Regulation of aquaporin‐4 expression in the central nervous system investigated using M23‐AQP4 null mouse

Abstract In astrocytes, unknown mechanisms regulate the expression of M1 and M23 isoforms of water channel aquaporin‐4 (M1‐AQP4 and M23‐AQP4). The ratio between these two isoforms controls the AQP4 assembly state in the plasma membrane known as orthogonal arrays of particles (OAPs). To give new insights into these mechanisms, here, we explore the regulation of AQP4 expression in the spinal cord of a CRISPR/Cas9 M23‐null mouse model (M23‐null). In the M23‐null spinal cord OAP assembly, the perivascular localization of AQP4 and M1‐AQP4 protein were drastically reduced. In heterozygous, M1‐AQP4 was proportionally reduced with M23‐AQP4, maintaining the isoform ratio unaffected. We hypothesize a role of the M23‐AQP4 in the regulation of M1‐AQP4 expression. M1‐AQP4 transcription, splicing and M1‐AQP4 protein degradation were found to be unaffected in M23‐null spinal cord and in M23‐null astrocyte primary culture. The translational control was investigated by mRNA‐protein pull down and quantitative mass spectrometry, to isolate and quantify AQP4 mRNA binding proteins (AQP4‐RBPs). Compared to WT, in M23‐null spinal cord, the interaction between AQP4 mRNA and polypyrimidine tract binding protein 1, a positive regulator of AQP4 translation, was higher, while interaction with the RNA helicase DDX17 was lower. In astrocyte primary cultures, DDX17 knockdown upregulated AQP4 protein expression and increased cell swelling, leaving AQP4 mRNA levels unchanged. Here, we identify AQP4‐RBPs and provide evidence that in mouse spinal cord M23‐AQP4 deletion changes the interaction between AQP4 mRNA and some RBPs involved in AQP4 translation. We describe for the first time the RNA helicase DDX17 as a regulator of AQP4 expression in astrocytes.

Other isoforms, named M1-AQP4ex and M23-AQP4ex, have recently been characterized as being able to modulate OAPs assembly and the astrocytic endfeet localization of AQP4 Palazzo et al., 2019;Palazzo et al., 2020). A large number of environmental, metabolic and hormonal stimuli have been reported to be able to change AQP4 expression, thereby regulating its role in physiological and pathological processes. In most of these studies, the authors have described the phenomena but have not provided any direct evidence concerning the molecular dynamics underlying AQP4 downregulation or upregulation (Costa et al., 2019;Li et al., 2019;Lichter-Konecki et al., 2008;Murillo-Carretero et al., 1999;Szpilbarg et al., 2018;Wang et al., 2018). Consequently, AQP4 regulation remains largely an unaddressed issue. This represents one of the main obstacles to developing a new strategy to fine-tune AQP4-mediated processes in which pathological AQP4 dysfunction plays an active and well documented role such as ischemia (Vella et al., 2015), brain edema (Clement et al., 2020), or neuromyelitis optica Pisani, Mastrototaro, et al., 2011). Some molecular dynamics related to the transcriptional regulation of AQP4 and to the microRNA-dependent AQP4 mRNA stability have been clarified (Abe et al., 2012;Kapoor et al., 2013;Sepramaniam et al., 2012;Yi et al., 2013;Zheng et al., 2017), as well as some translational regulation mechanisms such as leaky scanning (Rossi et al., 2010), reinitiation , read-through , and IRES-dependent translation (Baird et al., 2007). Most of these mechanisms were explored in vitro in transfected systems without providing direct evidence in AQP4-expressing tissues. Among these mechanisms, only the role of the AQP4 translational readthrough in the expression of AQP4ex has recently been confirmed in human, rat, and in mouse brain by in vivo study (Palazzo et al., 2019;Palazzo et al., 2020;Sapkota et al., 2019). Whether translational regulation of AQP4 expression plays an active role in the control of AQP4 expression levels in the central nervous system (CNS) and what molecular players are involved, remains still largely unknown.
We have recently developed a CRISPR/Cas9 mouse model in which the codon of the methionine in Position 23 (M23) was replaced with the isoleucin (I) codon (Rossi et al., 2010), eliminating the synthesis of OAP-forming M23-AQP4 isoforms (M23-null). We have reported that in M23-null mice OAPs are absent, the tetrameric form of AQP4 is upregulated and the M1-AQP4 protein expression is strongly downregulated. This demonstrates that M23-AQP4 is pivotal to preserve the normal AQP4 expression level in the CNS (de Bellis et al., 2020). In the present work, we have analyzed the same animal model with the aim of exploring the mechanism by which M23-AQP4 deletion controls AQP4 expression. The spinal cords of WT, M23-null, and heterozygous mice were used in this study. AQP4 protein localization, expression, and supramolecular assembly were analyzed by confocal microscopy and biochemical procedures. The AQP4 expression was investigated at the transcriptional, splicing, translational and posttranslational levels.
The main findings come from the analysis of the translational regulation. Here, we isolate AQP4 mRNA binding proteins (RBPs) and demonstrate that M23-AQP4 deletion changes the interaction between RBPs and AQP4 mRNA. Between these RBPs, we validate DDX17 in astrocyte primary culture as a new regulator of AQP4 expression.
The role of RBPs in the regulation of AQP4 expression has never been described before. This opens up a new perspective to clarify AQP4 regulation in physiological and pathological conditions.

| Ethics statement
In this study, no experiments were performed on live animals. Experiments were performed according to the European directive on animal use for research and the Italian law on animal care. The protocols were approved by the Italian Ministry of Health (Protocol No. 710/2017-PR and571/2018-PR). All experiments were designed to minimize the number of animals used and animal suffering. Mice were maintained under a 12 h dark to light cycle, at constant room temperature (RT) and humidity, with food and water provided ad libitum, and supplied with environmental. enrichment materials such as toys and shelters.

| Animals
The animal model used here was already published (de Bellis et al., 2020). OAP-null mice harboring the M23I point mutation were generated on the C57BL/6J background by Cyagen Biosciences (Santa Clara, CA) using CRISPR/Cas9-based targeting and homologydirected repair. AQP4 knockout (KO) mice with a CD1 genetic background and age-matched CD1 mice, used as WT mice, were kindly provided by Dr Hu (Nanjing Medical University, China). Genotyping was performed on tail DNA using standard protocols.

| Immunofluorescence and confocal microscopy
Spinal cord was isolated, fixed for 4 h in 4% PFA solution, washed in PBS, cryoprotected in 30% sucrose in PBS overnight, frozen at À80 C and sectioned in 10 μm thickness slices with a cryostat (CM 1900; Leica) at À20 C. After blocking in 3% BSA and 0.3% TRITON X-100 in PBS solution, sections were incubated with primary antibodies over- 2.6 | Astrocytes primary cell culture, RNA interference, and cycloheximide chase assay Mouse astrocyte primary cultures were prepared from newborn pups as previously described (Nicchia et al., 2000). Cells were cultured in DMEM medium supplemented with 10% fetal bovine serum (FBS), 100 U ml À1 penicillin and 100 mg ml À1 streptomycin, and maintained at 37 C in a 5% CO 2 incubator. All reagents were purchase from Euroclone. RNA interference experiments were performed as follows: P1 astrocytes were transfected with 30 pmol of DDX17 and PTB siRNA or scrambled siRNA by Lipofectamine 3000 (ThermoFisher) in a 12 multiwell format, according to the instruction manual in high glucose DMEM-Glutamax and analyzed after 5 days. Three independent astrocyte preparations were used. The short interfering RNAs used were: DDX17 silence selected predesigned siRNA (ThermoFisher, ID s2344296), PTB siRNA (sense sequence: UGCACCUCUCCAACAUCCCGCUU; antisense sequence: GCGGGAUGUUGGAGAGGUGCAUU) and a scrambled (target sequence: NNUGGAGAAGGCCAACUAGGG; sense: UGGAGAAG GCCAACUAGGGUU; antisense: CCCUAGUUGGCCUUCUCCAUU) (ThermoFisher). WT and M23-null astrocyte were treated with 30 μM cycloheximide (CHX; Sigma-Aldrich, Milan, Italy), harvested after 0, 4, and 8 h and analyzed by Western blotting.

| Cell line and transfection
The HEK293 cell line, derived from human embryonic kidney (ATCC CRL-1573), was grown in DMEM medium supplemented with 10% heat-inactivated FBS, 100 UI ml À1 penicillin, and 100 mg ml À1 streptomycin, and maintained at 37 C in a 5% CO 2 incubator.
Twenty-four hours before transfection, the cells at 70% confluence were plated using antibiotic-free medium. Transient transfection was carried out using Lipofectamine 2000 (Invitrogen, Milan, Italy, www. thermofisher.com) in OptiMEM growth medium according to the manufacturer's protocol and analyzed after 36 h. For protein stability assay, 36 h posttransfection, HEK293 cells were treated with 30 μM CHX (Sigma-Aldrich) or dimethyl sulfoxide then harvested for protein extraction and evaluation of AQP4 expression levels after CHX treatment.

| SDS-PAGE
Isolated mice spinal cords were immediately frozen in liquid nitrogen, pulverized, and weighed. For the extraction of different animal samples, the same weight of pulverized sample was dissolved in 10 volumes of Laemmli sample buffer (Bio-Rad), 50 mM dithiothreitol and Protease Inhibitor Cocktail (Roche), heated to 37 C for 10 min, resolved in a 13% polyacrylamide gel, and transferred onto polyvinylidene difluoride (PVDF) membranes (Immobilon PVDF; Millipore) for immunoblot analysis. Astrocyte protein extraction was carried out using TRIzol reagent (Invitrogen) according to the instruction manual, dissolved in 1%SDS and centrifuged at 10,000g for 10 min. The supernatants were collected, and the total protein content was measured using absorbance at 280 nm. This method was chosen to analyze the mRNA and protein of the same sample.
2.9 | Tissues sample preparation for twodimensional BN-SDS/PAGE Isolated mice spinal cords were dissolved in seven volumes of BN buffer (1% Triton X-100, 12 mM NaCl, 500 mM 6-aminohexanoic acid, 20 mM Bis-Tris, pH 7.0, 2 mM EDTA, 10% glycerol) plus Protease Inhibitor Cocktail (Roche). The tissues were lysed on ice for 1 h, and centrifuged at 21,000g for 30 min at 4 C. The supernatants were collected, and the total protein content was calculated using the BCA Protein Assay Kit (ThermoFisher).

| First dimension: Blue native-PAGE
Here, 50 μg of protein sample were mixed with 5% CBB G-250 (Coomassie blue G-250) and loaded onto a polyacrylamide native gradient gel (3-9%) (20153404). The running buffers were as follows: anode buffer (25 mM imidazole, pH 7) and blue cathode buffer (50 mM tricine; 7.5 mM imidazole; 0.02% Coomassie blue G-250; pH 7). Electrophoresis was performed at 6 mA and stopped when the tracking line of the CCB G-250 dye had left the edge of the gel. Proteins were blotted onto PVDF membranes (Millipore) for immunoblot analysis or alternatively the lanes were used for the second dimension.

| Second dimension: SDS-PAGE
For the 2D BN/SDS-PAGE analysis, lanes from the first dimension were cut into individual strips and equilibrated in denaturing buffer (1% SDS and 1% b-mercaptoethanol) for 1 h at RT and placed in a 2D SDS-PAGE of the same thickness. Separation of the second dimension was performed in a 13% SDS-polyacrylamide gel at 25 mA per gel. At the end of the run, the gel was blotted onto a PVDF membrane (Millipore) for Western blot analysis.

| Western blotting and densitometric analysis
After transfer, the membranes containing the blotted proteins were blocked and incubated with primary antibodies diluted as described in Section 2.3. After washing, the membranes were incubated with peroxidase-conjugated secondary antibodies and washed again. Reactive proteins were revealed with an enhanced chemiluminescent detection system (Clarity western ECL substrate, Bio-Rad) and visualized on a Chemi-Doc imaging system (Bio-Rad). Images were recorded and data analyzed with Image lab software (Bio-Rad). Actin or GAPDH were used as an internal control for protein loading. For the analysis of M1-AQP4 expression in spinal cords, many animals, and membranes were analyzed. To pool together data obtained from different animals and filters, the analysis of each filter was carried out comparing the M1-AQP4/actin ratio of heterozygous and M23-null mice with the M1-AQP4/actin ratio of the WT mice of the same filter.
Data are expressed as a percentage of the M1-AQP4/actin of the WT mice.
2.13 | RNA extraction, RT-PCR, and isoformspecific qPCR Tissues and cells RNA extraction was carried out using Trizol Reagent  Table 1. AQP4 total expression and AQP4-M1 was measured by the GAPDH normalized ΔΔCt quantification method. All reactions were run in triplicate. After statistical analysis, the data from the different experiments were plotted and averaged in the same bar graph. Each PCR was evaluated by melting-curve analysis.
2.14 | RNA-protein pull-down to isolate mouse M1-AQP4 RBP from mouse spinal cord Mouse spinal cord cDNA was obtained by First strand cDNA synthesis, using the SuperScript IV kit (ThermoFisher) with random esamers and amplified by M1-AQP4 mRNA-specific AQP4 primers to obtain a T7 promoter linked PCR.
Primers were designed to span 5 0 UTR, CDS and part of 3 0 UTR of mouse M1-AQP4 mRNA. T7 mM1-AQP4-For: TAATACGACTCACTA washing steps, proteins were eluted from beads by biotin elution buffer provided in the same kit. Eluates were analyzed by SDS-PAGE followed by silver staining (Mortz et al., 2001), by differential quantitative gel-free mass spectrometry (MS) or by Western blotting.
In particular, eluates obtained from four mice for each genotype were first analyzed by SDS-PAGE and silver staining to globally analyze the differences between the electrophoresis profile of eluates obtained using AQP4 mRNA-coated beads versus nude beads and versus control RNA-coated beads.
After the first analysis performed by SDS-PAGE, two eluates for each genotype obtained by AQP4 mRNA-coated beads and control RNA-coated beads were pooled and analyzed by differentially quantitative MS analysis (see next paragraph).
After the MS analysis, eluates obtained from another independent RNA-protein pull down experiment, in which other two mice for each genotype were used, were analyzed by Western blotting.
To further validate the specificity of the binding to AQP4 mRNA, another independent RNA-protein pull down experiment, followed by Western blotting analysis, were performed using six other mice and using the capped GAPDH mRNA as a negative control.
The number of animals is indicated in the figure legends for each type of analysis. All animals were age-matched.

| Quantitative MS of RNA-protein pull down eluates
The MS analysis was performed by Ceinge Biotecnologie Avanzate Briefly, eluates were first mixed with SDS and DTT, boiled, cooled to RT, and then alkylated with iodoacetamide in the dark for 30 min.
Subsequently, phosphoric acid was added to the samples with a final concentration of 1.2% and then the sample was diluted with six volumes of binding buffer (ammonium bicarbonate and methanol). After gentle mixing, the protein solution was loaded onto an S-Trap filter, spun at 2000 rpm, and the flow-through collected and reloaded onto the filter. This step was repeated three times, and then the filter was washed three times with binding buffer. Finally, the digestion buffer containing trypsin at 1:10 wt:wt was added onto the filter and digestion was carried out for 1 h. The peptide solution was pooled, lyophilized, and resuspended in 0.2% formic acid. The hypotonic challenge was applied 20 s after the beginning of data acquisition by adding an appropriate volume of NaCl-free DPBS to achieve a 60 mOsm/L osmotic gradient. The fluorescence signal increases following the hypotonic stimulus due to cell swelling and then decreases during the regulatory volume decrease (RVD) which tends to restore the isosmotic condition. Each well was read continuously over a 100 s period to record both the swelling phase and RVD phase.
Data acquisition was performed using SoftMax Pro software, and the data were analyzed with Prism software (GraphPad Software, La Jolla, CA). Graphs were obtained by fitting the data with an exponential function. The percentage of volume recovery was calculated from the maximum intensity of fluorescence reached after the osmotic shock (the amplitude of cell volume variation) and the level of fluorescence reached after the regulatory mechanism.

| Statistical analysis
Statistical analyzes were conducted using GraphPad Prism 6 software.
All data are reported as the mean ± SEM. We used the Student's t test for unpaired data and the one-way ANOVA with Tukey's multiple comparisons test to compare more than two groups. Differences were considered significant for p < .05.

| RESULTS
3.1 | M23-AQP4 deletion affects OAPs assembly, perivascular localization, and M1-AQP4 expression in mouse spinal cord AQP4 supramolecular assembly state in WT and M23-nullspinal cord was initially analyzed by BN/PAGE (Figure 1(a), left). In WT, OAPs of different size can be detected (arrowheads in Figure 1(a), left), while in M23-null, only the tetrameric form of AQP4 is visible and OAPs are completely absent. In spinal cord of M23-null mice, the tetrameric form is upregulated compared to WT (Figure 1(a), right), in agreement with the results we have recently reported in brain (de Bellis et al., 2020).
To investigate whether the M23-AQP4 absence affects AQP4 localization in spinal cord, AQP4 and GFAP co-immunofluorescence was performed followed by confocal microscopy analysis using anti-C-terminal AQP4 antibody as primary antibody. The single scan confocal microscopy shows a strong AQP4 signal in gray and white matter of WT (Figure 1(b)). In M23-null gray matter, the perivascular AQP4 was not detected. Interestingly, in M23-null, the residual AQP4 appears to be confined to the fibrous astrocytes of the white matter ( Figure 1(b), arrowheads in the magnified inset). No staining was detected in spinal cord sections of AQP4 KO mice.
To analyze the possible correlation between expression levels of M23-AQP4 and those of M1-AQP4, we performed SDS-PAGE and Western blot analysis in WT, heterozygous, and M23-null mice. We used two different antibodies, one directed against the AQP4 C-terminal, therefore able to recognize both isoforms (Figure 1(c)), and the other one specific for the M1-AQP4 isoform (Figure 1 ratio remains unchanged compared to WT (Figure 1(c), right), despite the M1-AQP4 isoform being significantly reduced (Figure 1(d)). This suggests that M1-AQP4 and M23-AQP4 isoforms are probably regulated in an interdependent way.
3.2 | M1-AQP4 mRNA levels, AQP4 mRNA alternative splicing, and M1-AQP4 protein degradation are unaffected in M23-null spinal cord and the absence of M23-AQP4 does not destabilize M1-AQP4 protein in astrocyte primary culture We have already shown that total AQP4 mRNA level is unchanged between WT and M23-null brain (de Bellis et al., 2020). Those data were obtained using qRT-PCR primers specific for the exon-junction II-III and for Exon IV. Considering that exons II-III and IV are identical in many different mRNAs able to express M23-and M1-proteins (NCBI Gene ID: 11829), the data obtained using the primers described above were not a direct measure of M1-mRNA. Here, to test the hypothesis of transcriptional control as a cause of M1 protein downregulation, WT and M23-null mice spinal cords were analyzed by M1-AQP4-specific qPCR. To this purpose an M1-and an AQP4-tot-specific qPCR were designed (Table 1) based on the reference sequence reported in the NCBI RefSeq Database (Figure 2(a)). As indicated in Figure 2 Tetramer is strongly upregulated in M23-null mice. ***p < .001 M23-null versus WT. n = 4 for each genotype. Student's t test for unpaired data. (b) AQP4 (green) and GFAP (red) localization in mouse spinal cord sections analyzed by single scan confocal microscopy. In WT, AQP4 is strongly expressed at both protoplasmic astrocytes of gray matter (GM) and fibrous astrocytes of white matter (WM). In M23-null mice, the AQP4 signal is strongly reduced in both sites and the perivascular staining is completely absent. M23-null mice only show low AQP4 staining in fibrous astrocytes of white matter (arrowheads). n = 6 for each genotype. No staining was observed in AQP4 KO mice spinal cord sections. (c) Western blotting analysis of M1-AQP4 and M23-AQP4 expression using C-terminal specific antibody. Densitometric analysis of M1-AQP4 expression (middle), and M23/M1 ratio (right). Data were analyzed by one-way ANOVA with Tukey's multiple comparisons test for M1/actin (%) and Student's t test for unpaired data for M23/M1 ratio analysis. ***p < .001 versus WT; ****p < .0001 versus WT; n.s. not statistically significant. Number of animals: WT (n = 6), heterozygous (n = 3), M23-null (n = 6). (d) Western blotting analysis of M1-AQP4 expression using M1-AQP4-specific antibody. Densitometric analysis of M1-AQP4 expression using M1-AQP4-specific antibody (right). Data are reported as a percentage of M1-AQP4 expression considering the M1-AQP4 expression in WT as 100% and analyzed by one-way ANOVA with Tukey's multiple comparisons test. ***p < .001 versus WT; ****p < .0001 versus WT; n.s. not statistically significant. Number of animals: WT (n = 6), heterozygous (n = 3), M23-null (n = 6). Samples from the same animals were analyzed both with C-terminal specific and with M1-AQP4-specific antibodies [Color figure can be viewed at wileyonlinelibrary.com] M1M23I coding plasmid and five parts of M23 expressing plasmid (or an empty vector control plasmid). The results obtained indicate that M1M23I protein expression was unchanged by the presence of M23-AQP4 in transfected cells (Figure 2(g)). To test the effect of the M23-AQP4 absence in the M1-AQP4 posttranslational stability in naturally AQP4 expressing cells, CHX chase assay was performed using WT and M23-null astrocyte primary culture. The data also show that the absence of M23-AQP4 does not affect M1-AQP4 posttranslational stability in primary cell culture that naturally express AQP4 ( Figure 2(h)).
F I G U R E 2 Legend on next page.
3.3 | RNA-protein pull down and differential quantitative MS analysis of M1 RBPs in spinal cord of WT and M23-null mice Starting from the evidence that AQP4 transcription, splicing and protein degradation were unaffected in M23-null mice and in M23-null astrocyte primary culture, we investigated the translational control.
Considering that RBPs actively control mRNA translation in vivo, we identified and quantified RBPs able to bind M1-AQP4 mRNA in WT and M23-null mice spinal samples. RNA-protein pull down assay followed by quantitative differential (M23-null vs. WT) LC MS/MS MS analysis (differential quantitative MS [dQMS]) were used for this scope (Figure 3(a)).
Eluates from M1-AQP4 mRNA-coated, control RNA-coated, and nude-beads were first analyzed by SDS-PAGE followed by silver staining. Specific bands were identified for M1-AQP4 mRNA-coated beads (Figure 3 We found that M1-AQP4 mRNA interacts with microtubule and tubulin-associated proteins (Figure 3(c) red boxes) which however remain mainly unchanged between WT and M23-null mice.
Interestingly, compared to WT, PTBP1, previously demonstrated as a positive regulator of M23-AQP4 translation (Baird et al., 2007), was the most abundant M1-AQP4 mRNA interactor in eluates from M23-null (M23-null/WT ratio 2.75, highlighted in red in Figure 3 (c)), while CMTR1 and DEAD-box RNA helicases 17 (DDX17) were the least abundant in eluates from M23-null (M23-null/WT ratio 0.22 for DDX17 and 0.15 for CMTR1, highlighted in red in Figure 3 (c)). Western blotting analysis performed by anti-PTBP1 antibody of RNA-protein pull-down eluates obtained using other WT and M23-null mice support data obtained by dQMS ( Supplementary Figure 3(a)).
To further test the binding specificity for M1-AQP4 mRNA, RNAprotein pulldown was performed using other mice spinal cord samples and full length capped GAPDH mRNA as negative control. Western blotting analysis of eluates confirms the binding specificity for M1-AQP4 mRNA ( Supplementary Figure 3(b)).
To analyze the expression levels of PTBP1 and DDX17 in WT and M23-null mice spinal cord, RNA-protein pull-down inputs were analyzed by Western blotting. The data show that expression levels of both proteins were unchanged in M23-null mice respect those measured in WT mice (Figure 3(d)). mRNA-specific AQP4 primers to obtain a T7 promoter linked PCR. Primers were designed to span 5 0 UTR, CDS and part of 3 0 UTR of M1-AQP4 mRNA. The PCR product was transcribed in a capped ( m7 G 5 0 PPP 5 0 G, similar to the Cap0 structure) mRNA. The capped M1-AQP4 mRNA was biotinylated to the 3 0 end and captured on streptavidin magnetic beads. Spinal cord extracts from M23-null and WT mice were incubated and eluates identified by LC MS/MS analysis. Negative control RNA coated beads and nude beads were used as control conditions. Eluates from control RNA coated beads were also identified and quantified by MS to identify non-specific binders, eliminated from the subsequently analysis. (b) Representative SDS-PAGE and silver staining analysis of spinal cord input and eluates from AQP4-RNA coated, control-RNA coated and nude beds. Arrowheads indicate bands specific for AQP4-RNA coated beads. n = 4 for each genotype. (c) Quantitative differential MS analysis of AQP4-mRNA eluates in M23-null versus WT. Only proteins specific for AQP4-RNA coated beads are shown in the graph. These proteins were identified as M1-AQP4-mRNA binders (M1-AQP4-RBPs). The quantitative differential analysis of M1-AQP4-RBPs abundance in M23-null and WT mice is shown as the M23-null/WT ratio. The unchanged proteins are in the range of 1 ± 0.3. The group of M1-AQP4-RBPs found to be more abundant in the M23-null eluates is indicated as upregulated while the group of M1-AQP4-RBPs found to be less abundant in the M23-null eluates is indicated as downregulated. PTBP1 was found to be 2.75-fold more abundant in KI (M23-null/WT = 2.75, red highlighted) while DDX17 and CMTR1 were found to be five to six times less abundant in M23-null (M23-null/WT = 0.22 for DDX17 and 0.15 for CMTR1, highlighted in red). n = 2 for each genotype, data were obtained using pooled samples. (d) Western blotting analysis of PTBP1 and DDX17 in WT and M23-null mice spinal cord showed no changes in expression levels of both proteins. n = 7 for each genotype [Color figure can be viewed at wileyonlinelibrary.com] siRNA-treated astrocytes. Western blotting and densitometric analysis (Figure 4(c,d)) show that DDX17 knockdown strongly upregulates AQP4 expression, while GFAP, PTBP1, and GAPDH were unaffected by DDX17 knockdown (Supplementary Figure 4). Interestingly, AQP4-qPCR analysis of cDNA prepared from the same samples shows that DDX17 knockdown does not change either AQP4-tot mRNAs or M1-AQP4 mRNA levels (Figure 4(e)). Furthermore, the amplification efficiency of AQP4-tot-and AQP4-M1-AQP4 mRNAspecific primers are identical (Supplementary Figure 1) and no differences emerge from AQP4-tot Ct and AQP4-M1-AQP4 mRNA Ct in F I G U R E 4 Legend on next page. astrocytes (Figure 4(f)). This indicates that astrocyte primary culture only expresses M1-AQP4 mRNA.

|
The role of PTBP1 as a positive M23-AQP4 translational regulator had previously been described in the literature by in silico prediction and luciferase assay in transfected cells (Baird et al., 2007). We found that the knockdown of PTBP1 in astrocyte primary cultures reduced both M23-AQP4 and M1-AQP4 expression (Figure 4(f,g)). Due to a basal very high level of AQP4 expression and water transport rate in WT astrocytes (Mola et al., 2016), swelling analysis in response to hypotonic shock of DDX17 knockdown astrocytes resulted in an increased magnitude of cell swelling and cell volume recovery during the RVD phase (Figure 4(i-l)).
Finally, CMTR1, the first transcribed nucleotide ribose 2 0 -O methyltransferase (Inesta-Vaquera & Cowling, 2017), which is the RBP that exhibits the lowest binding to AQP4-M1 mRNA in M23-null compared to WT (Figure 4(c)), was not here tested by RNAi because it is a key enzyme involved in the 5 0 CAP-mRNA synthesis and, as already shown (Inesta-Vaquera & Cowling, 2017), its knockdown could affect the 5 0 CAP-dependent protein translation of all genes.

| In silico analysis for M1-AQP4 mRNA regulatory elements
We have analyzed the mouse M1-AQP4 mRNA sequence searching for secondary structures, RBPs binding sites, IRES, G-RNA quadruplexes, and alternative ORFs. To this aim, the reference sequence of mouse M1-AQP4 mRNA was analyzed using free on-line tools: ORF finder, mFOLD, RBPsite, IRESPred, and QGRS. Figure 5(a) reports the most stable secondary structure of mouse M1-AQP4 mRNA as predicted by mFOLD (ΔG = À672.40 kcal/mol). The 5 0 UTR and M1-AQP4 and M23-AQP4 translation initiation signals are predicted in very stable stem-loop structures (red box and magnified region in Figure 5(a)). This region was deeply analyzed by IRESPred to search for potential IRES. A potential IRES is predicted in Positions 132-280. As indicated in Figure 5(b), this IRES is localized downstream of the M1-AQP4 AUG and included the M23-AQP4 AUG. This indicates the potential contribution of this IRES structure to a capindependent M23-AQP4 translation from M1-AQP4 mRNA. Figure 5 (b) reports data obtained in detail. In Positions 6 and 101, two different ORFs are found, an AUG-start out-of-frame uORF and an inframe non-AUG start uORF, respectively (first two black arrows from left to right in Figure 5(b)). Four different PTBP1 binding sites are identified in Positions 26, 36, 109, and 232 (yellow boxes in Figure 5 (b)). HNRNPK and HRNPL1 binding sites are predicted in Positions 39-47 (green box in Figure 5 Recently, the RNA sequence CACACCU was identified as a binding site for DDX17 (Ngo et al., 2019). The mouse M1-AQP4 mRNA contains exactly this sequence in two positions, 535 and 1,235 (blue boxes in Figure 5(b)).

| DISCUSSION
Physiological and pathological stimuli actively change gene expression through finely regulated mechanisms. The Mouse aqp4 gene expresses many AQP4 isoforms that control the AQP4 supramolecular assembly state and the AQP4 localization into microdomains of the astrocyte plasma membrane. How AQP4 expression is modulated in astrocytes remains still largely unknown.
F I G U R E 4 Functional validation of DDX17 and PTBP1 in the aquaporin-4 (AQP4) regulation by RNAi in mouse astrocyte primary cultures. (a) Confocal microscopy analysis of DDX17 (red) expression in control siRNA and DDX17 siRNA-treated astrocyte primary cultures. The immunofluorescence shows the main localization of DDX17 into the nucleus (N) and a lower dot-like cytoplasmic staining (C), enlarged in the red boxed area. (b) Wide-field immunofluorescence analysis of AQP4 (green) and DDX17 (red) shows a strong reduction in DDX17 staining in DDX17 RNAi treated astrocytes and the upregulation of AQP4 compared with control siRNA-treated astrocytes. (c) Representative Western blotting analysis of astrocytes treated with control and DDX17 siRNA. DDX17 is strongly downregulated while AQP4 expression is upregulated in DDX17 siRNA conditions. n = 9 from three independent astrocytes preparations. (d) Densitometric analysis of Western blotting represented in Panel (c). DDX17 knockdown strongly upregulates AQP4 expression. **p = .005, ***p = .0001, DDX17 siRNA versus CTRL siRNA; n = 9 from three independent astrocytes preparations. Student's t test for unpaired data. (e) Relative quantification of AQP4-tot and AQP4-M1 mRNA between control-and DDX17-siRNA-treated astrocytes obtained by RT-qPCR from the same samples reported in Panels (c) and (d). No differences were found for either target. n = 9 from three independent astrocyte preparations. Student's t test for unpaired data. (f) ΔCt (Ct AQP4-tot) -(Ct AQP4-M1-AQP4) analysis using cDNA from mouse astrocyte samples (n = 18 from three independent astrocyte preparations). (g,h) Functional validation of PTBP1 in the AQP4 regulation by RNAi in mouse astrocyte primary cultures. Astrocyte primary cultures treated with control and PTBP1 siRNA analyzed by Western blotting (g) and densitometric analysis (h). PTB knockdown strongly reduces AQP4 expression, ***p < .0001, PTBP1 siRNA versus CTRL siRNA; n = 10 from three independent astrocyte preparations. Student's t test for unpaired data. (i-l) Calcein-quenching assay for measurement of hypotonicity-induced volume changes in cultured WT astrocytes treated with scramble siRNA (CTRL) and DDX17  previously reported about the role of DDX17 in astrocytes and its role in AQP4 expression has not been described to date. DDX17 is a member of the large family of DEAD-box RNA helicase proteins necessary for proper function of RNA in many cellular processes. DDX17 contributes to the regulation of gene expression at many levels, mRNA splicing, export, stability, localization, and translation (Cordin et al., 2006;Giraud et al., 2018;Gustafson & Wessel, 2010;Linder & Jankowsky, 2011;Xing et al., 2019). The role of DDX17 in the regulation of AQP4 expression and AQP4-dependent water transport is validated here by RNAi in astrocyte primary culture. The data show that astrocytes express DDX17, that they express only M1-AQP4 mRNA, that in agreement with our previous findings M23-AQP4 protein is translated from M1-AQP4 mRNA   Little is known about the role of DDX17 in the CNS. Only few papers report data about DDX17 in brain (Kircher et al., 2002;Luo et al., 2020;Moon et al., 2018) and no data are available about spinal cord. In one of these papers Ip et al. (2000) report that DEAD Box Protein p72 (the alternative name of DDX17) is downregulated in brain and muscle (two AQP4 expressing tissues) during development.
Whether the DDX17 downregulation during brain development (Ip et al., 2000) contributes to the age-dependent regulation of AQP4 expression in polarized astrocytes and blood brain barrier development (Gautam et al., 2020;Trillo-Contreras et al., 2018) (Baird et al., 2007).We found that PTBP1 also controls the expression of M1-AQP4 in astrocytes, despite being characterized as a translational regulator of M23-protein translation (Baird et al., 2007). We speculate that astrocytes respond to M23-AQP4 regulation by regulating M1-AQP4 expression to maintain the physiological M1/M23 ratio.
Considering that in the spinal cord the M23-AQP4 is five-fold more abundant than M1-AQP4 (Figure 1(c)) and that the elimination of M23-AQP4 strongly also downregulates M1-AQP4, the total AQP4 levels are extremely low in M23-null mice. Starting from this evidence and from evidence obtained by RBP analysis, we hypothesize that in M23-null spinal cord PTBP1 and DDX17 play a role in a mechanism aimed to compensate for the missing AQP4. This could explain why, compared to WT, in M23-null mice PTBP1 is the most highly RBP and DDX17 is the lowest. It is possible to speculate that in M23-null mice the almost total loss of AQP4 protein expression activates a positive translational feedback aimed at compensating for the absence of AQP4. In this mechanism the PTBP1 and DDX17 are regulated in the opposite manner to WT. In M23-null mice the absence of the M23 AUG start codon completely and irreparably prevents M23-AQP4 protein synthesis. This could contribute to maintaining the positive translational feedback mechanism continuously activated in M23-null mice. The mechanistic scenario here proposed cannot be tested in AQ4 KO mice, because AQP4 KO mice express a small amount of truncated AQP4 mRNA (Ma et al. 1996). This highlights the unique opportunity provided by the M23-null model here analyzed to investigate how AQP4 expression affects the interaction between AQP4 mRNA and RBP.
It is generally accepted that RNA molecules can fold into intricate shapes that can provide an additional layer of modulator that shapes posttranscriptional control of gene expression beyond that of their sequence. Among these, alternative ORFs), IRES, and G-RNA quadruplexes play a key role in the translational regulation in a coordinated manner with RBPs (Leppek et al., 2018). Searching for these elements, here we show that mouse AQP4-M1 mRNA contains RNA binding sites for DDX17 and PTBP1, supporting wet data obtained by RNA-protein pull down. Furthermore, we show that the 5 0 region of AQP4-M1mRNA contains out-of-frame upstream ORF, in-frame upstream ORF, G-RNA quadruplexes, and one IRES. Despite the fact that all these predictions must be validated by wet-experiments to confirm a role in the regulation of AQP4 expression, we suggest that the alteration of the binding with DDX17 and PTBP1 found here could play a synergic role with these structures to regulate AQP4 expression. A similar scenario has amply already been shown for other genes (Arora et al., 2008;Hansel-Hertsch et al., 2017;Kumari et al., 2008;Leppek et al., 2018). The fact that spinal cord also expresses M23-AQP4 mRNAs poses another level of control in which DDX17 and PTBP1 could play a role in the regulation of AQP4 expression. Whether and how DDX17 and PTBP1 regulate the M23 protein expression by modulating M23-AQP4 mRNA structures must be further characterized.
In conclusion, the main novelty of this study is the characterization of M1-AQP4 MBPs and the identification of the role of DDX17 in the regulation of AQP4 expression. This lays the foundation to understand more deeply how AQP4 is regulated in physiological and pathological conditions.