RhoGTPase Regulators Orchestrate Distinct Stages of Synaptic Development

Small RhoGTPases regulate changes in post-synaptic spine morphology and density that support learning and memory. They are also major targets of synaptic disorders, including Autism. Here we sought to determine whether upstream RhoGTPase regulators, including GEFs, GAPs, and GDIs, sculpt specific stages of synaptic development. The majority of examined molecules uniquely regulate either early spine precursor formation or later maturation. Specifically, an activator of actin polymerization, the Rac1 GEF β-PIX, drives spine precursor formation, whereas both FRABIN, a Cdc42 GEF, and OLIGOPHRENIN-1, a RhoA GAP, regulate spine precursor elongation. However, in later development, a novel Rac1 GAP, ARHGAP23, and RhoGDIs inactivate actomyosin dynamics to stabilize mature synapses. Our observations demonstrate that specific combinations of RhoGTPase regulatory proteins temporally balance RhoGTPase activity during post-synaptic spine development.

Through these mechanisms RhoGTPases also impact learning and memory. Altered RhoGTPase signaling leads to abnormal spine morphology and synaptic development and appears to contribute to the pathology of neuronal disorders, such as Autism Spectrum Disorders and non-syndromic mental retardation, as well as neurodegenerative disorders, like Alzheimer's disease (AD) [12][13][14][15][16].
RhoGTPases are activated by Guanine Exchange Factors (GEFs) and inactivated by GTPase-Activating Proteins (GAPs), while Guanine Dissociation Inhibitors (GDIs) attenuate PLOS ONE | DOI: 10.1371/journal.pone.0170464 January 23, 2017 1 / 21 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 RhoGTPase signaling by binding and sequestering the inactive GDP-bound state in the cytosol [17]. Several of these upstream RhoGTPase regulatory proteins are implicated in neurodevelopmental disorders. For example, mutations in the RhoA-GAP OLIGOPHRENIN-1 result in non-syndromic mental retardation [16], through glutamatergic dysfunction, preventing dendritic spine development and synapse maturation [15,18]. Furthermore, RhoGTPases are disproportionately represented in copy number variants associated with Autism and schizophrenia, further highlighting the fundamental developmental role of RhoGTPases in shaping proper neuronal connections [19,20]. Despite a fundamental role for RhoGTPase signaling in neurons, how and when these signaling pathways are activated is not known. In this present study, we explored whether RhoGTPase regulators exhibit stage-specific roles in synaptic development. Our results suggest that RhoGTPase regulators function temporally at discrete stages of synaptic development. Individual regulators orchestrate actomyosin dynamics that support post-synaptic spine morphogenesis in either early spine precursor formation or later spine maturation. In addition to demonstrating that specific molecules uniquely remodel synaptic architecture at distinct developmental periods, this study also suggests the feasibility of targeted therapeutic intervention of actomyosin regulation during synaptic plasticity.

Expression of synaptic RhoGTPase regulatory proteins in neuronal development
While previous studies have largely focused on a particular stage of synaptic development, we sought to determine how RhoGTPase regulators function throughout synaptic development by evaluating their contribution to both early development, when immature filopodia-like spine precursors form, and later synapse formation, when spines mature into a mushroomshaped morphology. These stages of synapse development are observed in vitro with cultured rat hippocampal neurons; they exhibit robust spine precursor formation at approximately 1 week in culture and spine maturation after 2-3 weeks in culture ( Fig 1A) [9,21,22]. We used shRNA to acutely downregulate expression of select RhoGTPase regulators at these distinct stages of synaptic development (S1 Fig). We selected known synaptic RhoGTPase regulatory proteins, β-PIX, a GEF [5,23,24], and OLIGOPHRENIN-1, a GAP protein that is mutated in non-syndromic mental retardation [15,18]. We also included the three mammalian RhoGDI family members (α, β, and γ) [25], to develop a broader and more holistic picture of RhoGT-Pase regulation during synaptic development. All of these RhoGTPase regulators are either mutated or have copy number variants associated with known synaptic disorders, including non-syndromic mental retardation and Autism Spectrum Disorders (ASD) ( Table 1), although their specific contribution to disease pathology may not yet be determined. As a negative control for excitatory synapse development associated with spines, we included ARHGEF9, a regulator of gephyrin clustering and inhibitory synapse formation [26,27].
We also identified and included potential novel regulators of synaptic RhoGTPase activity by screening published synaptic proteomes [28-32] for targets containing either the RhoGAP domain or possessing GEF characteristics, e.g., a Dbl Homology (DH) domain followed by the Pleckstrin Homology (PH) domain [33]. This search revealed potential RhoGTPase regulatory proteins whose role in synaptic development is understudied or completely unknown. These include the GEFs, FRABIN/FGD4 and VAV2, as well as a GAP, ARHGAP23 (Table 1).
Using Real-Time RT-PCR, we assayed for mRNA expression in primary rat hippocampal neurons cultured for either 1 or 2 weeks (Fig 1B). Rat hippocampal neurons express all of the selected RhoGTPase regulators, except for RhoGDI-β, which is predominantly found in Expression of RhoGTPase regulators during synapse development. A) Primary rat hippocampal neurons expressing GFP were stained for the dendritic marker, MAP-2 (MAP-2). At~1 week in culture (days in vitro, DIV-10), filopodia-like spine precursors extend from the dendritic shaft. However, after 2 weeks, spines begin to mature into a polarized mushroom-shape, characterized by a bulbous spine head atop a thin spine neck. B) Real-Time RT-PCR of RhoGTPase regulators from RNA harvested from primary rat hippocampal neurons grown for either 1 or 2 weeks in culture (DIV-7, grey bars; DIV-14, black bars). RNA expression levels are expressed as a percentage of actin RNA. (n ! 5 neuron cultures for each time point; p<0.001 for Arhgdia expression at DIV-7 vs DIV-14 (Mann-Whitney Rank Sum Test) and p = 0.033 for Arhgdig expression at DIV-7 vs DIV-14 (t-test); RNA expression levels of all other regulators was not statistically significant between DIV-7 and DIV-14.
doi:10.1371/journal.pone.0170464.g001 hematopoietic cells [34], and was therefore incorporated as an additional negative control. Strikingly, the Rac1 GEF, β-PIX, exhibited the highest levels of expression at both time points. However, only Rho GDI-α ανδ-γ significantly increased during neuronal maturation, suggesting a specific role for RhoGDIs in synapse maturation.
The Rac1 GEF, β-PIX, drives spine precursor formation To assess the role of specific GEFs and GAPs in spine precursor formation and elongation during early stages of neuronal development, we performed acute shRNA knockdowns at approximately 1 week in culture (DIV6/7) and assessed spine morphology and density after 72 hours (DIV9/ 10). At one week in development, control neurons exhibit an abundance of immature filopodialike spine precursors with very few mature spines ( Fig 1A). These immature spine precursors are highly dynamic, allowing them to contact a pre-synaptic partner [9,22,35]. We therefore hypothesized that decreased expression of activators of actin polymerization, such as the Rac1 GEF, β-PIX, would reduce both spine precursor density and length. In support of this hypothesis, shRNA sequences targeting β-Pix decreased the density of spine precursors (Fig 2A and 2B). By contrast, we did not observe consistent, significant changes in spine density with shRNAs against the other GEFs or GAPs identified in Table 1 (data not shown). This is consistent with a role for Rac1 in the formation of immature filopodia-like spine precursors [5][6][7][8]. We confirmed this role for Rac1 in the formation of filopodia-like spine precursors using a genetically encoded photoactivatable Rac1 (PA-Rac) [36], which increased formation of spine precursors along the dendritic shaft after acute photoactivation (Fig 3A and 3C and S1 Video). Unexpectedly, shRNAs against β-Pix did not decrease spine length, but resulted in increased length (Fig 2A and 2C), similar to expression of dominant negative Rac1T17N [6]. Consistent with this finding, while chronic Rac1 activation increased spine length (Fig 3B and 3C 'Lit' Rac) acute Rac1 photoactivation did not alter spine length (Fig 3B and 3C PA-Rac). Together, these studies highlight a preferential role for Rac1 in initiating spine precursor formation. Balanced RhoA and Cdc42 activities regulate spine precursor elongation By contrast, shRNAs against the RhoA GAP, Oligophrenin-1, decreased spine length (Fig 2A  and 2C) as previously described for spines in the CA1 region of rat hippocampal slices [18]. Though primarily known for its role in myelination and Charcot-Marie Tooth Disease [37,38], shRNAs against the CDC42 GEF, Frabin, also decreased spine length (Fig 2A and 2C). We did not observe consistent, significant changes in spine precursor length with shRNAs against the other GEFs or GAPs identified in Table 1 (data not shown). These results lead us hypothesize that competition between RhoA-driven myosin II contractility and Cdc42-mediated actin polymerization may regulate spine precursor elongation.

ARHGAP23 promotes spine maturation
While actin polymerization drives spine precursor formation, RhoA/ROCK-mediated myosin II activation is necessary for spine maturation into a polarized mushroom-shape [9,10,41]. Furthermore, RhoA and Rac1 mutually inhibit each other's activity [42], even at synapses [10], leading us to hypothesize that spine maturation might be regulated by either activation of RhoA and/or inactivation of Rac1. To assess this, we performed acute shRNA knockdowns of specific regulators after 2 weeks in culture (DIV14) and assessed spine morphology and density after 48 hours (DIV16) (Fig 4). Regulators of spine precursor formation, including activators of actin polymerization, β-PIX and FRABIN, and OLIGROPHENIN-1, a GAP that inactivates RhoA-driven NMII activation, did not affect spine length or density during spine maturation ( Fig 4A, 4B and 4D). Interestingly, shRNAs against the GEF, Vav2, consistently resulted in increased spine length (Fig 4A and 4E). However, we observed the most prominent effect on mature spine morphology and density with shRNAs toward Arhgap23, a putative GAP identified by our in-silico screen. shRNAs against Arhgap23 increased spine density and resulted in immature spine morphologies with increased spine length similar to filopodia-like spine precursors (Fig 4A, 4C and 4E), leading us to hypothesize that ARHGAP23 may function to inactivate Rac-driven actin polymerization.

ARHGAP23 is a novel Rac1 GAP
ARHGAP23 is a novel RhoGTPase regulator identified in silico [43], but its cellular function is unknown. To determine whether it indeed functions as a Rac1 GAP, we used migratory CHO. K1 cells to readily visualize the effects of shRNA downregulation or expression on actin filament bundle organization and adhesion maturation, properties regulated by RhoGTPases. In migratory cells, Rac1-mediated actin polymerization contributes to the formation of the leading edge [17]. Similar to neurons, CHO.K1 cells also express ARHGAP23 (data not shown). ARHGAP23 knockdown resulted in extensive lamellipodia formation and nascent adhesions (Fig 5C and 5D). Conversely, cells expressing GFP-tagged ARHGAP23 exhibited increased adhesion maturation (Fig 5A and 5B, compare adhesions in control cells of S2 Video with adhesion maturation in a GAP23-GFP-expressing CHO.K1 cells in S3 Video). These results are consistent with a corresponding increase in RhoA-mediated actomyosin bundle formation in response to Rac1 inactivation [1,44,45]. A Rac FRET biosensor [39] confirmed that shRNA against ARHGAP23 increased Rac1 activity to levels indistinguishable from constitutively active Raichu Rac V12 (Fig 5D and 5E), demonstrating that ARHGAP23 functions as a novel Rac1 GAP in adhesion maturation of migratory CHO.K1 cells as well as synapse maturation in neurons.
GDI-mediated attenuation of RhoGTPase signaling maintains the mature mushroom-shape spine We next investigated the role of the three mammalian GDI isoforms (α, β, and γ) in spine development [25]. Expression of both α and γ significantly increases during neuronal maturation as measured by real-time RT-PCR, while β exhibited negligible RNA expression during both early (DIV7) and later (DIV14) neuronal development (Fig 1B). Consistent with these RNA expression profiles, knockdown of either GDI-α or -γ, but not β, during neuronal maturation resulted in a striking reversion to immature filopodia-like spine precursors (Fig 6), demonstrating that attenuation of RhoGTPase signaling is necessary for the maintenance of Rac drives spine precursor formation, while myosin-II and Cdc42 activity regulate spine length. A) Rac1 photoactivation increases spine precursor formation. DIV14-21 primary rat hippocampal neurons expressing either photoactivatable Rac1 or as a positive control, constitutively activated Rac1 ('Lit' PA-Rac), were kept in dark (black bars) or exposed to room lighting for 10min (white bars). The resulting spine density is expressed as percent control unactivated PA-Rac-expressing neurons. n = 24 PA-Rac neurons kept in dark, 23 PA-Rac light-exposed neurons, 8 'lit' PA-Rac neurons kept in dark, and 9 'lit' PA-Rac light-exposed neurons; p = 0.047 PA-Rac dark vs light-acivated (t-test). B) Acute Rac1 photoactivation does not affect spine length, unlike constitutive Rac1 activity ('Lit' control). C) mature-mushroom-shaped spines, and is the first evidence of a role for GDIs in post-synaptic spine plasticity.

Discussion
Our research demonstrates that unique combinations of Rho GTPase family members regulate distinct stages of synaptic development. Altered spine formation and maturation are increasingly implicated in neurodevelopmental disease pathology, whereas the maintenance of mature synaptic connections is lost in neurodegenerative disease pathology. In addition, regulatory pathways of the actin cytoskeleton are increasingly implicated in these neuronal disorders. Despite this association, how synaptic connections are temporally constructed and maintained has been unclear.
We showed that distinct RhoGTPase regulators sequentially orchestrate synapse formation and maturation. During spine formation, the Rac1 GEF β-PIX promoted spine precursor formation, while the RhoA GAP OLIGOPHRENIN-1 and Cdc42 GEF FRABIN cooperatively regulated spine precursor elongation. These results lead us to hypothesize that Rac1 initiates spine precursor formation, while competition between RhoA-mediated myosin II activation and Cdc42-driven actin polymerization determines spine length. In support of this hypothesis, acute Rac1 photoactivation drove spine precursor formation, but did not affect spine length. By comparison, myosin II inhibition promoted Cdc42 activation together with increased spine length. How can we understand these differential roles for Rac1 and Cdc42-mediated actin polymerization on spine precursor formation and subsequent elongation? In general, Rac1 drives Arp2/3-mediated branched actin arrays, generating lamellipodia-like veils along the dendritic shafts of neurons [8,10,46]. Intriguingly, EM of filopodia-like spine precursors suggests that they emanate from patches of branched actin [47] consistent with a role for Rac1 in the initiation of spine precursor formation. In contrast to these branched actin structures, Cdc42 promotes linear actin polymerization, generating filopodia in migratory cells, though the role for Cdc42 in synaptic plasticity and filopodia-like spine precursors is less clear. Here we show that, while not required for spine precursor formation, FRABIN-mediated Cdc42 activation drives spine precursor elongation. This spine autonomous role for Cdc42 is consistent with recent evidence for spine-restricted Cdc42 activity in response to glutamatemediated stimulation [48]. However, active Rac and RhoA spread through the dendrite into adjacent spines [48], suggesting either differential diffusion properties of RhoGTPases themselves or association of upstream regulators with distinct synaptic signaling scaffolds.
In contrast to spine formation, spine maturation was characterized by Rac1 inactivation via the novel RhoGTPase regulator, ARHGAP23. RhoGDI-mediated attenuation of RhoGTPase signaling pathways and subsequent actomyosin dynamics was also necessary for the maintenance of mature spines. These studies support a model for spine development whereby the switch from GEF-mediated Rac1/Cdc42 activation to GAP/GDI-mediated Rac1/RhoGTPase inactivation promotes spine maturation and the maintenance of excitatory synaptic connections. This Rac1 activity switch is consistent with previous work demonstrating that antagonism between the Rac1 GEF, TIAM1, and Rac1 GAP, BCR, control spine and synapse development [49]. Similarly, the Rac1 GEF, kalirin-7, is known to have reduced GEF activity when associated with the post-synaptic density [50], suggesting that maintenance of mature synapses involves Rac1 inactivation.
However, subsequent Rac1 and Cdc42 activation is necessary for spine head enlargement following excitatory neurotransmission [11,48]. Thus, it will be important to understand the mechanisms that transiently promote Rac1 and Cdc42 activity in response to excitatory     coincident activation of RhoA, Rac, and Cdc42 [48], suggesting that the GDI-mediated attenuation of RhoGTPase activity must also be transiently and dynamically regulated in response to excitatory stimulation to allow for structural synaptic plasticity. Thus, it will be of interest to perform a future screen with acute knockdown followed by NMDA receptor activation to assess the role of specific RhoGTPase regulators in response to excitatory stimulation leading to synaptic strengthening. Ultimately, our identification of distinct RhoGTPase regulators for particular stages of synaptic development suggests the feasibility of targeted intervention of actomyosin dynamics during critical periods of neuronal development.

Antibodies and reagents
A mouse monoclonal antibody (MAB378) against the dendritic marker, MAP-2, was purchased from Millipore. A mouse monoclonal antibody against paxillin (Clone 349) was purchased from BD Biosciences. Alexa-conjugated secondary antibodies were from Invitrogen. Rhodamine phalloidin was purchased from Cytoskeleton (Denver, CO) and used at a ratio of 1:100. Blebbistatin, was purchased from Calbiochem (La Jolla, CA) and used at a concentration of 50 μM.

Neuronal culture and transfection
Low-density hippocampal cultures were prepared from E19 rat embryos as described previously Full-length human Arhgap23 was obtained from Dharmacon ORFeome library and subcloned into the pEGFP-N1 vector (Clonetech). FRET biosensor probes, including Raichu Rac Cdc42 and WT Raichu Rac and constitutively active, Raichu Rac G12V (V12), were kindly provided by M. Matsuda (Osaka University, Osaka, Japan) [39], and consist of YFP, the CRIB domain of human PAK1 (amino acids 68-150), human Cdc42 or Rac1 (amino acids 1-176) and CFP cloned into a derivative of the pCAGGS vector [91]. The genetically encoded photoactivatable Rac was kindly provided by Dr. Klaus Hahn (UNC Chapel Hill, NC, USA) [36], and was activated by exposure to room lighting for 10min (Fig 1C) or acute activation with 458nm light for approximately 1 min using confocal imaging as described below. Full-length chicken paxillin-GFP cloned into the pEGFP-N3 vector (Clontech) has been described previously [92] and GFP was replaced by mCherry from R. Tsien (University of San Diego, San Diego, CA) [93].

Confocal imaging and analysis
Confocal images were acquired on a laser scanning Olympus Fluoview 1000 microscope (IX81 base) equipped with a 60X/1.35 NA oil objective (Olympus). Green fluorescent probes (GFP) were excited with a 488 nm laser line of a multi-Argon laser, while red probes (Rhodamine) were excited with the 543 nm laser line of a He-Ne laser; and the far-red probe Alexa647 was excited with the 635 nm line of a diode laser. Fluorescence emission was collected using the following dichroic mirror/filter combinations: SDM560/BA505-525 (GFP), SDM640/BA560-620 (mCherry, RFP, Alexa568 and Rhodamine) and BA655-755 (Alexa647). Fluorescent images were collected in a Z-stack and in sequential line scanning mode using Olympus Fluoview software. Image analysis was performed with Image J software on max intensity Z projections.

Statistical analysis
Statistical analysis was performed using Sigma Plot 11.0. In each case, we first assessed whether the data assumed a normalized distribution using the Shapiro-Wilk test and also if the two data sets under comparison presented with equal variance. If the data assumed a normalized distribution with equal variance, the t-test was used to assess statistical significance. However, if the data failed to have either a normalized distribution or equal variance, we used a nonparametric statistical analysis, the Mann-Whitney Rank Sum Test. We have indicated the test used to determine statistical significance in each figure legend.

FRET imaging and analysis
Confocal images were acquired with a Fluoview 1000 microscope using a 60× 1.35 NA oil objective (Olympus). For dual-emission ratio imaging of Raichu-Cdc42/Rac, we used the 458 nm line of a multi-Argon ion laser for CFP excitation and SDM510/BA480-495 (CFP) and BA535-565 (YFP/FRET) filters for the collection of CFP and YFP fluorescence emission, respectively. Ratiometric FRET images of max intensity Z projections of the acquired CFP and YFP images were created using the Biosensor Processing Software 2.1 available from the Danuser laboratory (University of Texas Southwestern, Dallas, TX). Image analysis of the resulting FRET images was performed with Image J software.

Total Internal reflection microscopy (TIRF)
TIRF images were acquired with an Olympus inverted microscope (IX70) fitted with a Ludl modular automation controller (Ludl Electronic Products, Hawthorne, NY) using an Olympus 60X/1.45NA oil Plan Apochromatic TIRFM objective and charge-coupled device (CCD) camera (Retiga Exi, QImaging). CHO.K1 cells were plated on 2ug/ml fibronectin-coated glass coverslips and imaged in HyClone CCM1 media (GE Healthcare) at 37˚C within~1hour of plating. Images were acquired with Metamorph Imaging Software (Molecular Devises, Sunnyvale, CA). Green fluorescent probes (GFP) were excited with a 488 nm laser line of a multi-Argon laser, while red probes were excited with a 561nm diode laser. Images were acquired every 5 sec for a total of 10 min. However, the two different shRNAs targeting FDG4 did not downregulate expression of the other GEFs, either β-pix or Vav2, and both shRNAs similarly reduced spine length (Fig 2D), and the involvement of Cdc42 regulation in spine length was further confirmed by a FRET biosensor (Fig 2E). Arhgdig, which was upregulated during neuronal development similar to Arhgdia (Fig 1B), similarly affected mature spine morphology (Fig 5). (TIF) S1 Video. Rac photoactivation promotes spine precursor formation. A DIV-14 primary rat hippocampal neuron transfected with a genetically encoded photoactivatable Rac was exposed to 458 nm light at the indicated region of photoactivation at t = 0 sec (continuous slow scan speed for~49 sec), and the resulting spine precursor formation was monitored by timelapse confocal microscopy. The magenta reference image represents neuron morphology at the time of photoactivation and is superimposed with GFP timelapse images before (t = -120 sec to 0 sec) and after photoactivation (t = 0 sec to 300 sec).