Alternative mRNA Splicing from the Glial Fibrillary Acidic Protein (GFAP) Gene Generates Isoforms with Distinct Subcellular mRNA Localization Patterns in Astrocytes

The intermediate filament network of astrocytes includes Glial fibrillary acidic protein (Gfap) as a major component. Gfap mRNA is alternatively spliced resulting in generation of different protein isoforms where Gfapα is the most predominant isoform. The Gfapδ isoform is expressed in proliferating neurogenic astrocytes of the developing human brain and in the adult human and mouse brain. Here we provide a characterization of mouse Gfapδ mRNA and Gfapδ protein. RT-qPCR analysis showed that Gfapδ mRNA and Gfapα mRNA expression is coordinately increased in the post-natal period. Immunohistochemical staining of developing mouse brain samples showed that Gfapδ is expressed in the sub-ventricular zones in accordance with the described localization in the developing and adult human brain. Immunofluorescence analysis verified incorporation of Gfapδ into the Gfap intermediate filament network and overlap in Gfapδ and Gfapα subcellular localization. Subcellular mRNA localization studies identified different localization patterns of Gfapδ and Gfapα mRNA in mouse primary astrocytes. A larger fraction of Gfapα mRNA showed mRNA localization to astrocyte protrusions compared to Gfapδ mRNA. The differential mRNA localization patterns were dependent on the different 3′-exon sequences included in Gfapδ and Gfapα mRNA. The presented results show that alternative Gfap mRNA splicing results in isoform-specific mRNA localization patterns with resulting different local mRNA concentration ratios which have potential to participate in subcellular region-specific intermediate filament dynamics during brain development, maintenance and in disease.


Glial fibrillary acidic protein (Gfap) is a component of the intermediate filaments in astrocytes together with Vimentin and
Nestin [1]. The intermediate filaments (diameter 8-12 nm) have an important function for signal transduction and structural properties of astrocytes and form the cellular cytoskeleton together with microtubules (diameter 25 nm) and actin microfilaments (diameter 7 nm) [2,3]. The intermediate filament proteins have a well conserved central helical rod domain involved in filament assembly through dimerization and multimerization and head and tail domains of variable size and amino acid sequence [2]. The head and tail domains can influence assembly of filaments. Gfap and Vimentin are classified as type III intermediate filament proteins due to their capacity to assembly into both homomeric and heteromeric intermediate filaments whereas Nestin belongs to type IV which requires heteromeric intermediate filament proteins for filament assembly [3,4].
In humans at gestational week 9-12 Gfap expression starts in the radial glial cells [5,6,7]. Radial glial cells are bipolar cells in the ventricular zone (VZ) which express Nestin and Vimentin and have neuronal stem cell potential [8,9]. During the second half of gestation Gfap expression increases and also becomes evident in the arising subventricular zone (SVZ) which persists into adulthood [5,7,10,11,12,13]. Gfap expression increases during the maturation and differentiation of the precursor cells whereas Nestin and Vimentin expression decreases. Some astrocytes maintain co-expression of Vimentin and Gfap [14,15]. Gfap expression is induced by brain damage and CNS degeneration and is also induced during ageing, and altered Gfap expression is associated with a variety of neurological diseases [16,17]. Increased expression of Gfap, together with Vimentin and Nestin, and enlargement of astrocytes is indicative of reactive gliosis [18]. Gfap missense mutations in the rod and tail domains are involved in Alexander disease where astrocytes accumulate Gfap containing cytoplasmic aggregates [19]. The differences in Gfap expression can alter the morphology of astrocytes with direct consequences for a variety of astrocyte functions during development and ageing, and also have indirect consequences for other CNS cell types [17].
The Gfap gene has nine exons and spans 10 kb in the human genome. At least eight different Gfap mRNA isoforms exist which are generated as a consequence of alternative mRNA splicing and polyadenylation signal selection [16,17,20,21,22,23,24]. Corresponding Gfap proteins can be expressed in specific astrocyte subtypes and possess the capability to modify the astrocyte intermediate filament network. In humans the most abundant Gfap isoform in the CNS is Gfapa (432 amino acids) [17,21]. Alternative mRNA splicing combined with alternative polyadenylation of the human Gfap gene intron 7 generates a novel exon, E7a, which together with exons 1 to 7 encodes the Gfapd isoform (431 amino acids) (Fig. 1A) [21,22,25]. We previously abbreviated this isoform Gfape [20,21,26] but accept the nomenclature by Middeldorp and Hol and will following use the nomenclature Gfapd [17,22,25]. The mRNA expression level of Gfapd is in the order of 10-fold lower than Gfapa [21,22,23,27]. The Gfapd isoform has a novel tail domain and thereby lacks the capability to assembly into homomeric Gfapd intermediate filaments but forms heteromeric intermediate filaments with Gfapa and Vimentin [23,26,27]. An increased Gfapd level can result in a Gfap and Vimentin intermediate filament collapse [26,27]. Moreover, Gfapd has a specific capacity to form interactions with the Presenilin proteins [21]. Thus, Gfapd has the potential to function as a modulator of the Gfap filament structure and associations with other proteins, which could influence astrocyte function. In mouse the Gfapd tail domain is 41 amino acids and 71% homologous to the human Gfapd tail domain [21,28]. The rat Gfapd tail domain is 33 amino acids and 75% homologous to the human Gfapd tail domain [23,25,28]. The degree of homology between mammalian Gfapd tail domains are lower than for the other Gfap protein sequences [28]. The Gfapd isoform is only identified in mammals and the alternative exon E7a is proposed to be under a different evolutionary selection pressure than other Gfap gene exons [28].
A detailed examination of Gfapd expression has been performed in the developing human brain showing that at 13-15 weeks of gestation Gfapd and Gfapa are expressed in parallel in the VZ radial glial cells [11]. From around week 17 of gestation, the neuronal progenitors of the SVZ also express Gfapd and this expression continues until birth [11]. In human and mouse postnatal Gfapd expression is observed in a specific subpopulation of astrocytes located along the ventricles in the subventricular zone (SVZ) and in the subpial zone [11,22,23,29]. The SVZ astrocytes express more Gfapd than astrocytes in adjacent locations and the Gfap expression is highly variable [30]. The Gfapd positive cells are proposed to constitute neural stem cells of the developing and adult brain [17,23]. Gfapd expression is also determined in the rostral migratory stream and the olfactory bulb [23,30]. In addition, subpopulations of astrocytes express Gfapd and Vimentin such as fibrillary astrocytes in the molecular layer of the supratentorial cortex and astrocytes of the glia limitans [31]. Together, these studies indicate that in the developing and adult human brain a subgroup of Gfapd positive cells can represent proliferating neurogenic astrocytes. A recent report showed that all astroglia cells in the developing and adult mouse brain express Gfapd regardless of the neurogenic potential indicating that in mice Gfapd is not a neural stem cell marker as in humans [32].
We here extend the characterization of Gfapd in the mouse brain and describe that the Gfapd and Gfapa mRNA isoforms have distinct subcellular localization patterns which indicates a novel mechanism for regulation of Gfap intermediate filament dynamics.

Ethics Statement
Gfap(2/2) and Gfap(+/+) mouse tissue slides were a generous gift from Dr. S. Itohara and the animal experiments to prepare these tissue slides [33,34] have been approved and performed according to the guidelines of the Animal Care, Regulation of Infectious Agents and Experimental Committee, National Institute of Animal Health, Japan.
Killing of experimental animals (for example for collection of organs for further examinations) without prior intervention (surgery, medical treatments, etc.) is not considered an animal experiment and does not require prior approval.

RNA Purification and RT-qPCR of Brain Samples
Mouse cortices were dissected from P0, P2, P4, P8, P10, P12, P70, and P400 mice. RNA was purified by standard phenol chloroform extraction methods using 1 mg of glycogen for precipitation. cDNA was made using an iScript cDNA synthesis kit (Bio-Rad). 1 ml of total RNA solution was used per reaction. Real time quantitative PCR (RT-qPCR) was performed using SYBR Green 480 master mix (Roche). RT-qPCR analyses were performed using a Roche Lightcycler TM 480 with annealing temperature of 58uC. A common forward primer, Gfap-all, F, ACATCGAGATCGCCACCTACA, was used with the following Gfap isoform specific reverse primers: Gfapd, R, CCATTTT-CAATCTGGTGAGCCTG; Gfapa, R, CCTTCACATCAC-CACGTCCTTG; Gfapk, R, AGATGCATGCCCTAG-GATCCT [20,21,23]. The primers were designed as intron spanning and PCR primers were determined to have amplification efficiencies close to 100% and specificity and integrity of amplicons were verified by gel electrophorese and melting curve peak analysis. All Ct values (for the presented mouse P0 to P400 brain samples) used for quantitation of transcript levels were within the detection limit. RT-qPCR amplifications were made in triplicates for each gene and the Ct values were converted into linear values using the X o method [35]. Results are presented as mean +/2 standard deviation (SD) and represent three independent RT-qPCR experiments. To select for normalization genes in RT-qPCR analysis the expression levels of potential reference genes in the postnatal brain cDNA samples were assessed and used for GeNorm analysis [36]: Gapdh, F, GGTGAAGGTCGGTGT-GAACG, R, CTCGCTCCTGGAAGATGGTG; Actb, F, ACA-CAGTGCTGTCTGGTGGT, R, CTGGAAGGTGGACAGT-GAGG; beta-2-Microglobulin (b2m), F, AGACTGATACATACGCCTGCAG, R, GCAGGTTCAAAT-GAATCTTCAG; Arpc3, F, TTTCCTCTCAACGCCATTTA, R, ACCTTCTCACACAGCCTCAG. b2m was superior in fulfilling the GeNorm criteria and the presented RT-qPCR data for postnatal brain samples are normalized to the b2m mRNA expression level.

Mouse Primary Astrocyte Cultures, Boyden Chamber Analysis and Direct RNA Sequencing (DRS)
Mouse primary astrocyte cultures were prepared from the cerebral cortex of P0 NMR1 mice (Taconic, Denmark) as previously described [37]. RNA and protein from primary astrocyte protrusions were isolated using a modified Boyden chamber assay as described [37]. The RNA expression levels in the cell protrusion fraction (PF, Boyden chamber membrane underside) and in the cell body fraction (CF, Boyden chamber membrane upper side) of Actb, Gfapk, Gfapd and Gfapa mRNA were determined by RT-qPCR. The values were used to calculate the relative mRNA localization ratio between PF and CF. Arpc3 mRNA is not differentially localized in mouse primary astrocytes and the localization ratio was given the value 1 [37]. The calculated mRNA localization ratios were normalized to the Arpc3 mRNA localization ratio. Results are presented as mean +/2 SD. P-values for differences in the localization ratios were calculated by a Student's unpaired two tailed t-test. All experiments were repeated three times.
RNA isolated from PF and CF pools from three independent Boyden Chamber assays were sequenced by the direct RNA sequencing (DRS) method using the Helicos Biosciences platform (Helicos Biosciences) as previously described [38]. Expression values were processed as RNA transcripts per million reads (tpm). The RNA enrichment in cell protrusions was presented as the relative expression value in PF compared to CF. P-values for DRS mRNA localization data were calculated by Fisher's exact test.

FISH
Single RNA FISH was essentially made as described in previous protocols [39]. Briefly, probes consisting of 50-mer single stranded DNA oligonucleotides were synthesized and labeled with Cy3 fluorophore for Gfapa FISH analysis or Cy3.5 fluorophore for Gfapd FISH analysis. A total of eight various oligonucleotides were hybridized to each target mRNA. The Gfap FISH probes were synthesized by an automated DNA/RNA synthesizer Model 392/ 394 (Applied Biosystems). Sequences of FISH probes are available in Table S1. Mouse primary astrocytes were seeded onto 18 mm Ø, 0.17 mm thick coverslips (Marienfeld) and cultured as described above. At approximately 60% confluence cells were fixed in 4% paraformaldehyde for 20 min at room temperature, and washed and stored in PBS at 4uC. Before hybridization cells were permeabilized using 0.5% triton X-100 in PBS for 10 min at room temperature, washed in PBS, and then incubated in prehybridization solution, 50% formamide (Sigma; F4761) and 26SSC (Ambion), for 15 min at room temperature. The probes were hybridized in prehybridization solution supplemented with 2 mg/ml BSA (Roche), 0.2 mg/ml Escherichia coli tRNA (Roche), and 0.2 mg/ml sheared salmon sperm DNA (Sigma; D7656) for 3 h at 37uC. 10 ng DNA probe was used per coverslip. Cells were washed twice with prehybridization solution for 20 min at 37uC, then 10 min in 2 6 SSC at room temperature, and in PBS for 10 min at room temperature. Cell nuclei were counterstained with DAPI (0.5 mg/L in PBS). After a final wash in PBS, coverslips were rinsed in double distilled water to remove excess salt, dried and mounted using ProLong gold (Invitrogen).

Immunohistochemistry
Rabbit monoclonal antibodies were raised against a peptide with a sequence from the mouse Gfapd tail, GGKSTKE-samples. RNA was isolated from cortices at the indicated time points. cDNA was used in RT-qPCR experiments with primer combinations specific for Gfapd and Gfapa cDNA. Results are normalized to the b2m mRNA expression level which is determined as reference gene by GeNorm analysis. Results are presented as mean +/2 SD. P-values for the expression levels compared to the P12 expression level were calculated by a Student's unpaired two tailed t-test. The results represent three independent RT-qPCR experiments from one brain sample cohort. Another independent brain sample cohort gave similar results. (D) The post-natal expression ratio between Gfapa and Gfapd mRNA is constant. The expression ratio at different time points using the values from panels B and C was calculated and presented as mean +/2 SD. doi:10.1371/journal.pone.0072110.g001 GEGQKVTRPLKRL, on a commercial basis (Thermo Scientific). The antibodies were affinity purified before usage. By western blotting and immunohistochemical the Gfapd antibody PA2190 was selected for use in the subsequent analysis. We note the presence of a DBA and 129X1 mice SNP which changes the P residue to H in the mouse Gfapd tail sequence from which the antigen peptide was selected.
Mouse brains from ages P3, P10 and P70 were immersion fixed in formalin, and paraffin-embedded tissue blocks were produced from various brain regions. 10 mm coronal sections were mounted on coated glass slides. The sections were de-paraffinized and boiled in PBS for 25 min to obtain antigen retrieval. The slides were treated with peroxidase block (DAKO) for 5 min, and blocked with bovine serum albumin (BSA) (1 mg/ml) for 10 min. Immunohistochemical analysis were performed using the EnVi-sion+ System-HRP-DAB (DAKO). Pan-Gfap antibody (Polyclonal Rabbit Anti-Gfap, Z0334, Dako) was diluted 1:3000 incubation 30 min; Nestin antibody (Ms x Nestin, MAB353, Millipore) was diluted 1:200 incubation 30 min; Vimentin antibody (Mouse monoclonal [VI-10], ab20346, Abcam) was diluted 1:1500 incubation30 min; Gfapd antibody (PA2190) was diluted 1:400 incubation 2 h. The pan-Gfap antibody recognizes both Gfapd and Gfapa. The brain sections were counterstained with haematoxylen solution. The slides were finally cover slipped with Faramount Aqueous Mounting Medium (DAKO), and analysed by a Leica DM 2500 microscope using Leica IM50 4.0 software. Gfap(2/2) and Gfap(+/+) mouse tissue slides were a generous gift from Dr. S. Itohara and prepared as described [33,34]. Briefly, brains from 3 month mice were fixed by intracardiac perfusion with neutral buffered formalin followed by rehydration and embedding in paraffin. Coronal sections (4 mm) including the hippocampal region were used for immunohistochemical staining as described above.

siRNA and DNA Vector Transfections
For siRNA experiments 1000000 primary mouse astrocytes were immediately before the transfection plated into 10 cm dishes in DMEM with 10% FCS. In 640 ml serum free medium was mixed siRNA to a final concentration of 2 mM and incubated for 5 min. 13 ml Dharmafect was mixed with 1267 ml serum free medium and incubated 5 min. The two solutions were mixed and incubated 20 min, added to the cells, and incubated 24 h. The medium was changed to serum free medium and cells incubated for further 48 h. For Gfapd depletion we used a mix of two siRNA both targeting the Gfapd 39-UTR. siRNA sequences (sense): Gfapd (1), GGUUAUACCGAUAGAGCUA(dTdT); Gfapd(2), GAUU-CAGCCCAGAGGGUUA(dTdT); non-specific control: AG-GUAGUGUAAUCGCCUUG(dTdT) (Eurofins MWG Operon). In the mouse Gfapd NM_001131020 reference sequence Gfapd(1)-siRNA target nucleotides 1669-1687 and Gfapd(2) siRNA target nucleotides 2517-2535, both in the 39-UTR. We note that these siRNA also will target the Gfapk isoform (data not shown) since the entire Gfapd mRNA sequence is included in the Gfapk mRNA [40]. Results are presented as mean +/2 SD and represent three independent RT-qPCR experiments. Data were normalized for the gapdh expression level or actb expression level with similar results (data not shown). The siRNA transfections and examination of mRNA and corresponding protein levels were verified in a biological replicate.
The Gfap minigene in the background of pTAG4 was previously described [20]. For each transfection were used 150000 NIH3T3 cells or primary mouse astrocytes cells in 6-well plates using 200 ml serum free medium and 3 ml Xtreme Gene 9 DNA transfection reagent version 03 (Roche). Cells were incubated 24 h before a medium shift to serum free medium and a subsequent incubation for 24 h. The cells were used in a standard Boyden chamber assay with 1.0 mM membranes and RNA was purified from the protrusion fraction (PF) and cell body fraction (CF) from 3 membranes for each transfection and pooled. Primers for RT-qPCR analysis of Gfapd and Gfapa chimerical mRNA expressed from pTAG4 were previously described [20]. Results are presented as mean +/2 SD and P-values for differences in the localization ratios were calculated by a Student's unpaired two tailed t-test.

Western Blotting
A P2 mouse brain extract and protein fractions isolated from the Boyden chamber upper and lower sides were mixed with 56 Loading buffer (Fermentas) and 206 Reducing agent (Fermentas) to a final concentration of 16. The samples were heated to 95uC for 5 min and centrifuged 1 min at 16000 rpm at room temperature. Samples were loaded onto a Tris-HCl Ready Gel 4-15% (Biorad) and processed at 45 mA until the loading buffer had reached the bottom. Proteins were transferred to a hybond-P membrane (GE Healthcare) at 75V for 30 min at 4uC, and the membrane blocked in 10% skimmed milk powder (Difco) mixed with PBS and 1% Tween 20 (Sigma-Aldrich) for 4 h at room temperature. The membrane was incubated with primary antibodies diluted in PTM buffer (PBS containing 0.5% skimmed milk power and 0.1% Tween 20) ON at 4uC. The membrane was washed in PTM three times and incubated with secondary polyclonal HRP-conjugated antibodies (Dako) diluted 1:10000 in PTM-buffer 1 h at room temperature, and washed 56 in PTMbuffer. For signal detection was used BM Chemo-luminescence blotting substrate (Roche) and the signal developed (AGFA Curix 60) and monitored with X-ray film (Konica Minolta).

Immunofluorescence and Double Immunofluorescence
Cells were grown on 1861860.17 mm coverslips until 60% confluence, fixed in 4% paraformaldehyde for 20 min at room temperature, washed in PBS and stored in PBS at 4uC. Coverslips were incubated in 0.5% triton X-100 in PBS for 10 min at room temperature followed by several washes in PBS. Blocking was done using 1% BSA in PBS for 30 min at room temperature. Primary antibodies were dissolved in blocking buffer and incubated for 1 h at room temperature. Cells were washed 3 times in PBS and incubating with secondary antibody dissolved in blocking buffer. After 3 washes in PBS double immunofluorescence was performed as described above with a second treatment of primary and secondary antibodies. After final secondary antibody incubation cell were washed 26 and cell nuclei were stained with DAPI, then washed 16in PBS. Coverslips were rinsed in double distilled water to remove salt, dried, and mounted with ProLong gold. Antibodies used were: Rabbit anti-Gfapd (PA2190) diluted 1:100; mouse anti-Vimentin (Abcam; ab20346) diluted 1:100; and goat anti-Gfapa (GFAP (C-19), sc-6170, Santa Cruz) diluted 1:100 [11]. Alexa conjugated secondary antibodies (InVitrogen) were diluted 1:2000.

Microscopy and Image Processing
All images for FISH and IF analysis were made on a Zeiss axiovert 200 m microscope, with a plan apochromatic 6361.4 NA objective, a HBO 100 W mercury light source, and a CoolSNAP-HQ cooled CCD camera (Photometrics) operated by Meta-MorphH. Filters were from Chroma, Cy3 (41003), FITC (41001), and DAPI (31000), and narrow band pass filters were used for dual labeled RNA FISH Cy3 (SP-102v1) and Cy3.5 (SP-103v1). For RNA FISH we took z-stacks with 20 sections 0.2 mm step size and 500 ms exposure. For detection of RNA on a single molecule resolution, images were processed using the open source software Image J. Background noise was reduced by convolving all frames of a z-stack using a 969 gaussian kernel. Z-stacks were collapsed to a 2D maximum intensity projection, and single RNA molecule spots were detected using the ''Find Maxima'' function in Image J. For the shown immunofluorescence images a single, best in focus, 2D image was selected from a 20 sections z-stack.

Characterization of the Gfapd Isoform in Mouse
The mouse Gfapd mRNA isoform from the Gfap gene was identified due to the homology to human Gfapd [21] and the mRNA and protein expression was recently described in the mouse brain [23,32]. A schematic description of the alternative Gfap splicing resulting in Gfapd and Gfapa is shown in Fig. 1A. To examine Gfapd expression during brain development mouse brain RNA was purified from different developmental times. In brain samples from embryonic mice Gfapd expression was below the cutoff detection limit of the used RT-qPCR quantification method.
From P0 to P12 an increased Gfapd mRNA expression was detected (Fig. 1B). Expression was also present at the age of 14 months which was the last time point examined (Fig. 1B). The expression profile of Gfapd was similar to Gfapa (Fig. 1C). In accordance, the ratio between the Gfapa and Gfapd mRNA expression levels was constant at the examined time points (Fig. 1D). RT-qPCR analysis with different Gfapd and Gfapa specific primer combinations systematically indicated an expression level in mouse of Gfapa 10-fold to 25-fold higher than Gfapd (data not shown). This is in accordance with previous results [21,23,32].

Gfapd is Expressed in Mouse Brain Astrocytes with Proximity to Ventricles
Due to the relatively low conservation of the Gfapd tail domain between human, rat, and mouse, the cross reactivity of already developed human Gfapd antibodies has not been satisfactory for immunocytochemically based analysis. To monitor the expression of Gfapd protein we raised polyclonal antibodies against a peptide sequence corresponding to a mouse Gfapd tail specific sequence. Western blotting analysis of mouse brain extracts with the antibody PA2190 showed reactivity towards protein with the expected Gfapd size (Fig. 2A). Other generated antibodies showed similar recognition patterns in mouse brain protein extracts but the PA2190 Gfapd antibody was used in subsequent analysis due to highest sensitivity. Mouse primary astrocytes were transfected with siRNA targeting Gfapd mRNA, which resulted in a ten-fold reduction of the Gfapd mRNA expression (Fig. 2B). We note a significant down regulation of Gfapa mRNA level, which could be explained by the siRNA complementarity to the Gfapa intron 7 (Fig. 2B). Notably, we observed reduced Gfapd protein amounts upon siRNA treatment supporting the specificity of the Gfapd antibody PA2190 (Fig. 2C). However, pan-Gfap Western blotting showed no significant decrease in Gfap protein despite the observed reduction in also Gfapa mRNA (Fig. 2B and 2C). For further control of Gfapd antibody specificity we performed immunohistochemical staining on brain tissue slides from Gfap knock-out mice, Gfap(2/2), and corresponding Gfap(+/+) control with same genetic background ( Figure 2D). Gfapd staining that was seen in some cells close to ventricles in the wt mouse was absent in the Gfap(2/2) mouse (Fig. 2D). The observed staining at the cellular level was in accordance with staining of intermediate filament structures (Fig. 2D). Altogether these results supported that the used immunohistochemical staining protocol was specific for Gfapd.
We performed a more detailed immunohistochemical analysis on mouse brain tissue from P3, P10 and P70. Consecutive brain slides were stained for Gfapd, pan-Gfap, Vimentin, and Nestin expression (Fig. 3). The overall pattern of Gfapd staining appeared as few positive cells in the P70 and P10 brains whereas Gfapd immune reactivity was hardly detected in the P3 brain. In P70 and P10 brains most of the detected Gfapd positive cells were present in regions with proximity to ventricles or directly along ventricles in a location corresponding to the SVZ (Fig. 3). The distribution of the Gfapd positive cells along the ventricles was rather nonuniform (Fig. 3). The overall Gfapd staining pattern was relatively similar at P70 and P10 (Fig. 3). Sections from P70 and P10 were stained with a pan-Gfap antibody which primarily targets Gfapa according to the highest absolute expression of this isoform. The pan-Gfap antibody also recognizes Gfapd and we selected the use of pan-Gfap to visualize total amount of Gfap due to the very low efficiency of tested Gfapa specific antibodies in our immunohistochemical analysis. pan-Gfap showed a more widespread staining pattern than the Gfapd antibody. The regions positive for Gfapd were also positive in the pan-Gfap staining (Fig. 3). At P3 a clear pan-Gfap staining was observed (Fig. 3). Vimentin staining was observed at P70, P10 and P3. In the ventricular areas positive for Gfapd some Vimentin staining was observed in agreement with co-expression in these cells of the intermediate filament proteins (Fig. 3). Nestin staining in the mouse P70, P10, and P3 brains was more restricted than Vimentin and pan-Gfap staining (Fig. 3).
Only in a few groups of ependymal and ventricular cell areas Nestin and Gfap expression were overlapping (Fig. 3). In conclusion, the Gfapd staining pattern in the post-natal mouse brain resembles findings in the adult human brain and supports that Gfapd positive mouse brain regions also are Gfapa positive [11,22,29,30,31].

Immunofluorescence Analysis of Gfapd in Mouse Primary Astrocytes
To examine the subcellular localization of mouse Gfapd we performed immunofluorescence analysis on cultured primary murine astrocytes. Most of the cells showed a filamentous localization with a Gfapa specific antibody [11]. We used a Gfapa specific antibody instead of pan-GFAP to allow detection of eventual lack in co-localization of Gfapa and Gfapd. In the fraction of astrocytes with most Gfapa a filamentous Gfapd localization also was observed ( Fig. 4A and data not shown). Colocalization studies showed largely overlapping Gfapd and Gfapa immunostaining patterns (Fig. 4A). The immunohistochemical analyses (Fig. 3) were pointing towards the existence of some brain regions with co-expression of Vimentin and Gfapd [11]. Immunofluorescence analysis of the mouse primary astrocytes showed the existence of cells with a filamentous expression of both Gfapd and Vimentin (Fig. 4B). Co-localization analysis showed only a partial overlap between Vimentin and Gfapd containing filaments (Fig. 4B). In conclusion, immunofluorescence analysis showed also the presence of Gfapd in the most Gfapa positive cells supporting that the two Gfap mRNA isoforms are transcriptional co-expressed.

Subcellular mRNA Localization Analyses of Gfapa and Gfapd
Subcellular mRNA localization is mediated by a combination of mRNA cis-elements, as for example the Zip-code and G-rich sequences, and associated transport involved protein factors [41,42]. Cis-regulatory elements are often residing in the 39-UTR, and alternative splicing and polyadenylation consequently generate different coding regions and 39-UTRs of the Gfapa and Gfapd isoforms [21]. Previous studies indicated that a fraction of the Gfap mRNA is actively localized to cell protrusions [43]. Therefore, we examined if mRNA localization could be different between Gfapa and Gfapd. We have recently described a modified Boyden chamber based method to isolate mRNA present in astrocyte protrusions [37]. The mRNA present in protrusions from mouse primary astrocytes was isolated from the underside of the Boyden chamber membrane, whereas total cellular mRNA was represented by cell material from the membrane upper side. By RT-qPCR we examined for Gfapd and Gfapa mRNA accumulation in the protrusion fraction (PF) and compared it to the entire cell represented by the Boyden chamber upper side cell body fraction (CF). The primer combinations described in Fig. 1 were used for the RT-qPCR analysis. In addition, we analyzed the expression of Actb mRNA and Arpc3 mRNA where the latter have uniform distribution in the mouse primary astrocyte cytoplasm [37]. Relative to arpc3 mRNA, a 13-times increase in Gfapa mRNA accumulation was observed in primary astrocyte protrusions (Fig. 5A). To this end, it is very important to note that only a small fraction of the total amount of Gfapa mRNA localizes to the protrusions and the localization ratio indicates the grade of Gfapa mRNA enrichment relatively to Arpc3 mRNA. The localization ratio for Gfapd mRNA was similar to Actb and Arpc3 mRNA (Fig. 5A). This supports that the exon E7a present in Gfapd lacks mRNA localization signals present in the 39-sequence of Gfapa mRNA. The Gfapk mRNA isoform includes the intron 7a sequence compared to the Gfapd mRNA [40] (Fig. 1A) and accordingly Gfapk mRNA also lacks Gfapa exon 8 and exon 9 which could contribute to mRNA localization. By RT-qPCR we determined that Gfapk mRNA, similar to Gfapd mRNA, lacked mRNA localization in astrocyte protrusions (Fig. 5A). Thus, the different Gfap mRNA isoforms display different mRNA localization patterns in astrocyte protrusions.
We performed a direct RNA sequencing (DRS) of polyadenylated RNA isolated from primary mouse astrocyte protrusions and cell bodies. In this experiment no amplifications of material is introduced and RNA sequences are obtained corresponding to the polyadenylated region. Extracting Gfap mRNA data from the DRS experiment again showed a relative increased Gfapa mRNA  (Table 1). It should be noted that the Gfapk isoform share a common mRNA 39-terminal exon sequence with Gfapd, and therefore sequence reads from both isoforms will be uniformly aligned and counted in the DRS experiment. Notably, Gfapd is expressed to a higher level than Gfapk pointing that Gfapd will be the primary detected isoform in DRS [20,23,40].
Protein expression analyses of cell fractions from protrusions (PF) and cell bodies (CF) from Boyden chamber experiments showed the same relative distribution of Actb, Gfapd and Gfapa (Fig. 5B). Thus, the differences in Gfap mRNA localization were not equivalently reflected in major differences in protein localization patterns. This was in accordance with only a minor amount of the total Gfapa mRNA population specifically localizing to protrusions.
To determine if the different compositions of 39-exons in Gfapd and Gfapa were involved in mediating the localization of Gfapa mRNA to cell protrusions we performed cell transfection analyses with a Gfap minigene including the Gfap gene sequences from exon 6 to 1000 bp downstream the exon 9 located polyadenylation signal used for generation of Gfapa mRNA [20]. Mouse primary astrocytes were transfected with the minigene. Following a Boyden assay RNA was purified from protrusions and cell bodies and synthesized cDNA examined by RT-qPCR. The ectopic mRNA was measured by a common forward primer recognizing chimeric exon sequences from the minigene vector pTAG4 and reverse primers specifically for Gfapd or Gfapa. Calculation of mRNA localization ratios showed 3.5-fold increased Gfapa mRNA localization relatively to Gfapd (Fig. 5C). We observed a systematically inefficient transfection efficiency of mouse primary astrocytes with the Gfap minigene and accordingly also performed the localization experiment in a readably transfectable mouse cell line. For this we transfected the Gfap minigene in mouse NIH3T3 cells which previously were used as model in mRNA localization studies [41]. In this cellular background we observed 20-fold increase in Gfapa mRNA localization relatively to Gfapd (Fig. 5C). From the minigene transfection experiments we conclude that the Gfapa specific 39-exons include sequence determinants for Gfapa mRNA localization.
The RNA localization studies were extended by single molecule RNA fluorescence in situ hybridization (FISH) analysis. Primary astrocytes were analyzed simultaneous with 8 Gfapa and 8 Gfapd mRNA FISH probes labeled with Cy3 and Cy3.5, respectively. The FISH probe sets were designed to have specificity towards each Gfap isoform (Table S1). Gfapd mRNA FISH analysis resulted in a relative faint number of total signals (148 and 303 single RNA foci in the representative pictures shown in Fig. 5D and 5E, respectively) and with mRNA signals mostly restricted to the cell soma ( Fig. 5D-E). Gfapa mRNA FISH analysis resulted in a larger number of signals (808 and 1469 single RNA foci in the representative pictures shown in Fig. 5D and 5E, respectively) dispersed through the cell including the cytoplasm, rim and protrusions (Fig. 5D-E). The Increased appearance of Gfapa FISH signals compared to Gfapd supports the presence of a lower amount of Gfapd mRNA compared to Gfapa mRNA. Incubating the cells with either Gfapa or Gfapd mRNA probes individually gave similar results (data not shown). The FISH analysis further supported the localization of a fraction of Gfapa mRNA in astrocyte protrusions in alignment with the Boyden chamber analysis. The low level of Gfapd mRNA hampered significant conclusions from the FISH analyses concerning a relative lower localization ratio of this isoform compared to Gfapa mRNA in astrocyte protrusions but were indicative of relative more Gfapd mRNA in the central part of the cells.

Discussion
We here describe the characterization of the mouse Gfapd isoform at the levels of mRNA and protein. The alternative processing of Gfap mRNA which results in generation of the Gfapd protein isoform is conserved between human and mouse [20,21,23,28]. In the Gfapd transcript the exon 8 and 9 sequences used for generating the Gfapa mRNA isoform are skipped whereas a novel exon E7a generated from intron 7 is included [17,21]. Exon E7a includes a novel polyadenylation signal whereby eliminating the use of the exon 9 located polyadenylation signal used for generation of Gfapa [20,21]. Of notable difference in the Gfap gene structure between human and mouse is the presence of repetitive elements immediately after the Gfapd exon E7a stop  Mouse primary astrocytes and NIH3T3 cells were transfected with a Gfap minigene inserted in the pTAG4 vector. Following a Boyden chamber analysis purified RNA samples from protrusions and cell bodies were analyzed by RT-qPCR with primer combinations specific for Gfapa and Gfapd mRNA expressed from the pTAG4 minigene and mRNA localization ratios determined by division. Results are presented as mean +/ codon in mouse. The relative mRNA expression level of Gfapd compared to Gfapa was previously estimated to be in the order of 10% [21,23,27,32]. Our RT-qPCR analysis and direct sequence analysis are also in support of an mRNA expression ratio of this magnitude for these two Gfap isoforms. By screening postnatal mouse brain tissue for Gfapa and Gfapd mRNA expression we found a coordinated expression. In embryonic tissue, the mRNA expression level of Gfapd was undetectable with the sensitivity by our used RT-qPCR approach, whereas Gfapa expression was significantly detected, but to a relative low level (Data not shown).
Vimentin and Nestin mRNA have expression levels in the mouse brain which gradually decreases post-natal. The Gfap expression results are concordant with previous analysis from the human, pig and mouse brains in where Gfapd and Gfapa expression emerges from mid-gestation [11,17,22,23,32,40].
The low degree of homology between the Gfapd protein tail domains in mouse and human has hampered the use of antibodies developed to detect the human Gfapd isoform in mouse tissue. Recently, a mouse Gfapd antibody was described and Gfapd expression in the adult and developing mouse brain carefully characterized [23,32]. We here describe the generation of another antibody detecting mouse Gfapd. In the characterization of the antibody specificity in Gfapd siRNA transfected mouse primary astrocytes we observed that Gfapd siRNA also was targeting the Gfapa mRNA expression level despite only recognizing intron 7 in the Gfapa primary RNA. Gfapd and Gfapa expression levels were 10-fold and 3-fold down regulated, respectively. A coordinated Gfapd protein down regulation was observed, but we did not observe such an effect for the Gfapa protein level. An effect of the siRNA for Gfapa mRNA expression could be dependent on nuclear events at the level of transcription and RNA processing and thereby affecting the de novo generation of Gfapa mRNA. This scenario is similar to the described transcriptional gene silencing by intron targeting exogenous siRNA [44]. In addition, the siRNA effect on the Gfapd mRNA level could be on stability and translation of already present Gfapd mRNA at the time of siRNA transfection and thereby have stronger effect on the Gfapd protein level in the time frame of the siRNA experiment. Possible different protein stabilities of Gfapa and Gfapd could also participate to the observation. We performed a characterization of Gfapd expression at the protein level in the developing post-natal mouse brain and at the subcellular level in mouse primary astrocytes. In line with a coordinated expression of Gfapd and Gfapa the immunofluorescence studies in mouse primary astrocytes showed that all filamentous Gfapd positive cells also were filamentous Gfapa positive, and that a high Gfapa expression was prognostic for Gfapd expression. Cells with an intermediate or low Gfapa expression had only faint or no filamentous Gfapd staining. Gfapd and Vimentin immunofluorescence studies showed that a subset of the primary astrocytes expresses both proteins, but that only partial co-localization was evident.
Immunohistochemistry studies showed rather identical Gfapd staining patterns in P10 and P70 mouse brains. Immunohistochemistry analysis showed that Gfapd expression in our assays was only at the limit of detection in the P3 mouse brain whereas at P3 we observed Gfapa, Vimentin and Nestin expression. We note that Gfapd expression recently was described to increase from E18 to P5 and then decrease until plateauing at P25 [32]. One explanation of the faint Gfapd staining at P3 in the immunohistochemical analysis could be the use of formalin fixed brain samples and not cryosections as described which could interfere with intermediate filament protein detection [32]. Gfapd staining P10 and P70 was evident in proximity to the ventricles and was similar to Gfapd staining patterns in the adult human brain [11,22,29,30,31] and in the adult mouse brain [23,32]. The Gfapd positive regions were also positive for Gfapa, shown by a pan-Gfap antibody, but this pan-Gfap staining was also present throughout other brain regions in accordance with a general astrocyte staining. The lack of detectable Gfapd staining in such cells could be a consequence of sensitivity according to the relative low level of Gfapd expression. Vimentin and Nestin staining was only incompletely overlapping in the Gfapd positive regions. The lineage and function of Gfap positive astrocytes and ependymal cells of the postnatal SVZ have been carefully described [45,46]. Radial glial cells transform into Gfap positive cells and ependymal cells in the SVZ and astrocytes at other locations. The Gfap positive cells are progenitors for neuroblasts and glioblasts. The glioblasts can differentiate into astrocytes or oligodendrocytes during development and following brain injury. In many ways, ependymal cells resemble astrocytes: they express Gfap, are 2 SD and represent three independent RT-qPCR experiments from 2 independent Boyden chamber inserts. P-values were calculated by a Student's unpaired two tailed t-test. (D-E) FISH analyses showing representative examples of Gfapa and Gfapd mRNA localization in mouse primary astrocytes. The cells were probed with a mixture of 8 Gfapd mRNA Cy3.5 labeled probes and 8 Gfapa mRNA Cy3 labeled probes (left panels). FISH images were analyzed for spot detection (central panels) as described in the materials and methods section. The nuclei were counterstained with DAPI. A merged image is shown with Gfapa labeled red, Gfapd labeled green and DAPI labeled blue. Arrowheads show astrocyte protrusions and for all panels scale bars represent 20 mm. doi:10.1371/journal.pone.0072110.g005 derived from radial glial transformation, and maintain glycogen as a functional energy store. However, they are unique in that they possess cilia and do not contact neuroblasts. We found that Gfapd is highest expressed in the SVZ in the mouse post-natal brain. This is largely concordant with other observations concerning Gfapd [11,22,23,29,30,31]. Recently it was shown that all astroglia cells in the developing and adult mouse brain express Gfapd regardless of the neurogenic potential indicating that in mice Gfapd is not a neural stem cell marker as in humans [32]. Our studies, together with previous observations, suggest that Gfapd positive cells always express a high level of Gfapa and that the relative expression ratio is constant [20,21,23,27,31,32]. That Gfapd was detected only in a subset of the pan-Gfap positive cells could be a matter of detection sensitivity. Much attention has been denoted to understanding how intermediate filaments are regulated during cell growth, migration and morphology changes. Transcriptional regulations, posttranslational modifications, capability to assembly into filaments, and protein localization have been carefully addressed. Moreover, it has been more and more established from studies in various cell models that mRNA localization regulates local protein synthesis during cell growth and migration. Many localized mRNA species have been described and their localization identified in several cell types [42,47,48]. Only few studies have described localization of mRNA for intermediate filament proteins in astrocytes. Dahlstrand et al. demonstrated subcellular localization of the nestin mRNA in processes of pial end-feet of radial glial cells by in situ hybridization assay on tissue sections from embryonic E10 mouse brains [49]. Other studies have demonstrated localization of Gfap mRNA in the branch points and distal parts of astrocyte protrusion by In situ hybridization studies on cultured rat type-2 astrocytes and in Müller cells of the rat eye retina [43,50]. We here show that the different 39-exon sequences included in Gfapd and Gfapa mRNA determines different subcellular mRNA localization patterns. The Gfapa mRNA is present at a higher relative level in cell processes compared to the cell body if compared to the Gfapd mRNA. It is important to note that only a very minor fraction of the total amount of Gfapa mRNA is localized to protrusions. However, the ratio between Gfapa and Gfapd mRNA will be different in astrocyte protrusions compared to cell bodies by this different mRNA localization capability. Whereas the general expression ratio between Gfapa and Gfapd mRNA is constant during development and in different astroglia cell types, variations in ratios between the different Gfap mRNA isoforms can still exist at the subcellular level. The differences in mRNA localization are determined by mRNA sequences present in the distal part of the transcripts and in accordance with the general pattern for RNA localization cis-elements to be located in the 39-UTR [42], yet, the exact mechanisms involved in the localization of Gfapa mRNA need further examination. In general, Gfap expression is gradually up-regulated as Nestin and Vimentin expression decreases during brain development. Vimentin and Nestin containing intermediate filaments have potential to act as scaffolds for the establishment of long term Gfap intermediate filaments [1,49]. It is straightforward to make a simple model wherein Gfap filaments within astrocytes initially can be locally synthesized at specific subcellular regions as a consequence of mechanisms which at least to some extend includes specific mRNA localization, and during astrocyte maturation, Gfap filaments are expanded and become more uniformly distributed throughout the cytoplasm. In such a scenario it can be important to decrease the relative concentration of the Gfapd isoform at specific intracellular localizations due to specific functional properties of Gfapd in terms of filament forming capacity and interaction with other proteins [21,23,26]. This could generate Gfap intermediate filaments with distinct surface structures and accordingly functional capacities at specific subcellular regions within astrocytes. How Gfap gene alternative splicing and associated differences in mRNA localization is linked to functional consequences for the intermediate filament dynamics will require further studies.

Supporting Information
Table S1 Sequences of FISH probes for Gfapd and Gfapa mRNA. (DOC)