in cooled isopentane on dry ice before sectioning by cryostat to 20 or 40 μm sections on glass slides. Fresh frozen sections were thawed for 5 min and fixation was carried out using 4% formaldehyde in 0.1 M PB at pH 6.0. For investigations of pSmad1/5/9 expression, animals (n = 4) were anesthetized with zoletil mixture (0.2 mL/10 g; i.p.) and transcardially perfused fixed with 2% ice cold dextran, followed by 15 min of 4% formaldehyde (FA), freshly depolymerized from paraformaldehyde (PFA), dissolved in 0.1 M phosphate buffer (PB). Perfused brains were post fixed for 24 h and treated with a 10–30% sucrose gradient before cryostat sectioning. Sections from perfused animals were thawed for 15 min before use. Sections were rinsed with phosphate buffer saline (PBS; 0.01 M) and blocked using blocking solution (10% normal donkey serum, 1% bovine serum albumin (BSA; w/v), 0.5% triton in PBS) for 60 min. The sections were incubated overnight with primary antibodies (Table 1) diluted in blocking solution with 0.01% sodium azide overnight Sections were washed with PBS and incubated with secondary antibodies at 1:500 dilution for 1 h. Secondary antibody concentrations, sections used and technical details are indicated in Table 3. Nuclear staining was performed by incubating the sections with Hoechst 33258, (1:5,000 dilution; Thermo Fisher Scientific; Cat#: H3569; RRID:AB_2651133) for 5 min, before three 5 min washes in distilled water. Sections were mounted using ProLong Gold Antifade Mountant (Thermo Fisher Scientific; Cat#: P36934; RRID:AB_015961).

Images were acquired using a LSM 510 META Confocal Microscope (Zeiss, Jena, Germany) and a LSM 710 confocal microscope (Zeiss, Jena, Germany) at 20× (air objective), 40× (oil objectives) or 63× (water objective) magnification. AxioScan Z1 (Carl Zeiss) was used to acquire scanned immunofluorescence images from midbrain. The ZEN 2.6 (blue edition) (Carl Zeiss Microscopy GmbH, 2018) software was used for image processing.

We treated mouse embryonic stem cell (mESC)-derived astrocytes with 50 ng/μL BMPs for 6 days, immunostained with antibody to AQP4 and performed a Plate Runner assay (Figure 1A). Assessment of fluorescence intensity revealed a significant increase in AQP4 staining in cells treated with BMP2 (1.36-fold, p < 0.0001), BMP4 (1.36-fold, p < 0.0001), BMP5 (1.33-fold, p < 0.0001), BMP6 (1.35-fold, p < 0.0001), BMP7 (1.34-fold, p < 0.0001), BMP10 (1.37-fold, p < 0.0001), whereas no change was found in cells treated with the non-canonical BMP1 and BMP3, compared to untreated controls (Figure 1B).

To investigate whether matured primary astrocytes in culture show similar responses to BMP treatment, primary mouse astrocytes were matured 2 weeks after the first split (P1) and treated with the same panel of BMPs. FGF2 was included as a negative control (Savchenko et al., 2019). Cells were then fixed and immunostained with antibodies to AQP4 and Glial fibrillary acidic protein (GFAP), the latter allowing to specifically identify astrocytes (Figure 1C). Immunofluorescence signal for AQP4 was increased in cells treated with the canonical BMPs (BMP 2, 4, 5, 6, 7, and 10; Figure 1D). The strongest increases were found after treatment with BMP2, BMP4 and BMP10. Notably, BMP2 (1.46-fold, p < 0.0001), BMP4 (1.54-fold, p < 0.0001) and BMP10 (1.66-fold, p < 0.0001) increased AQP4 expression by ∼50% compared to untreated control cultures. GFAP expression was also affected by these BMPs (Figure 1D), showing a 20–30% increase after BMP2 (1.21-fold, p < 0.0001), BMP4 (1.26-fold, p < 0.0001) and BMP10 (1.29-fold, p < 0.0001) treatments, compared to controls. BMP3 did not significantly affect AQP4 expression (p = 0.658), but led to a 1.05-fold increase in GFAP (p = 0.029). FGF2 did not significantly affect AQP4 (FGF2 p = 0,083) or GFAP expression (p = 0.206).

Western blots of astrocyte lysates confirmed the results of the immunofluorescence analysis, showing that BMP2, BMP4, and BMP10 were the most effective in causing an increase in AQP4 protein levels, followed by BMP6 (Figure 1E). As with the mESC cells, BMPs 1 and 3 did not alter AQP4 expression in primary astrocytes indicating a conserved mechanism in embryonic and matured primary astrocytes.

To assess whether the effects of BMPs on AQP4 and GFAP occur at the transcriptional level, we performed RT-qPCR on mRNA from primary astrocyte cultures treated with BMPs for 12 and 72 h (Figure 2) and untreated controls. Our data showed that Aqp4 was rapidly and consistently upregulated by BMP2 (12 h 1.89-fold, p < 0.001; 72 h 1.87-fold, p = 0.002), BMP4 (12 h 1.92-fold p < 0.001; 72 h, 2.79-fold, p < 0.001), BMP6 (12 h 1.98-fold p < 0.001; 72 h 2.0-fold p < 0.001) and BMP10 (12 h 1.89-fold p < 0.001; 72 h 2.96-fold p < 0.001), compared to untreated control cells. BMP5 (1.59-fold p = 0.026) and BMP7 (1.68-fold p < 0.001) induced upregulation at 72 and 12 h, respectively. The strongest inductive effect after 72 h was observed with BMP10, which causes a threefold increase in Aqp4 expression (Figure 2A). Our data further showed that at 12 h after treatment, Gfap is strongly upregulated by BMP2 (1.61-fold, p = 0.014), and by 72 h, BMP2 (1.78-fold p = 0.001), -BMP4 (1.96-fold p < 0.001), BMP6 (1.57-fold p = 0.01) and BMP10 (1.96-fold p < 0.001) upregulated Gfap transcript levels compared to untreated cells (Figure 2B).

Next, we investigated the effect of BMPs on expression of the DAP complex genes products which are required for anchoring of AQP4 to astrocytic endfoot membrane domains. We found that the expression of α-syntrophin (Snta1) was upregulated by BMP2 (1.39-fold p = 0.013), BMP4 (1.36-fold, p = 0.02), BMP6 (1.37-fold p = 0.017) and BMP7 (1.36-fold p = 0.019), compared to untreated cells at 12 h and by BMP10 (1.46-fold p = 0.007) at 72 h compared to untreated cells (Figure 2C). We also found significant upregulation of Dag1 by BMP5 (1.27-fold p = 0.002) and BMP7 (1.17-fold p = 0.037) at 12 h compared to untreated cells, with no detectable effects at 72 h (Figure 2D). The Dag1 transcript gives rise to two protein products, α- and β-dystroglycan which might be differentially regulated and cannot be separately assayed by RT-qPCR (Holt et al., 2000).

Interestingly, expression level of both α-dystrobrevin (Dtna) and dystrophin isoform DP71 (Dmd) were downregulated by BMPs (Figures 2E,F). In the case of Dtna, all BMPs tested caused a downregulation at 12 h (BMP2 0.54-fold, BMP4 0.54-fold, BMP5 0.68-fold, BMP6 0.73-fold p = 0.001, BMP7 0.86-fold p = 0.008, and BMP10 0.49-fold, p < 0.001) compared to untreated controls, while the effects of BMP4 (0.64-fold p < 0.001), BMP6 (0.81-fold p = 0.016) and BMP10 (0.59-fold p < 0.001) persist at 72 h (Figure 2E). The Dmd transcript variant DP71 was significantly downregulated by BMP5 (0.66-fold p = 0.022) and BMP10 (0.66-fold p = 0.024) at 12 h compared to controls, whereas no significant effects were detected at 72 h (Figure 2F).

These results indicate that BMPs can regulate astrocytic gene expression, and that this regulation has different temporal profiles.

As BMP2, BMP4, and BMP10 consistently showed strong induction of AQP4 and GFAP transcripts at both time points, we chose to validate these effects at the protein level after 72 h treatment (Figure 3A). We found that AQP4 protein expression was significantly increased after treatment with BMPs that signal through the Smad1/5/9 signaling pathway. Both BMP2 (1.54-fold, p = 0.003) and BMP4 (1.55-fold, p = 0.003) increased AQP4 levels compared to controls (Figure 3B). In the case of GFAP, protein levels were doubled after treatment with BMP2 (1.86-fold p = 0.041), BMP4 (2.16-fold p = 0.01) and BMP10 (2.57-fold p = 0.002) compared to controls (Figure 3C). No such effect was induced by BMP3 (Smad2/3 signaling pathway) on AQP4 (p = 0.130) or GFAP (p = 0.328), which we used as a negative control (Figures 3B,C).

We next investigated whether astrocytes expressed BMP receptors and whether the associated Smad1/5/9 pathway was activated in them. Using RT-qPCR, we found that primary astrocytes in culture robustly express Bmpr1a, Bmpr1b, Bmpr2, Acvr1, Acvr1b, and Acvr2a BMP receptors (Figure 4A). Interestingly, expression of Acvr2b was low, and that of Acvr1l was absent.

Furthermore, we found that Smad1 is present in astrocytes and that Smad1/5/9 is phosphorylated upon treatment with BMPs –2, –4, and 10 but not by BMP3 which is known to signal through the Smad2/3 pathway (Figure 4B). Quantification shows that Smad1 expression is significantly lower (Figure 4C) and Smad1/5/9 phosphorylation is significantly higher (Figure 4D) in cells treated with BMP2, –4, –10 compared to the untreated or BMP3 treated cells. This is consistent with the idea that both AQP4 and GFAP expression is regulated by the Smad1/5/9 signaling pathway in astrocytes.

In light of our in vitro data, we further examined whether the relevant BMP receptors were expressed in adult mouse brain. We found that all BMP receptors were expressed (Figure 5A). Bmpr1a and Bmpr1b showed moderate to high expression. Notably, Bmpr2 expression was high in all regions, while Acvr1, Acvr1b, Acvr2a, Acvr2b, and Acvrl1 were weakly to moderately expressed.

We then asked whether the adult mouse cells in brain expresses BMPs known to be associated with the Smad1/5/9 pathway. Bmp transcripts were readily detected in the adult mouse brain, however, regional differences in their expression could be identified (Figure 5B). The RT-qPCR analysis showed moderate to strong expression of Bmp2, Bmp4, Bmp5, Bmp6, and Bmp7 in all brain regions. Bmp10 expression was not detected in any brain region. Notably, Bmp2 and Bmp6 expression was particularly high in the striatum and the midbrain regions. Bmp4 transcripts appeared to be rather evenly distributed across regions. The Bmp5 expression level was much higher in the cerebellum than in any other region. To verify that the Bmp transcripts were translated into protein in the adult brain, we selected BMP2 and BMP4 for analysis by SDS-PAGE and Western blot. We detected the proteins in all brain regions (Figure 5C).

We next assessed the microscopic distribution of BMP2 and BMP4 by use of confocal microscopy combined with various cell type markers in cerebral cortex, hippocampus, cerebellum and midbrain regions. Our results show strong intensity staining of these BMPs in neurons. Indeed, BMP co-localized with neuronal nuclei (NeuN), throughout the brain (Supplementary Figure 1A). The cerebellar Purkinje neurons and granule cells showed also strong BMP2 and BMP4 immunofluorescence (Supplementary Figure 1B). In the midbrain, we found colocalization of BMP2 and BMP4 within both DAT-positive (dopaminergic neurons) and DAT-negative neurons (Figure 6).

By co-labeling with BMP antibodies and the endothelial cell marker lectin, we found that BMPs are also present in vessels (Figure 7). BMP4 was expressed in endothelial cells in each of the brain structures analyzed, while BMP2 was identified in neocortical vessels but lacking or very weakly expressed in vessels of the hippocampus and the midbrain.

By co-labeling with the astrocyte marker GFAP, we identified individual GFAP-positive astrocytes with BMP4 labeling in cell bodies and processes (Figure 8). However, the labeling intensity in individual astrocytes was weak compared with the labeling intensity in adjacent neurons. Moreover, the possibility remains that some of the punctate BMP2 and BMP4 labeling in the neuropil can be attributed to GFAP negative astrocyte processes.

Finally, we examined localization and activation of Smad proteins. Brain sections were immunostained with phospho-Smad1/5/9 antibody recognizing the activated Smad1/5/9 complex. Immunofluorescence signals mainly occurred in the nuclei of GFAP-positive astrocytes near the brain surface and close to large vessels (Figure 9). Astrocytes near the brain surface typically exhibit high expression of AQP4 both in their subpial endfeet and in the processes extending into the subpial neuropil (Supplementary Figure 2). Only weak or no phospho-Smad1/5/9 signal was observed in endothelial cells and the nuclei of GFAP-negative cells (Figure 8).

BMPs can activate both canonical and non-canonical signaling pathways (Moustakas and Heldin, 2005, 2009). In the canonical signaling pathway, phosphorylated Smad1/5/9 forms a complex with Smad4 and migrates to the nucleus to take part in gene regulation by binding to BMP response elements. The BMPs that are effective when it comes to increase in AQP4 expression led to phosphorylation of Smad1/5/9—suggesting that they act through the canonical Smad1/5/9 pathway. Further, we observed that BMP treatment also led to a dysregulation of members of the dystrophin associated protein (DAP) complex that are equipped with a BMP response element (von Bubnoff et al., 2005). Moreover, in the intact brain, phosphorylated Smad1/5/9 signal was observed mainly in the astrocytic nuclei associated with the pial surface or large vessels. These astrocytes exhibit high levels of AQP4 in their endfoot neuropil processes. Our data strongly suggest that BMPs act through the canonical pathway. However, we cannot rule out the involvement of additional pathways, such as the non-canonical pathway activating p38 MAPK, JNK, and ERK (Moustakas and Heldin, 2005). A study in mice has shown that hypoxia-induced activation of p38 MAPK pathway takes place only in neurons and microglia and not in astrocytes (Bu et al., 2007). However, in rat astrocytes, activation of the p38 MAPK pathway has been linked to upregulation of AQP4 under pathological conditions (Nito et al., 2012; Li et al., 2019). According to one study, rat cortical astrocytes in culture show concomitant upregulation of AQP4 and phosphorylated p38 (pp38) after 16 h of oxygen-glucose deprivation (Nito et al., 2012). In the same study, rats subjected to middle cerebral artery occlusion showed significantly reduced edema formation and infarction size when treated with SB203580, a selective p38 MAPK inhibitor (Nito et al., 2012). While the possible contribution of the p38 MAPK pathway merits further studies it should be emphasized that this pathway may converge with the canonical Smad pathway through its activation of Smad1, as shown in lung cancer cells (Su et al., 2011). Taken together with available literature data our findings suggest that BMPs may play a pivotal role in regulating AQP4 expression in brain.

It has long been known that the DAP complex is critically involved in anchoring of AQP4 to specific astrocyte membrane domains (Frigeri et al., 2001; Neely et al., 2001; Amiry-Moghaddam et al., 2004a). This protein complex interacts with basal lamina components such as laminin and agrin (Hoddevik et al., 2017). AQP4 is lost from endfoot membranes in animals that lack the dystrophin complex (Dmd) or the genes encoding the dystrophin associated protein α-syntrophin (Snta1) (Frigeri et al., 2001; Neely et al., 2001; Amiry-Moghaddam et al., 2004a,b) or α-dystrobrevin (Lien et al., 2012). As alluded to above, it has previously been shown that the DAP complex members α-syntrophin, DP71 dystrophin and dystroglycan contain BMP response elements (BREs) in their promoter regions, and these are conserved between mouse and human (von Bubnoff et al., 2005). We found that genes encoding for the DAPC members α-dystrobrevin (Dtna) and dystrophin (Dmd) were downregulated by BMPs, while Snta1 and Dag1 were upregulated. This could possibly indicate a “decoupling” of the DAP complex upon activation of the canonical BMP pathway, which would fit with the known loss of astrocyte polarity in neuroinflammation and reactive astrogliosis that is observed in neurodegenerative disorders (Smith et al., 2019). α-syntrophin is intimately coupled to AQP4, presumably through C-terminal PDZ-domains (Neely et al., 2001). The concomitant upregulation of α-syntrophin might provide an important clue to the mechanisms responsible for upregulation of AQP4. In the early postnatal brain, the expression of α-syntrophin precedes the expression of AQP4 and in the mature brain α-syntrophin is the single most important factor determining the size of the AQP4 pool in astrocytic endfeet (Lunde et al., 2015; Hoddevik et al., 2017). In our experiments, Dtna was the member of the DAP complex with the most robust downregulation by the canonical BMPs. Dtna is a key member of the DAP complex keeping other members of the complex at the perivascular astrocytic endfeet. Mice with genetic deletion of Dtna show a complete loss of perivascular AQP4 and a pronounced downregulation of α-syntrophin and dystrophin at this site (Bragg et al., 2010).

Taken together our data suggest that BMP signaling regulates AQP4 expression in astrocytes. BMPs may impact AQP4 along two separate paths: 1, an upregulation of AQP4 expression on the transcriptional level; 2 affecting AQP4 microdistribution through altering the expression levels of DAP complex proteins that under normal circumstances anchor AQP4 to endfoot membranes. Loss of AQP4 polarization combined with a preserved or upregulated AQP4 level is seen in a number of pathophysiological conditions (Eid et al., 2005; Yang et al., 2011; Reeves et al., 2020) and the present study opens for the possibility that this may well be an effect—partly or fully—of perturbed BMP activation. To our knowledge, the present study is the first to demonstrate a mechanism that could be responsible for dysregulation of the DAPC/AQP4 complex and thus explain AQP4 mislocalization in pathological conditions. Our data showed that the same BMPs that caused AQP4 to upregulate also enhanced the expression of GFAP in mature astrocytes. This is in line with previous studies showing an association between BMP4 and astrogliosis (Hart et al., 2020) and a concomitant upregulation of AQP4 and GFAP in several neurodegenerative disorders (Ikeshima-Kataoka, 2016).