Small GTPases and BAR domain proteins regulate branched actin to make clathrin and dynamin independent endocytic vesicles

Numerous endocytic pathways operate simultaneously at the cell surface. Here we focus on the molecular machinery involved in the generation of endocytic vesicles of the clathrin and dynamin-independent CLIC/GEEC (CG) pathway. This pathway internalises many GPI-anchored proteins and a large fraction of the fluid-phase in different cell types. We developed a real-time TIRF assay using pH-sensitive GFP-GPI to identify nascent CG endocytic sites. The temporal profile of known CG pathway modulators showed that ARF1/GBF1 (GTPase/GEF pair) and CDC42 (RhoGTPase) are recruited sequentially to CG endocytic sites, ∼60s and ∼9s prior to scission. Using a limited RNAi screen, we found several BAR domain proteins affecting CG endocytosis and focused on IRSp53 and PICK1 that have interactions with CDC42 and ARF1 respectively. IRSp53, an I-BAR domain containing protein, was recruited to the plasma membrane at the site of forming CG endocytic vesicles and in its absence, nascent endocytic CLICs, did not form. The requirement for actin polymerization in the CG pathway suggested a role for nucleators of actin polymerization, and ARP2/3 was found enriched at the site of the forming endocytic vesicle. PICK1, a BAR domain containing protein and the ARP2/3 inhibitor is recruited at an early stage along with ARP2/3, but is removed from the endocytic site coincident with CDC42 recruitment and a burst of Factin polymerization. This study provides a spatio-temporal understanding of the molecular machinery necessary to build a CG endocytic vesicle.

provided a reliable real-time assay for studying the spatio-temporal dynamics of the 137 internalisation of GPI-anchored proteins (See, S.I.).

139
Newly formed CG endocytic vesicles recruit GBF1, ARF1, and CDC42 140 We visualised the temporal dynamics of co-expressed CDC42 at the site of formation 141 of nascent SecGFP-GPI endocytic vesicles. TagRFPt-CDC42 recruitment was quantified as 142 average fold accumulation of CDC42 relative to the local background in all the endocytic 143 events recorded over a 40s time window straddling the endocytic event. A significant change 144 in the intensity of CDC42 began at around -9s and peaked around 0 to +3s ( Figure S2a). 145 Based on the presence or absence of a co-detected CDC42 spot (within -18 to +18s time 146 window) at the endocytic site, we identified two populations via manual classification (See 147 S.I. for a detailed description). We called them CDC42 Coloc (Co-detection of CDC42 and 148 SecGFP-GPI) and CDC42 NoColc (the remainder). We compared the fold accumulation of 149 all the CDC42 spots (CDC42 All), CDC42 Coloc and CDC42 NoColoc with Random. While 150 the CDC42 Coloc profile was similar to that of CDC42 All, the CDC42 NoColoc profile was 151 comparable to Random ( Figure S2b). Thus, the endocytic sites detected by our assay 152 consisted of two populations wherein one fraction exhibited an accumulation of CDC42 153 while the second fraction failed to show a discernable accumulation. CDC42 Coloc 154 corresponded to the 56% (of the CDC42 All) SecGFP-GPI endocytic sites. As the removal of 155 the events which did not coincide with the presence of CDC42 did not alter CDC42 156 recruitment profile (compare, Figure 1d & S1a), they were discarded from further analysis. currently serves as the earliest initiators of the CG endocytic pathway. Furthermore, when we 171 assessed the ultra-structure of newly formed fluid-filled endosomes by electron microscopy 172 (EM) 18 in cells treated with the small molecule inhibitor of GBF1, LG186 34 , the number of 173 CLICs and fluid-uptake was drastically reduced, whereas, CME derived vesicles and uptake 174 appear relatively unaffected (Figure S1e-f). In contrast to ARF1, GBF1 and CDC42, we did 175 not observe a frequent recruitment of clathrin and dynamin to the CG endocytic sites. The 176 recruitment profile was comparable to random for >65% endocytic events for both mCherry-  Table 2). The remainder of the profiles exhibited high 178 levels of clathrin and dynamin at SecGFP-GPI sites in a fashion unlike that observed for 179 CME (Ref. 5 & Figure S2c). This may account for the endocytic uptake of a fraction of 180 SecGFP-GPI via a CME like process as shown previously (Ref. 12, and might 181 explain why we failed to quantitatively localise CDC42, ARF1 and GBF1 to all the SecGFP-182 GPI-endocytic events as noted above. 183 The pH pulsing analysis revealed a surprising facet of the recruitment of ARF1, which 184 suggested a CDC42-independent function(s) for ARF1 in CG endocytosis. Despite being 185 responsible for the timed removal of CDC42 from the plasma membrane 15 , ARF1 was 186 recruited long before the recruitment of CDC42. Taken together these results establish a 187 reliable real-time imaging assay to follow newly formed CG endocytic vesicles containing 188 SecGFP-GPI, correlated with recruitment of its known regulators, ARF1, GBF1 and CDC42. IRSp53 to the plasma membrane was compromised following ARF1 depletion 48 . Hence,

214
IRSp53 was a good candidate to act as a signalling platform, linking CDC42 activation, 215 membrane curvature and actin regulation for CG endocytosis. 216 To address the function of IRSp53 we compared endocytosis between mouse 217 embryonic fibroblasts (MEFs) generated from IRSp53-/-mice (IRSp53-/-MEFs) and  This led us to hypothesise that CG cargo would traffic via CME in the absence of 231 IRSp53. Therefore, we looked at the trafficking of GPI-AP (GFP-GPI), fluid-phase and TfR 232 (CME) in the absence of IRSp53 at a higher resolution. We first, we counted the number of 233 GFP-GPI and fluid endosomes and found them to be significantly lower in IRSp53-/-MEFs specifically affected fluid-phase and GPI-AP endocytosis, while CME remained unaffected.

239
Moreover, in cells lacking IRSp53, GPI-AP and the residual fluid-phase is endocytosed via 240 CME. 241 We next analysed the contribution of different domains of IRSp53 on CG endocytosis 242 by re-introducing into IRSp53-/-MEFs, GFP-tagged IRSp53WT and a number of mutants of 243 IRSp53 specifically defective in various domains (Figure 4d, schematic). We found that GFP-

244
IRSp53WT & GFP-IRSp53V522G rescued endocytosis while the rest of the mutants failed to 245 do so (Figure 4d). In conclusion, these results indicated that IRSp53 is an essential and 246 specific regulator of CG endocytosis that requires functional I-BAR, CRIB and SH3 domains.

248
IRSp53 is recruited to forming CG endocytic vesicles 249 The complete absence of CG endocytosis in IRSp53-/-led us to hypothesise that 250 IRSp53 has a direct role to play in CG vesicle formation. Hence, we used pH pulsing assay  Table 2). Since the I-BAR domain of IRSp53 has been shown to 254 sense/induce negative curvature in a membrane tension and protein concentration dependent 255 manner 49 , we looked at changes in its spatial distribution during the formation of endocytic 256 vesicle using two types of masksa spot and a ring mask (Figure 5a, schematic). Unlike the 257 intensity profiles of CDC42 that did not exhibit any differential temporal patterns of  Table 1).

266
To visualise the recruitment of GFP-IRSp53 at high spatial resolution, we 267 coexpressed a GBP-APEX reagent (GFP binding protein-soybean ascorbate peroxidase) and 268 processed for EM as described previously 50 (Figure 5c Table 2) began earlier than -35s and peaked at -290 6s. This was unexpected since CDC42, a key regulator of ARP2/3 56,57 was not recruited until 291 -9s. Instead, the ARP3 profile was correlated with IRSp53 between -9s and +9s (r = 0.8; 292 Table 1). This indicated that, at least initially, the ARP2/3 complex may be recruited in a 293 CDC42-independent manner. On the other hand, F-actin recruitment began around -9s and 294 continued even after the scission event (Figure 6c, S6e & Table 2), highly correlated to the 295 CDC42 profile (r = 0.6; Table 1). Thus, F-actin is generated coincident with the recruitment 296 of CDC42, a known regulator of actin polymerization 56,57 . These observations suggest that 297 ARP2/3 complex might be first recruited as an inactive ARP2/3 complex, and then activated 298 following the arrival of CDC42.  forming CG endosome at the early stage, rendering ARP2/3 inactive.

324
To explore this possibility, we utilised the pH pulsing assay to directly visualise 325 PICK1 at the CG endocytic site. We found that TagRFP-PICK1 was recruited to forming CG

335
CG endocytosis was initially discovered as a route for the entry of toxins 62 , and fluid-336 phase and GPI-anchored proteins when CME 12,13,63 was perturbed, raising some concerns 337 regarding its physiological role in unperturbed cells 19 . Using pH pulsing assay we show, 338 here, that a majority of SecGFP-GPI-containing endocytic vesicles form due to a 339 stereotypical and temporally orchestrated recruitment of the key molecular machinery namely, CDC42, ARF1 and GBF1, which, in turn, mediate the coordinated assembly of 341 specific BAR-containing, membrane deforming and actin regulatory proteins, IRSp53 and 342 PICK1. Thus, a dedicated complex protein machinery drives CG internalisation, similarly to 343 what is observed in CME. Notably, however, the vast majority of the endocytic vesicles are 344 devoid of clathrin and dynamin. The ability of the pH pulsing assay to provide a temporal 345 profile for the recruitment dynamics of the molecular players has considerably extended our 346 understanding of CG endocytic vesicle formation.

347
The GBF1/ARF1 pair is the earliest module that is assembled and judging by their 348 recruitment profiles, it takes around 1 minute and more, to assemble the molecular machinery  The role of ARP2/3 in CG endocytosis is reminiscent of the endocytic process 362 occurring in the budding yeast. In this system, the endocytic machinery strictly depends on 363 Las17, the yeast homologue of N-WASP but not so much on clathrin and dynamin 64,65 . 364 There are however important differences. In CG endocytosis, the ARP2/3 complex appears to   Figure 6e). By contrast, in CME, both ARP2/3 complex and N-WASP are 369 recruited to budding CME vesicles, and influence endocytosis in some cell types 27 .

370
The unexpected recruitment profile of ARP2/3, and the identification of two BDPs, 371 PICK1 and IRSp53 as upstream regulators of ARP2/3 activity also suggests a biphasic 372 mechanism. PICK1 operates in the early phases, and may function as an inhibitor of ARP2/3, 373 consistent with the modest recruitment of F-actin in the presence of PICK1 observed here and as reported previously 58,59 . PICK1 recruitment occurs via its BAR domain since when 375 mutated and overexpressed, it acts as a dominant negative for CG endocytosis ( Figure S5c).

376
PICK1 recruitment is not only rapid but also transient. We find that the localisation of PICK1 377 to the membrane was negatively correlated to the activity of ARF1, similar to that observed 378 in neuronal cells 58,66 wherein GTP-ARF1 interaction with PICK1 rendered PICK1 incapable 379 of inhibiting ARP2/3 complex. This is followed by the second phase characterised by the where it would regulate the actin machinery necessary to trigger CG vesicle scission. The 393 spatio-temporal dynamics of IRSp53 at CG sites in consistent with this model. Furthermore, 394 the complete and specific loss of CG endocytosis (but not CME) in the absence of IRSp53 395 makes IRSp53 essential for CG endocytic process.

396
In summary, we propose that CG endocytic vesicle formation begins with 397 GBF1/ARF1 concentrating at sites of endocytic pits. Following this, initial event ARP2/3 is       191-196 (2009   Laboratory) for making the schematic in Figure 1a and Figure 8.         Table 2). The random traces were derived from randomly assigned spots of the same radius  (1 pixel = 84nm). Random (red) are generated randomly within the cell mask.
Step 2 -Spot 819 mask (green filled) and background mask (green dashed) considered around the centroid the 820 new spot identified in the step -1 [also see pH 5 frame in the middle panel of (Figure 1a)].

821
The process is repeated for Random spot mask (red filled) and background mask (red dashed).
Step 3 -schematic of the temporal profile of a new spot (black) and a random spot. NoColoc; 44%) and its corresponding random intensity trace (n,