Protease induced plasticity: matrix metalloproteinase-1 promotes neurostructural changes through activation of protease activated receptor 1

Matrix metalloproteinases (MMPs) are a family of secreted endopeptidases expressed by neurons and glia. Regulated MMP activity contributes to physiological synaptic plasticity, while dysregulated activity can stimulate injury. Disentangling the role individual MMPs play in synaptic plasticity is difficult due to overlapping structure and function as well as cell-type specific expression. Here, we develop a novel system to investigate the selective overexpression of a single MMP driven by GFAP expressing cells in vivo. We show that MMP-1 induces cellular and behavioral phenotypes consistent with enhanced signaling through the G-protein coupled protease activated receptor 1 (PAR1). Application of exogenous MMP-1, in vitro, stimulates PAR1 dependent increases in intracellular Ca2+ concentration and dendritic arborization. Overexpression of MMP-1, in vivo, increases dendritic complexity and induces biochemical and behavioral endpoints consistent with increased GPCR signaling. These data are exciting because we demonstrate that an astrocyte-derived protease can influence neuronal plasticity through an extracellular matrix independent mechanism.

Matrix metalloproteinases (MMPs) are a family of zinc-dependent endopeptidases that remodel synaptic circuits and extracellular matrix (ECM) proteins through proteolytic processing. Regulated MMP activity contributes to physiological synaptic plasticity, while dysregulated activity stimulates neuronal injury 1,2 . In physiological plasticity, MMPs have traditionally been shown to influence the remodeling and growth of neuronal cellular elements through proteolytic cleavage of ECM proteins 3,4 . This generates soluble integrin binding ligands as well as a permissive environment for reorganization [5][6][7] . At the cellular level, dendrites or branched neuronal projections structurally change shape in response to extrinsic and intrinsic stimuli 8 . Dendritic growth and remodeling impacts the number of synaptic connections, which are crucial for ongoing cognitive processes, such as learning and memory 9 . Additionally, recent evidence that suggests astrocyte-derived factors greatly affect neuronal morphology 10 . In fact, a single astrocyte in the mouse cortex physically contacts several neuronal cell bodies, up to 600 dendrites, and approximately 100,000 synapses 11,12 . Notably, astrocytes secrete specific MMP family members in response to neuronal activity or pro-inflammatory stimuli 13,14 . Yet, very little is known about the role specific astrocyte-derived MMPs play in neuronal plasticity. Addressing this question is important to the field because it may uncover therapeutic targets for several disorders associated with increased MMP activity, astrocyte activation, and synaptic deficits including autism, fragile X syndrome, and Alzheimer's disease 15,16 .

Concurrent expression of hMMP-1 mRNA and GFAP protein in situ.
To verify that GFAP positive cells express the transgene, we explored the colocalization of hMMP-1 mRNA using fluorescent in situ hybridization (FISH) combined with GFAP immunofluorescent chemistry (IFC) (Fig. 3). We find that hMMP-1 mRNA is restricted to GFAP positive cells in hMMP-1 Tg animals but not WT littermate controls. A low magnification image taken from hMMP-1 Tg brain with an area of interest in cortex and hippocampus highlighted in white ( Fig. 3a,b, respectively). Higher magnification views for cortical and hippocampal areas of interest are shown in Fig. 3a1-a3 and Fig. 3b1-b3, respectively. Cell morphology indicates that double positive cells are most likely astrocytes. Further analysis reveals that cells in the dentate granule cell layer (Fig. 3c, low magnification) are positive for both GFAP and hMMP-1 mRNA (Fig. 3c1-c3, high magnification). Based on localization in this layer, double positive cells could represent astrocytes and/or neural progenitor cells 32 . Lastly, a GFAP positive A schematic diagram of the construct used for pronuclear injections is shown in (a). The murine GFAP promoter drives human MMP-1 expression in the transgenic model. To verify transgene presence, hMMP-1 was amplified by genomic PCR and a representative result is shown in (b). GFAP immunofluorescent reactivity in WT and hMMP-1 Tg brains are shown in (c,d) GFAP (Green), Dapi (blue). Mean hMMP-1 protein levels (ng/mg of total protein) in discrete brain regions were measured by ELISA and are quantified in (e). hMMP-1 levels in WT (n = 7) brain regions are not detected. Mean and ± SEM for hMMP-1 Tg (n = 8) brain regions: Cortex (CTX) 1.18, ± 0.23; Cerebellum (CBL) 4.47, ± 2.80; Hippocampus (HP) 11.71, ± 2.29; Striatum (STR) 10.96, ± 3.22 (Student's t-test; ***p value ≤ 0.001, **p value ≤ 0.01, ns denotes a p value > 0.05). To ensure that astrocytes secrete hMMP-1, supernatants from primary astrocyte cultures were tested for the presence of hMMP1 by ELISA in (f) (WT n = 4, levels not detected; hMMP1 Tg n = 4, hMMP-1 mean, ± SEM: 17.60 ng/mL, ± 4.14; Student's t-test, **p value = 0.005. The inset contains a representative image of cultured cortical astrocytes; GFAP immunocytochemistry confirmed 95% of the cells were GFAP positive. Scale bars = 200 μm (c), 100 μm (d), 25 μm (f).
cell lacking hMMP-1 mRNA in WT littermate control brain is presented for comparison in Fig. 3d (low magnification). Higher magnification views for the WT hippocampal region of interest, outlined in white, are shown in Fig. 3d1-d3. Notably, results from quantitative real-time PCR (qRT-PCR) reveal mRNA for the murine orthologue of MMP-1, Mmp-1a, is undetectable in both WT and hMMP-1 Tg mouse brains (Supplementary Table 1). The primers detected Mmp-1a transcripts in our positive control (primary astrocytes activated with IL-1β ). In addition, transcript levels of the functionally similar mouse MMP, Mmp-13, do not differ significantly between genotypes (Supplementary Table 1).
hMMP1 Tg animals show changes in brain metabolites relevant to gliosis as determined by in vivo MRS imaging. In our model, astrocytes secrete hMMP-1, which could influence varied cell populations. To investigate the effects of hMMP-1 on the living brain, we examined specific brain metabolites including myo-Inositol because these levels increase in the setting of astrocyte and/or microglial activation 33 . To Brain coronal sections were assayed for the presence of hMMP-1 mRNA (red), GFAP protein (green), and DAPI nuclei (blue) in hMMP-1 Tg and WT animals. We highlight several regions of interest (a-d). Low magnification images show the distribution of cells that concurrently express hMMP-1 mRNA as well as GFAP protein in hMMP-1 Tg mouse brain (a-c). White boxes (a-d) indicate the region of interest for high power views. These high magnification views show cortical GFAP positive astrocytes with a hMMP-1 mRNA signal (a1-a3), and a stereotypical astrocytic foot process surrounding a blood vessel in the hippocampus that espressses both GFAP and hMMP-1 mRNA (b1-b3). Lastly, GFAP positive cells expressing hMMP-1 mRNA in the hippocampal dentate gyrus are also shown (c1-c3). Notably, hMMP-1 mRNA is not present in WT littermate control brain (d), GFAP positive cells lacking hMMP-1 mRNA are shown in (d1-d3). Scale bars: 200 μ m (a,b,d), 50 μ m (c), and 20 μ m (a1-a3, b1-b3, c1-c3, d1-d3). evaluate regional changes in metabolites, we use in vivo MRS to analyze both hippocampal and cortical regions of hMMP-1 Tg mice and WT littermate controls (region of interest depicted in Fig. 4a inset image). We find a significantly higher level of myo-Inositol in hMMP-1 Tg brains when compared to WT littermate controls. Genotype does not significantly alter levels of other metabolites measured in this study. Metabolite levels from in vivo 1 H MR spectra analysis, are normalized to total creatine (Fig. 4a,b)    m-Ins, myo-Inositol; Tau, taurine; GPC, glycerophosphocholine; PCH, phosphocholine). To examine neuronal morphology in vivo, the Golgi impregnation technique was performed. We find increased spine density on apical dendrites of CA1 hippocampal pyramidal cells in hMMP-1 Tg when compared to WT littermate controls (c) (WT n = 56 dendrites from 15 neurons per animal from 4 animals, excluded 4 dendrites because out of focus; hMMP1 Tg n = 80 dendrites from 16 neurons per animal from 5 animals; Mean followed by ± SEM: WT 1.01 ± 0.03, hMMP-1 Tg 1.13 ± 0.03; Student's t-test *p value = 0.02). Also, hMMP-1 Tg animals exhibited a significantly higher number of secondary and tertiary dendrites from pyramidal neurons in layer IV/V of the somatosensory cortex (d) (hMMP-1 Tg n = 3 animals and WT n = 3 animals; mean followed by ± SEM: WT primary 5.33 ± 0.22, secondary 6.70 ± 0.41, tertiary 3.41 ± 0.63, quaternary 0.30 ± 0.14, hMMP-1 Tg primary 5.48 ± 0.26, secondary 8.60 ± 0.42, tertiary 5.41 ± 0.66, quaternary 0.96 ± 0.28; ANOVA, *p value ≤ 0.05).

hMMP-1 transgenic animals displayed deficits in behaviors associated with synaptic plasticity.
Because structural changes in dendritic complexity and spine number have consequences for cognitive processes, we next investigated deficits in plasticity-related behaviors. We performed a battery of behavioral assays pertinent to cognitive processes that require activity dependent synaptic plasticity, including sociability, anxiety, and hippocampal learning and memory. Notably, disruptions in all three of these behavioral domains have been reported animals models of autism and autism-related disorders, which have altered spine density and dendritic complexity 34 . In our tests, behavioral assays were performed in order from least to most stressful (refer to Fig. 5a for experimental timeline). During sociability testing, habituated mice were able to freely explore a three-chambered box that contained a social stimulus at one end and a non-social stimulus at the other. Results show that hMMP-1 Tg mice display decreased sociability (Fig. 5b). WT animals spent significantly more time in the social quadrant (WT n = 10, Student's t-test *p value = 0.05, 266 s ± 25 s compared to 196 s ± 22). In contrast, hMMP-1 Tg animals showed no preference toward the social quadrant, (hMMP-1 Tg n = 14, Student's t-test p value = 0.18, 256 s ± 22 s compared to 215 s ± 20 s).
Next, we examined performance on the elevated plus maze (EPM). This assay has been pharmacologically proven to measure anxiety in mice 35 . We find that hMMP-1 Tg animals enter the open arm of the maze more times that WT littermate controls ( Fig. 5c; WT n = 12, hMMP-1 Tg n = 15; 18 ± 1.70 entries compared  To test hippocampal-dependent spatial memory deficits, we utilized the Morris water maze task (WT n = 11; hMMP-1 Tg n = 15). During the training phase, animals were trained to find an invisible platform in a pool of opaque water for four trials over four days. The latency to platform was recorded. On the fifth day (probe trial), the invisible platform was removed and time and distance traveled in the quadrant that previously housed the platform was recorded. Notably, both groups were able to perform the task as no significant differences were observed during training, which indicates appropriate training efficiency ( Fig. 6a; ANOVA, p value ≥ 0.05). Additionally, we report in Fig. 6b,c that both swim speed (WT 0.26 cm/s ± 0.01 compared to hMMP-1 Tg 0.24 cm/s ± 0.01) and total distance traveled (WT 15.82 m ± 0.53 compared to hMMP-1 Tg 14.23 m ± 0.66) during the probe trial did not differ between groups (Student's t-test p value ≥ 0.05). These three control measures suggest that hMMP-1 Tg animals do not display motor impairments and/or task induced anxiety that would prevent accurate interpretation of the experimental results. We find a trend toward decreased path efficiency ( Fig. 6d; WT 0.54, Figure 6. hMMP-1 Tg animals display deficits in learning and memory on the Morris water maze. Results from Morris water maze testing (WT n = 11; hMMP-1 Tg n = 15) are presented in (a-i). (a) We find no statistical differences in the latency to locate the hidden platform during training (a) (4 trials administered for 4 consecutive days) suggesting both groups were able to perform task and indicating an appropriate level of training (2-way ANOVA p value > 0.05). On the fifth day, the hidden platform was removed and several endpoints were measured. Swim speed (b) and total distance travelled (c) during the probe trial are not significantly altered between the WT and hMMP-1 Tg suggesting that motor impairments do not account for differences observed during probe trial (Student's t-test p value > 0.05). hMMP-1 Tg exhibited decreased path efficiency (d) Student's t-test p value = 0.07) as well as crossings over the area that formerly housed the platform (e) Student's t-test p value = 0.09). Notably, hMMP-1 Tg animals travel significantly less in the quadrant that formerly housed the invisible platform (f) (Student's t-test **p value = 0.01) and spend less time in this quadrant (g) (Student's t-test *p value = 0.03) when compared to WT littermate controls. Times spent in all quadrants during the probe trial, which tests reference memory, are presented for WT animals in (h) and hMMP-1 Tg animals in (i).  Consistent with these results, data from the probe trial, which is 24 hours after the last training session and measures memory 36 , reveal hMMP-1 Tg animals travel significantly less distance in the quadrant that formerly housed the hidden platform ( Fig. 6f Evidence supporting PAR1 activation in hMMP-1 Tg animals. To address the question of increased PAR1 GPCR signaling in hMMP-1 Tg animals, we examined protein levels of PAR1 in cortical/hippocampal lysates. It has been shown that following activation (i.e. tethered ligand binding), PAR1 is rapidly internalized with only 25% trafficked back to the cell membrane 37 . Moreover, in a disease model with increased PAR1 signaling in the lung, receptor protein levels are decreased 38 . Consistent with receptor activation, PAR1 protein levels are significantly decreased in cortical/hippocampal lysates from hMMP-1 Tg when compared to WT littermate controls as shown by Western blot analyses (Fig. 7a,b; WT n = 4, hMMP-1 Tg n = 4; mean followed by ± SEM: WT 1.16 a.u. ± 0.10, hMMP-1 Tg 0.81 a.u. ± 0.10; Student's t test, *p value = 0.03).
Scientific RepoRts | 6:35497 | DOI: 10.1038/srep35497 Pharmacological inhibition and genetic deletion of PAR1 reverse hRecMMP-1 effects on neuronal morphology and Ca 2+ flux. The potential for MMPs to act on specific targets depends on several factors including substrate availability and proximity. MMP-1 is most active at the cell surface 43 and is a potent agonist for cell surface PAR1. Interestingly, PAR1 has been found in synaptosomal preparations, which most likely includes PAR1 located on astrocytic processes that cradle synapses as well as pre-and post-synaptic neuronal membranes 44 . To interrogate the role the MMP-1/PAR1 signaling pathway plays in dendritic arborization, we derived neuron-enriched cultures from MMP-1 Tg and WT animals. These cells were treated with DMSO vehicle control or a PAR1 inhibitor (SCH79797; SCH). Subsequent analysis shows that neurons cultured from hMMP-1 Tg animals have significantly increased number of intersections from 0 μm to 90 μm from cell soma when compared to hMMP-1 Tg + SCH, WT + SCH, and WT + DMSO groups (Fig. 8a,b,c)  Our results suggest MMP-1 activation of PAR1 is responsible for the increased dendritic complexity because hMMP-1 Tg group treated with the PAR1 inhibitor do not significantly differ from WT neurons. Notably PAR1 inhibition does not affect the results of Sholl analysis in WT group suggesting aberrant over activation of the receptor as seen in the hMMP-1 Tg group is required for the morphological change.
To investigate whether the MMP-1/PAR1 axis is also involved in Ca 2+ flux, which may influence dendritic arbor patterning, we performed live cell Ca 2+ imaging with cultures enriched for neurons derived from PAR1 KO pups at DIV18. Representative images taken during the application of control, 40 nM hRecMMP-1, and 50 nM N-methyl-D-aspartate (NMDA) are shown in Fig. 8d. The mean peak of Δ F/F 0 Ca 2+ responses from each cell is quantified in Fig. 8e (N = 65 cells per each condition, (Fig. 8f). We find that genetic deletion of the GPCR PAR1 prevents the hRecMMP-1-mediated increase in Ca 2+ in 95% of NMDA responsive cells. Five percent of NMDA responsive cells responded to treatment with an increase in Ca 2+ suggesting that MMP-1 acts on a small subset of cells through a PAR1 independent pathway.

Discussion
Our results demonstrate a novel mechanism through which brain-derived MMP-1 directs neuronal dendritic organization by activating the GPCR PAR1. Interestingly, this work also strengthens previous findings that show astrocytic factors affect structural remodeling, which underlie physiological synaptic plasticity. Studies in primary culture reveal that MMP-1-mediated PAR1 activation enhances the complexity of neuronal dendritic arbors and Ca 2+ flux. Additionally, hMMP-1 Tg mice display increased spine density in CA1 hippocampal neurons and increased complexity of dendritic arbors in layer IV/V of somatosensory cortex. At the behavioral level, astrocyte-derived MMP-1 induces several phenotypes associated with altered synaptic plasticity including decreased anxiety and deficits in sociability and hippocampal dependent memory. Taken together, we show here that a single MMP driven by GFAP expression can influence neuronal plasticity at least in part through an ECM independent mechanism. Modulation of this axis could provide a therapeutic target for disorders in which astrocyte activation and elevated levels of MMPs are observed 45 .
Disentangling the role individual MMPs play in synaptic plasticity has been especially difficult for the field because these proteases share overlapping structure and function and activate one another. Indeed, redundancy in proteolytic processing of substrates is especially strong among functionally related MMPs. MMP-1 is a soluble collagenase named for its processing of extracellular fibrillar collagen and includes MMP-8 and -13 as sub-family members 46 . MMP-8 exhibits differences in function and expression profile when compared to MMP-1 and -13 47 . Results from real-time quantitative reverse transcription PCR (qRT-PCR) reveal mRNA transcripts for murine MMPs that are functionally similar to hMMP-1, Mmp-1a and Mmp-13, do not differ significantly in the brains of hMMP-1 Tg animals when compared to WT littermate controls. These results suggest they are not upregulated in our model, and thus unlikely contribute to differences reported herein.
Though our work is the first to focus on MMP-1 and neuroplasticity, it complements previous studies that implicate other MMP family members in structural and functional changes, including stromelysin (MMP-3) and gelatinases (MMP-2 and -9). For example, in Drosophila models, deletion of mmp2 completely blocks dendrite reshaping attributed to local degradation of the basement membrane 48,49 . Similarly, in murine systems, deletion of Mmp2 and Mmp3 induce a reduction in dendritic arbor surrounding purkinje cells of the cerebellum and apical dendritic length in layer 5 pyramidal neurons of the visual cortex, respectively 50,51 . In addition, mature cortical neurons treated with a pan-metalloproteinase inhibitor show reduced neurite outgrowth 52 . These studies, although experimentally rigorous, do not focus on specific substrates nor do they evaluate the role of cell-type specific MMP expression. It has been hypothesized that pan-MMP processing of the ECM generates a permissive environment for the rearrangement of dendritic structures. MMPs liberate integrin binding laminin fragments and soluble adhesion molecule ectodomains, which have the potential to influence actin dependent structural plasticity 53,54 . Yet, the specific MMPs and mechanisms that are critical for these changes remain unclear. Complicating the interpretation of results even further is the use of inhibitors that target highly conserved catalytic domains across family members, and therefore lack specificity 55 .
Herein, we show pharmacologic inhibition of a specific substrate, PAR1, reverses increases in dendritic branching complexity in hippocampal neurons due to hMMP-1 overexpression. Likewise, genetic deletion of PAR1 prevents MMP-1 induced flux of intracellular Ca 2+ . Astrocyte-derived hMMP-1 increases branching complexity in layer IV/V neurons of the somatosensory cortex. Further, we confirm that astrocytes derived from hMMP-1 Tg animals secrete active hMMP-1. Taken together, these results support a critical role for PAR1 in Scientific RepoRts | 6:35497 | DOI: 10.1038/srep35497 synaptic transmission because dendritic patterning is one determinant of the amount of innervation that a neuron receives 8 .
Emerging evidence also suggests that MMPs are important modulators of dendritic spine number and structure. Spines are membranous protrusions that extend from the dendritic shaft and are the major sites of excitatory synaptic transmission. Inhibition of the gelatinase, MMP-9, corrects the immature dendritic spine phenotype found in in murine models of the autism-related syndrome, fragile X (FXS) [56][57][58] . Similarly, oral administration of minocycline, a broad-spectrum MMP inhibitor, to FXS animals promotes a shift in the spine profile from more immature to mature 59 . In agreement with a role for MMPs in spine remodeling, we find an increase in dendritic spine density in hippocampal CA1 pyramidal neurons. Given that astrocyte activation is observed with autism and FXS 16,60 , it is tempting to speculate that astrocyte derived proteases contribute to altered neuroplasticity in these conditions and more specifically, that the MMP-1/PAR1 signaling axis could represent a therapeutic target.
Changes in dendrite morphology and spine density impact the number of synaptic connections made on a neuron, which can influence learning and memory. Thus, MMP mediated structural changes could have consequences for behaviors affected by synaptic plasticity. Consistent with this concept, both protein and mRNA transcripts of MMP-3 and -9 are elevated during acquisition trials of Morris water maze, a spatial-hippocampal learning and memory task 7 . Further, MMP activity is critical to varied forms of hippocampal LTP 6,61,62 . Previous studies of MMPs in the setting of learning and memory have focused on family members that generate integrin binding ligands, which include MMP-2, -3, and -9 6,61 . Though neuronal activity dependent release of PAR1 activating MMP-1 or -13 has not been well studied, it is important to note that PAR1 signaling can enhance integrin avidity 63 . This leads us to speculate that PAR-activating MMPs may have additive or synergistic effects with those that generate specific integrin binding ligands.
Interestingly, however, the ability of MMPs to positively influence LTP appears to be a tightly regulated process: too much or too little MMP activity is inimical to the maintenance of LTP 1 . Here, we too provide data that overexpression of single MMP-1, which is observed in pathological conditions 13,15 , impairs performance in the Morris water maze. Two other behaviors are disrupted as well: decreased sociability and increased anxiety.
PAR1 is detected in neurons and astrocytes in hippocampus, cerebral cortex, and striatum of humans and mice with immunohistochemical techniques [19][20][21] . Electron microscopy studies in rat tissue also localize PAR1 to synaptic astrocyte endfeet 44 . Thus, PAR1 localization patterns suggest the receptor plays an important role in modulating plasticity. PARs belong to a unique 4-member family of GPCRs that are activated by site-specific proteolytic cleavage in the N-terminal extracellular region, which uncovers a tethered ligand that folds back onto the receptor conferring an active conformation 64 . Upon irreversible activation, the receptor is quickly internalized and degraded 37 with sustained chronic activation leading to lower protein levels 38 . Consistent with these results, we too find decreases in PAR1 protein in lysates of mixed cortical/hippocampal regions in hMMP-1 Tg mice.
In terms of specific mechanisms by which PAR1 signaling affects neuronal structure, much of what we know about the role PAR1 plays in synaptic plasticity comes from studies that use thrombin, a potent peripheral activator that gains entry into the brain after blood brain barrier disruption and neurotoxicity 65 . It appears that thrombin induced PAR1 activation in the hippocampus and CNS cell types modulates synaptic transmission and plasticity through the enhancement of NMDA receptor currents and Ca 2+ flux 66,67 . With respect to the study of MMP-1 as an agonist, it is important to note that PAR1 cleavage by thrombin and MMP-1 occurs at distinct locations in the N-terminus. Moreover, evidence suggests that thrombin and MMP-1generated activating peptides differentially bias receptor signaling. In endothelial cells, MMP-1 and thrombin induce expression of different groups of pro-angiogenic genes through PAR1 68 . Once activated, PAR1 can signal through several G proteins including Gα q/11 , Gα i/o , or Gα 12/13 in addition to β -arrestin-mediated endosomal signaling 69 . Biased intracellular signaling depends on the activating ligand, availability of G proteins within the cell-type, and heterodimerization with other family members 70 . Interestingly, protease concentration can have opposing actions: high concentrations induce cell death whereas low doses appear to be neuroprotective 71,72 .
What has yet to be determined is whether astrocyte-derived protease activation of PAR1 might also selectively activate this receptor in a biased manner. This will be investigated in future studies.
We find evidence supporting the selective activation of non-canonical G-protein independent signaling. Once activated, GPCRs are quickly targeted for internalization through phosphorylation of the C-terminus, which permits scaffolding proteins, such as β -arrestins, to bind and initiate internalization through endocytic complexes. In addition to its role in receptor desensitization, β -arrestin promotes G-protein independent signaling. Work focused on non-canonical signaling of dopamine receptors suggests that D2 receptor activation triggers the formation of a signaling complex through of β -arrestin 2, protein phosphatases, and Akt. This signaling pathway results in dephosphorylation of Akt and GSK-3β , which activates the latter 73 . We find evidence of selective activation of this pathway in hMMP-1 Tg animals as hippocampal lysates have increased levels of activated GSK-3β. High levels of active GSK-3β are linked to an increase in the number of thin spines in the dentate gyrus 41 . Additionally, using a broad spectrum MMP inhibitor the same study reversed the increased spine phenotype in constitutively active GSK-3β mice 41 .
In summary, we show that a protease driven by the expression of GFAP can influence neuronal plasticity through an ECM independent mechanism. These findings add to an emerging appreciation of glial derived factors as important effectors of brain structure and function, and have important implications for neurological disorders. Though outside the scope of the present study, future work will focus on sorting out the role of cell type specific contributions of PAR1 signaling and determining whether cell morphology changes are Ca 2+ -dependent. Lastly, identification of astrocyte derived synaptogenic factors and their mechanism of action provide the field with a more complete view of the complex interactions between neurons and glia during synaptic plasticity.

Materials and Methods
Animals. Experiments were conducted in accordance with the ethical guidelines of the National Institutes of Health and approved by the Institutional Animal Care and Use Committee at Georgetown University. All animals were group housed with littermates and permitted free access to food and water. Experimental results compare transgenic with littermates backcrossed to C57BL/6J (The Jackson Laboratory, stock #000664; http://jaxmice.jax. org/strain/000664.html) for at least 10 generations, unless stated otherwise.
Preparation of primary cultures enriched for neurons. Cells were derived from WT and hMMP-1 Tg post-natal day 0 mouse pups. After the removal of adherent meninges, the hippocampi were removed and dissociated into a single cell suspension with trituration following incubation for 5 minutes with 0.1% trypsin as outlined in ref. 74. Dissociated cells were plated onto 18-mm sterilized glass coverslips or plastic 12-well culture dishes.
Cultures were maintained in neural basal media supplemented with B27, glutamine, and penicillin/streptomycin antibiotic and incubated at 37 °C in a humidified 5% CO 2 -containing atmosphere. The neuronal enriched cultures also contained astrocytes as confirmed with GFAP and MAP2 immunocytochemistry (Anti-GFAP Millipore 5804 and Anti-MAP2 PhosphoSolutions, Aurora, CO).

Dendritic Sholl analysis. Primary neuron-enriched cultures were treated with 8 nM human recombinant
MMP-1 protein (hRecMMP-1) at DIV5 and DIV14. Cultures were fixed for 15 minutes with a 4% paraformaldehyde/sucrose solution two-hours after the second treatment at DIV14. Immunofluorescent techniques were used to label MAP2 cytoskeleton of neurons as previously described in ref. 75. Briefly cells were washed three times in PBS, permeabilized in PBS containing 2% (vol/vol) Triton X-100 (Sigma) for 10 min at RT followed by a 2-hour incubation in blocking solution (PBS containing 2% (vol/vol) Triton X-100 (Sigma) and 10% (vol/ vol) goat serum). The primary antibody against MAP2 (chicken, 1: 2000, Phosphosolutions, Aurora, CO) was incubated overnight at 4 °C in a PBS antibody solution containing 1% (vol/vol) goat serum. On the second day, Alexa Fluor 488 conjugated goat anti-chicken secondary antibody (1: 2000, Invitrogen) was incubated at RT for 2 hours. Nuclei were counterstained with DAPI (1:10000, Sigma). Fluorescent mounting medium (Electron Microscopy Sciences, Hatfield, PA) was applied to slides as antifading agent prior to addition of coverslips. Images were acquired on Axioplan 2 Zeiss microscope. Next, we employed the use of a semi-automated Sholl analysis 27  Genomic PCR. Genomic PCR for hMMP-1 transgene was performed on DNA samples isolated from mouse-tail biopsies collected upon weaning at post-natal day 21. DNA was purified using a traditional phenol-chloroform extraction protocol. The following primer sequences were used (forward: 5′ AGC ACA TGA CTT TCC TGG AAT TGG C and reverse: 5′ ATT TTG TGT TAG AAG AGT TAT CC). Tail biopsies were also sent to Transnetyx, Inc. (Cordova, TN) for genotyping.
Preparation of primary astrocytes. Astrocytes derived from WT and hMMP-1 Tg mouse pups were prepared at post-natal day 1-2 as previously reported by our laboratory 79 . Specifically, after removal of adherent meninges, cortices were microdissected, incubated for 5 minutes with 0.1% trypsin, and dissociated into a single cell suspension with trituration. Dissociated cells were plated onto 18-mm sterilized glass coverslips or plastic 12-well culture dishes. Cells were maintained in MEM complete with Earle's salts and L-glutamine (Gibco, catalog #11095) supplemented with 10% heat inactivated fetal bovine serum and penicillin-streptomycin. Cultures were maintained at 37 °C in a humidified 5% CO 2 -containing atmosphere. GFAP immunocytochemistry confirmed the presence of astrocytes in the cultures (≥ 95% of total cells were GFAP positive; anti-GFAP antibody, Millipore, 5804 or Cell Signaling, 3670 at a 1:1000-dilution).
Western blot. Brains were dissected and lysed in RIPA buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 0.1% SDS, 1% NP-40, and 1X protease and phosphatase cocktail (Thermo Scientific 1861281)). Lysates were sonicated for 10 seconds, placed on ice for 20 minutes, and centrifuged for 15 minutes at 14,000 rpm in 4 °C. Supernatants were recovered and used for future western blotting experiments. Protein concentrations were determined using BCA protein assay (Pierce Biotechnology, Inc.) and equal amounts of protein were used in all subsequent assays. Supernatants were incubated with Laemmli sample buffer (Bio-Rad, Hercules, CA, USA, catalog #161-0737) containing 5% β -mercaptoethanol and boiled for 5 minutes at 95 °C before denaturing electrophoresis on tris-glycine polyacrylamide gradient gels (Bio-Rad, Hercules, CA, USA). Proteins were transferred to nitrocellulose membranes, blocked in phosphate-buffered saline (PBS) containing 5% non-fat dry milk and 0.1% Tween (PBST) for 1 hour and subsequently probed with primary antibody at 4 °C overnight (PAR1: Santa Cruz H-111 at a 1:100-dilution). The following day membranes were washed 3 times for 15 minutes in PBST, species specific HRP-conjugated secondary antibody was applied at a 1:1000-dilution for 2 hours at room temperature, and Scientific RepoRts | 6:35497 | DOI: 10.1038/srep35497 followed by a second set of washes immunoreactive bands were visualized on film after incubation with chemiluminescence reagents (Perkin Elmer, NEL602).

RNAscope Fluorescent in situ hybridization combined with immunohistochemistry. hMMP-1
Tg mice and WT littermate controls were sacrificed and their brains were removed, fixed in 4% formol, embedded in paraffin and then sliced into 10-μ m-thick sections and mounted onto Super Frost Plus slides (Fisher). FISH was performed according to the RNAscope 2.0 Red Fluorescent kit for formalin-fixed, paraffin-embedded brain sections (Advanced Cell Diagnostics (ACD)) according to the manufacturer´s instructions. Brain sections were dehydrated by 50%, 70%, and 100% ethanol gradually for 5 minutes, treated with pretreatment 2 solution during 15 minutes, and incubated for hMMP-1 probe (accession number: NM_001145938.1, manufactured by ACD) for 2 hours at 40 °C in the HybEZ humidified incubator. Following probe hybridization, brain sections experienced sequentially a series of probe signal amplification steps, rinsed in ACD Wash Buffer (2 × 2 minutes) and incubated in reagents after fluorescently labeled probes designed to target the red channel associated with hMMP-1 mRNA. Brain sections were washed in PBS (3 × 5 minutes) and blocked using Normal Goat Serum during 1 hour. Immunohistochemistry was performed using a primary antibody against rabbit anti-GFAP (Dako, 1:200) incubated overnight. Alexa Fluor 488-conjugated goat anti-rabbit IgG was applied for 2 hour (1:500) and slides were washed three times with PBS, counterstained with DAPI, and coverslips were mounted with Fluorogel with Tris Buffer mounting medium (Electron Microscopy Sciences). Images were obtained on an Axioplan 2 Zeiss microscope.

MMP-1 activity assay.
MMP-1 activity was tested using a fluorometric assay (AnaSpec) with a 60-minute endpoint reading. Prior to testing, supernatant samples were concentrated 40 times using 3 kDa cutoff centrifugal filters (VWR). The assay was performed according to the manufacturer's instructions except that the 4-aminophenylmercuric acetate (APMA) sample activation step was omitted to permit measurement of endogenous secreted hMMP-1 activity. The standard curve was generated using recombinant human MMP-1 (R & D systems) and was linear within its range (0-8 ng).
In vivo magnetic resonance imaging and spectroscopy. Magnetic resonance imaging (MRI) was performed at the Preclinical Imaging Research Laboratory of the Lombardi Comprehensive Center at Georgetown University Medical Center on a 7.0 Tesla Bruker horizontal bore Magnetic Resonance Imager run by Paravision 5.0 software as previously described 80 . Mice were anesthetized using 1.5% isoflurane and 30% nitrous oxide, positioned in a custom-made mouse stereotaxic device with temperature and respiration control and imaged in a 23 mm mouse volume coil. Two-dimensional MR anatomical locator images were acquired with a T2-weighted RARE protocol with the following parameters: TR: 3000 ms, TE: 24 ms, FOV: 2.25 cm, Matrix: 256 × 256, Averages: 4, Slice thickness: 0.5 mm. Magnetic resonance spectroscopy (MRS) allowed for the quantification of metabolic biomarkers. We used single voxel proton MRS with volume-localized PRESS sequence with the following parameters: TE: 20 ms, TR: 2500 ms, averages: 1024, spectral width of 4 kHz, and 512 k complex data points and 6 Hz line broadening, using a voxel of 2 mm on edge. The voxel was localized on the hippocampus and cortex based on the previously obtained locator image. All in vivo peak integrated areas were analyzed by visual inspection using the using the LC Model (see http://www.s-provencher.com/pages/lcmodel.html) software.
Golgi staining and dendritic spine analysis. Golgi staining was performed on hMMP-1 Tg animals and WT littermate controls aged 3-4 months old using FD Rapid GolgiStain kit (FD NeuroTechnologies, Inc.; Columbia, Maryland) according to the manufacturer's instructions. The brains were sliced on a vibratome (VT1000S; Leica) at 150 μ m. Images of CA1 pyramidal neurons were taken in bright-field on an Axioplan2 Zeiss microscope at 63X or 100X. Images were coded, and dendritic spines counted in a blinded manner similar to previous protocols 81 . Dendritic branching was assessed in a blinded manner according to previously described methods 82 . Fully impregnated cerebral somatosensory cortical layer IV/V neurons were selected. Using a Zeiss Axioplan 2 microscope at 40X, a blinded investigator imaged multiple focal planes so that each basilar branch could be followed in its entirety. For each animal, a total of 9 neurons were evaluated to determine the number of primary, secondary, tertiary and quaternary branches.
Behavioral testing. Animals were housed in temperature-controlled rooms with a 12-hour light/dark cycle. Cages were changed weekly; care was taken not to test animals on the same day as when animal facility changed cages. To further minimize confounding results due to stress, the mice were handled for 3 consecutive days prior to the start of testing. For all studies, the experimenter was blinded to genotype. Mice were permitted a minimum 30-minute habituation period to the testing room, and testing was performed at the same time of day except where noted. All comparisons were made between male littermates. Animals were 3-5 months of age at the time of testing. Three-chambered sociability assay. The social approach assay was slightly modified from previously described protocols 83 . Testing occurred in a 3-chambered apparatus where animals were permitted free access to all chambers during testing. The test was divided into two phases: habituation and social preference. During the habituation phase test animals were placed in center of apparatus box that contained two identical clear, Plexiglass cylinders with multiple holes to allow for air exchange at each end chamber. Animals were monitored for 10 minutes. Next, a stimulus mouse (gonadectomized A/J mice, The Jackson Laboratory, stock #000646; http://jaxmice. jax.org/strain/000646.html) was placed in one cylinder. The time spent in the social or non-social chamber was recorded over a 10-minute testing period. Analysis was performed with ANY-Maze (San Diego Instruments, San Diego, CA). Elevated plus maze. The elevated plus maze was performed by placing mice in the center of the apparatus and subsequently monitoring their movements over the course of 5 minutes. The plus shaped elevated platform consisted of two closed quadrants (wall height: 12 inches) and two open quadrants (no protective wall). Analysis was performed with ANY-Maze (San Diego Instruments, San Diego, CA). Morris water maze. Hippocampal learning and memory deficits were evaluated using the Morris water maze paradigm as previously described (Washington et al. 2012), with additional modifications. Specifically, the water maze apparatus consisted of a 4-foot-diameter pool (San Diego Instruments, San Diego, CA) filled with opaque water (25 °C; colored with non-toxic Crayola ® white paint). Visual cues were placed on the walls surrounding the pool and a platform (4 inches in diameter) was hidden below the surface of the water (1 cm). Initial training consisted of four trials per day over four days. Mice were introduced into the pool at variable entry points, with every entry point used over the course of the day. The location of the platform remained constant throughout training period. The mice were given 60 seconds to locate platform. On the fifth day of testing, a probe trial was conducted in which the platform was removed over one 60-second trial. Tracking software (ANY-Maze; San Diego Instruments, San Diego, CA) was used to record swim speed, total distance traveled, distance traveled in platform quadrant, time spent in platform quadrant, and platform crossings.
Statistical analyses. Statistics were performed using Prism 5.0 (GraphPad Software). Individual statistical tests are listed in the figure legends as well as the number of samples. A p value ≤ 0.05 was considered significant and the following structure was applied to the figures in this paper ****≤ p value 0.0001, ***≤ p value 0.001, **≤ p value 0.01, and *≤ p value 0.05. Bonferroni's, Dunnett's or Tukey's post hoc tests were performed when appropriate to correct for multiple hypothesis testing.