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

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).

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).

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) 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.

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).

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 (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.

(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 (Figure 2–figure supplement 1B).” 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? 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?

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 Figure 5B, 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 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.

(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).

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 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.

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.