Dynamic recruitment of the curvature-sensitive protein ArhGAP44 to nanoscale membrane deformations limits exploratory filopodia initiation in neurons.

In the vertebrate central nervous system, exploratory filopodia transiently form on dendritic branches to sample the neuronal environment and initiate new trans-neuronal contacts. While much is known about the molecules that control filopodia extension and subsequent maturation into functional synapses, the mechanisms that regulate initiation of these dynamic, actin-rich structures have remained elusive. Here, we find that filopodia initiation is suppressed by recruitment of ArhGAP44 to actin-patches that seed filopodia. Recruitment is mediated by binding of a membrane curvature-sensing ArhGAP44 N-BAR domain to plasma membrane sections that were deformed inward by acto-myosin mediated contractile forces. A GAP domain in ArhGAP44 triggers local Rac-GTP hydrolysis, thus reducing actin polymerization required for filopodia formation. Additionally, ArhGAP44 expression increases during neuronal development, concurrent with a decrease in the rate of filopodia formation. Together, our data reveals a local auto-regulatory mechanism that limits initiation of filopodia via protein recruitment to nanoscale membrane deformations.

Introduction

During the development of the central nervous system, neuronal progenitor cells proliferate, migrate, and finally differentiate into functional units to form a multi-cellular neuronal network (Ayala et al., 2007). In culture, differentiation of individual neurons occurs in a stereotypic pattern starting with the formation of an axon, followed by the creation of an elaborate dendritic tree and culminating with the initiation and maturation of trans-neuronal synaptic contacts (Dotti et al., 1988). The formation of synaptic connections is often facilitated by dynamic exploratory filopodia that extend out of thicker dendritic branches to sample the environment and thereby increase the probability that selective pre-to-postsynaptic connections are established (Ziv and Smith, 1996; Marrs et al., 2001). Exploratory filopodia are dynamic finger-like membrane structures containing actin cables formed out of actin patches along the dendritic shaft (Lau et al., 1999; Matus, 2000). Extension of filopodia is driven by local activation of formins, Ena/VASP proteins, small GTPases, and likely other steps (Krugmann et al., 2001; Lebrand et al., 2004). Intriguingly, the frequency of filopodia formation dramatically drops once high synapse density is established (Ziv and Smith, 1996), suggesting that the initiation of these structures is controlled by opposing negative regulators. However, the identity of these inhibitors has remained elusive.

Here, we provide evidence that ArhGAP44 limits the initiation of exploratory dendritic filopodia. Consistent with previous reports, we observe that formation of actin patches precedes filopodia extension. Notably, we find that within actin patches Myosin II-mediated pulling on plasma membrane (PM)-associated actin cables induces highly curved membrane sections that trigger ArhGAP44 recruitment. The resulting enrichment of ArhGAP44 then reduces local actin polymerization due to the Rac GAP activity of ArhGAP44, preventing the formation of filopodia. ArhGAP44 expression increases as the neuronal network is established and the frequency of exploratory filopodia formation is diminished, suggesting that ArhGAP44 may facilitate the transition of neurons from a dynamic exploratory mode to a mature more static state, a hallmark of nervous system development.

Results

ArhGAP44 is predominantly expressed in the brain and increases with age

Formation of filopodia depends on proteins that regulate polymerization of actin filaments (Krugmann et al., 2001; Lebrand et al., 2004). To identify new regulators of exploratory dendritic filopodia formation, we performed a literature search and identified 286 genes that were previously associated either directly or indirectly with actin reorganization. We then clustered these genes according to expression pattern using published microarray data and found 89 of the 286 genes to be expressed predominantly in neuronal tissues (Figure 1—figure supplement 1A and ‘Materials and methods’ and Supplementary file 1). As we were interested in regulators of actin dynamics selective for the brain, we ranked these 89 genes for high expression in the brain compared to the spinal cord and tested the validity of the ranking using a set of control genes expressed only in one of the respective tissues (Figure 1—figure supplement 1B and Table 1 and ‘Materials and methods’). Among the five actin regulators with the highest brain vs spinal cord ratio, we found ArhGAP44 (also known as Rich2 or Nadrin2), a membrane-curvature-sensing GTPase Activating Protein (GAP) selective for the small Rho GTPases Rac1 and Cdc42 (Richnau and Aspenstrom, 2001). In previous studies, ArhGAP44 has been associated with the maintenance of apical microvilli in polarized epithelial cells as well as postsynaptic maturation and vesicle release (Rollason et al., 2009; Nahm et al., 2010; Raynaud et al., 2013, 2014). Considering the role of Rho GTPases during filopodia formation (Krugmann et al., 2001), we decided to further investigate the role of ArhGAP44 in developing neurons.

Figure 1—figure supplement 1.

Cluster analysis of putative actin-regulating genes.

(A) Identification of neuron-enriched putative regulators of actin dynamics. 286 genes identified in a NCBI data-search (see ‘Materials and methods’) were clustered according to expression pattern across 74 different tissue samples. A group of 89 genes enriched in neuronal tissue is highlighted in red. (B) ArhGAP44 is enriched in the whole brain but not the spinal cord. The 89 genes identified to be enriched in neuronal tissues were sorted according to their relative whole brain vs. spinal cord expression ratio. Reference genes (Table 1) expressed selectively in Schwann cells (green) or the brain (red) is shown below the ranked list.

DOI: http://dx.doi.org/10.7554/eLife.03116.005

Table 1.

Reference genes expressed predominantly in the adult brain or in the spinal cord

DOI: http://dx.doi.org/10.7554/eLife.03116.003

Gene	Adult brain	Spinal cord	Ratio	Reference
PMP22*	720.55	2516.35	0.286347	(Snipes et al., 1992)
MPZ*	5.5	181.5	0.030303	(Su et al., 1993)
BSN†	56.7	3.45	16.43478	(tom Dieck et al., 1998)
GRIA2†	355.1	11.55	30.74459	(Martin et al., 1993)

enriched in the spinal cord.

enriched in the adult brain.

To validate the microarray expression pattern of ArhGAP44, we first isolated various organs and brain regions and probed protein levels with an antibody directed against ArhGAP44. Consistent with previous work (Richnau and Aspenstrom, 2001), western blot analysis showed expression of ArhGAP44 in the brain while being below detection level in all other tested organs (Figure 1A and Figure 1—figure supplement 2A). Within the brain, immunoblotting directed against ArhGAP44 showed increased protein levels in the frontal cortex and olfactory bulb (Figure 1B). The same expression pattern was found in sagittal brain sections stained against ArhGAP44 (Figure 1—figure supplement 2B).

Figure 1.

The brain-enriched ArhGAP44 regulates exploratory dendritic filopodia formation.

(A) Microarray data for ArhGAP44 across 74 tissue samples show predominant expression in neuronal tissues (red box). Immunoblot to the right shows ArhGAP44 expressed in the brain while being below detection level in other tissues. Note that individual tissue samples are likely composed of a variety of different cell types. (B) Total extracts of individual brain regions probed with an antibody directed against ArhGAP44 (top) and tubulin (bottom). (C) Expression of ArhGAP44 in cultured hippocampal neurons. Total extracts of individual neuronal samples were isolated at DIV3, DIV10, and DIV17 and probed with an antibody directed against ArhGAP44 (top), the synaptic proteins PSD95 and Bassoon (middle lanes), and tubulin (bottom). (D) Scanning electron micrographs of cultured neurons. Dendritic surface structures are classified based on morphology as convoluted nodes (red), elongated protrusion without contact (light gray), or stubby protrusions that contact adjacent neurites (dark gray). Analysis is shown for DIV3 (n = 10 neurons, 2 independent experiments), DIV10 (n = 10 neurons, 2 independent experiments), and DIV17 (n = 9 neurons, 2 independent experiments). (E) Overexpression of ArhGAP44 decreases filopodia density. Representative examples of neurons (green) stained with anti-MAP2 antibody (red) are shown. Analysis of filopodia density upon overexpression of control (black; n = 67 neurons, 3 independent experiments), ArhGAP44(wt) (dark blue; n = 73 neurons, 3 independent experiments), and mutant ArhGAP44(R291M) (light blue; n = 53 neurons, 3 independent experiments) is shown below. (F) Rac-GAP activity of individual ArhGAP44 mutants. Note that ArhGAP44(R291M) shows higher GTP-Rac1 hydrolysis than control. (G) Knockdown of ArhGAP44 increases filopodia density. Analysis of filopodia density upon expression of control (black; n = 67 neurons, 3 independent experiments), control siRNA (gray; n = 67 neurons, 3 independent experiments), and knockdown of ArhGAP44 (siRNA #1, light yellow; n = 83 neurons; siRNA #2, dark yellow; n = 85 neurons; both 3 independent experiments) is shown below. (H) Control western blot analysis testing the effectiveness of individual siRNA pools. Scale bars (D), 1 µm; (E and F), 20 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.004

(A) Identification of neuron-enriched putative regulators of actin dynamics. 286 genes identified in a NCBI data-search (see ‘Materials and methods’) were clustered according to expression pattern across 74 different tissue samples. A group of 89 genes enriched in neuronal tissue is highlighted in red. (B) ArhGAP44 is enriched in the whole brain but not the spinal cord. The 89 genes identified to be enriched in neuronal tissues were sorted according to their relative whole brain vs. spinal cord expression ratio. Reference genes (Table 1) expressed selectively in Schwann cells (green) or the brain (red) is shown below the ranked list.

DOI: http://dx.doi.org/10.7554/eLife.03116.005

(A) Loading control of tissues used in Figure 1A. Same amounts of total protein were loaded into the different lanes. Note the non-uniform protein composition. (B) ArhGAP44 is enriched in the cerebral cortex and olfactory bulb. Adult rat brain was sectioned and stained with DAPI (blue) and an antibody directed against ArhGAP44 (yellow).

DOI: http://dx.doi.org/10.7554/eLife.03116.006

(A) ArhGAP44 is expressed in the adult but not the fetal human brain. The 89 genes previously identified to be enriched in neuronal tissues were sorted for strong expression in the adult brain vs. fetal brain. Reference genes (Table 2) expressed during neuronal migration (green), trans-synaptic contact formation (yellow), and synapse maturation (red) are shown along the x-axis. Note that ArhGAP44 (red) is expressed 20-fold higher in the adult compared to the fetal brain. (B) Expression levels of ArhGAP44 in the prefrontal cortex increase with age. Individual samples from rat prefrontal cortex were fixed at the age of 6, 18, 24, 34, and 39 months and stained against ArhGAP44. Scale bar (B), 20 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.007

Scanning Electron Micrographs of neurons fixed at DIV3 (n = 10 images, 2 independent experiments), DIV10 (n = 10 images, 2 independent experiments), and DIV17 (n = 9 images, 2 independent experiments) as well as quantification of total dendrite length are shown. Scale bar, 20 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.008

(A) Dendritic surface analysis. Neurons were imaged at low resolution and individual neurons were identified. The proximal 100 µm of dendritic arbors were then imaged at high resolution (boxes #1–#7) and surface structures in high resolution images were then grouped as nodes, elongated protrusions, and stubby protrusions. Note in image #6 (red box) protrusion emerging from the dendritic surface. (B) Examples of nodes that form on the dendritic surface. Scanning electron micrographs of neurons that were fixed on DIV11. Individual nodes are highlighted in yellow. Scale bar, 500 nm.

DOI: http://dx.doi.org/10.7554/eLife.03116.009

Cells were transfected with control plasmid, ArhGAP44(wt) and ArhGAP44(291), respectively, and cell morphology was assessed 12 hr (n = 161 neurons, 3 independent experiments), 24 hr (n = 112 neurons, 3 independent experiments), and 48 hr (n = 160 neurons, 3 independent experiments) after transfection. Neurons were classified based on cell morphology showing no effect (white), reduced protrusion density (red), and varicosity formation (black). Scale bar, 10 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.010

Figure 1—figure supplement 2.

Ponceau loading control of various tissues and ArhGAP44 protein expression in the brain.

(A) Loading control of tissues used in Figure 1A. Same amounts of total protein were loaded into the different lanes. Note the non-uniform protein composition. (B) ArhGAP44 is enriched in the cerebral cortex and olfactory bulb. Adult rat brain was sectioned and stained with DAPI (blue) and an antibody directed against ArhGAP44 (yellow).

DOI: http://dx.doi.org/10.7554/eLife.03116.006

Table 2.

Reference genes expressed predominantly in the adult or in the fetal brain

DOI: http://dx.doi.org/10.7554/eLife.03116.011

Gene	Adult brain	Fetal brain	Ratio	Reference
RELN*	34.85	93.35	0.373326	(D'Arcangelo et al., 1995)
DCX*	10.55	2577	0.004094	(des Portes et al., 1998)
NRXN1*	72.1	143.15	0.503667	(Ushkaryov et al., 1992)
NLGN1*	7.5	24.7	0.303644	(Ichtchenko et al., 1995)
CAMK2B†	818.15	321.25	2.54677	(Omkumar et al., 1996)
MUNC13†	26.3	9.75	2.697436	(Betz et al., 1998)
MECP2†	818.15	321.25	2.54677	(Amir et al., 1999)
PSD95†	362.1	12.7	28.51181	(Kornau et al., 1995)
PACSIN1†	90.8	20.1	4.517413	(Qualmann et al., 1999)

enriched in the fetal brain.

enriched in the adult brain.

We then explored the expression of ArhGAP44 in the brain during development and found ArhGAP44 among the genes with the highest adult-to-fetal ratio (Figure 1—figure supplement 3, and Table 2 and ‘Materials and methods’). To validate the observed increase in ArhGAP44 expression with age, we measured protein levels in rat primary hippocampal neurons by western blot, using samples isolated during neurite extension (3 days in vitro, i.e., DIV3), the peak of exploratory filopodia formation (DIV10), and after initial synaptic contacts were formed (DIV17). Consistent with the micro-array data, ArhGAP44 protein levels increased with time (Figure 1C).

Figure 1—figure supplement 3.

ArhGAP44 expression increases over time.

(A) ArhGAP44 is expressed in the adult but not the fetal human brain. The 89 genes previously identified to be enriched in neuronal tissues were sorted for strong expression in the adult brain vs. fetal brain. Reference genes (Table 2) expressed during neuronal migration (green), trans-synaptic contact formation (yellow), and synapse maturation (red) are shown along the x-axis. Note that ArhGAP44 (red) is expressed 20-fold higher in the adult compared to the fetal brain. (B) Expression levels of ArhGAP44 in the prefrontal cortex increase with age. Individual samples from rat prefrontal cortex were fixed at the age of 6, 18, 24, 34, and 39 months and stained against ArhGAP44. Scale bar (B), 20 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.007

As N-BAR domains present in ArhGAP44 and other proteins can bind to inward-curved plasma membranes (Galic et al., 2012), we used high-resolution Field Emission Scanning Electron Microscopy to examine potential changes to local neuronal membrane curvature during maturation. As expected, we observed an increase in overall neuronal complexity with time (Figure 1—figure supplement 4). High magnification micrographs further showed convoluted membrane sections (i.e., convoluted nodes) along the dendrite at DIV10 that may reflect such local regions of high curvature (Figure 1D and Figure 1—figure supplement 5).

Figure 1—figure supplement 4.

Neuronal complexity increases over time.

Scanning Electron Micrographs of neurons fixed at DIV3 (n = 10 images, 2 independent experiments), DIV10 (n = 10 images, 2 independent experiments), and DIV17 (n = 9 images, 2 independent experiments) as well as quantification of total dendrite length are shown. Scale bar, 20 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.008

Figure 1—figure supplement 5.

Electron micrographs of neurons.

(A) Dendritic surface analysis. Neurons were imaged at low resolution and individual neurons were identified. The proximal 100 µm of dendritic arbors were then imaged at high resolution (boxes #1–#7) and surface structures in high resolution images were then grouped as nodes, elongated protrusions, and stubby protrusions. Note in image #6 (red box) protrusion emerging from the dendritic surface. (B) Examples of nodes that form on the dendritic surface. Scanning electron micrographs of neurons that were fixed on DIV11. Individual nodes are highlighted in yellow. Scale bar, 500 nm.

DOI: http://dx.doi.org/10.7554/eLife.03116.009

ArhGAP44 negatively regulates filopodia density

To determine the function of ArhGAP44 in neurons, we transfected primary rat hippocampal neurons. Compared to control cells, overexpression of ArhGAP44 caused a significant reduction in the density of dendritic filopodia at DIV12 (Figure 1E, dark blue). Prolonged expression increased the fraction of cells forming varicosities, likely due to increased RhoGAP activity associated with elevated ArhGAP44 levels (Figure 1—figure supplement 6). To decrease the GAP activity of ArhGAP44, we substituted a conserved arginine in the catalytic cleft with a methionine, which was shown to reduce but not eliminate the enzymatic activity of GAP proteins (Muller et al., 1997; Graham et al., 1999). Consistent with partial activity, expression of the ArhGAP44(R291M) mutant reduced but did not abolish GTP hydrolysis of the small GTPase Rac1 (Figure 1F), showed a milder loss in filopodia density (Figure 1E, light blue), and delayed the onset of varicosity formation (Figure 1—figure supplement 6).

Figure 1—figure supplement 6.

Overexpression phenotypes of ArhGAP44 in cultured neurons.

Cells were transfected with control plasmid, ArhGAP44(wt) and ArhGAP44(291), respectively, and cell morphology was assessed 12 hr (n = 161 neurons, 3 independent experiments), 24 hr (n = 112 neurons, 3 independent experiments), and 48 hr (n = 160 neurons, 3 independent experiments) after transfection. Neurons were classified based on cell morphology showing no effect (white), reduced protrusion density (red), and varicosity formation (black). Scale bar, 10 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.010

To further validate the observed effect on filopodia density, neurons were transfected with siRNA directed against ArhGAP44. Markedly, knockdown of ArhGAP44 augmented the density of dendritic filopodia (Figure 1G, yellow). Control experiments using a second siRNA pool directed against a different region of ArhGAP44 mRNA confirmed the phenotype, thus showing that siRNA knockdown and overexpression of ArhGAP44 have opposing effects. Both siRNAs were effective since protein levels of ArhGAP44 were significantly reduced in cells co-transfected with either one of the siRNA pools directed against ArhGAP44 (Figure 1H).

ArhGAP44 negatively regulates de novo filopodia formation

As dendritic filopodia frequently extend, reorient, and collapse (Ziv and Smith, 1996), filopodia density reflects the product of de novo formation frequency and stabilization rate. To determine which of these parameters are controlled by ArhGAP44, we performed time-lapse imaging of dendritic filopodia dynamics (Figure 2A and Figure 2—figure supplement 1). Compared to control, neither knockdown of ArhGAP44 nor expression of ArhGAP44(R291M) showed significant changes in the density of static protrusions that persisted longer than 10 min (Figure 2B and Figure 2—figure supplement 2A and Video 1) In contrast, expression of ArhGAP44(R291M) reduced (Figure 2B, blue and Video 2), whereas knockdown of ArhGAP44 increased (Figure 2B, yellow and Video 3 and Video 4) the formation of dynamic protrusions. Most of these newly formed protrusions were short-lived, while stabilization of extending protrusions or collapse of previously static protrusions was observed only infrequently (Figure 2B and Figure 2—figure supplement 2B). Together, these results argue that ArhGAP44 primarily limits filopodia formation with little effect on the stabilization of existing filopodia.

Figure 2.

Knockdown of ArhGAP44 and overexpression of Rac both increase de novo filopodia formation.

(A) Color-coded overlay of image-series. Note increased protrusion dynamics upon knockdown of ArhGAP44. (B and C) ArhGAP44 negatively regulates de novo protrusions. Analysis of protrusion dynamics in neurons transfected with control (black; n = 85 protrusions, 10 neurons, 3 independent experiments), ArhGAP44(R291M) (blue; n = 106 protrusions, 11 neurons, 3 independent experiments), and upon knockdown of ArhGAP44 (yellow; siRNA #1 = 131 protrusions, 12 neurons, 3 independent experiments; siRNA #2 = 215 protrusions, 15 neurons, 3 independent experiments). Compared to controls, overexpression of ArhGAP44(R291M) reduces while knockdown of ArhGAP44 increases formation of transient protrusion (B). Both increase node formation (C). (D) Color-coded overlay of image-series. Note increased protrusion dynamics upon overexpression of Rac1. (E and F) Co-expression of Rac1 reverses ArhGAP44-dependent reduction in protrusion dynamics. Compared to control (black; n = 85 protrusions, 10 neurons, 3 independent experiments), overexpression of Rac1 (red, n = 123 protrusions, 9 neurons, 3 independent experiments) increases the formation of transient protrusion (E) and nodes (F). For both parameters, co-expression of Rac1 with ArhGAP44(R291M) (purple, n = 126 protrusions, 12 neurons, 3 independent experiments) can compensate for the reduction observed for ArhGAP44(R291M) alone (blue, n = 106 protrusions, 11 neurons, 3 independent experiments). Scale bars (A and D), 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.012

Analysis method used to measure filopodia kinetics. A set of images (left column) was binarized (middle column) and individual images were super-imposed to create a color-coded overlay of the image-series (right image). Scale bar, 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.013

(A) Protrusions that persist throughout the acquisition interval were considered static. (B) Dynamic protrusions were divided into three groups: protrusions that extend and retract during acquisition (top panel); protrusions that extend and remain extended for at least 3 min (middle panel); and protrusions that persisted for at least 3 min before collapsing (bottom panel). Scale bars (A and B), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.014

(A) Nodes are identified as thickened regions in the dendritic shaft. A section of the dendrite with a node but no filopodia (red arrow) and a node with a filopodium (white arrow) are shown enlarged. (B) Dendritic nodes (red arrow) are often observed without a subsequent extension of a filopodia. Scale bars (A), 20 µm; (B), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.015

(A) Time lapse of a dendritic branch from which a protrusion (red) emerges and collapses. Note that not all nodes (yellow arrows) can be identified when using a 60 s interval (e.g., compare columns 2 and 4). (B) The majority of protrusions emerge from nodes. Quantification of percentage of filopodia that emerge from nodes using acquisition intervals of 60 s (black; 45 nodes form 11 neurons, 3 independent experiments) and 15 s (red; 75 nodes form 12 neurons, 3 independent experiments), respectively. Note that the fraction of protrusions emerging from nodes increases with shorter intervals as nodes can be short-lived. Scale bar (A), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.016

(A) ArhGAP44 is a RhoGAP with dual selectivity for Rac1 and Cdc42. GTPase pull-down assay show hydrolysis of GTP-Cdc42 (green) and GTP-Rac1 (red) by ArhGAP44. (B) Overexpression of Rac1 but not of Cdc42 increases protrusion density. Neurons were transfected with the wild type form of human Rac1 and Cdc42. Note that only Rac1(red) increased protrusion density. (C) Rac1 but not Cdc42 reverse ArhGAP44(R291M)-dependent slow-down in protrusion dynamics. Neurons were transfected either with a control plasmid (black, n = 18 neurons, 3 independent experiments), Cdc42 (green, n = 15 neurons, 3 independent experiments), Rac1 (red, n = 13 neurons, 3 independent experiments), ArhGAP44(R291M) (blue, n = 18 neurons, 3 independent experiments), or co-transfected with ArhGAP44(R291M) together with Cdc42 (green-blue, n = 11 neurons, 3 independent experiments) or Rac1 (violet, n = 13 neurons, 3 independent experiments). (D) Overexpression of Rac1 but not of Cdc42 triggered node formation observed upon knockdown of ArhGAP44. Time-lapse images of neurons co-transfected with a siRNA directed against ArhGAP44 together with a fluorescence marker (top), expressing wild-type Rac1 (second panel), expressing wild-type Cdc42 (third panel), and expressing control plasmid (bottom). Note that ectopic nodes are formed from which filopodia emerge upon knockdown of ArhGAP44 and overexpression of Rac1 but not after expression of Cdc42. Scale bars (D, left panels), 20 µm; (D, right panels), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.017

Figure 2—figure supplement 1.

Filopodia analysis tool and GTPase overexpression control.

Analysis method used to measure filopodia kinetics. A set of images (left column) was binarized (middle column) and individual images were super-imposed to create a color-coded overlay of the image-series (right image). Scale bar, 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.013

Figure 2—figure supplement 2.

Examples of prototypic dendritic protrusions.

(A) Protrusions that persist throughout the acquisition interval were considered static. (B) Dynamic protrusions were divided into three groups: protrusions that extend and retract during acquisition (top panel); protrusions that extend and remain extended for at least 3 min (middle panel); and protrusions that persisted for at least 3 min before collapsing (bottom panel). Scale bars (A and B), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.014

Video 1.

Example of a stable dendritic protrusion.

Neuron was transfected with a fluorescence marker at DIV11 and imaged 24 hr later. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 720× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.018

Video 2.

ArhGAP44(R291M) transfection causes increased dendritic node and reduced protrusion formation.

Neuron was transfected with ArhGAP44(R291M) at DIV11 and imaged 24 hr later. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.019

Video 3.

Knockdown of ArhGAP44 increases dendritic node and protrusion formation.

Neuron was co-transfected with diced RNA directed against ArhGAP44 and a cytosolic reference at DIV7 and imaged at DIV12. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.020

Video 4.

Knockdown of ArhGAP44 increases dendritic node formation.

Neuron was co-transfected with diced RNA directed against ArhGAP44 and a cytosolic reference at DIV7 and imaged at DIV12. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.021

For both, overexpression and knockdown of ArhGAP44, we further observed an increase in the number of transiently formed nodes along dendritic arbors (Figure 2C and Figure 2—figure supplement 3 and Videos 2–5). Intriguingly, we find that the majority (83% ± 7%) of de novo protrusions extended from such node-like structures (Figure 2—figure supplement 4 and Video 6). We thus consider these nodal structures to represent nascent filopodia sites, where filopodia formation is either initiated or aborted. Given their spacing and their presence at the same time in culture, they likely correlate with the convoluted nodes along dendrites seen in electron microscopy studies (Figure 1D).

Figure 2—figure supplement 3.

Examples of dendritic nodes.

(A) Nodes are identified as thickened regions in the dendritic shaft. A section of the dendrite with a node but no filopodia (red arrow) and a node with a filopodium (white arrow) are shown enlarged. (B) Dendritic nodes (red arrow) are often observed without a subsequent extension of a filopodia. Scale bars (A), 20 µm; (B), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.015

Figure 2—figure supplement 4.

Acquisition interval of 60 s is sufficient to detect dynamic protrusions but not all nodes.

(A) Time lapse of a dendritic branch from which a protrusion (red) emerges and collapses. Note that not all nodes (yellow arrows) can be identified when using a 60 s interval (e.g., compare columns 2 and 4). (B) The majority of protrusions emerge from nodes. Quantification of percentage of filopodia that emerge from nodes using acquisition intervals of 60 s (black; 45 nodes form 11 neurons, 3 independent experiments) and 15 s (red; 75 nodes form 12 neurons, 3 independent experiments), respectively. Note that the fraction of protrusions emerging from nodes increases with shorter intervals as nodes can be short-lived. Scale bar (A), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.016

Video 6.

Knockdown of ArhGAP44 triggers formation of protrusion from dendritic nodes.

Neuron was co-transfected with RNA directed against ArhGAP44 and a cytosolic reference at DIV7 and imaged at DIV12. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.023

Consistent with previous reports (Richnau and Aspenstrom, 2001), we find that the GAP domain of ArhGAP44 inhibits GTPase activity of Rac and Cdc42 (Figure 2—figure supplement 5A). We thus aimed to investigate to which extent the two possible targets of ArhGAP44 contribute to filopodia formation. We find that overexpression of wild-type Rac (human Rac1) but not Cdc42 phenocopied protrusion density (Figure 2—figure supplement 5B) as well as protrusion kinetics (Figure 2—figure supplement 5C and Videos 7–10) observed upon knockdown of ArhGAP44. Similar to knockdown of ArhGAP44, Rac1 overexpression increased protrusion dynamics and node formation (Figure 2D–F and Figure 2—figure supplement 5D and Videos 7,8). These observations are consistent with previous work showing Rac activity associated with increased actin patch formation and filopodia dynamics in axons (Spillane et al., 2012) and dendrites (Korobova and Svitkina, 2010; Cheadle and Biederer, 2012). We thus considered that ArhGAP44 might act by inhibiting Rac to limit initiation of exploratory filopodia formation.

Figure 2—figure supplement 5.

The ArhGAP44 knockdown phenotype is phenocopied by the small GTPase Rac1 but not Cdc42.

(A) ArhGAP44 is a RhoGAP with dual selectivity for Rac1 and Cdc42. GTPase pull-down assay show hydrolysis of GTP-Cdc42 (green) and GTP-Rac1 (red) by ArhGAP44. (B) Overexpression of Rac1 but not of Cdc42 increases protrusion density. Neurons were transfected with the wild type form of human Rac1 and Cdc42. Note that only Rac1(red) increased protrusion density. (C) Rac1 but not Cdc42 reverse ArhGAP44(R291M)-dependent slow-down in protrusion dynamics. Neurons were transfected either with a control plasmid (black, n = 18 neurons, 3 independent experiments), Cdc42 (green, n = 15 neurons, 3 independent experiments), Rac1 (red, n = 13 neurons, 3 independent experiments), ArhGAP44(R291M) (blue, n = 18 neurons, 3 independent experiments), or co-transfected with ArhGAP44(R291M) together with Cdc42 (green-blue, n = 11 neurons, 3 independent experiments) or Rac1 (violet, n = 13 neurons, 3 independent experiments). (D) Overexpression of Rac1 but not of Cdc42 triggered node formation observed upon knockdown of ArhGAP44. Time-lapse images of neurons co-transfected with a siRNA directed against ArhGAP44 together with a fluorescence marker (top), expressing wild-type Rac1 (second panel), expressing wild-type Cdc42 (third panel), and expressing control plasmid (bottom). Note that ectopic nodes are formed from which filopodia emerge upon knockdown of ArhGAP44 and overexpression of Rac1 but not after expression of Cdc42. Scale bars (D, left panels), 20 µm; (D, right panels), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.017

Rac1 overexpression causes abnormal dendritic node and protrusion formation.

Neuron was transfected with Rac1(wt) at DIV11 and imaged 24 hr later. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.024

Neuron was transfected with Rac1(wt) at DIV11 and imaged 24 hr later. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.025

Cdc42 overexpression causes abnormal dendritic protrusion formation.

Neuron was transfected with Cdc42(wt) at DIV11 and imaged 24 hr later. Note the elongation of a preexisting dendritic protrusion. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.026

Neuron was transfected with Cdc42(wt) at DIV11 and imaged 24 hr later. Note the elongated filopodia emerging from the dendrite. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.027

To test this hypothesis, we performed a synthetic compensation experiment, co-expressing Rac1 with ArhGAP44(R291M). Although the resulting protrusions were shorter than control filopodia, co-expression of ArhGAP44(R291M) together with Rac1 reversed the observed decrease in the frequency of protrusion formation and node formation caused by expression of ArhGAP44(R291M) (Figure 2E,F purple and Video 11).

Video 11.

Rac1 synthetically rescues ArhGAP44(R291M)-dependent reduction in protrusion formation.

Neuron was co-transfected with Rac1(wt) and ArhGAP44(R291M) at DIV11 and imaged 24 hr later. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.028

ArhGAP44 localizes to patches that precede filopodia extension

To analyze the subcellular protein localization, we cultured hippocampal neurons and stained against endogenous ArhGAP44. We found ArhGAP44 to be absent from the nucleus and present in patches along dendrites (Figure 3—figure supplements 1 and 2A). We then expressed fluorescently tagged ArhGAP44 in cultured neurons. Like the endogenous protein, fluorescently tagged ArhGAP44 was excluded from the nucleus, distributed uniformly through the cytosol, and formed distinct ArhGAP44 patches along dendrites (Figure 3—figure supplement 2B,C), arguing that the fluorescent tag did not interfere with its localization.

Figure 3—figure supplement 2.

ArhGAP44 localization in neurons.

(A) Immunostaining directed against endogenous ArhGAP44 in cultured neurons (top). Primary hippocampal neurons were cultured for 11 days and stained with antibodies directed against ArhGAP44. ArhGAP44 is absent from the nucleus and enriched in distinct patches along the dendritic shaft (red arrows). (B) Overexpressed ArhGAP44 is absent from the nucleus and enriched at distinct patches along the dendritic arbors (bottom). Primary hippocampal neurons were transfected with ArhGAP44(R291M) (red) and a cytosolic reference (green) at DIV10, and imaged 24 hr after transfection. ArhGAP44(R291M) is absent from the nucleus (bottom panels), but present at distinct patches along the dendrite (top panels, red arrow). (C) Overexpression of the isolated N-BAR domain of ArhGAP44 for 48 hr causes bright protein aggregates that are distinct from ArhGAP44 patches. Colocalization analysis (right panels) indicates that expression of ArhGAP44 for extended time causes formation of bright protein aggregates (yellow, n = 24 aggregates, 9 neurons, 2 independent experiments) that do not co-localize with actin patches (white, 34 actin patches form 25 neurons; 3 independent experiments). (D) Nodes that form along the dendrite surface display highly curved membrane sections. Average diameter of individual membrane ruffles in nodes is shown next to it. Scale bars (A and B), 20 µm; (C), 2 µm; (D), 500 nm.

DOI: http://dx.doi.org/10.7554/eLife.03116.031

To investigate the molecular mechanism that caused ArhGAP44 patch formation, we compared the subcellular localization of various deletion mutants of ArhGAP44 (Figure 3A). ArhGAP44 contains an N-BAR domain that has been reported to bind to positively (i.e., inward) curved lipid membranes (Peter et al., 2004). Full-length protein and the isolated N-BAR domain of ArhGAP44 both enriched in patches along dendritic arbors, while deletion of the amino-terminal amphipathic helix (ΔN-BAR), a critical sequence motif for binding and stabilization of curved membranes (Peter et al., 2004), showed no enrichment over a cytosolic reference (Figure 3B). Intriguingly, none of the ArhGAP44 constructs enriched in extended filopodia (Figure 3C). Time-lapse imaging showed that filopodia often emerged from ArhGAP44 patches (Figure 3D). We considered that ArhGAP44 patches might reflect the convoluted, node-like dendritic membrane sections we previously observed in electron micrographs (Figure 1D) and fluorescence images (Figure 2—figure supplement 3A). Consistently, measurement of individual membrane folds within nodes (Figure 3—figure supplement 2D) showed sufficient deformation to trigger curvature-dependent protein recruitment to the plasma membrane (Bhatia et al., 2009). Together with the previous overexpression and knockdown experiments (Figure 2A,B), the transient localization of ArhGAP44 to patches but not to extended filopodia argues that ArhGAP44 limits initiation rather than elongation of newly formed filopodia.

Figure 3.

ArhGAP44 recruitment to dendritic nodes precedes filopodia extension.

(A) ArhGAP44 protein structure and deletion mutants. ArhGAP44 is composed of an amino-terminal curvature-sensing N-BAR domain (AA 1–254), followed by the RhoGAP domain (AA 255–445) and a stretch of ∼350 amino acids (AA 445–818) with no annotated domain structure. Full length (left, dark red), the isolated N-BAR domain (middle, red), and the N-BAR domain lacking the initial 18 amino acids (right, white) were tested. (B) Enrichment of full length (dark red; 25 patches, 12 neurons, 3 independent experiments) and the isolated N-BAR domain of ArhGAP44 (red, n = 34 patches, 11 neurons, 3 independent experiments) in dendritic patches. No enrichment is observed for the N-BAR domain lacking the first 18AA that encode an amphipathic helix critical for curvature sensing (white; 20 patches, 11 neurons, 3 independent experiments). Note, ArhGAP44(R291M) was used to reduce the compromised cell health caused by overexpression of equal levels of active wild-type protein. (C) No enrichment of full length (dark red; 26 filopodia, 14 neurons, 3 independent experiments), the isolated N-BAR domain of ArhGAP44 (red, n = 40 filopodia, 13 neurons, 3 independent experiments), or the N-BAR domain lacking the first 18AA (white; 22 patches, 10 neurons, 3 independent experiments) in dendritic filopodia. (D) Filopodia emerge from ArhGAP44-rich patches. Scale bars (B, C, D), 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.029

(A) Western blot of whole rat brain with antibody directed against ArhGAP44(818AA) (unspecific bands are visible). (B) Increase immunostaining directed against ArhGAP44 after overexpression of ArhGAP44. Primary hippocampal neurons were transfected with a fluorescently tagged ArhGAP44(R291M) and fixed 24 hr later. Quantification of fluorescence intensity of non-transfected neurons (black, n = 20 cells, 2 independent experiments) and neurons transfected with ArhGAP44 (red, n = 12 cells, 2 independent experiments) after staining with an antibody directed against ArhGAP44 is shown below. (C) Control experiment showing reduced immunostaining of ArhGAP44 after knockdown of ArhGAP44. Primary hippocampal neurons were transfected at DIV10 with diced RNA directed against ArhGAP44 and a fluorescence reference (empty pEYFP plasmid) and fixed 48 hr later. Relative fluorescence intensity was measuerd in dendritic stretches and normalized to the adjacent background (arrows). Quantification of fluorescence intensity after staining with an antibody directed against ArhGAP44 of cells transfected with control siRNA (black, n = 16 cells, 2 independent experiments) and siRNA directed against ArhGAP44 (red, n = 18 cells, 2 independent experiments) is shown below. Scale bars (B and C), 20 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.030

(A) Immunostaining directed against endogenous ArhGAP44 in cultured neurons (top). Primary hippocampal neurons were cultured for 11 days and stained with antibodies directed against ArhGAP44. ArhGAP44 is absent from the nucleus and enriched in distinct patches along the dendritic shaft (red arrows). (B) Overexpressed ArhGAP44 is absent from the nucleus and enriched at distinct patches along the dendritic arbors (bottom). Primary hippocampal neurons were transfected with ArhGAP44(R291M) (red) and a cytosolic reference (green) at DIV10, and imaged 24 hr after transfection. ArhGAP44(R291M) is absent from the nucleus (bottom panels), but present at distinct patches along the dendrite (top panels, red arrow). (C) Overexpression of the isolated N-BAR domain of ArhGAP44 for 48 hr causes bright protein aggregates that are distinct from ArhGAP44 patches. Colocalization analysis (right panels) indicates that expression of ArhGAP44 for extended time causes formation of bright protein aggregates (yellow, n = 24 aggregates, 9 neurons, 2 independent experiments) that do not co-localize with actin patches (white, 34 actin patches form 25 neurons; 3 independent experiments). (D) Nodes that form along the dendrite surface display highly curved membrane sections. Average diameter of individual membrane ruffles in nodes is shown next to it. Scale bars (A and B), 20 µm; (C), 2 µm; (D), 500 nm.

DOI: http://dx.doi.org/10.7554/eLife.03116.031

Myosin-dependent contraction of PM-associated actin filaments induces membrane curvature and ArhGAP44 recruitment

Next, we aimed to investigate what caused the convoluted membrane surface at dendritic nodes. Consistent with previous reports (Lau et al., 1999; Spillane et al., 2012), ratio-imaging of the filamentous actin marker f-tractin (Johnson and Schell, 2009) to a cytosolic reference showed formation of actin patches that preceded extension of exploratory filopodia (Figure 4—figure supplement 1A and Video 12). This was the case in 89% ± 6% of all filopodia (Figure 4—figure supplement 1B). To directly test whether actin patches cause convoluted node-like PM subsections (Figure 3—figure supplement 2D), we performed correlative light and electron microscopy (Figure 4—figure supplement 2). Consistent with a role of actin in node-formation, we find that actin intensity in nodes was significantly higher compared to adjacent dendritic sections (Figure 4A). We then expressed f-tractin together with the isolated N-BAR domain of ArhGAP44 and found significant enrichment of f-tractin in ArhGAP44 patches (Figure 4B and Figure 4—figure supplement 3), arguing that ArhGAP44 and actin localize to the same structure. Consistently, the N-BAR domain of ArhGAP44 enriched in dendritic actin-patches by 80% ± 8% compared to adjacent dendritic regions (Figure 4—figure supplement 3B). To test for Myosin II-dependent contractile forces in ArhGAP44 patches, we co-expressed the isolated N-BAR domain of ArhGAP44 and non-muscle Myosin Heavy Chain IIB (NMHC-2B) and found a significant enrichment of NMHC-2B in ArhGAP44 patches (Figure 4C). Together, this suggests that myosin-dependent contraction of actin patches triggers inward membrane deformations within nodes to which ArhGAP44 is enriched. Consistently, we find increased staining for the phosphorylated form of myosin light chain (pMLC), a marker for active Myosin II (Tan et al., 1992), in dendritic actin patches (Figure 4D).

Figure 4—figure supplement 1.

Time-lapse of exploratory dendritic filopodia emerging from actin patches.

(A) Neurons were transfected at DIV10 with the filamentous actin marker f-tractin (red) together with a cytosolic reference (green) and imaged 24 hr later. Red arrows indicate relative accumulation of f-tractin compared to a cytosolic reference at nodes that form before filopodia initiation and persist after filopodia collapse. (B) Filopodia emerge from pre-existing actin patches. Quantification using an acquisition interval of 15 s show that 89% ± 6% of filopodia emerge from actin patches (24 patches form 10 neurons; 3 independent experiments). Scale bar (A), 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.034

Video 12.

Actin enriches at patches that precede exploratory filopodia initiation.

Neurons were transfected with a marker for filamentous actin (f-tractin, red) and a cytosolic reference (green). Note the formation of actin patches prior to filopodia formation (white boxes). Individual frames were taken every 5 s. Video is 50× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.032

Figure 4—figure supplement 2.

Workflow to identify individual neurons for correlative SEM/IF microscopy.

Example of pattern used for navigation on the glass slide is shown at the top. Gold deposit can be detected on the SEM using the back-scatter or inlens detector as well as in immunofluorescence. Scale bar, 10 mm.

DOI: http://dx.doi.org/10.7554/eLife.03116.035

Figure 4.

ArhGAP44 is recruited to convoluted dendritic PM sections enriched in polymerized actin and myosin.

(A) Correlative fluorescence and scanning electron microscopy shows actin enrichment in convoluted nodes that form along dendritic arbors. Quantification of the relative fluorescent intensity of phalloidin in individual nodes compared to adjacent dendritic sections is shown to the right. (B) F-tractin and ArhGAP44 co-localize. Neuron was transfected with the N-BAR domain of ArhGAP44 (red), f-tractin (blue), and a cytosolic reference (green). A magnified section (white box) and quantification of relative f-tractin intensity in ArhGAP44 patches are shown next to it (34 patches form 25 neurons; 3 independent experiments). (C) NMHC-2B and ArhGAP44 co-localize. Neurons were transfected with the N-BAR domain of ArhGAP44 (red), NMHC-2B (blue), and a cytosolic reference (green). Quantification of relative NMHC-2B intensity in ArhGAP44 patches is shown next to it (n = 20 patches form 13 neurons; 3 independent experiments). (D) Phosphorylated regulatory myosin light chain (green) is enriched in dendritic actin patches (red). Scale bars (A, C, D), 5 µm; (B), 10 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.033

(A) Neurons were transfected at DIV10 with the filamentous actin marker f-tractin (red) together with a cytosolic reference (green) and imaged 24 hr later. Red arrows indicate relative accumulation of f-tractin compared to a cytosolic reference at nodes that form before filopodia initiation and persist after filopodia collapse. (B) Filopodia emerge from pre-existing actin patches. Quantification using an acquisition interval of 15 s show that 89% ± 6% of filopodia emerge from actin patches (24 patches form 10 neurons; 3 independent experiments). Scale bar (A), 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.034

Example of pattern used for navigation on the glass slide is shown at the top. Gold deposit can be detected on the SEM using the back-scatter or inlens detector as well as in immunofluorescence. Scale bar, 10 mm.

DOI: http://dx.doi.org/10.7554/eLife.03116.035

(A) Neurons were fixed and stained with an antibody directed against ArhGAP44(red) and phalloidin (green). Note that the antibody directed against ArhGAP44 is enriched in phalloidin-rich patches. (B) The isolated N-BAR domain of ArhGAP44 is enriched in dendritic actin-patches. Neurons were transfected with the isolated N-BAR domain of ArhGAP44, a marker for filamentous actin (f-tractin) and a cytosolic reference. The average enrichment of the isolated N-BAR domain of ArhGAP44 over a cytosolic reference in dendritic actin patches is 80% ± 8% (34 patches form 25 neurons; 3 independent experiments). Scale bar, 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.036

Figure 4—figure supplement 3.

ArhGAP44 is enriched at actin patches.

(A) Neurons were fixed and stained with an antibody directed against ArhGAP44(red) and phalloidin (green). Note that the antibody directed against ArhGAP44 is enriched in phalloidin-rich patches. (B) The isolated N-BAR domain of ArhGAP44 is enriched in dendritic actin-patches. Neurons were transfected with the isolated N-BAR domain of ArhGAP44, a marker for filamentous actin (f-tractin) and a cytosolic reference. The average enrichment of the isolated N-BAR domain of ArhGAP44 over a cytosolic reference in dendritic actin patches is 80% ± 8% (34 patches form 25 neurons; 3 independent experiments). Scale bar, 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.036

To test the hypothesis that ArhGAP44 enrichment relies on acto-myosin-dependent pulling forces to the PM, we altered either actin integrity or myosin-dependent actin contraction. Notably, both treatments led to a significant reduction of ArhGAP44 concentration in actin patches (Figure 5A,B and Figure 5—figure supplement 1), suggesting that Myosin II-dependent forces within dendritic actin patches trigger local inward deformation of the PM, mediating a recruitment of curvature-sensing protein ArhGAP44. Notably, we find ArhGAP44 to be also enriched at inward plasma membrane deformation created by retracting lamellipodia (Figure 5—figure supplement 2 and Video 13), as well as in response to membrane bending by artificial nanocone structures (Figure 5—figure supplements 3,4; and Videos 14,15), arguing that inward membrane deformation is sufficient for binding of ArhGAP44 to the plasma membrane.

Figure 5.

Myosin dependent recruitment of the N-BAR domain of ArhGAP44 to actin patches.

(A) Inhibition of acto-myosin dependent forces decreases relative concentration of the N-BAR domain of ArhGAP44 in actin patches. Neurons were transfected with the N-BAR domain of ArhGAP44 (red), f-tractin (blue), and a cytosolic reference (green). Neuron is shown before (left panels) and after addition of the MLCK inhibitor ML-7 (right panels). (B) Cumulative distribution and average values of the relative intensity of the N-BAR domain of ArhGAP44 to a cytosolic reference in actin patches are shown before (black) and after (red) addition of CytoD (n = 49 patches; 12 cells, 3 independent experiments), LatA, (n = 27 patches; 10 cells, 3 independent experiments), ML-7 (n = 118 patches; 15 cells, 3 independent experiments), or the vehicle DMSO (n = 62 patches; 12 cells, 3 independent experiments). Scale bar, 5 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.037

Neurons were transfected with the N-BAR domain of ArhGAP44 (red), f-tractin (blue), and a cytosolic reference (green), and imaged 24 hr later. Note that actin patches persist treatment with CytoD (red arrow). Scale bar, 5 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.038

(A) Rapamycin-induced dimerization assay. Neurons were quadruple-transfected with a membrane-anchored FRB, a fluorescently labeled FKBP that was associated with the Rac-GEF TIAM, a cytosolic reference, and the fluorescently labelled N-BAR domain of ArhGAP44. Addition of rapamycin triggered dimerization of FRB and FKBP, which led to a rapid translocation of TIAM to the plasma membrane. Enrichment of TIAM at the plasma membrane augmented local Rac activity and actin dynamics. (B) Neurons quadruple-transfected with the constructs before and after addition of rapamycin. Note the formation of ectopic actin-rich structures. (C) Time lapse images show enrichment of the N-BAR domain of ArhGAP44 at retracting actin-rich structures. Cells were quadruple-transfected with Lyn-FRB, CFP-FKBP-TIAM, YFP-ArhGAP44(N-BAR), and the empty mCherry plasmid as a cytosolic reference. Note the enrichment of the N-BAR domain of ArhGAP44 at the collapsing lamellipodia. (D) Kymograph of ArhGAP44(N-BAR) (top panel) and the cytosolic reference mCherry (middle panel) show relative enrichment of the N-BAR domain of ArhGAP44 at a retracting actin-rich structures (bottom panel). (B), 10 µm; (C), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.039

(A) Inward membrane deformation by acto-myosin-dependent contraction of membrane-associated actin cables (top panel) is mimicked by artificially applied external forces via cone-shaped nanostructures (bottom panel). (B) Brightfield image of glass slide where cone-shaped nanostructures (i.e., nanocones) were deposited in 3-µm wide stripes. Nanocones are depicted as small triangular structures at the bottom of the image. (C) Atomic force microscope images of the surface of such cone-shapes nanostructures. (D and E) Control experiment showing cells cultured on stripes of nanocones transfected with a cytosolic marker (D) and with a membrane marker (E). (F) Control experiments testing for correlation between ArhGAP44(R291M) and membrane (CFP-CAAX) puncta over individual nanocones. Since positions and amplitudes of the respective puncta show low correlation, an increase in total membrane cannot account for the formation of YFP-ArhGAP44(R291M) puncta (G) Control experiments testing for a possible local actin polymerization induced by nanocones. Cells expressing the actin marker Ruby-LifeAct (left) together with CFP-ArhGAP44(R291M) (right) over nanocones. Individual LifeAct and ArhGAP44(R291M) puncta show no significant correlation. This is consistent with the hypothesis that nanocone-induced membrane deformation and not binding to local actin structures is responsible for the observed N-BAR translocation. (H) Control experiment testing for co-localization of ArhGAP44(R291M) with itself. Scale bars (B, D, E), 3 µm; (C), 5 µm; (F–H), 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.040

(A and B) The curvature-sensitive N-BAR domain of ArhGAP44 is enriched to nanocone-induced membrane deformation at the basal membrane. Neurons were cultured on a glass-slide patterned with nanocones and transfected with the isolated N-BAR domain of ArhGAP44. Note in (B) that puncta are present only on the dendritic stretch to the left that is in contact with nanocones but not in the section of the dendrite to the right that is not touching the nanomaterial. (C) Neurons cultured on a glass-slide patterned with stripes of nanocones and transfected with the isolated N-BAR domain of ArhGAP44. Quantification of N-BAR puncta on glass (black, n = 17 neurons, 2 independent experiments) and on nanocones (red, n = neurons, 2 independent experiments) is shown below. Scale bars (A and B), 10 µm; (C), 50 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.041

Primary hippocampal neurons were transfected with a fluorescently tagged N-BAR domain of ArhGAP44 (red), a marker for filamentous actin (blue), and a cytosolic reference (green) at DIV6 and imaged 24 hr later. Red arrows depict dendritic tips. Quantification of fluorescence intensity show a relative enrichment of the isolated N-BAR domain over a cytosolic reference at actin-rich structures at the tip of dendrites (n = 14 dendritic tips from 10 neurons; 2 independent experiments). Scale bar, 5 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.042

Primary hippocampal neurons were transfected with fluorescently tagged ArhGAP44 (red), a marker for filamentous actin (blue), and a cytosolic reference (green) at DIV16 and imaged 24 hr later. Quantification of fluorescence intensity show a relative enrichment of the full-length ArhGAP44(R291M) (dark red) and for the isolated N-BAR domain of ArhGAP44 (red) over a cytosolic reference at spine-shaped, actin-rich dendritic structures (n = 21 structures form 13 neurons for ArhGAP44(R291) and 24 structures from 12 neurons for the N-BAR domain of ArhGAP44; both from 2 independent experiments). Scale bar, 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.043

(A) Increased filopodia dynamics after knockdown of ArhGAP44. Primary hippocampal neurons were transfected at DIV14 with diced RNA directed against ArhGAP44 (or a control) and a fluorescence marker and imaged 72 hr later. Note the increase in node and filopodia formation (red arrows) upon knockdown of ArhGAP44. (B) Reduction of spine-shaped protrusions upon knockdown of ArhGAP44 (n = control, 18 neurons; siRNA ArhGAP44 (diced pool #1), 20 neurons; siRNA ArhGAP44 (diced pool #2), 23 neurons; both from 2 independent experiments). Scale bar, 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.044

Figure 5—figure supplement 1.

Inhibition of actin polymerization decreases relative concentration of the N-BAR domain of ArhGAP44 but does not completely dissolve actin patches.

Neurons were transfected with the N-BAR domain of ArhGAP44 (red), f-tractin (blue), and a cytosolic reference (green), and imaged 24 hr later. Note that actin patches persist treatment with CytoD (red arrow). Scale bar, 5 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.038

Figure 5—figure supplement 2.

ArhGAP44 is recruited to collapsing artificial lamellipodia in neurons.

(A) Rapamycin-induced dimerization assay. Neurons were quadruple-transfected with a membrane-anchored FRB, a fluorescently labeled FKBP that was associated with the Rac-GEF TIAM, a cytosolic reference, and the fluorescently labelled N-BAR domain of ArhGAP44. Addition of rapamycin triggered dimerization of FRB and FKBP, which led to a rapid translocation of TIAM to the plasma membrane. Enrichment of TIAM at the plasma membrane augmented local Rac activity and actin dynamics. (B) Neurons quadruple-transfected with the constructs before and after addition of rapamycin. Note the formation of ectopic actin-rich structures. (C) Time lapse images show enrichment of the N-BAR domain of ArhGAP44 at retracting actin-rich structures. Cells were quadruple-transfected with Lyn-FRB, CFP-FKBP-TIAM, YFP-ArhGAP44(N-BAR), and the empty mCherry plasmid as a cytosolic reference. Note the enrichment of the N-BAR domain of ArhGAP44 at the collapsing lamellipodia. (D) Kymograph of ArhGAP44(N-BAR) (top panel) and the cytosolic reference mCherry (middle panel) show relative enrichment of the N-BAR domain of ArhGAP44 at a retracting actin-rich structures (bottom panel). (B), 10 µm; (C), 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.039

Video 13.

Enrichment of the N-BAR domain of ArhGAP44 at retracting actin-rich structures in neurons.

Neuron was transfected with Lyn-FRB, CFP-FKBP-Tiam1 (blue), the N-BAR domain of ArhGAP44 (red), and a cytosolic reference (green) at DIV11. Images show formation and retraction of artificial lamelipodia-like structures that formed along the dendritic shaft upon rapamycin-triggered recruitment of the Rac GEF Tiam1 to the plasma membrane. Individual frames were taken every 15 s. Scale bar is 2 μm. Video is 120× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.045

Enrichment of the N-BAR domain of ArhGAP44 at basal membrane sections indented by nanocones.

3D rotation of neuron plated on nanocone-coated glass slide and transfected with the N-BAR domain of ArhGAP44. Note the punctate enrichment at the basal membrane below the soma. Scale bar is 10 μm.

DOI: http://dx.doi.org/10.7554/eLife.03116.046

Enrichment of the N-BAR domain of ArhGAP44 at the dendritic section that is in contact with nanocones.

3D rotation of neuron plated on nanocone-coated glass slide and transfected with the N-BAR domain of ArhGAP44. Note the punctate enrichment at the top where the dendrite is touching the nanocone surface. Scale bar is 10 μm.

DOI: http://dx.doi.org/10.7554/eLife.03116.047

Finally, we investigated whether ArhGAP44 is also enriched to other actin-rich structures in neurons. We find the N-BAR domain enriched at the end of dendrites (dendritic tips; Figure 5—figure supplement 5), as well as in spine-shaped dendritic protrusions (Figure 5—figure supplement 6). Knockdown of ArhGAP44 in aged neurons facilitates re-emergence of exploratory dendritic protrusions (Figure 5—figure supplement 7). Thus, considering that ArhGAP44 expression increases with time, this suggests that ArhGAP44 may facilitate the transition of neurons from a dynamic exploratory mode to a mature more static state.

Figure 5—figure supplement 5.

ArhGAP44 is enriched at dendritic tips.

Primary hippocampal neurons were transfected with a fluorescently tagged N-BAR domain of ArhGAP44 (red), a marker for filamentous actin (blue), and a cytosolic reference (green) at DIV6 and imaged 24 hr later. Red arrows depict dendritic tips. Quantification of fluorescence intensity show a relative enrichment of the isolated N-BAR domain over a cytosolic reference at actin-rich structures at the tip of dendrites (n = 14 dendritic tips from 10 neurons; 2 independent experiments). Scale bar, 5 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.042

Figure 5—figure supplement 6.

ArhGAP44 is enriched at dendritic spines.

Primary hippocampal neurons were transfected with fluorescently tagged ArhGAP44 (red), a marker for filamentous actin (blue), and a cytosolic reference (green) at DIV16 and imaged 24 hr later. Quantification of fluorescence intensity show a relative enrichment of the full-length ArhGAP44(R291M) (dark red) and for the isolated N-BAR domain of ArhGAP44 (red) over a cytosolic reference at spine-shaped, actin-rich dendritic structures (n = 21 structures form 13 neurons for ArhGAP44(R291) and 24 structures from 12 neurons for the N-BAR domain of ArhGAP44; both from 2 independent experiments). Scale bar, 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.043

Figure 5—figure supplement 7.

Knockdown of ArhGAP44 alters protrusion morphology and dynamics in aged neurons.

(A) Increased filopodia dynamics after knockdown of ArhGAP44. Primary hippocampal neurons were transfected at DIV14 with diced RNA directed against ArhGAP44 (or a control) and a fluorescence marker and imaged 72 hr later. Note the increase in node and filopodia formation (red arrows) upon knockdown of ArhGAP44. (B) Reduction of spine-shaped protrusions upon knockdown of ArhGAP44 (n = control, 18 neurons; siRNA ArhGAP44 (diced pool #1), 20 neurons; siRNA ArhGAP44 (diced pool #2), 23 neurons; both from 2 independent experiments). Scale bar, 2 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.044

Discussion

Our study shows that recruitment of ArhGAP44 to actin patches, which seed exploratory filopodia along dendritic branches, is mediated by Myosin II-dependent contraction of membrane-associated actin cables. These actin patches have been shown to serve as precursors for the formation of filopodia in axons (Lau et al., 1999; Spillane et al., 2012) and dendrites (Korobova and Svitkina, 2010). Studies in non-neuronal cells showed that individual actin filaments within the actin cortex form bundles prior to filopodia elongation (Svitkina et al., 2003). Consistently, electron micrographs of neurons showed that these filopodial actin bundles are embedded in the underlying dendritic actin meshwork (Korobova and Svitkina, 2010). It has been proposed that once enough actin filaments are bundled to generate the force required to protrude the PM, the resulting outward membrane deformation triggers recruitment of actin bundling/Cdc42 activating proteins, which then further increase the polymerization rate within the extending filopodia (Krugmann et al., 2001; Disanza et al., 2006). Given the localization of Myosin II in actin patches (Figure 4D), it is reasonable to conjecture that Myosin II-dependent contraction of individual actin filaments within actin patches provides structural integrity to counter the force generated by the extending filopodia. However, considering that individual actin filaments within a bundle are oriented with the barbed end to the PM, Myosin II-dependent contraction also exerts an inward directed pull forces that curve the PM inward which triggers increased recruitment of ArhGAP44. We propose that the highly convoluted membrane topography associated with actin patches (Figure 4A) reflect such Myosin II-dependent contraction of membrane-associated actin cables, at which ArhGAP44 becomes enriched in a curvature-dependent manner. In support of this hypothesis, we find not only that inward membrane deformation is sufficient for ArhGAP44 and N-BAR domain recruitment (Figure 5—figure supplements 3 and 4), but that deletion of ArhGAP44 curvature-sensitivity (Figure 3A,B) or reducing action-myosin-dependent contractile forces (Figure 5A) both prevented ArhGAP44 enrichment.

A second major finding of our study is that recruitment of ArhGAP44 to plasma membrane deformations in nodes (i.e., acto-myosin patches) limits initiation of exploratory filopodia. We show that ArhGAP44 can hydrolize the small GTPase Rac and Cdc42 (Figure 2—figure supplement 5). Thus, a likely function of local ArhGAP44 at actin patches is to suppress GTPase-mediated local actin polymerization. We propose that ArhGAP44 is limiting Rac-dependent formation of actin patches, which provide the structural integrity required for filopodia to protrude outwards (Figure 6). This is consistent with previous reports in neurons, showing that dynamic rearrangements within actin patches relies on Rac activity (Andersen et al., 2005; Spillane et al., 2012), and that elevated Rac levels increase filopodia dynamics (Luo et al., 1996; Nakayama et al., 2000; Zhang and Macara, 2006; Cheadle and Biederer, 2012). In support of this notion, we find Rac1 localized at actin patches (Figure 6—figure supplement 1A) and show that artificial decrease of ArhGAP44 concentration at actin patches either by reducing overall ArhGAP44 levels by knockdown (Figure 1F) or by preventing ArhGAP44-recruitment to actin patches via inhibition of acto-myosin contraction (Figure 6—figure supplement 1B, see also [Ryu et al., 2006; Hodges et al., 2011]) both increased the number of exploratory filopodia.

Figure 6.

Proposed model of ArhGAP44-dependent regulation of exploratory dendritic filopodia initiation.

Scanning electron micrographs of dendritic protrusions with and without filopodial protrusions aligned in a hypothetical time-line. Proposed model that myosin-dependent contractions in actin patches triggers local plasma membrane indentations (i.e., node formation) with curved membranes to which ArhGAP44 is recruited (frame 4–6, red). Increased local ArhGAP44 concentration limits Rac1-dependent actin polymerization and weakens or dissociates the node. Low ArhGAP44 concentration at nodes allows filopodia initiation that relies on Cdc42 and other factors (frame 7–11, gray). Scale bar, 500 nm.

DOI: http://dx.doi.org/10.7554/eLife.03116.048

(A) Rac1 is enriched in dendritic actin patches. Neurons were fixed at DIV12 and stained with an antibody directed against Rac1 (green) and phalloidin (red). (B) Inhibition of myosin light chain kinase increases filopodia density. Filopodia density on neurons was measured before (black) and 120 min after (red) the addition of the MLCK inhibitor ML-7 (n = 21 neurons, 3 independent experiments). Scale bars (A and B), 10 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.049

Figure 6—figure supplement 1.

Control experiments supporting the proposed model.

(A) Rac1 is enriched in dendritic actin patches. Neurons were fixed at DIV12 and stained with an antibody directed against Rac1 (green) and phalloidin (red). (B) Inhibition of myosin light chain kinase increases filopodia density. Filopodia density on neurons was measured before (black) and 120 min after (red) the addition of the MLCK inhibitor ML-7 (n = 21 neurons, 3 independent experiments). Scale bars (A and B), 10 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.049

Cdc42 acts as an activator of Irsp53 (Kast et al., 2014), promoting IRSp53-dependent enrichment and clustering of VASP and other factors to drive actin assembly in elongating filopodia (Disanza et al., 2013). Consistently, knockdown of Cdc42 substantially reduces filopodia formation in neurons (Garvalov et al., 2007). Intriguingly, overexpression of Cdc42 is not sufficient to initiate filopodia formation in neurons (Figure 2—figure supplement 5, see also [Hotulainen et al., 2009]) or in other cell lines (Kast et al., 2014). This has led to the hypothesis that elongation of filopodia is a combinatorial process requiring multiple factors (Kast et al., 2014). We propose that signal integration at actin patches controls this decision of filopodia elongation. Considering that actin-patch formation occurs before filopodia elongation, this argues for a 2-step process where Rac1-induced patch formation (and ArhGAP44-dependent regulation thereof) precedes Cdc42-induced filopodia elongation (Figure 6). However, since ArhGAP44 shows dual specificity for Rac1 and Cdc42, both steps will be limited by recruitment by ArhGAP44 to actin patches.

Taken together, we propose that ArhGAP44 mediates a localized negative feedback that becomes upregulated as neurons mature to reduce the frequency with which neurons initiate new exploratory filopodia. Considering that acto-myosin initiated PM deformation is a ubiquitous process and that a high number of curvature-sensing proteins are known to modify actin dynamics, this suggests that the local feedback mechanism for ArhGAP44 described in our study likely exemplifies a more general principle for receptor-independent signaling whereby signal transduction is initiated by the transient recruitment of regulatory proteins to actin- and force-dependent nanoscale PM indentations.

Materials and methods

Clustering of microarray data

A set of 286 genes relating to the actin cytoskeleton and GTPase signaling was identified with a search on the NCBI Gene database with the query ‘actin AND GTPase AND human[orgn]’. The Human U133A/GNF1H Gene Atlas data set (gnf1h-gcrma unaveraged) was downloaded from the biogps.org website (Su et al., 2004) (only the U133A data were analyzed). The data were renormalized using the median intensity on each array. To generate a focused expression data set for actin- and GTPase-related genes, we extracted the data from all probe sets corresponding to the 286 genes described above. The data were then log-transformed, and the mean log-expression for each probe across all tissue types was subtracted to yield relative expression values. The values were then hierarchically clustered using the Cluster 3.0 software, with the Pearson correlation distance, and average linkage.

Ranking and validating of microarray data

To test the validity of the brain vs spinal cord ranking in Figure 1—figure supplement 1B, we characterized genes known to be expressed exclusively in the brain or in the spinal cord (Table 1 and [Snipes et al., 1992; Martin et al., 1993; Su et al., 1993; tom Dieck et al., 1998]). We found for the Schwann cell-specific genes MPZ and PMP22 an adult brain/spinal cord ratio smaller than 0.3 while the brain enriched presynaptic vesicle fusion protein Bassoon (BSN), and the postsynaptic AMPA receptor 2 (GRIA2) both showed an adult brain/spinal cord ratio greater than 15. Of the 89 neuron-enriched genes identified in Figure 1—figure supplement 1B the top five hits were: the RhoGEF Kalirin that was shown to contribute to EphB receptor-dependent spine maturation (Penzes et al., 2003), the synaptic vesicle-associated protein Amphiphysin (David et al., 1996), the small GTPase K-Ras that translocates from the PM to the Golgi complex, and early/recycling endosomes in response to neuronal activity (Fivaz and Meyer, 2005), the microtubule tip-tracking protein EB3 that is a modulator of spine morphology (Jaworski et al., 2009), and ArhGAP44.

To test the validity of the ranking for genes critical for specific steps during neuronal developmental in Figure 1—figure supplement 3A, we characterized where genes known to be involved in various aspects of neuronal maturation (Table 2 and [Ushkaryov et al., 1992; D'Arcangelo et al., 1995; Ichtchenko et al., 1995; Kornau et al., 1995; Omkumar et al., 1996; Betz et al., 1998; des Portes et al., 1998; Amir et al., 1999; Qualmann et al., 1999]). We observed a clear separation of DCX and RELN, which are involved in neuronal migration, and NLGN1 and NRXN1, which initiate trans-synaptic contact, from MUNC13, MECP2, PSD95, and PACSIN1which all contribute to synapse function. Of the 89 neuron-enriched genes identified in Figure 1—figure supplement 3A, genes with the highest adult-to-fetal ratio included the microtubule tip-tracking protein EB3 (Jaworski et al., 2009), the MAP kinase ERK1 as well as the MAP kinase kinase MEK1 which both control dendrite development (Crino et al., 1998) and synaptic plasticity (Dash et al., 1990), the Armadillo-like protein PKP4 (Wolf et al., 2006), the Lowe syndrome protein OCRL (Attree et al., 1992), and ArhGAP44.

Western blot analysis of rat tissue samples

Tissue samples were isolated from female Wistar rat and suspended in ice-cold lysis buffer containing 1% Tween and protease inhibitors (Roche [Indianapolis, IN], 11873580001). Each sample was homogenized and absolute protein concentration was measured, using the BCA Protein Assay Kit (Thermo Scientific [Rockford, IL], 23225), and adjusted to equal levels for each sample. Next, 6× SDS was added, and the samples were heated to 90°C for 5 min. Finally, the samples were vortexed and loaded on a gel. 20 µg total protein was loaded for each sample in Figure 1A and probed with an antibody directed against ArhGAP44 (Sigma-Aldrich [St. Louis, MO], HPA038814). No single protein was used as reference for comparison of ArhGAP44 expression level across organs, as the expression of conventional housekeeping proteins (e.g., tubulin or GAPDH) can vary between tissues by up to an order of magnitude (Figure 1—figure supplement 2A and http://biogps.org/#goto=genereport&id=37238). In Figure 1B 20 µg total protein was loaded and probed with an antibody directed against ArhGAP44 as well as beta-tubulin (Sigma, T8578) as a reference. For detection, secondary antibodies from Invitrogen and SuperSignal West Femto Maximum Sensitivity Substrate (Pierce [Thermo Scientific, Rockford, IL], 34095) were used.

Westernblot analysis of cultured neurons

Neurons were harvested at DIV3, DIV10, and DIV17 in ice-cold lysis buffer containing 1% Tween and protease inhibitors (Roche, 11873580001). Absolute protein concentration was immediately measured using the BCA Protein Assay kit (Thermo 23225) and adjusted to equal levels for each time point. Relative protein levels were probed using specific antibodies directed against ArhGAP44 (Abcam [Cambridge, MA], ab93627), the postsynaptic marker PSD95 (EMD Millipore [Billerica, MA], MAB1596), the presynaptic protein Bassoon (Abcam, 76065), and the loading control beta-tubulin (Sigma, T8578). For detection, we used secondary antibodies from Invitrogen and SuperSignal West Femto Maximum Sensitivity Substrate (Pierce, 34095).

Scanning electron micrographs of cultured neurons

Neurons were cultured on Poly-L-Lysine-coated glass coverslips and fixed using 2% Glutaraldehyde (8% stock-EM grade) and 4% p-Formaldehyde in NaCacodylate buffer pH 7.4 for 10 min. Neurons were rinsed with 0.1 M NaCacodylate buffer (pH 7.4) after primary fixation and post-fixed for 1 hr with aqueous 1% OsO4, washed briefly with water and dehydrated in an ascending ethanol series (50, 70, 90, and 100% [twice] for 20 min each) before critical point drying with liquid CO2 in a Tousimis 815B (Tousimis, Rockville, MD, USA). Samples were mounted on colloidal Graphite on 15-mm aluminum stubs (Ted Pella, Redding, CA, USA) and sputter-coated with 70A of Au/Pd using a Denton Desk 11 Sputter Coater. Visualization of samples was performed with a Zeiss Sigma FESEM (Zeiss Microscopy LLC, Thornwood, NY) operated at 2–3 kV, working distance 4–6 mm and an in-lens SE detector under high vacuum conditions. Images were captured in TIFF format.

Quantification of protrusion types via scanning electron micrographs

Neurons were cultured on glass slides for various periods of time (3, 10, and 17 days), fixed and prepared for SEM as described above. Using low resolution (1000× magnification), individual neurons were identified (Figure 1—figure supplement 5A, left panel). Starting from the soma, initial segments of the dendritic arbors were imaged at high resolution (10,000×), and individual protrusions were classified based on morphology (Figure 1—figure supplement 5A, right panel). Only the proximal 50–60 μm of the dendritic arbors that can clearly be associated to a particular neuron were analyzed. Examples of dendritic nodes are shown in Figure 1—figure supplement 5B.

Culturing and immunostaining of primary hippocamal neurons

Rat hippocampal neurons were prepared as previously described (Fink et al., 2003). Neurons were transfected using Lipofectamine 2000 (Invitrogen, Carlsbad, CA) according to the manufacturer's protocol. For live imaging, neurons were plated in chambers (Lab-Tek 155383; Thermo, Rockford, IL) using NBM (Neurobasal Medium, Gibco [Life Technologies, Carlsbad, CA], 21103-049), supplemented with SM1 (StemCell Technologies [Vancouver, Canada], 05711), Pen/Strep (Gibco, 15070-063) and 20 mM HEPES (Gibco, 15630). For immunostaining, cells were fixed in PBS (Gibco, 10010-023) containing 4% Formaldehyde (Ted Pella, 18505) and 120 mM sucrose, stained, and imaged. ArhGAP44 antibody was from Sigma (1:150, HPA038814), MAP2 antibody was from Chemicon (1:1000, AB 5622; Chemicon, EMD Millipore, Billerica, MA), pMLC antibody was from Cell Signaling (1:200, 3671S; Cell Signalling Technology, Danvers, MA), and Rac1 antibody was from Cytoskeleton (1:200, ARC03; Cytoskeleton, Denver, CO).

Fluorescence microscopy

All experiments were performed on a spinning disc confocal microscope. CFP, YFP, and mCherry excitations were obtained by a 442-nm helium cadmium laser (100 mM; Kimmon Electrics, Centennial, CO), a 514-nm argon laser (300 mW; Melles Griot, Carlsbad, CA), and a 594-nm solid-state laser (80 mW; CNI Laser, Changchun, China), respectively. Images were captured using an EMCCD camera (QuantEM 512SC [Photometrics, Tucson, AZ]), driven by Micromanager mounted on the side port of an inverted microscope (model IX-71; Olympus, Center Valley, PA).

Analysis of filopodia density

Primary cultured rat hippocampal neurons were transfected 7 days after plating with siRNA directed against ArhGAP44 together with a fluorescent marker and fixed at DIV12. Fluorescently tagged ArhGAP44(R291M) was transfected at DIV11 and fixed at DIV12. For each condition, the sample was fixed and individual neurons were imaged. Filopodia density was measured manually, analyzing only the proximal 100 µm of each dendrite. For each condition tested, >20 cells were used.

Rac1 activation assay

HeLa cells were grown in DMEM (high glucose) supplemented with 10% FCS and Pen/Strep until they reached 80% confluency and then transfected either with CFP-ArhGAP44(wt), CFP-ArhGAP44(R291M), or empty CFP plasmid (control) using LF2000 according to the manufacturer's protocol for 4 hr in DMEM in the absence of FCS and Pen/Strep. Cells were then serum starved for 18 hr. For all conditions, live cell fluorescence 18 hr post transfection showed transfection efficiency of >80%. Cells were stimulated with 50 ng/ml EGF for 5 min. Next, cells were scratched and protein levels were measured and adjusted for all samples to equal levels. Of each sample 900 µl were used for pulldown and 100 µl for loading control. GTP levels were probed using Rac1 Activation Assay Kit (Cell Biolabs [San Diego, CA], STA-404) according to the manufacturer's protocol. In brief, GTP-bound Rac was eluted from cell lysates using PAK PBD agarose beads and detected by western blot using α-Rac1 (Cell Biolabs, 240106) antibody. Relative intensities were compared to loading controls.

Generation of diced siRNA pools

The protocol used to synthesize siRNA has been previously reported (Liou et al., 2005). Specific primers for ArhGAP44 were automatically designed and used to amplify from a cDNA library an approximate 600-bp PCR fragment of the 3′ region of the coding sequence. A second amplification was performed with a set of nested primers bearing a T7 promoter sequence on their 5′ extension. Nested PCR products were transcribed in vitro (T7 MEGA script kit; Ambion, Austin, TX) and the resulting double-stranded RNAs were annealed and processed with 30 units per reaction of human recombinant Dicer (Invitrogen) for 15 hr at 37°C. The 21mer siRNAs were separated from incompletely digested fragments using a succession of isopropanol precipitations and filtration on glass fiber plates (Nunc, Rochester, NY).

Filopodia dynamics analysis

Neurons were imaged for 10 min every 60 s using a 63× objective. Changes in filopodia dynamics were assessed manually using ImageJ. In detail, dynamic protrusions (i.e., nodes and dynamic filopodia) were counted and normalized to filopodia that remained over the course of the acquisition (i.e., static filopodia). Dynamics was visualized using the Temporal Color-Code designed by Kota Miura (http://fiji.sc/wiki/index.php/Temporal-Color_Code).

Ratiometric images of fluorescence intensity in dendritic nodes and actin patches

The software used for ratio-metric imaging has been previously described (Tsai and Meyer, 2012). In brief, a low-pass Gaussian filter was first applied to all images to suppress the noise while retaining the details of the fluorescent signals. Background subtraction was subsequently performed by (the value of each pixel)—(the mean value of the background within 40 μm of that pixel). To determine relative intracellular ArhGAP44 levels, ratio images were created dividing ArhGAP44 fluorescence over the cytosolic fluorescence.

Analysis of relative protein concentration in dendritic patches

To measure the relative protein concentration at patches, average fluorescent intensities of the protein of interest (POI) and the cytosolic reference were measured in the patch and in the adjacent dendritic stretch. The background (BG) was determined separately for both channels using the average of four sectors outside the cell adjacent to the region of interest. The relative intensity of POI’s was determined as [(POIPatch − POIBG)/(CytosolPatch − CytosolBG)]/[(POIDendrite − POIBG)/(CytosolDendrite − CytosolBG)].

Correlative IF/SEM image analysis

A micro-pattern was generated on glass slides and coated over night with PLL (0.1 mg/ml). Neurons were plated for 11 days and then fixed as described above for the FESEM analysis. Using the micro-pattern as reference points, individual neurons were then labeled with fluorescently tagged phalloidin and imaged on a 63× objective with a 1.5 Optovar. Cells were then sputter-coated with 70A of Au/Pd using a Denton Desk 11 Sputter Coater. Using the backscatter detector, individual micro-patterns were identified and used to navigate and identify individual previously imaged neurons as shown in Figure 4—figure supplement 2. SEM Images were then taken at 10,000× magnification and aligned with the immunofluorescent images. Finally, individual nodes were identified using the SEM images, and the average fluorescent intensity of phalloidin was measured in nodes and the adjacent dendritic stretches.

Constructs and drugs

Full-length and the N-BAR domain constructs of ArhGAP44, F-tractin, NMHC-2B (Addgene Plasmid No. 11348), Rac1 and Cdc42 were previously described (Wei and Adelstein, 2000; Heo and Meyer, 2003; Johnson and Schell, 2009; Galic et al., 2012). The point mutation in ArhGAP44(R291M) was introduced using the site-specific mutagenesis kit (200518; Stratagene, Cedar Creek, TX). All constructs were sequenced prior to use. ML-7 (sc-200557; Santa Cruz Biotechnology) was used at 10 µM (Figure 5A) and 50 µM (Figure 6—figure supplement 1B). Latrunculin A (428026; Cal Biochem, EMD Millipore, Billerica, MA) was used at 4 µM. Cytochalasin D (PHZ1063; Invitrogen) was used at 5 µM.

Dug-induced changes in ArhGAP44 intensity

For individual neurons, z-stacks were acquired before and 20–30 min after addition of drugs. For the analysis, a maximal projection was made and individual actin patches (= sites along the dendrite where the actin/cytosol ratio was >200% above the average ratio) were identified. Using a mask, the ArhGAP44/cytosol intensity within the actin patches was determined and normalized to ArhGAP44/cytosol ratio in the dendritic shaft.

FKBP-Tiam1 assay

Dynamic translocation of Tiam1 with the FRB-FKB system in non-neuronal cells has been previously described (Inoue et al., 2005). Here, we cultured hippocampal neurons were quadruple-transfected with Lyn-FRB, CFP-FKBP-Tiam1 the YFP-tagged N-BAR domain of ArhGAP44 and the cytosolic reference mCherry using Lipofectamine 2000 (according to manufacturer's protocol). 24 hr later, individual cells were imaged before and after addition of 100 nM rapamycin (B0560; Sigma–Aldrich) with a 63× objective and a 1.5× Optovar module.

Nanocone assay

Nanocone production has been previously described (Jeong et al., 2011; Galic et al., 2012). In brief, a 35–50 nm thin film of tin was deposited by heat evaporation on a glass coverslip at room temperature. The glass with the deposited tin was then exposed to a nitrogen gas environment with a low concentration of oxygen (about 1 part per million) at 350°C for 90 min. The annealing to the glass and the formation of the replicate nanocone shapes occurred during this heating step. In order to make the nanocone structures transparent, the glass coverslip with nanocones was further heated to a temperature 400°C for 3 hr in air. Labtek chambers were then mounted with nanocoated glass slides as previously described (Jeong and Galic, 2014), and neurons were subsequently cultured, transfected on DIV10, and imaged alive 24 hr later. 3D rotation (Videos 14,15) of confocal stacks was done in ImageJ.

Atomic force microscopy

AFM images of nanocones were done in tapping mode using commercial cantilevers on a JPK Nanowizard II instrument. Height analysis was performed using JPK image processing software.

Statistics

p-values in all figures depict pair-wise comparisons and were evaluated using the Student's t test, with two tails and two-sample unequal variance. Error bars in all images represent SEM of the mean value. **p < 0.01.

eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.

[Editors’ note: this article was originally rejected after discussions between the reviewers, but the manuscript was accepted after revisions and re-review.]

Thank you for choosing to send your work entitled “Dynamic Recruitment of Nadrin2 to Nanoscale Membrane Deformations Limits Exploratory Filopodia Initiation in Neurons” for consideration at eLife. Your full submission has been evaluated by Richard Losick (Senior editor), a Reviewing editor, and 2 peer reviewers, and the decision was reached after discussions between the reviewers.

After careful examination of the results and multiple consultations, the reviewers and the Reviewing editor decided that the results did not reach the level of significance claimed by the authors and despite tackling an interesting problem, the main interpretations would require a lot more work to meet our publication standard. The reviewers’ comments follow for your consideration.

Reviewer #1:

The authors analyze here the Rac1 GAP and BAR-domain containing protein Nadrin2, which they identified as an actin regulatory protein enriched in brain vs spinal cord and that they show to be enriched in cortex. Overexpression and knock-down studies demonstrate that this protein is a negative regulator of dendritic filopodia number. The GAP activity contributes to this function, but is not required. Analyzing filopodia dynamics, live imaging data upon knock-down and in overexpressing neurons support that this protein causes newly formed filopodia to be unstable. Nadrin2 localizes to actin rich dendritic patches (which are nicely analyzed using correlative scanning EM/immunostaining), and imaging of the less active Nadrin2 GAP mutant supports that filopodia can emerge from these dendritic Nadrin2 patches. Interestingly, Nadrin2 appears to be recruited dynamically and in a myosin light chain kinase-dependent manner to these actin patches, which also contain Rac1. The model that emerges from these results is that myosin-actin interactions at dendritic membranes cause membrane indentations to form, which recruit Nadrin2 via its BAR domain to dynamically regulate Rac1 and restrict filopodia formation. This provides evidence that stable filopodia require not only positive signals to form but also the removal of negative regulators such as Nadrin2, which is an interesting concept. Several key experimental questions need to be addressed, however, to further strengthen this study.

1) How frequently do filopodia emerge from Nadrin2-positive patches? Are filopodia more likely to emerge from these patches than from Nadrin-2 negative dendritic shaft areas? This information will allow to assess how important a player Nadrin2 is in the control of filopodia formation. If most emerging filopodia do not contain Nadrin2, this protein is only relevant to control a subset of these protrusions.

2) Does overexpression or loss of Nadrin2 impact the number of dendritic spines? This is important to assess the biological role of Nadrin2-mediated filopodia formation.

3) Can the authors show more directly that Nadrin2 is recruited to membranes in a curvature-dependent manner?

4) Figure 2 only shows that Rac1 overexpression promotes the formation of transient filopodia independent of Nadrin2. The statement that “Nadrin2 limits Rac-dependent de novo filopodia formation” is not supported by these data.

5) Knock-down neurons should be stained for Nadrin2 to confirm antibody specificity in immunostaining.

6) The tissue distribution blot in Figure1 needs to be repeated for better signals.

Reviewer #2:

The paper identifies Nadrin2 as brain-enriched regulator of filopodia outgrowth in neuronal dendrites, as overexpression of the wt strongly suppresses filopodial number. More refined manipulations of Nadrin2 have a number of more moderate effects: it shows a mild (50%) enrichment in actin patches along the dendrite compared to GFP, and its knockdown prompts an increase in the number of quickly collapsing protrusions and a 20% increase in steady state filopodia density. Its recruitment to patches is shown to depend on its N-BAR domain, suggesting it is recruited to the membrane curvature observed in detail by SEM in these patches. Pharmacological inhibition of Myosin II or application of LatA or CytoD acutely reduced Nadrin2 at the patches, suggesting that the status of actin and myosin in the patches is also involved in Nadrin2 targeting. The model is that Nadrin2 recruitment suppresses Rac1 activity important for filopodial extension, and this model is supported by weaker phenotypes of a Nadrin2 mutant that retains only weak Rac1 GAP activity.

The paper presents a new model for suppression of protrusion formation, and postulates a role for the membrane curvature seen in the filopodial birthplaces. These observations highlight previously overlooked aspects of neuronal process formation. The combination of live-cell imaging and correlative SEM is potentially quite powerful. However, the live-cell approaches are not used to test the Nadrin2 model very directly, and though the phenotype is compelling, overall I find the explanation offered by the model only weakly supported by the available data.

My major concerns are as follows:

1) The authors overuse the terms “transient” and “dynamic” to describe the Nadrin2 enrichment at patches. For example, “A major finding of our study is that transient recruitment of Nadrin2 to actin patches...”… “deletion of Nadrin2 curvature-sensitivity (Figure 3A) as well as limiting action-myosin dependent contractile forces (Figure 5A) both prevented dynamic Nadrin2 recruitment.” “both treatments led to a significant reduction of Nadrin2 recruitment”

In fact, the time course of recruitment is never analyzed in the paper. There is only one Figure (3C) that shows any transience. No treatment was analyzed in a manner that would have assessed the dynamic nature of the recruitment. Perhaps the decreased but remaining colocalization is just as transient but weak, or more short-lived but just as pronounced. Regarding myosin, the ML7 results (Figure 5A) seem to suggest only that retention, not recruitment, of Nadrin2 are mediated by myosin II. It seems that examining the time course of accumulating Nadrin2 with respect to the time of filopodial extension would be required to evaluate the model. A more specific test would be to compare the time course of active Rac1 in the patch with Nadrin2 levels. Similarly, the curvature aspect could be addressed in detail specifically by timelapsing the N-BAR construct during extension.

2) The authors state “the transient localization of Nadrin2 to patches but not to extended filopodia argues that Nadrin2 limits initiation rather than elongation of newly formed filopodia.”

Why doesn't this argue equally well that recruitment of Nadrin2 initiates outgrowth? Loss of Nadrin in an elongating filopodium would seem to indicate that it might in fact suppress elongation, not initiation. With the few examples available for viewing, initiation seems to occur before loss of Nadrin. Clearly, an additional step of mechanism would be required to explain the overexpression and knockdown phenotypes, but I think it is too facile to say the live imaging supports this model.

3) The authors state “Our experiments identified the small GTPase Rac as the principal target of Nadrin2 within actin patches (Figure 2–figure supplement 1B).”

The indicated figure does not demonstrate this at all. The assumption is that Nadrin2 acts through Rac1 only, not Cdc42, but the alternative (interaction with Cdc42) is not tested. Can this really be dismissed out of hand? Figure 1E tests the relative levels of GTP-Rac1, but does not test whether Nadrin2 or its mutant alter levels of other potential targets (e.g. Cdc42).

A related but unstated presumption is that the dynamic assays of Figure 2 predict total protrusion number, and expressing Cdc42 would provide a test of that as effect of its overexpression is opposite to Rac1. Does Cdc42 overexpression alter the dynamic measures in a way opposite to Rac1? Does it decrease the number of nodes?

Presumably, dynamics were not measured following Nadrin2(wt) overexpression because there are so few filopodia left, but shouldn't the Rac1+Nadrin2(wt) case be examined, since Nadrin2(wt) should suppress the Rac1 phenotype? That is the claim in the model of Figure 5B, left column.

4) Surprisingly, the movies do not make a great case for this paper, and in fact in my mind raise more questions than they answer. Technically, the intensity levels fluctuate substantially, making me wonder whether small or thin structures such as nascent filopodia or nodes could really be identified with confidence. One troubling example is in the upper right panel, going from frame 1 to 2. The structure seemingly of most interest is the node or protrusion in the middle of the process (under “n2”), which appears to pop into existence. But the entire image brightens in this panel, making it unclear whether to trust the relative intensity of the node. Related to that type of phenomenon, some of the fluctuation is apparently due to an inconsistent z span; e.g. in Movie 1, the overexpression of Nadrin2 panel clearly shows the middle one of the processes fade out and then return, as it was insufficiently spanned by the planes of the z sectioning. In the upper right panel, that particular node could just have been captured more completely in bounds of the z stack rather than appear or brighten in reality.

In terms of observations, it doesn't seem as though the control example in movie 1 demonstrates any protrusion formation at all. Am I missing something? Based on the quantification in Figure 2A', there should be just as many forming and collapsing processes as there are processes that persist through the movie, but I don't see this at all. At the very least, I'd say that readers will need some help: formation, collapse, and unclassified changes should probably be pointed out with markers of some sort.

Raw data (movies) of the Rac1 and Rac1+Nadrin2(R291M) experiments would be good to see.

More raw data (stills or movies) from the key experiments in Figure 3A and 3B is needed.

5) In Figure 3D, the accumulation of a set of SEM images into a numbered series of panels portrayed as “stages” is unnecessary and misleading-there is no evidence presented here or cited that the stages correspond to a growth sequence. There is a near-total lack of information about how these images were assessed or quantified, and what constituted a node or a protrusion or any other feature.

We have substantially expanded description and analysis of filopodia initiation from dendritic nodes, showing now in detail morphological rearrangements and kinetic behavior of ArhGAP44 and actin at such sites. We also included in the revised version of the manuscript experiments to investigate a possible regulation of Cdc42 by ArhGAP44 during exploratory dendritic filpodia initiation. These experiments further validate the proposed model that ArhGAP44 targets Rac1 in actin-patches that precede exploratory filopodia formation. To directly demonstrate that inward membrane deformation is sufficient for ArhGAP44 recruitment, we artificially indented the plasma membrane in living neurons using cone-shaped nanostructures (nanocones) as well as a chemical inducer of actin dynamics (a dimerization system we developed to rapidly pull a Rac GEF from the cyctosol to the plasma membrane). Finally, we also included experiments to test for a possible function of ArhGAP44 at other actin-rich neuronal structures, such as dendritic spines and growth cones.

Together, the added studies further strengthened the validity of our major finding that ArhGAP44 acts in a localized negative feedback that allows neurons to tune the frequency with which new exploratory filopodia are initiated.

Comment on changed protein name: in compliance with the official protein nomenclature, we are using in the revised version of the manuscript the protein symbol ArhGAP44 instead of Nadrin2.

Reviewer #1: […] Several key experimental questions need to be addressed, however, to further strengthen this study.

1) How frequently do filopodia emerge from Nadrin2-positive patches? Are filopodia more likely to emerge from these patches than from Nadrin-2 negative dendritic shaft areas? This information will allow to assess how important a player Nadrin2 is in the control of filopodia formation. If most emerging filopodia do not contain Nadrin2, this protein is only relevant to control a subset of these protrusions.

In the revised version we now investigate the origins of dendritic protrusions in more detail. The following three experiments were added:

We show that 83% ± 7% of all protrusions emerge from dendritic nodes (Figure 2–figure supplement 4B).

We show that 89% ± 6% of all protrusions emerge from actin-rich patches (Figure 4–figure supplement 1C). This further strengthens the argument that dendritic nodes visible by light microscopy and electron microscopy are local actin patches in dendrites.

100% of dendritic actin patches show enrichment for ArhGAP44. The average enrichment of ArhGAP44 in dendritic actin-patches is 80% ± 8% over a cytosolic reference (Figure 4–figure supplement 3B).

Together, these added experiments argue that the majority of dendritic protrusions are emerging from dendritic nodes that have convoluted membrane invaginations and that show a relative increase for both, actin and ArhGAP44 concentration.

2) Does overexpression or loss of Nadrin2 impact the number of dendritic spines? This is important to assess the biological role of Nadrin2-mediated filopodia formation.

To further explore the function of ArhGAP44 in aged neurons, we added the following two experiments to the revised manuscript:

Full-length and the isolated N-BAR domain of ArhGAP44 both enrich in dendritic spines in neurons (Figure 5–figure supplement 6).

Knockdown of ArhGAP44 in aged neurons (DIV17) increased the fraction of dynamic filopodia-shaped vs. spine-shaped dendritic protrusions (Figure 5–figure supplement 7).

We did not include this data in the initial submission, as we felt that it does not strengthen the core findings of this manuscript that is focused on the initiation of exploratory filopodia from dendrites.

3) Can the authors show more directly that Nadrin2 is recruited to membranes in a curvature-dependent manner?

Yes, we can. We have previously established an assay that relies on cone-shaped nano-scale structures (nanocones) to deform the plasma membrane (PM) in live cells (Ayala et al., 2007): When cultured on nanocones, adherent cell transiently deform the basal PM. This creates local sites with positively curved PMs of up to 50nm diameter that allows investigating the functional consequences of nano-scale membrane deformation under physiological conditions in live cells. In the revised version we added the following experiments:

We show that the isolated N-BAR domain of ArhGAP44 forms puncta selectively above nanocone-induced membrane-deformation in the basal PM of primary hippocampal neurons (Figure 5–figure supplement 4 and Videos 14, 15).

Figure 5—figure supplement 4.

N-BAR domain of ArhGAP44 enriched at nanocone-induced dendritic plasma membrane deformations in neurons.

(A and B) The curvature-sensitive N-BAR domain of ArhGAP44 is enriched to nanocone-induced membrane deformation at the basal membrane. Neurons were cultured on a glass-slide patterned with nanocones and transfected with the isolated N-BAR domain of ArhGAP44. Note in (B) that puncta are present only on the dendritic stretch to the left that is in contact with nanocones but not in the section of the dendrite to the right that is not touching the nanomaterial. (C) Neurons cultured on a glass-slide patterned with stripes of nanocones and transfected with the isolated N-BAR domain of ArhGAP44. Quantification of N-BAR puncta on glass (black, n = 17 neurons, 2 independent experiments) and on nanocones (red, n = neurons, 2 independent experiments) is shown below. Scale bars (A and B), 10 µm; (C), 50 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.041

We show that enrichment of ArhGAP44 over nanocone-induced PM-deformations does not correlate with enrichment of the PM (Figure 5–figure supplement 3F) or of actin (Figure 5–figure supplement 3G).

Figure 5—figure supplement 3.

Artificial membrane deformations recruit ArhGAP44 to the plasma membrane.

(A) Inward membrane deformation by acto-myosin-dependent contraction of membrane-associated actin cables (top panel) is mimicked by artificially applied external forces via cone-shaped nanostructures (bottom panel). (B) Brightfield image of glass slide where cone-shaped nanostructures (i.e., nanocones) were deposited in 3-µm wide stripes. Nanocones are depicted as small triangular structures at the bottom of the image. (C) Atomic force microscope images of the surface of such cone-shapes nanostructures. (D and E) Control experiment showing cells cultured on stripes of nanocones transfected with a cytosolic marker (D) and with a membrane marker (E). (F) Control experiments testing for correlation between ArhGAP44(R291M) and membrane (CFP-CAAX) puncta over individual nanocones. Since positions and amplitudes of the respective puncta show low correlation, an increase in total membrane cannot account for the formation of YFP-ArhGAP44(R291M) puncta (G) Control experiments testing for a possible local actin polymerization induced by nanocones. Cells expressing the actin marker Ruby-LifeAct (left) together with CFP-ArhGAP44(R291M) (right) over nanocones. Individual LifeAct and ArhGAP44(R291M) puncta show no significant correlation. This is consistent with the hypothesis that nanocone-induced membrane deformation and not binding to local actin structures is responsible for the observed N-BAR translocation. (H) Control experiment testing for co-localization of ArhGAP44(R291M) with itself. Scale bars (B, D, E), 3 µm; (C), 5 µm; (F–H), 1 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.040

Together, these experiments argue that nanocones locally deform the PM and that the resulting local high PM-curvature is sufficient for ArhGAP44 enrichment in neurons.

4) only shows that Rac1 overexpression promotes the formation of transient filopodia independent of Nadrin2. The statement that “Nadrin2 limits Rac-dependent de novo filopodia formation” is not supported by these data.

We thank the referee for pointing out that the title of the figure legend is misleading. We have rephrased the figure title to: ´Knockdown of ArhGAP44 and overexpression of Rac1 both increase de novo filopodia formation´.

5) Knock-down neurons should be stained for Nadrin2 to confirm antibody specificity in immunostaining.

The following two experiments were added to confirm the antibody specificity:

Overexpression control: Neurons transfected with ArhGAP44(R291M) for 24h, fixed and stained with an antibody directed against ArhGAP44 show a >8-fold increase in fluorescence intensity compared to non-transfected cells (Figure 3–figure supplement 1B).

Figure 3—figure supplement 1.

ArhGAP44 antibody controls.

(A) Western blot of whole rat brain with antibody directed against ArhGAP44(818AA) (unspecific bands are visible). (B) Increase immunostaining directed against ArhGAP44 after overexpression of ArhGAP44. Primary hippocampal neurons were transfected with a fluorescently tagged ArhGAP44(R291M) and fixed 24 hr later. Quantification of fluorescence intensity of non-transfected neurons (black, n = 20 cells, 2 independent experiments) and neurons transfected with ArhGAP44 (red, n = 12 cells, 2 independent experiments) after staining with an antibody directed against ArhGAP44 is shown below. (C) Control experiment showing reduced immunostaining of ArhGAP44 after knockdown of ArhGAP44. Primary hippocampal neurons were transfected at DIV10 with diced RNA directed against ArhGAP44 and a fluorescence reference (empty pEYFP plasmid) and fixed 48 hr later. Relative fluorescence intensity was measuerd in dendritic stretches and normalized to the adjacent background (arrows). Quantification of fluorescence intensity after staining with an antibody directed against ArhGAP44 of cells transfected with control siRNA (black, n = 16 cells, 2 independent experiments) and siRNA directed against ArhGAP44 (red, n = 18 cells, 2 independent experiments) is shown below. Scale bars (B and C), 20 µm.

DOI: http://dx.doi.org/10.7554/eLife.03116.030

Knockdown control: When stained with an antibody directed against ArhGAP44, neurons transfected with siRNA directed against ArhGAP44 show reduced fluorescence intensity compared to cells transfected with a control siRNA (Figure 3–figure supplement 1C).

6) The tissue distribution blot in Figure1 needs to be repeated for better signals.

The blot shown in Figure 1A has been replaced.

Reviewer #2: My major concerns are as follows:

1) The authors overuse the terms “transient” and “dynamic” to describe the Nadrin2 enrichment at patches. For example, “A major finding of our study is that transient recruitment of Nadrin2 to actin patches...”…“deletion of Nadrin2 curvature-sensitivity () as well as limiting action-myosin dependent contractile forces () both prevented dynamic Nadrin2 recruitment.”… “both treatments led to a significant reduction of Nadrin2 recruitment”

In fact, the time course of recruitment is never analyzed in the paper. There is only one that shows any transience. No treatment was analyzed in a manner that would have assessed the dynamic nature of the recruitment. Perhaps the decreased but remaining colocalization is just as transient but weak, or more short-lived but just as pronounced. Regarding myosin, the ML7 results () seem to suggest only that retention, not recruitment, of Nadrin2 are mediated by myosin II. It seems that examining the time course of accumulating Nadrin2 with respect to the time of filopodial extension would be required to evaluate the model. A more specific test would be to compare the time course of active Rac1 in the patch with Nadrin2 levels. Similarly, the curvature aspect could be addressed in detail specifically by timelapsing the N-BAR construct during extension.

(A) Overuse of the terms ‘transient’ and ‘dynamic’. We have substantially reduced the use of both terms in the text, using them now only 2 times in the whole manuscript to describe ArhGAP44.

(B) Curvature-aspect of protein recruitment. In the revised version of the manuscript, we added a series of experiments to investigate the curvature-dependence of ArhGAP44 enrichment (see also Referee #1, point 3):

We show that the isolated N-BAR domain of ArhGAP44 forms puncta selectively above nanocone-induced membrane-deformation in the basal PM of primary hippocampal neurons (Figure 5–figure supplement 4 and Videos 14, 15).

We show that enrichment of ArhGAP44 over nanocone-induced PM-deformations does not correlate with enrichment of the plasma membrane (Figure 5–figure supplement 3F) or of actin (Figure 5–figure supplement 3G).

These experiments provide evidence that inward plasma membrane deformation is sufficient for enrichment of ArhGAP44 to inward deformed plasma membranes in neurons.

(C) Time course of ArhGAP44 enrichment. The following four experiments were added: ArhGAP44 is recruited to contracting structures: We added a synthetic approach that we developed earlier, where we used a small molecule to rapidly recruit a Rac GEF to the PM and activate Rac in order to increase actin polymerization (Figure 5–figure supplement 2 and Video 13). In this approach, we observe enrichment of the isolated N-BAR domain of ArhGAP44 at retracting actin-rich structures induced along the dendritic shaft. Together with the experiment where we inhibit MLCK using ML-7 (Figure 5) and show that recruitment of ArhGAP44 requires myosin contraction, this argues that the increase in local ArhGAP44 concentration is caused by increased formation of acto-myosin-dependent inward membrane deformation.

Kinetic analysis of actin and ArhGAP44 during filopodia initiation. The following experiments were added: we find that 83% ± 7% of all protrusion emerge from dendritic nodes (Figure 2–figure supplement 3B), that 89% ± 6% of all protrusions emerge from actin patches (Figure 4–figure supplement 1B), and that ArhGAP44 is enriched in 100% of actin patches (Figure 4–figure supplement 3B). These experiments argue that the majority of dendritic protrusions are emerging from dendritic nodes that are enriched both in actin and in ArhGAP44.

Together, these added experiments further strengthen the statement that ArhGAP44 is recruited to contracting actin patches within nodes that precede filopodia elongation.

(D) Retention vs. Recruitment. We propose that ArhGAP44 diffuses trough the cytosol by Brownian movement, and that enrichment of ArhGAP44 at nodes is caused by binding of the protein to inward membrane deformations that transiently form at such sites due to myosin-dependent contraction of membrane-associated actin cables. Binding of N-BAR domain proteins (such as ArhGAP44) has been shown to depend (i) on increased electrostatic interactions between negatively charged lipid head groups in curved membranes and positively charged amino acids of the banana-shaped protein dimer facing the membrane, and (ii) the insertion of an amphipatic helix present in all N-BAR domain proteins into membrane-imperfections that predominantly occur in curved lipid bilayers (Dotti et al., 1988; Ziv & Smith, 1996). Enrichment of N-BAR domain proteins at curved membranes is thus believed to be due to an increase in binding affinity that may result molecularly from a reduced off-rate and possibly an increased on-rate. As the term ‘recruitment’ does not imply active directed transport (it is often used to describe a diffusion mediated retention at local sites), we feel that it is appropriate to use this term since it describes the observed dynamic increase in local concentration of ArhGAP44 when new actin patches are formed.

2) The authors state “the transient localization of Nadrin2 to patches but not to extended filopodia argues that Nadrin2 limits initiation rather than elongation of newly formed filopodia.”

Why doesn't this argue equally well that recruitment of Nadrin2 initiates outgrowth? Loss of Nadrin in an elongating filopodium would seem to indicate that it might in fact suppress elongation, not initiation. With the few examples available for viewing, initiation seems to occur before loss of Nadrin. Clearly, an additional step of mechanism would be required to explain the overexpression and knockdown phenotypes, but I think it is too facile to say the live imaging supports this model.

We agree with the referee that it is a valid question whether ArhGAP44 functions more as an inhibitor of filopodia initiation vs. an inhibitor of filopodia elongation/outgrowth: We considered that an inhibitor of filopodia elongation/outgrowth would be expected to localize together with positive regulators of actin dynamics (e.g. MENA or IRSp53) at the tip of extending filopodia where actin-polymerization in growing filopodia occurs. In contrast, an inhibitor of filopodia initiation would be expected to be present at the ´birthplace´ of filopodia, and to dissociate once elongation/outgrowth begins. The localization of ArhGAP44 all across actin patches, which our data argues is caused by acto-myosin dependent inward curved plasma membrane deformation, suggests that it reduces Rac and Cdc42 activity globally across the patch which we think explains the lower frequency of filopodia extension. The dissociation from extending filopodia (with outward membrane deformation) can best be explained by the change in curvature (all negative) in the extending membrane tubes. Based on the selectivity of N-BAR domain proteins to positively curved membranes, we therefore think that our data is consistent with a role of ArhGAP44 in regulating filopodia initiation. In the revised version of the manuscript we discuss this now in the text.

As an added note: the possibility that ArhGAP44 may have additional functions in nodes (e.g. inhibit filopodia initiation AND elongation/outgrowth) will be discussed in more detail below in point 4.

3) The authors state “Our experiments identified the small GTPase Rac as the principal target of Nadrin2 within actin patches ().” The indicated figure does not demonstrate this at all.

We agree. The statement has been replaced.

4) The assumption is that Nadrin2 acts through Rac1 only, not Cdc42, but the alternative (interaction with Cdc42) is not tested. Can this really be dismissed out of hand? tests the relative levels of GTP-Rac1, but does not test whether Nadrin2 or its mutant alter levels of other potential targets (e.g. Cdc42).

A related but unstated presumption is that the dynamic assays of predict total protrusion number, and expressing Cdc42 would provide a test of that as effect of its overexpression is opposite to Rac1. Does Cdc42 overexpression alter the dynamic measures in a way opposite to Rac1? Does it decrease the number of nodes?

This is an excellent question. Considering the dual specificity of ArhGAP44 to Rac1 and Cdc42 (as we cite in the paper), ArhGAP44 also likely has an added function in regulating Cdc42 during filopodia initiation (it is well established that Cdc42 is critically involved in filopodia formation (Marrs et al., 2001; Matus, 2000; Lau et al., 1999)). In the revised version, we have added 3 experiments as well as raw material to characterize the potential interplay between ArhGAP44 and Cdc42:

ArhGAP44 can hydrolyze GTP-Cdc42: Consistent with previous reports, we find hydrolysis of GFP-Cdc42 by ArhGAP44 (Figure 2–figure supplement 5A).

Overexpression of Rac1(wt) but not of Cdc42(wt) phenocopies protrusion dynamics upon knockdown of ArhGAP44: We have previously shown that knockdown of ArhGAP44 increases density (Figure 1G, yellow) and dynamics (Figure 2B, yellow) of dendritic protrusions, while overexpression of ArhGAP44 reduces protrusion density (Figure 1E, blue) and protrusion dynamics (Figure 2B, blue). As ArhGAP44 is a RhoGAP, overexpression of the small GTPase that is targeted by ArhGAP44 should phenocopy the knockdown of ArhGAP44. We previously showed that overexpression of Rac1(wt) increases density (Figure 2–figure supplement 5B, red) and dynamics (Figure 2E, red) of dendritic protrusions, and rescues reduced protrusion dynamics upon overexpression of ArhGAP44 (Figure 2E, purple). For Cdc42, we showed in the initial submission that overexpression of Cdc42(wt) did not increase protrusion density (Figure 2–figure supplement 5B, green). We have now added experiments that show that overexpression of Cdc42(wt) has no effect on protrusion dynamics, and does not rescues reduced protrusion dynamics upon overexpression of ArhGAP44(R291M) (Figure 2–figure supplement 5C). The latter suggests that Rac regulation is more important but does not exclude an additional regulatory role of Cdc42.

Overexpression of Rac1(wt) but not of Cdc42(wt) phenocopies protrusion morphology upon knockdown of ArhGAP44: In the revised version of the manuscript, we added raw material showing that overexpression of Rac1 but not of Cdc42 phenocopied the ArhGAP44 knockdown causing the formation of dynamic dendritic nodes and filopodia (Figure 2–figure supplement 5D and Videos 3, 4, 6–10). Again, this suggest for a more important role of Rac but does not exclude an added regulation of ArhGAP44 on Cdc42.

Together, while these results argue for ArhGAP44-dependent regulation of Rac1 in actin patches, we agree with the referee that a discussion about the function of Cdc42 has been missing in the initial manuscript. Following discussion on the role of Cdc42 have been added to the text: Cdc42 acts as an activator of Irsp53 (Lebrand et al., 2004), promoting IRSp53-dependent enrichment and clustering of VASP and other factors to drive actin assembly in elongating filopodia (Krugmann et al., 2001). Consistently, knockdown of Cdc42 substantially reduces filopodia formation in neurons (Richnau & Aspenstrom, 2001). Intriguingly, overexpression of Cdc42 is not sufficient to initiate filopodia formation in neurons (Figure 2–figure supplement 5, see also (Rollason et al., 2009)) or in other cell lines (Lebrand et al., 2004). This has led to the hypothesis that elongation of filopodia is a combinatorial process requiring multiple factors (Lebrand et al., 2004). We propose that signal integration at actin patches controls this decision of filopodia elongation. Considering that actin-patch formation occurs before filopodia elongation, this argues for a 2-step process where Rac1-induced patch formation (and ArhGAP44-dependent regulation thereof) precedes Cdc42-induced filopodia elongation (Figure 6). However, since ArhGAP44 shows dual specificity for Rac1 and Cdc42, both steps will be limited by recruitment by ArhGAP44 to actin patches.

5) Presumably, dynamics were not measured following Nadrin2(wt) overexpression because there are so few filopodia left, but shouldn't the Rac1+Nadrin2(wt) case be examined, since Nadrin2(wt) should suppress the Rac1 phenotype? That is the claim in the model of , left column.

Expression of ArhGAP44(wt) rapidly triggers varicosity formation and cell death (Figure 1–figure supplement 6), likely due to excessive levels of the enzyme. Consequentially, either enzymatic efficiency or expression levels need to be reduced to study ArhGAP44 function in neurons. Since western blot analysis and live cell experiments show that ArhGAP44(R291M) is less potent than ArhGAP44(wt) but still active (Figure 1F and Figure 1–figure supplement 6), we decided to use ArhGAP44(R291M) for the synthetic rescue experiments. In contrast to a system that would rely on reduced expression of ArhGAP44(wt), this approach has the additional advantage of providing a strong fluorescence signal that is critical to identify transfected cells and study sub-cellular protein localization.

6) Surprisingly, the movies do not make a great case for this paper, and in fact in my mind raise more questions than they answer. Technically, the intensity levels fluctuate substantially, making me wonder whether small or thin structures such as nascent filopodia or nodes could really be identified with confidence. One troubling example is in the upper right panel, going from frame 1 to 2. The structure seemingly of most interest is the node or protrusion in the middle of the process (under “n2”), which appears to pop into existence. But the entire image brightens in this panel, making it unclear whether to trust the relative intensity of the node. Related to that type of phenomenon, some of the fluctuation is apparently due to an inconsistent z span; e.g. in Movie 1, the overexpression of Nadrin2 panel clearly shows the middle one of the processes fade out and then return, as it was insufficiently spanned by the planes of the z sectioning. In the upper right panel, that particular node could just have been captured more completely in bounds of the z stack rather than appear or brighten in reality.

In terms of observations, it doesn't seem as though the control example in movie 1 demonstrates any protrusion formation at all. Am I missing something? Based on the quantification in , there should be just as many forming and collapsing processes as there are processes that persist through the movie, but I don't see this at all. At the very least, I'd say that readers will need some help: formation, collapse, and unclassified changes should probably be pointed out with markers of some sort.

Raw data (movies) of the Rac1 and Rac1+Nadrin2(R291M) experiments would be good to see.

More raw data (stills or movies) from the key experiments in is needed.

(A) Added raw material: We agree with the reviewer that addition of more raw data would be helpful to (i) explain what types of dendritic protrusions exist, and (ii) show how these protrusions are affected by overexpression and knockdown of ArhGAP44 or the small GTPases Rac1 and Cdc42. To better explain and illustrate the main findings of the analysis shown in Figure 2, the following 11 raw data movies and 3 figure supplements have been added:

To illustrate different types of protrusions formed on dendrites, and better explain the analysis used in Figure 2:

Timelapse examples of static and dynamic protrusion types characterized in Figure 2 (Figure 2–figure supplement 2A and B).

Raw data movie showing stable dendritic filopodia (Video 1, filopodia lasts for >40 minutes).

Raw data showing formation of dynamic dendritic nodes (Figure 2-figure supplement 3 and Video 5).

Video 5.

Example of dendritic node formation.

Neuron was transfected with a fluorescence marker at DIV11 and imaged 24hr later. Individual frames were taken every 60 s. Scale bar is 2 μm. Video is 360× real-time.

DOI: http://dx.doi.org/10.7554/eLife.03116.022

To illustrate the findings of Figures 2B and 2C:

Raw data movie showing increased dendritic node and reduced protrusion formation upon ArhGAP44(R291M) overexpression (Videos 2).

Raw data showing increased dendritic node and protrusion formation upon knockdown of ArhGAP44 (Figure 2–figure supplement 5D and Videos 3, 4).

Raw data movie showing dendritic protrusion emerging from dendritic nodes upon knockdown of ArhGAP44 (Video 6).

To illustrate the findings of Figures 2E and 2F:

Raw data examples showing abnormal dendritic node and filopodia formation upon Rac1 overexpression (Figure 2–figure supplement 5D and Videos 7, 8).

Raw data examples showing abnormal dendritic filopodia formation upon Cdc42 overexpression (Figure 2–figure supplement 5D and Videos 9, 10).

Raw data movie showing synthetic rescue of ArhGAP44(R291M)-dependent reduction in filopodia formation by co-overexpression of Rac1(wt) (Video 11).

(B) Protrusions ‘pop into existence’? As the referee correctly state, we only observe structures in the illuminated plane of the confocal microscope. This means that dynamic dendritic sections that move in z-direction during the 10 minute long acquisition period may cause filopodia to leave the confocal plane. To investigate the frequency and origins of protrusions that ‘pop into existence, the following experiments were added:

Timelapse analysis of filopodia formation using a shorter acquisition interval: As shown in Figure 2–figure supplement 4B, 57% ± 8% of all filopodia elongate from nodes, while the remaining filopodia appear to emerge directly from the dendritic shaft (i.e. ‘pop into existence’) using an acquisition interval of 60 seconds (between frames). However, when the acquisition interval was reduced from 60 seconds to 15 seconds (i.e. between frames), the fraction of filopodia emerging from nodes increased to 83% ± 7%.

Time series showing dendritic filopodia emerging from node (Figure 2–figure supplement 4A). This example shows that the acquisition interval of 60 seconds is sufficient to detect protrusions, but may in some cases miss nodes (Figure 2–figure supplement 4A, compare columns 2 and 4).

These added experiments provide evidence that the majority of filopodia that ‘pop into existence’ are reflective of filopodia emerging from dendritic nodes that were to short-lived to be captured using a 60-second interval. We consider the remaining 17% of filopodia that ‘pop into existence’ to be a mixure of (i) filopodia emerging from nodes in less than 15 seconds (i.e. between frames), (ii) filopodia emerging directly from the dendritic shaft, or, more likely, (iii) filopodia that appear due to changes in the focal plane of the dendritic tree.

7) In , the accumulation of a set of SEM images into a numbered series of panels portrayed as “stages” is unnecessary and misleading-there is no evidence presented here or cited that the stages correspond to a growth sequence. There is a near-total lack of information about how these images were assessed or quantified, and what constituted a node or a protrusion or any other feature.

The reviewer raised a valid point. As the sequence of electron micrographs that was previous shown in Figure 3D is an interpretation of data, it has now been integrated in the proposed model (Figure 6). To better describe how the images in Figure 1D were quantified, the following supplemental figure panels and methods section have been added:

Figure 1–figure supplement 5 illustrates how dendritic protrusions were quantified in scanning electron micrographs.

New chapter in the Material and methods section: Quantification of Protrusion Types using Scanning Electron Micrographs. Neurons were cultured on glass slides for various periods of time (3, 10 and 17 days), fixed and prepared for SEM as described above. Using low resolution (1000x magnification), individual neurons were identified (Figure 1–figure supplement 5A, left panel). Starting from the soma, initial segments of the dendritic arbors were imaged at high resolution (10’000x), and individual protrusions were classified based on morphology (Figure 1–figure supplement 5, right panel). Only the proximal 50-60 μm of the dendritic arbors that can clearly be associated to a particular neuron were analyzed. Examples of dendritic nodes are shown in Figure 1–figure supplement 5B.