The EIIIA domain from astrocyte‐derived fibronectin mediates proliferation of oligodendrocyte progenitor cells following CNS demyelination

Central nervous system remyelination by oligodendrocyte progenitor cells (OPCs) ultimately fails in the majority of multiple sclerosis (MS) lesions. Remyelination benefits from transient expression of factors that promote migration and proliferation of OPCs, which may include fibronectin (Fn). Fn is present in demyelinated lesions in two major forms; plasma Fn (pFn), deposited following blood‐brain barrier disruption, and cellular Fn, synthesized by resident glial cells and containing alternatively spliced domains EIIIA and EIIIB. Here, we investigated the distinctive roles that astrocyte‐derived Fn (aFn) and pFn play in remyelination. We used an inducible Cre‐lox recombination strategy to selectively remove pFn, aFn or both from mice, and examined the impact on remyelination of toxin‐induced demyelinated lesions of spinal cord white matter. This approach revealed that astrocytes are a major source of Fn in demyelinated lesions. Furthermore, following aFn conditional knockout, the number of OPCs recruited to the demyelinated lesion decreased significantly, whereas OPC numbers were unaltered following pFn conditional knockout. However, remyelination completed normally following conditional knockout of aFn and pFn. Both the EIIIA and EIIIB domains of aFn were expressed following demyelination, and in vitro assays demonstrated that the EIIIA domain of aFn mediates proliferation of OPCs, but not migration. Therefore, although the EIIIA domain from aFn mediates OPC proliferation, aFn is not essential for successful remyelination. Since previous findings indicated that astrocyte‐derived Fn aggregates in chronic MS lesions inhibit remyelination, aFn removal may benefit therapeutic strategies to promote remyelination in MS. GLIA 2015;63:242–256


Introduction
M ultiple sclerosis (MS) is a central nervous system (CNS) disease, of which inflammation, demyelination, and axonal loss are the main pathological features. Regeneration of myelin (remyelination) involves proliferation and migration of activated oligodendrocyte progenitor cells (OPCs) to demyelinated lesions, and their subsequent differentiation into myelinating oligodendrocytes Zawadzka et al., 2010). The extent of remyelination in MS is variable, but often is insufficient to prevent chronic axonal loss (Franklin et al., 2012;Patrikios et al., 2006). Promoting endogenous remyelination provides a means to reduce axonal degeneration and thereby slow progression of clinical disability (Duncan et al., 2009;Franklin et al., 2012).
Fibronectin (Fn) is an extracellular matrix (ECM) protein, that is essential for embryonic development (George et al., 1993). In healthy adults, Fn is continuously produced by hepatocytes and circulates in the plasma. Fn is absent from the healthy CNS, but expressed following injury, which includes demyelination. Plasma Fn (pFn) leaks across a disrupted blood-brain barrier (Satoh et al., 2009;Sobel and Mitchell, 1989;van Horssen et al., 2005), and cellular Fn (aFn) is primarily synthesized by astrocytes (Hibbits et al., 2012;Stoffels et al., 2013a), but also by microglia/macrophages and endothelial cells (Stoffels et al., 2013a). In contrast to pFn, cellular Fn may contain alternatively spliced domains, named EIIIA, EIIIB, and the V-region in rodents (EDA, EDB, and IIICS, respectively, in humans;Paul et al., 1986;Schwarzbauer et al., 1987). In vitro, pFn stimulates migration and proliferation of OPCs at low growth factor levels via av integrin receptors, the only integrin receptors for Fn present on OPCs Blaschuk et al., 2000;Milner and ffrench-Constant, 1994;Milner et al., 1996). The expression of both Fn and av integrins is transiently increased in demyelinated lesions (Stoffels et al., 2013a;Zhao et al., 2009). Therefore, Fn may contribute to remyelination by promoting OPC recruitment. However, at later stages of oligodendrocyte maturation, removal of Fn is required for remyelination to proceed to completion, since Fn inhibits myelin sheet formation (Buttery and ffrench-Constant, 1999;Maier et al., 2005;Siskova et al., 2006Siskova et al., , 2009). In addition, Fn aggregates impair OPC differentiation in vivo, and likely contribute to remyelination failure in MS (Stoffels et al., 2013a).
Here, we explored the role of Fn in more depth by asking what distinctive roles the two Fn variants, pFn and astrocyte-derived Fn (aFn), play in remyelination. Using an inducible Cre-lox recombination strategy to selectively remove pFn, aFn or both (Hirrlinger et al., 2006;Sakai et al., 2001) in combination with a well-established model of CNS deand remyelination (Blakemore and Franklin, 2008), we found that conditional knockout of aFn, but not pFn, reduced the density of OPCs. In vitro analysis revealed that this was likely mediated by the EIIIA domain, which mediates OPC proliferation on aFn at sufficient growth factor levels. However, although conditional knockout of aFn was associated with reduced OPC numbers following demyelination, it was not sufficient to affect the remyelination outcome. The translational implication of our data therefore is that elimination of aFn may be amenable in MS to prevent the formation of remyelination-inhibiting Fn aggregates. This will likely be beneficial in promoting endogenous remyelination (Stoffels et al., 2013a).

Mice
Mice were housed under standard conditions. All experiments were performed in compliance with United Kingdom Home Office regulations. Plasma Fn (pFn) inducible, conditional knockout mice (hereafter referred to as pFn cKO ) were a kind gift from Dr. R. F€ assler, Max Planck Institute for Biochemistry, Martinsried, Germany. Inducible, conditional knockout (hereafter referred to as "conditional knockout") of pFn was created as described (Sakai et al., 2001). Briefly, floxed Fn mice were crossed with mice expressing Cre recombinase under the control of the polyinosinicpolycitidic acid (poly I:C) responsive Mx promoter (Mx-Cre). On Cre-mediated recombination at the loxP sites, the start codon, signal sequence and the exon/intron border of exon 1 are removed from the Fn gene to generate the null allele (Sakai et al., 2001). Cremediated recombination was induced in hepatocytes from the 6week old mice carrying Mx-Cre by two intraperitoneal injections of poly I:C (GE Healthcare, Amersham, UK) with a 48 h interval as previously described (Sakai et al., 2001). Wild type (WT) control mice received vehicle only (phosphate-buffered saline; PBS). Mice were subjected to lysolecithin-induced demyelination at 2-3 weeks following induction of the knockout.
Conditional knockout mice devoid of aFn (astrocyte Fn; aFn cKO ) were created by crossing Fn floxed mice with mice expressing Cre recombinase driven by human glial fibrilary acid protein (GFAP), with its nucleus translocation controlled by a modified estrogen receptor (hGFAP-CreERT2; Hirrlinger et al., 2006). The hGFAP-CreERT2 mice were a kind gift from Dr. F. Kirchhoff, Max Planck Institute of Experimental Medicine, Goettingen, Germany. To induce conditional knockout of Fn from astrocytes, tamoxifen (100 mg/kg in corn oil, Sigma-Aldrich, Gillingham, UK) was injected intraperitoneally daily for 5 consecutive days, starting from 10 days prior to demyelination (Hirrlinger et al., 2006;Leone et al., 2003). The littermate WT control group was injected with corn oil.
Compound astrocyte and pFn conditional knockout (a 1 pFn cKO ) was achieved by breeding mice heterozygous for MxCre and hGFAP-CreERT2, and homozygous for the floxed Fn gene. The induction protocol for these mice was the combination of that described for single conditional knockout strains above.

Lysolecithin-Induced Demyelination of the Spinal Cord and Tissue Processing
Surgery and tissue processing were performed as described . Briefly, mice at about 9-10 weeks old were anaesthetized with isoflurane, and spinal cord lesions were created by direct injection of 1 mL 1% lysolecithin into the ventral funiculus through a gap between two thoraco-lumbar vertebrae. In the conditional knockout mice, lesions were induced 1-2 weeks after completing the induction protocol. Blood samples were obtained for isolating plasma from the tail at the time of lesion, and collected in citratedextrose solution (Sigma-Aldrich, Dorset, UK), then stored at 280 C until use.
At the designated time post lesion, mice were euthanized by intraperitoneal injection of pentobarbital followed by appropriate protocols of tissue processing. For immunohistochemistry and in situ hybridization, mice were perfusion fixed with 4% phosphatebuffered paraformaldehyde (PFA) solution via the left ventricle, after which the dissected spinal cord containing the lesions was either directly frozen at 280 C for later RNA extractions, or treated with 20% sucrose in PBS overnight. Cords were embedded in OCT (Thermo Fisher Raymond Lamb, Loughborough, UK), and cut in coronal sections at 12 mm thickness. These sections were mounted on poly-L-lysine (PLL)-coated slides (Polysciences Europe GmbH Eppelheim, Germany) and stored at 280 C until further use. For resin embedding and semi-thin sectioning, mice were perfused with 4% phosphate-buffered gluteraldehyde and subjected to a standard resin embedding process . Semi-thin sections of 1 mm were cut and stained with alkaline toluidine blue. Ranking analysis on semi-thin sections of remyelinated lesions was performed by two independent, blinded researchers as described (Ibanez et al., 2004).

Glial Cell Cultures
Primary glial cell cultures were generated from 1 to 2 day old Wistar rats (Harlan, the Netherlands) as previously described Bsibsi et al., 2012). Briefly, after 10-12 days in culture on PLL (5 mg/mL, Sigma-Aldrich, St. Louis, MO) coated flasks, OPCs and astrocytes were isolated via a shake-off procedure (McCarthy and de Vellis, 1980). Contaminating microglia were removed by shaking the flasks at 150 rpm for 1 h at 37 C on an orbital shaker. Subsequently, flasks were shaken at 240 rpm overnight at 37 C. Floating OPCs were further purified by differential adhesion. Then, OPCs were cultured in defined Sato medium (Maier et al., 2005) on appropriate cell culture plastics for the different assays (see below). Purity of the OPC cultures was routinely assessed by immunocytochemistry for Olig2 (1:1000; Millipore, Billerica, MA) and cultures used were >97% pure.
To obtain astrocytes, a subsequent overnight shake off was performed at 240 rpm and the remaining astrocyte monolayer was removed by trypsin. Astrocytes were cultured in T162 flasks (Corning Costar, Lowell, MA) in DMEM (Life Technologies, Paisley, UK) containing 10% fetal calf serum (FCS; Bodinco, Alkmaar, The Netherlands) and antibiotic supplements (Life Technologies), and passaged once before experimental use. Regular immunocytochemistry for glial fibrillary acidic protein (GFAP; Millipore; 1:500) was performed to assure sufficient purity of the astrocyte cultures (>97%).

Plasma and Astrocyte-Derived Fibronectin Coatings
To obtain aFn, astrocytes were cultured on 10 cm dishes (Corning) for 48 h, followed by water-lysis at 37 C. The remaining astroglial matrix was scraped in sterile PBS containing Complete Mini protease inhibitor cocktail (Roche Diagnostics, Mannheim, Germany), followed by protein concentration determination with Bio-Rad DC protein assay (Bio-Rad Laboratories, Hercules, CA) using bovine serum albumin (BSA) as standard. Then, 8-well chamber-slides were incubated with 8 mg pFn from serum (Sigma-Aldrich) or 8 mg astroglial matrix protein ("aFn") per well in PBS for 3 h at 37 C. Similarly, both sides of membranes on transwell microchambers (Becton-Dickinson Labware, Bedford, MA) were coated with 10 mg of pFn or aFn per well, and 3.5 mg pFn or aFn was used per well of a 96wells plate. When indicated, Fn coatings were incubated with functional blocking antibodies (2 mg/mL) for 1 h at 37 C, followed by a gentle wash with PBS and seeding of OPCs. Blocking antibodies used were: mouse anti-Fn IST9 (directed against the EIIIA domain, Abcam, Cambridge, UK) and mouse anti-Fn C6 (directed against the EIIIB domain, Abcam). Notably, the aFn preparation also con-tains other astroglial matrix proteins, such as laminin (Liesi et al., 1983(Liesi et al., , 1984 and chondroitin sulphate proteoglycans (Lau et al., 2012), but although these proteins may affect OPC adhesion, proliferation and migration also (Hu et al., 2009;Lau et al., 2012;Milner et al., 1996), EIIIA and EIIIB are exclusively present in aFn.

SDS-PAGE and Western Blotting
Equal amounts of plasma from pFn cKO and a 1 pFn cKO mice and their controls were mixed with standard SDS sample buffer, heated for 10 min at 95 C, and separated by electrophoresis on SDS-PAGE gels (8%, Expedeon, Cambridge, UK) for 1 h at 150 V. Protein was transferred to a PVDF membrane (Invitrogen, Paisly, UK) using a wet blotting system (Expedeon, Cambridge, UK) and according glycine-Tris-methanol buffer. After three washes with 0.1% Tween-20 in PBS, membranes were blocked with 5% non-fat milk solution and incubated with a rabbit anti-Fn antibody (Millipore, Watford, UK; 1:10,000) in blocking buffer, overnight at 4 C. Membranes were washed, and incubated with HRP-conjugated anti-rabbit antibody (Roche, Lewes, UK; 1:1000) in washing buffer. Signals were detected using enhanced chemiluminescence plus (ECL plus; GE Healthcare, Amersham, UK) followed by exposure on suitable X-ray film (Thermo Scientific, Rockford, IL). For detection of mouse IgG as a control, membranes were washed and incubated with biotinylated donkey antimouse IgG antibody (Jackson ImmunoResearch Laboratories, Newmarket, UK; 1:1000) for 1 h, after which the Vectastain ABC Elite kit (Vector Laboraties, Peterborough, UK) was used according to the manufacturer's instructions, and signals were developed as described above.
Images from immunohistochemistry with either protocol were acquired using a Zeiss Observer A1 fluorescent microscope, and images from immunochemistry on coatings were acquired using a Zeiss Axioskop 2 microscope with Leica Application Suite V3 software. For counting cell numbers, in each animal 3 demyelinated lesion levels were selected, spanning the centre of lesions, at 120 mm distance from each other. The outline of each lesion was defined based on the increase in DAPI cellularity inside the lesion using Zeiss Axovision 4.8 software, which corresponds to the demyelinated area as stained for with Sudan Black (Stoffels et al., 2013a). The numbers of markerpositive cells inside the lesions were manually counted three times and averaged for each lesion. There were 4-5 mice per group. To allow for quantitative comparison, sections were stained in parallel and image acquisitions were performed on the same day using identical camera settings. Individual Fn-and Iba1-immunereactive cells could not be discerned reliably, hence these were quantified by measuring the optical densities from immunofluorescence using FIJI software.

In Situ Hybridization
The DIG-labeled PLP probe was obtained as described (Chari et al., 2006), using mouse PLP cDNA (a kind gift from I.R. Griffiths, Glasgow, UK) that was cloned into the pGEM4 plasmid (Promega, Southampton, UK). To obtain the Fn probe, a cDNA fragment was generated by reverse transcription PCR, using total RNA isolated from WT mouse liver and the following Fn primers: forward: 5 0 -CACCACCCAGAACTACGATG-3 0 , reverse: 5 0 -GGACACCATGCA CAAACTTC-3 0 . The cDNA was inserted into the pGEM4 plasmid (Promega, Southampton, UK). For both the PLP and Fn probes, antisense digoxygenin labeled cRNA probes (riboprobes) were then synthesized with appropriate RNA polymerase using a digoxigenin (DIG) RNA labeling kit (Roche, Lewes, UK). The size of the riboprobes was checked by agarose gel electrophoresis. In situ hybridization was then performed on air dried spinal cord sections as described previously (Chari et al., 2006).

Real-Time PCR
Total RNA was isolated from tissue homogenates using the RNeasy Mini kit (Qiagen). Reverse transcription of 0.5 mg total RNA was performed in the presence of oligo(dT)12-18 and dNTPs (Invitrogen) with SuperScript_II reverse transcriptase (Invitrogen) according to the manufacturer's instructions. Real-time qPCR was performed using the Applied Biosystems 7900HT Real-Time PCR System. Each reaction contained 5 ng cDNA, 10 pM primers (listed in Table  1) and ABsolute SYBR Green Rox mix (Thermo Scientific, Landsmeer, NL). No-template controls were performed to ensure that amplification was not a result of contamination with genomic DNA. Gene expression levels were analyzed using the 2 2DDct method (Livak, 2001), and relative expression to HPRT1 or GAPDH. Similar results were obtained for both housekeeping genes.

BrdU Incorporation Assay
OPCs were plated on 8-well Permanox chamber-slides (Nunc, Naperville, IL), precoated with 5 mg/mL PLL followed by appropriate aFn or pFn coatings (described above), at a density of 30,000 cells per well. OPCs were allowed to incorporate 5-bromo-2deoxyuridine (BrdU; 10 mM; Roche) for 24 h in the presence of 10 ng/mL PDGF-AA and 10 ng/mL FGF-2. Then, cells were fixed in 4% PFA for 20 min, and additionally fixed in 5% acetic acid in ethanol for 20 min. BrdU was detected using reagents from the BrdU labeling and Detection Kit I (Roche) according to the manufacturer's instructions with the addition of the oligodendrocyte lineage marker Olig2 (Abcam) and Alexa FluorV C 546-conjugated antirabbit antibody, and visualization of nuclei with DAPI (1 mg/mL). To compare the percentages of BrdU-incorporating cells between the conditions, the numbers of double BrdU-and Olig2-positive cells were counted relative to the Olig2-positive cells (at least 150 cells per condition) from images captured with a Leica TCS SP8 Confocal Laser Scanning Microscope.

Transwell Migration Assay
OPCs were seeded on pFn or aFn coated-polyethylene terephthalate membranes of 8 mm pore size (Becton-Dickinson Labware) in 12well modified Boyden transwell microchambers at a density of 80,000 cells per insert. OPCs were allowed to migrate through the membranes for 4 h using 10 ng/mL PDGF-AA as a chemoattractant in the bottom of the well. Nonmigrating cells were removed from the top compartment with a cotton swab. Remaining cells in the

Statistical Analyses
Statistical analyses were performed in GraphPad Prism (GraphPad, La Jolle, CA). First, the Kolmogorov-Smirnov test was applied to test for a normal distribution of the data. Multiple group comparisons of data, which could thus be assumed to have a normal distribution (data sets of Fn cKO mice, derived from immunohistochemistry and in situ hybridization experiments, except for the data describing immunofluorescence intensity of Fn-and Iba1-staining), were performed using one-way ANOVA followed by Tukey's Multiple Comparison Test (* P < 0.05; ** P < 0.01; *** P < 0.001). Multiple group comparisons of data that were not compatible with a normal distribution (data sets derived from in vitro assays and real-time qPCR on demyelinated tissue, as well as of the data describing immunofluorescence intensity of Fn and Iba1 stainings) were performed using the Kruskal-Wallis test followed by Dunn's Multiple Comparison Test (*P < 0.05; ** P < 0.01; *** P < 0.001). Nonparametric ranking data, derived from blindly ranking alkaline toluidine blue stained resin sections for estimated degrees of remyelination, were statistically analyzed using the Mann-Whitney test (Ibanez et al., 2004). For the in vivo mice data from immunohistochemistry, in situ hybridization and real time qPCR studies, a representative graph of 2-3 independent experiments is shown, displaying absolute means of 3-4 mice per group and 3 lesions levels per animal, each 120 mm from each other, with the exception of immunochemistry for Iba1, data of which are shown as a relative to the mean immunofluorescence levels of littermate WT controls, set at 100% for each independent experiment. Quantifica-tion methods are described under "immunochemistry." For in vitro proliferation, migration, and adhesion assays, graphs display mean values relative to a control condition (aFn or pFn) from 3 to 5 independent experiments. All error bars represent standard deviations.

Results
Fibronectin in Demyelinated Lesions is Predominantly Produced by Astrocytes Transient Fn expression during demyelination is hypothesized to benefit remyelination (Hibbits et al., 2012;Stoffels et al., 2013a;Zhao et al., 2009). The aim of the present study was therefore to better define the potential involvement and underlying mechanism(s) of the different Fn sources in remyelination. We used a well-established rodent model of experimental demyelination, in which focal, primary demyelination is created by the injection of lysolecithin into the white matter of the spinal cord ventral funiculus (Blakemore and Franklin, 2008). Spontaneous and complete remyelination of these lesions proceeds over a period of 21-28 days in young adult rodents, which involves proliferation and migration of local, activated OPCs to the demyelinated lesions ["recruitment"; 1-10 days post lesion (DPL)], followed by differentiation of oligodendrocytes and myelin sheath formation (10-21 DPL) . To eliminate pFn and cellular Fn from lysolecithin-demyelinated lesions, we used Cre/lox technology, creating conditional knockout adult mice devoid of plasma Fn (pFn cKO ), astrocyte Fn (aFn cKO ) or both (a 1 pFn cKO ). The pFn null allele was generated by activation of the Mx-Cre promoter with poly I:C as described (Sakai et al., 2001), which removes the start codon, signal sequence and the exon/intron border of exon 1 from the pFn gene. To remove cellular Fn, we targeted astrocytes, because they are considered a major source of Fn in the CNS following injury (Hibbits et al., 2012), and synthesize pathological Fn aggregates in MS (Stoffels et al., 2013a). We thus used mice expressing Cre recombinase under control of a GFAP promoter, which was rendered active by exposure to tamoxifen via a modified estrogen receptor. The induction efficiency was tested in a reporter line, as described in a previous report (Hirrlinger et al., 2006). In our own experiments, this approach showed that at 5 days after a course of tamoxifen administration in unlesioned adult mice, of astrocytes in spinal cord white matter on average 78.5% 6 2.4% SD (n 5 4) expressed the reporter (yellow fluorescent protein), indicating an efficient recombination.
Using this approach, pFn was successfully eliminated from plasma in pFn cKO mice, as revealed by Western blot analysis (Fig. 1A). Similar results were obtained on plasma from a 1 pFn cKO mice, whereas pFn remained present in the plasma of aFn cKO mice as expected (Fig. 1B). Given the virtual absence of aFn from the healthy CNS and its increased 246 Volume 63, No. 2 expression after demyelination (Stoffels et al., 2013a;Zhao et al., 2009), we assessed local Fn expression in aFn cKO and a 1 pFn cKO mice after lysolecithin-induced demyelination of the spinal cord ventral funiculus white matter (Fig. 1C). First, we confirmed that astrocytes are a source of cellular Fn in mouse demyelinated lesions, using immunhistochemistry against GFAP and in situ hybridization for Fn mRNA on WT lesions (Fig. 1D). Further, as shown in Fig. 1E,F, Fn expression was increased in demyelinated lesions of littermate WT mice at 5 DPL, as previously reported for rat models (Stoffels et al., 2013a;Zhao et al., 2009). Moreover, this increase was transient, and Fn levels were decreased in mice at 14 DPL (cf Fig. 5, data not shown), similar to the transient expression pat-tern of Fn observed in rats (Stoffels et al., 2013a;Zhao et al., 2009). In pFn cKO mice, Fn immunostaining was unaltered compared with expression in littermate WT mice (Fig. 1E,F), despite the strong reduction in Fn plasma levels (Fig. 1A).
These data indicate that either pFn is not a major source of Fn expression within lysolecithin-induced lesions, or that aFn expression may compensate for the loss of pFn. In contrast, Fn expression was markedly reduced in demyelinated lesions of aFn cKO and a 1 pFn cKO mice (Fig. 1E,F), corroborating that astrocytes are a major source of cellular Fn after demyelination.
Having reduced Fn levels from plasma and astrocytes, we next analyzed how remyelination was affected at both the recruitment and differentiation stages of remyelination. , astrocytes (aFn cKO ) or astrocytes, and plasma (a 1 pFn cKO ). Note the virtual absence of Fn from plasma after pFn cKO and a 1 pFn cKO . C: Demyelination was induced by injection of lysolecithin into the white matter of the spinal cord ventral funiculus (red area). D: Immunohistochemistry for DAPI (cell nuclei) and GFAP (astrocytes) combined with in situ hybridization for Fn mRNA on lysolecithin-demyelinated lesions of WT mice at 5 DPL, showing that Fn mRNA is present in astrocytes. E: Immunohistochemistry for Fn on lysolecithin-demyelinated lesions at 5 DPL after pFn cKO , aFn cKO , a 1 pFn cKO , or in WT mice. F: Quantification of immunofluorescent staining intensities in (E) from 3 mice per group, and 3 lesion levels per animal,~120 lm distant from each other from images that were taken on the same day using the exact same camera settings. Note the reduction in intensity of immunofluorescence relative to WT after aFn cKO and a 1 pFn cKO , but not following pFn cKO . Bars represent means. Error bars show standard deviations. Statistical analysis was performed using the Kruskal-Wallis test followed by Dunn's Multiple Comparison Test (* P < 0.05; ** P < 0.01; *** P < 0.001). Demyelinated lesion outlines were measured in Zeiss Axovision 4.8 software based on the increase in DAPI staining inside the lesions, and outlines are represented by dashed white lines. Scale bar is 20 lm (D) and 50 lm (E). [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Proliferation of OPCs in Response to Demyelination is Decreased in Astrocyte Fibronectin Conditional Knockout Mice
To determine whether the density of OPCs in the demyelinated area is affected by pFn cKO and/or aFn cKO , we analyzed OPC numbers after lysolecithin-induced demyelination in the different Fn cKO mice. Immunostaining for Olig2 was used as a marker for oligodendrocyte lineage cells (Arnett et al., 2004;Fancy et al., 2009) and Sox2 and Olig2 double immunohistochemistry to identify OPCs (Kondo and Raff, 2004;Shen et al., 2008; Fig. 2A-D). This approach revealed a small, significant reduction in Olig2-positive (Olig21) cells and Sox21Olig21 cells after aFn cKO , but not pFn cKO at 5 DPL ( Fig. 2A-D). A similar reduction in Olig21 and Sox21Olig21 cells was detected after a1pFn cKO ( Fig. 2A-D), indicating that the decrease in OPC density was associated with the decrease in aFn. An additional decrease in pFn levels in a 1 pFn cKO did not further reduce the number of OPCs. In support of these observations, by determining OPC density using the transcription factor Nkx2.2 as a marker (Watanabe et al., 2004), we similarly detected that the OPC density was reduced after aFn cKO and a 1 pFn cKO (Fig.  2E,F). The lesions were of comparable sizes in the different Fn cKO samples relative to WT (data not shown). In tissue culture, pFn promotes OPC migration and proliferation Hu et al., 2009;Milner et al., 1996). To examine whether the reduction in OPC numbers in aFn cKO mice reflected an impairment in proliferation, we next analyzed the numbers of proliferating OPCs at 5 DPL, determined by co-labeling with Olig2 and the cell proliferation marker KI67 (Gerdes et al., 1983). This revealed a reduction in KI671Olig21 cells relative to Olig21 cells in aFn cKO and a 1 pFn cKO , but not pFn cKO (Fig. 2G,H), indicating that the decrease in OPC numbers is, at least partly, a result of impaired proliferation. In contrast, we did not observe a substantial change in microglia/macrophage numbers after Fn cKO , as assessed by immunohistochemistry for ionized calcium-binding adaptor molecule 1 (Iba1) (Imai et al., 1996; Fig. 3A,B). Similarly, astrocyte numbers were not significantly affected by Fn cKO , as reflected by comparable numbers of cells that stained positive for aldehyde dehydrogenase 1, member L1 (Aldh1L1; Cahoy et al., 2008;Lovatt et al., 2007;Fig. 3C,D). Hence, aFn cKO resulted in a reduced number of OPCs after demyelination, whereas microglia/macrophage and astrocyte cell numbers were unaffected.

Oligodendrocyte Differentiation and Remyelination are not Affected by Conditional Knockout of Fibronectin from Astrocytes
We next examined whether the decreased OPC density observed in aFn cKO and a 1 pFn cKO during demyelination resulted in altered CNS remyelination following lysolecithininduced demyelination. Oligodendrocyte lineage cell numbers were analyzed by immunohistochemistry for Olig2 at 14 days post lysolecithin-induced demyelination. As shown in Fig.  4A,B, Olig21 cells were still decreased in a 1 pFn cKO and aFn cKO as compared with littermate WT mice. In pFn cKO mice, Olig21 cell numbers did not differ from the WT mice in the lesions. Furthermore, analysis of differentiated oligodendrocytes, determined by immunohistochemical staining for CC1 (Bhat et al., 1996), showed that the density of CC11 cells was similar in all groups (Fig. 4C). This was confirmed using an alternative marker of mature oligodendrocytes, in situ hybridization for proteolipid protein (PLP) mRNA, a major myelin protein, which revealed no differences in PLP mRNA1 cells (Fig. 4D,E). Therefore, despite a reduction of OPCs after demyelination in aFn cKO and a 1 pFn cKO , it was not sufficient to result in an impairment of oligodendrocyte generation at 14 DPL.
To assess whether myelin sheath formation was affected, we examined toluidine blue stained semithin resin sections of lesions from littermate WT and a 1 pFn cKO mice at 21 DPL. We could not detect any gross morphological differences between remyelinated lesions from littermate WT and a 1 pFn cKO mice (Fig. 4F). Ranking analysis of the degree of remyelination (Ibanez et al., 2004) did not reveal significant differences between WT and a 1 pFn cKO lesions either (Fig.  4G). Therefore, aFn, although involved in OPC recruitment, is not essential for remyelination.

Expression of the Alternatively Spliced Domains EIIIA and EIIIB mRNA of Cellular Fibronectin is Increased Following Demyelination
We next investigated the specific contribution of the alternatively spliced domains EIIIA and EIIIB, exclusively localized within the cellular Fn structure (Paul et al., 1986;Schwarzbauer et al., 1987) in the regulation of OPC proliferation following demyelination. We first determined the expression of both domains following demyelination of the mouse spinal cord ventral funiculus by quantitative, real-time PCR analysis of mRNA expression. Compared with unlesioned tissue, EIIIA, EIIIB, and total Fn mRNA levels were increased in demyelinated tissue at 5 DPL relative to expression of the housekeeping gene GAPDH (Fig. 5A-C; 5 DPL), similar to our previous report for total Fn protein and Fn EIIIA protein (Stoffels et al., 2013a). Furthermore, during remyelination (14 DPL), mRNA levels of EIIIA, EIIIB, and total Fn returned to lower levels ( Fig. 5A-C; 14 DPL). Because both EIIIA and EIIIB mRNA levels are increased at 5 DPL, when OPCs are recruited, and since these elements are exclusively present in Fn derived from local cellular sources, such as astrocytes, we next examined the extent to which aFn EIIIA, FIGURE 2: OPC numbers in lysolecithin-induced demyelinated lesions decrease after conditional knockout of astrocyte fibronectin, but not after plasma fibronectin conditional knockout alone. A-I: After 5-days post lysolecithin-demyelinated lesions (5 DPL) of the mouse spinal cord ventral funiculus, the density of Olig2-positive ("Olig21"; A,B), Sox21 Olig21 (C,D), Nkx2.21 (E,F), and KI671Olig21 (G,H) cells were determined by immunohistochemistry. Note the decrease in Olig21, Sox21Olig21, Nkx2.21, and KI671Olig21 cells number after conditional knockout of fibronectin from astrocytes (aFn cKO ) or from astrocytes and plasma (a 1 pFn cKO ), but not from plasma alone (pFn cKO ) relative to littermate WT mice. Images in A,C,E and G are representative images of demyelinated areas in the different groups. Insets show higher power magnifications of the double positive cells that were counted, with the blue color representing DAPI staining. The outline of the demyelinated lesions was measured in Zeiss Axovision 4.8 software based on increased DAPI stainings' inside lesions and outlines are represented by dashed white lines. Double (C,G) or single (A,E) positive cell numbers were manually counted 3 times for 4 mice per group, and 3 lesion levels per animal,~120 lm distant from each other. Representative graphs of 2-3 independent experiments are shown. Bars represent means. Error bars show standard deviations. Statistical analyses were performed using one-way ANOVA, followed by Tukey's Multiple Comparison Test (* P < 0.05; ** P < 0.01; *** P < 0.001; NS: not significant). Scale bars are 50 lm. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.] EIIIB, or both are involved in proliferation and/or migration of OPCs in vitro.

The EIIIA Domain of Astrocyte-Derived Fibronectin Mediates Proliferation of OPCs, but not Their Migration and Adhesion
To examine the effect of pFn and aFn on OPC proliferation, we used commercially available pFn from bovine plasma and aFn derived from extracellular deposits of primary astrocytes (Fig. 6A, "Fn"). The aFn preparation contained the alternatively spliced domain EIIIA (Fig. 6A, "EIIIA") and EIIIB (Fig.  6A, "EIIIB"), as confirmed by immunostaining with domainspecific antibodies. Functional blocking antibodies against EIIIA ("IST9", Liao et al., 1999) and EIIIB ("C6", Balza et al., 2009) were used to eliminate signals from these domains. OPCs were allowed to proliferate on the different substrates in the presence of relatively high levels of PDGF-AA and FGF-2 for 24 h, after which percentages of BrdUincorporating Olig21 cells were determined by immunocyto-chemistry as a measure of OPC proliferation. As expected under these conditions , Colognato et al., 2004, OPCs proliferated equally well on either aFn (16 6 5% SD) or pFn (15 6 8% SD; Fig. 6B), and to comparable levels when grown on PLL (data not shown). However, proliferation of OPCs was markedly reduced when the cells were cultured on aFn with the EIIIA domain blocked (Fig. 6B; "aFn 1 IST9"). In contrast, addition of the EIIIA blocking antibody to pFn did not affect OPC proliferation ( Fig. 6B; "pFn 1 IST9"), indicating that the reduction in OPC proliferation following the blocking of EIIIA with the IST9 antibody was attributable to the specific effect(s) of the antibody. Moreover, blocking EIIIB from aFn ( Fig. 6A; "EIIIB") with the C6 antibody ( Fig. 6B; "aFn 1 C6") did not affect OPC proliferation. To determine whether EIIIA and/or EIIIB of aFn are also important for OPC migration, we allowed OPCs to migrate through transwell microchambers, containing membranes coated with pFn or aFn on both sides, using PDGF-AA as an attractant in the bottom well. As shown in Fig. 6C, migration FIGURE 3: Microglia and astrocyte cell numbers in lysolecithin-demyelinated lesions are not affected by conditional knockout of plasma and/or astrocyte fibronectin. A-D: After 5 DPL of the mouse spinal cord ventral funiculus from littermate WT mice and after conditional knockout of Fn from plasma (pFn cKO ), astrocytes (aFn cKO ), or astrocytes and plasma (a 1 pFn cKO )., the density of Iba11 cells was determined by measurement of relative immunofluorescence (A,B), and Aldh1L11 cells were stained by immunohistochemistry, then counted (C,D). Representative images of demyelinated areas in the different groups are shown. Insets show higher power magnifications of the double positive cells, with the blue color representing DAPI staining. The outline of the demyelinated lesions was measured in Zeiss Axovision 4.8 software based on increased DAPI stainings' inside lesions and outlines are represented by dashed white lines. Immunofluorescence levels of Iba1 cells were measured using FIJI software (A,B) and Aldh1L11 cell numbers were manually counted 3 times for 4 mice per group, and 3 lesion levels per animal,~120 lm distant from each other. Representative graphs of 2-3 independent experiments are shown. Bars represent means. Error bars show standard deviations. Statistical analyses were performed using the Kruskal-Wallis test followed by Dunn's Multiple Comparison Test (A,B) or one-way ANOVA, followed by Tukey's Multiple Comparison Test (C,D) (*P < 0.05; ** P < 0.01; *** P < 0.001; NS: not significant). Scale bars are 50 lm. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

250
Volume 63, No. 2 Statistical analyses were performed using one-way ANOVA, followed by Tukey's Multiple Comparison Test (** P < 0.01; *** P < 0.001; NS: not significant). Scale bars are 50 lm. F,G. At 21 DPL, semi-thin sections of resin-embedded WT and a 1 pFn cKO mice were stained with alkaline toluidin blue to analyze the myelin structure (F), and blindly ranked according to estimated percentage of remyelination by two independent researchers, after which statistical differences were analyzed using the Mann Whitney test (G). Scale bar is 100 lm. NS means "not significant." [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.] of OPCs was similar on aFn and pFn, and was not affected by blocking the EIIIA or EIIIB domains. These experiments demonstrate that EIIIA, but not EIIIB, is primarily required for proliferation of OPCs on aFn, and that neither domain is essential for migration of OPCs on aFn.
Proliferation of OPCs on pFn at low levels of PDGF-AA is largely mediated by the integrin receptors avb3 and to a lesser extent by avb1 Blaschuk et al., 2000;Milner et al., 1996). Although EIIIA does not directly bind to avb1 or avb3, it has been reported that EIIIA may indirectly promote adhesion of cellular Fn to these integrins, thereby facilitating proliferation (Manabe et al., 1999). To test this possibility, we assessed whether adhesion of OPCs to aFn indirectly involved integrin b binding by using blocking antibodies against integrins b1, b3 and b5, and whether such adhesion was facilitated by functional EIIIA and EIIIB domains. Compared with adhesion to pFn ( Fig. 6D; "pFn 1 b3" and "pFn 1 b5"), adhesion of OPCs to aFn was only modestly reduced when the integrins b3 ( Fig. 6E; "aFn 1 b3") and b5 ( Fig. 6E; "aFn 1 b5") were blocked. In contrast, integrin b1 had no effect on adhesion of OPCs to both pFn and aFn (Fig. 6D,E; "pFn 1 b1" and "aFn 1 b1"). Furthermore, functional blocking of EIIIA and EIIIB domains did not alter adhesion of OPCs to aFn (Fig. 6F). Thus, whereas the integrin b3 and b5 receptors likely mediate adhesion of OPCs to aFn, adhesion of OPCs does not require functional EIIIA or EIIIB domains of aFn.

Discussion
In this study, we investigated how Fn, derived from plasma (pFn) or cellular Fn derived from astrocytes (aFn), modulates remyelination following CNS demyelination. After conditional knockout of Fn from astrocytes (aFn cKO ) or both astrocytes and plasma (a 1 pFn cKO ), the density of (proliferating) OPCs in demyelinated lesions was reduced. Our in vitro analyses of OPCs, cultured on either pFn or aFn, revealed that the alternatively spliced EIIIA domain, exclusively present in cellular Fn, is instrumental in proliferation. Our data further showed that this control of proliferation by the EIIIA domain was likely not related to an ability of EIIIA to enhance adhesion of OPCs, although integrins b3 and b5 mediate adhesion of OPCs to aFn. Furthermore, we did not observe an effect of aFn on migration of OPCs. Despite a reduction in oligodendrocyte lineage cells on a 1 pFn cKO at later stages of the remyelination process, both pFn and aFn were dispensable for oligodendrocyte differentiation and complete remyelination.
Although both EIIIA and EIIIB are largely absent from healthy tissue after embryonic development, reappearance of cellular Fn with the alternatively spliced EIIIA and EIIIB domains occurs after injury of different tissue types (ffrench-Constant et al., 1989;Jarnagin et al., 1994;Nickeleit et al., 1996;Ulrich et al., 1997). Injury-induced expression of cellular Fn containing EIIIA is associated with increased cell proliferation, as demonstrated in a variety of cell types (Manabe et al., 1999;Olsen et al., 2012;Stenzel et al., 2011). In addition, while proliferation of OPCs is known to be independent Real-time PCR on spinal cord RNA from control mice (unlesioned), at 5 days (5 DPL) or 14 days (14 DPL) post lysolecithininduced demyelination. Note the increase in fibronectin EIIIA and EIIIB mRNA at 5 DPL. Gene expression levels were analyzed using the 2 2DDct method (Livak, 2001), with relative expression to HPRT1 or GAPDH. Similar results were obtained for both housekeeping genes; graphs shown the relative expression to GAPDH. Statistical analysis was performed using the Kruskal-Wallis Test, followed by Dunn's Multiple Comparison Test (* P < 0.05; ** < P 0.01).

252
Volume 63, No. 2 of pFn at sufficient levels of mitogenic growth factors , Colognato et al., 2004, which are likely also sufficient in lysolecithin-induced demyelinated lesions (Hinks and Franklin, 1999), our data indicate that the EIIIA domain is essential for proliferation of OPCs on aFn under these conditions. This suggests that proliferation of OPCs on aFn involves a different mechanism from pFn. The EIIIA could theoretically mediate proliferation of OPCs on aFn via bind-ing to the integrins a9b1 (Olsen et al., 2012;Ou et al., 2013;Sun et al., 2014) and a4b1 (Liao et al., 2002), or by enhancing the binding affinity of other integrin receptors for Fn, most notably a5b1 and avb3 (Manabe et al., 1997(Manabe et al., , 1999Xia and Culp, 1995). However, integrins a4b1 and a9b1 are not expressed by OPCs (Milner and ffrench-Constant, 1994) and the EIIIA domain was not important for adhesion of OPCs to aFn, implying that EIIIA mediates FIGURE 6: Blocking the EIIIA domain from astrocyte-derived fibronectin affects OPC proliferation, but not migration and adhesion. A: Astroglial matrix derived from neonatal rat astrocytes (aFn) immunostained for total fibronectin ("Fn"), and the alternatively spliced EIIIA ("EIIIA") or EIIIB ("EIIIB") domains. Scale bar is 50 lM. B: OPCs from neonatal rats were cultured for 24 h on plasma fibronectin (pFn) or aFn, alone or in the presence of blocking antibodies against the EIIIA or EIIIB domain and subjected to a BrdU incorporation assay combined with Olig2 immuncytochemistry. On aFn, 16% of OPCs proliferated on average 6 5% SD. Note that addition of an EIIIA blocking antibody ("aFn 1 IST9") decreased the number of Olig21 OPCs incorporating BrdU on aFn, but not on pFn ("pFn 1 IST9"), and BrdU incorporation did not differ between the other conditions. C: OPCs were allowed to migrate through transwell inserts coated with aFn or pFn for 4 h, alone or in the presence of blocking antibodies against EIIIA ("pFn 1 IST9"; "aFn 1 IST9") or EIIIB ("aFn 1 C6") antibodies. Note that the numbers of migrated cells did not differ between the conditions. D-F: OPCs were allowed to adhere to pFn (D) or aFn (E,F) in the presence of blocking antibodies against integrins b1 ("pFn 1 b1," "aFn 1 b1"), b3 ("pFn 1 b3," "aFn 1 b3"), or b5 ("pFn 1 b5, "aFn 1 b5"; D,E) or after blocking the EIIIA ("aFn 1 IST9") or EIIIB ("aFn 1 C6"; F) domains of aFn. Note that whereas integrins b3 and b5 likely mediate adhesion of OPCs, particularly to aFn (E), adhesion does not require a functional EIIIA or EIIIB domain on aFn (F). Bars in the graphs represent means of 3-4 independent experiments. The horizontal, dashed line represents pFn (D) or aFn-treated (B, C, E, F) cells, and this condition was set to 100% in each experiment. Error bars show standard deviations. Statistical analyses were performed using the Kruskal-Wallis Test, followed by Dunn's Multiple Comparison Test (** P < 0.01; NS: not significant). [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.] proliferation of OPCs through an alternative mechanism. This mechanism may involve other domains of Fn, since alternative splicing of Fn can change the conformation of Fn, affecting the presentation of binding sequences and exposing cryptic binding sites (Pickford and Campbell, 2004;Ventura et al., 2010). In addition to EIIIA, expression of the EIIIB domain was also upregulated after demyelination. Whereas EIIIB appears not to have a role in proliferation and migration of OPCs, its functional involvement in processes other than proliferation and migration cannot be excluded and merits further investigation. The apparent redundancy of pFn for achieving a normal cell density of OPCs observed in our in vivo studies was unexpected, given that stimulation of cell proliferation and migration by pFn is well documented for several cell types (To and Midwood, 2011;von Au et al., 2013), including OPCs at low growth factor levels Colognato et al., 2004;Hu et al., 2009;Milner et al., 1996). Our in vitro proliferation studies suggested that OPCs proliferate equally on pFn and aFn in the presence of PDGF-AA and FGF-2. After pFn cKO , a compensatory increase in cellular Fn may have occurred, mediated by astrocytes, microglia/macrophages, and endothelial cells (Stoffels et al., 2013a). From these cell types, astrocytes are thought to represent the major source of Fn after toxin-induced demyelination (Hibbits et al., 2012;Stoffels et al., 2013a), which is further supported in this study, given the pronounced reduction in Fn levels after aFn cKO . Therefore, a compensatory increase in aFn could explain the absence of a clear phenotype after pFn cKO , in agreement with studies of other tissues (Sakai et al., 2001). However, the observation that a 1 pFn cKO does not substantially amplify the reduction in OPC numbers during demyelination may support an alternative explanation, namely that excessive leakage of pFn to the demyelinated area does not occur on lysolecithin-induced demyelination, despite breakdown of the blood-brain barrier (Ford et al., 1990). Hence, in contrast to previous studies, in which pFn was considered a predominant source of the Fn matrix in tissue (Moretti et al., 2007), our findings indicate that aFn rather than pFn is the major component of the Fn matrix expressed after lysolecithininduced CNS demyelination. In MS lesions, where immunemediated blood-brain barrier disruption is more diffuse, pFn may be a more prominent source of Fn. Also, pFn could potentially modulate OPC proliferation when growth factor levels are low Colognato et al., 2004).
Although aFn mediates OPC cell numbers in demyelinated lesions, it is dispensable for remyelination. Moreover, at the stage of oligodendrocyte differentiation, the absolute numbers of differentiated oligodendrocytes did not differ between aFn cKO or a 1 pFn cKO and WT animals. After lysolecithin-induced demyelination in WT animals, the amount of differentiating oligodendrocytes remains relatively stable from this stage onwards until the completion of remyelination our unpublished observations). This indicates that despite the reduced numbers of OPCs after aFn cKO or a 1 pFn cKO , sufficient numbers of differentiated oligodendrocytes are still generated to reach the maximal level of differentiation necessary to remyelinate the lesion. In this way, relative redundancy of aFn, in spite of decreased OPC proliferation after aFn cKO , is consistent with the concept that OPCs are normally recruited in excess relative to the numbers required for remyelination following small, focal, toxin-induced lesions Stidworthy et al., 2004). In addition, functional compensation may result from the increased expression of several other ECM proteins in demyelinated lesions . In support of this, osteopontin, another ECM molecule expressed after demyelination, is also redundant for remyelination (Zhao et al., 2008). In contrast to modulation of OPC proliferation by aFn, which likely promotes recovery, persistent expression of Fn variants in pathology mediates failure of tissue regeneration (Stoffels et al., 2013b). In particular, the EIIIA domain is involved in adverse remodeling after tissue injury (Arslan et al., 2011;Hirshoren et al., 2013;Kohan et al., 2010;Muro et al., 2008). In MS lesions, Fn assembles into aggregates, which is likely mediated by inflammatory factors. Astrocytes are an important source of Fn aggregates, and these Fn aggregates contribute to remyelination failure (Stoffels et al., 2013a). Since our findings indicate that aFn is a nonessential element for remyelination, developing therapeutic approaches that remove aggregates of Fn from MS lesions represent a legitimate translational objective.