Giant ankyrin-B mediates transduction of axon guidance and collateral branch pruning factor sema 3A

Variants in the high confident autism spectrum disorder (ASD) gene ANK2 target both ubiquitously expressed 220 kDa ankyrin-B and neurospecific 440 kDa ankyrin-B (AnkB440) isoforms. Previous work showed that knock-in mice expressing an ASD-linked Ank2 variant yielding a truncated AnkB440 product exhibit ectopic brain connectivity and behavioral abnormalities. Expression of this variant or loss of AnkB440 caused axonal hyperbranching in vitro, which implicated AnkB440 microtubule bundling activity in suppressing collateral branch formation. Leveraging multiple mouse models, cellular assays, and live microscopy, we show that AnkB440 also modulates axon collateral branching stochastically by reducing the number of F-actin-rich branch initiation points. Additionally, we show that AnkB440 enables growth cone (GC) collapse in response to chemorepellent factor semaphorin 3 A (Sema 3 A) by stabilizing its receptor complex L1 cell adhesion molecule/neuropilin-1. ASD-linked ANK2 variants failed to rescue Sema 3A-induced GC collapse. We propose that impaired response to repellent cues due to AnkB440 deficits leads to axonal targeting and branch pruning defects and may contribute to the pathogenicity of ANK2 variants.


Introduction 19
collateral axon branches ( Figure 3B,C). As expected, AnkB440 signal was absent from axons of AnkB440 KO neurons 134 stained with the AnkB440 antibody ( Figure 3Di). Neither the overall axonal distribution nor the localization of AnkB440 at 135 GCs changed in neurons selectively lacking AnkB220 stained with AnkB440-specific ( Figure 3Dii) or pan-AnkB (Figure 3Diii) 136 antibodies. Interestingly, AnkB440 localized to both microtubule-and F-actin-enriched axonal domains ( Figure 3C,D), 137 respectively stained with βIII-tubulin and phalloidin, suggesting that it might promote the organization and/or dynamics 138 of both cytoskeletal networks. 139 140 AnkB440 interacts with L1CAM (Yang et al., 2019), which is required for transducing the repulsive growth and guidance 141 effects of the soluble guidance cue class III semaphorin A (Sema 3A) (Castellani et al., 2000;Castellani et al., 2002;142 Castellani et al., 2004). Sema 3A promotes GC collapse (Luo et al., 1993;Kolodkin et al., 1993) and inhibits axon branching 143 in vitro (Dent et al., 2004), which likely underlies its roles in repelling cortical axons and in pruning axonal branches in vivo 144 and in vitro (Polleux et al., 1998;Bagri et al., 2003). Thus, we next evaluated the response of AnkB deficient cortical 145 neurons to Sema 3A. We observed that roughly 60% of GCs of both control and AnkB220 KO cortical neurons collapsed 146 after 30 minutes of Sema 3A treatment, up from a baseline of 20% GC collapse in untreated cells, without any appreciable 147 difference between the two groups ( Figure 4A,B). In contrast, GCs of total AnkB KO and AnkB440 KO neurons significantly 148 failed to collapse when exposed to Sema 3A ( Figure 4A,C,D). The failure of GCs to collapse upon Sema 3A treatment in 149 AnkB440 KO neurons was restored by transfection with AnkB440 but not with AnkB220 plasmids . To rule 150 out that the inability of Sema 3A to collapse GCs in AnkB440 KO neurons was due to a generalized loss of transduction of 151 chemorepulsive cues, we treated control and AnkB440 neurons with Ephrin A5, which also induces GC collapse in cortical 152 neurons (Meima et al., 1999). Ephrin A5 promoted GC collapse in AnkB440 KO cultured cortical neurons at levels 153 indistinguishable from control neurons (Figure 4-figure supplement 4). These results indicate a selective requirement for 154 AnkB440 for cortical neuron responses to the GC collapse-inducing effect of Sema 3A. 155 156 The AnkB440-L1CAM complex promotes GC collapse in response to Sema 3A in control neurons, but was virtually lost in AnkB440 KO neurons and in L1CAM neurons harboring the p.Y1229H variant 161 ( Figure 5A), which lacks AnkB440 binding activity, consistent with previous findings in vivo and in vitro (Yang et al., 2019). 162 We further confirmed the specific interaction of L1CAM with AnkB440 using a cellular assay where co-expression of each 163 GFP-tagged AnkB isoform with L1CAM in HEK293T cells resulted in the effective recruitment of GFP-AnkB440, but not of 164 GFP-AnkB220, from the cytoplasm to the plasma membrane ( Figure 5- We next sought to identify the mechanisms by which AnkB440 and its interaction with L1CAM promotes GC collapse in 174 response to Sema 3A. First, we established that AnkB440 loss does not alter L1CAM expression in total lysates from PND1 175 AnkB440 KO cortex or from AnkB440 KO cortical neuron cultures ( Figure 6A-D, see Figure 6-source data 1-10). Using a 176 biotinylation assay that labels and captures surface proteins in cortical neuron cultures, we determined that AnkB440 loss 177 markedly reduced surface levels of L1CAM relative to its membrane abundance in control neurons, while surface levels of 178 the AMPA receptor subunit GluR1 remained unchanged ( Figure 6C,D, Figure 6-source data 1-10). These findings indicate 179 that AnkB440 is required to maintain normal levels of L1CAM at the cell surface of neurons. L1CAM is not a Sema 3A 180 receptor (Castellani et al., 2000). Instead, L1CAM directly binds the Sema 3A receptor neuropilin-1 (Nrp1) (Chen et al., 181 1998), which due to its short cytoplasmic domain requires the formation of complexes with transmembrane co-receptors 182 to be stably anchored at the cell surface and propagate repulsive Sema 3A signals (Castellani et al., 2000;Castellani et al., 183 2002). In addition to the L1CAM-Nrp1 complex, Plexin A1 also takes part in the transduction of Sema 3A signals (Bechara 184 et al., 2008). Thus, we next evaluated whether these two receptors participate in AnkB440-L1CAM modulation of Sema 8 total expression ( Figure 6C,D). In contrast, Plexin A1 levels remained unchanged both in AnkB440 KO brains ( Figure 6A,B) 188 and at the surface of AnkB440 KO neurons ( Figure 6C,D). Noticeably, Sema 3A levels were increased in the cortex of 189 AnkB440 KO PND1 mice ( Figure 6A,B), which together with the smaller increase in Nrp1 ( Figure 6A-D) points towards 190 protein upregulation as a likely mechanism to compensate for the observed loss of Sema  To confirm whether surface levels of L1CAM-Nrp1 complexes were altered in GCs of AnkB440 KO cortical neurons, we 194 selectively labeled surface Nrp1 in non-permeabilized DIV3 neurons with an Nrp1 antibody that recognizes an extracellular 195 epitope. Then, we labeled all L1CAM molecules upon cell permeabilization and visualized surface  using the PLA assay. PLA signal between surface Nrp1 and L1CAM, indicative of surface was 197 abundant at the GC and in the axon of control neurons, but significantly reduced in AnkB440 KO neurons (Figure 6E,F). 198 Consistent with the proposed role of AnkB440 in promoting L1CAM, and consequently, Nrp1 localization at the cell surface, 199 p.Y1229H L1CAM neurons in which the AnkB440-L1CAM association is disrupted, also showed a dramatic reduction in 200 surface L1CAM-Nrp1 PLA signal ( Figure 6E,F). Both AnkB440 KO and p.Y1229H L1CAM neurons exhibited reduced 201 internalization of surface Nrp1 in response to Sema 3A, consistent with previous reports showing that Nrp1 internalization 202 is required during Sema 3A-induced GC collapse (Castellani et al., 2004). Together, these results indicate that AnkB440 203 promotes the membrane surface localization of the L1CAM-Nrp1 holoreceptor complex, which is required to respond to 204 Sema 3A repulsive cues. 205 206 AnkB440 does not require βII-spectrin or binding to microtubules to transduce Sema 3A signals during GC collapse 207 AnkB directly binds βII-spectrin (Davis and Bennett, 1984;Davis et al., 2009), which is widely distributed along axons and 208 required for axonal elongation and organelle transport (Lorenzo et al., 2019). βII-spectrin binds F-actin and tetramers of 209 βII-spectrin/αII-spectrin organize a periodic submembrane network of F-actin and associated proteins throughout all 210 axonal domains (Xu et al., 2013). Thus, we next investigated whether βII-spectrin is required to propagate Sema 3A signals 211 to the F-actin network during GC collapse. GCs of cortical neurons harvested from mice lacking βII-spectrin in the brain 212 (βII-spectrin KO) (Galiano et al., 2012;Lorenzo et al., 2019) collapsed normally in response to Sema 3A treatment ( Figure 6-figure supplement 6A,B). This result indicates that AnkB440 promotes Sema 3A-induced GC collapse independently of 214 βII-spectrin. 215 216 AnkB440 also binds and bundles microtubules through a bipartite microtubule-interaction site located in its NSD (Chen et 217 al., 2020). This microtubule-binding activity is required to suppress ectopic axon branching in cultured neurons Chen et 218 al., 2020). The site of microtubule interaction in AnkB440 comprises a module of 15 tandem imperfect 12-aa repeats that 219 includes highly conserved residues in the third (Pro, P), fifth (Ser, S), and ninth positions (Lys, K) (Chen et al., 2020). Point 220 mutations in each of these residues in full length AnkB440 (PSK mutant) is sufficient to impair microtubule-binding activity 221 and cause axon hyperbranching in vitro (Chen et al., 2020). To test whether AnkB440-microtubule binding is required for 222 Sema 3A-induced GC collapse, we transfected Halo-tagged cDNA of AnkB440-PSK into AnkB440 KO cortical neurons and 223 treated them with Sema 3A. Like in control neurons, about 60% of GCs of neurons expressing AnkB440-PSK collapsed in 224 response to Sema 3A, indicating that the AnkB440-microtubule interaction is not required for normal transduction of Sema 225 3A signals ( Figure 6-figure supplement 6C,D). 226 227 AnkB440 and its interaction with L1CAM are required to modulate cofilin activity in response to Sema 3A 228 GC motility and collapse involves fast turnover and reorganization of actin filaments (Omotade et al., 2017). F-actin 229 dynamics is regulated by the actin depolymerizing and severing factor (ADF)/cofilin (Carlier et al., 1997;Maciver, 1998). 230 Cofilin phosphorylation by the Ser/Thr kinase LIM-kinase (LIMK) at the Ser3 site (Arber et al., 1998;Yang et al., 1998) 231 inactivates cofilin and prevents its F-actin severing activity (Agnew et al., 1995;Moriyama et al., 1995). Similarly, LIMK 232 phosphorylation of cofilin and the rapid subsequent cofilin activation have been shown to be critical steps in the 233 disassembly of actin filaments during Sema 3A-induced GC collapse (Aizawa et al., 2001). Thus, we evaluated whether loss 234 of AnkB440 led to changes in expression of LIMK and cofilin, or in cofilin phosphorylation. We found that total levels of 235 LIMK and cofilin were not altered in cortical lysates from PND1 AnkB440 brains ( Figure 7A,B, see Figure 7-source data 1-236 5). However, AnkB440 loss decreased the ratio of inactive phospho-cofilin (pcofilin S3 ) to total cofilin by 40% relative to 237 control ( Figure 7A,B, see Figure 7-source data [1][2][3][4][5]. This increase in active cofilin in AnkB440 mice during early brain 238 development might provide a more dynamic actin pool and could underlie the surges in axonal actin patches and emerging 239 filopodia observed in AnkB440 KO cortical neurons. 240 To determine whether AnkB440 participates in Sema-3A-induced regulation of actin dynamics through the action of 242 cofilin, we examined the effect of Sema-3A on the localized activation/inactivation of cofilin at GCs through confocal 243 microscopy. As previously observed in cultured dorsal root ganglion neurons (Aizawa et al., 2001), levels of p-cofilin rose 244 rapidly above 3-fold in the GC of control cortical neurons during the first minute after exposure to Sema 3A, but underwent 245 a sharp reduction during the next four minutes to 43% of basal levels ( Figure 7D,E). This sharp F-actin stabilizing period 246 followed by a fast increase in F-actin depolymerization is thought to reflect the reorganization of the actin cytoskeleton 247 during GC collapse (Aizawa et al., 2001). Consistent with this plausible signaling cascade and lesser GC collapse due to a 248 diminished response to Sema 3A, GCs of both AnkB440 KO and p.Y1229H L1CAM neurons showed a different pattern of 249 cofilin phosphorylation. First, basal p-cofilin signal per GC area was roughly 40% and 30% lower in AnkB440 KO and 250 p.Y1229H L1CAM neurons, respectively, relative to control ( Figure 7D,E), which is consistent with lower levels of p-cofilin 251 in AnkB440 brains ( Figure 7A,C). Second, in contrast to the higher than three-fold increase in control neurons, the rise in 252 p-cofilin during the first minute of Sema 3A treatment was below two-fold in both AnkB440 KO and p.Y1229H L1CAM GCs, 253 which represented only 30-40% of control levels. Lastly, in AnkB440 KO and p.Y1229H L1CAM GCs the reduction of p-254 cofilin five minutes past Sema 3A treatment was approximately only 20% lower than its peak at one minute and remained 255 around 50% above basal levels ( Figure 7D,E). These aberrant patterns of cofilin regulation indicate that AnkB440 and its 256 association with L1CAM promote steps of the Sema 3A signal transduction pathways upstream of changes in cofilin 257 activation and F-actin disassembly during GC collapse. 258 259 Autism-linked ANK2 variants affect the transduction of Sema 3A repulsive cues 260 The results above support a mechanism wherein AnkB440 is required to modulate the initiation, growth, and 261 establishment of axon collateral branches and to properly respond to Sema 3A repulsive cues that modulate GC collapse, 262 axon guidance, and pruning. Consequently, AnkB440 deficiencies can result in aberrant structural and functional axon 263 connectivity, which in turn may contribute to the pathogenicity of ANK2 variants in ASD. AnkB440 R2589fs mice, which models 264 the de novo p.(R2608fs) frameshift variant in exon 37 of ANK2 found in an individual diagnosed with ASD, exhibit ectopic 265 axonal connections assessed by brain DTI and axonal hyperbranching in vitro (Yang et al., 2019). Instead of full-length the entire death and C-terminal regulatory domains (Yang et al., 2019) (Figure 8A,B asterisk, see Figure 8-source data 1, 268 2, 5). Given that this truncated protein fails to associate with L1CAM in vivo (Yang et al., 2019), we tested whether its 269 expression could also disrupt Sema 3A-induced GC collapse. Like AnkB440 KO and p.Y1229H L1CAM neurons, 270 AnkB440 R2589fs neurons had diminished GC collapse response to Sema 3A ( Figure 8C,D). Thus, in addition to altered 271 microtubule stability, impaired responses to Sema 3A during GC collapse likely contribute to the axonal branching and 272 connectivity deficits observed in AnkB440 R2589fs brains and neurons. 273

274
Over 70 ANK2 variants have been identified in individuals diagnosed with ASD (De Rubeis et al., 2014;Iossifov et al., 2014;275 Iossifov et al., 2015). ASD-linked ANK2 variants target the NSD or domains shared by both AnkB440 and AnkB220 isoforms. 276 We previously reported that de novo ASD variants p.(P1843S) and p.(E3429V) in the NSD of AnkB440 ( Figure 8A) failed to 277 rescue axon hyperbranching of AnkB440 KO cortical neurons (Yang et al., 2019). Therefore, we evaluated whether these 278 AnkB440-specific variants could restore GC responses to Sema 3A of AnkB440 KO neurons. We also tested expression of 279 Halo-tagged AnkB440 bearing the de novo variant p.(R1145Q) ASD variant in exon 30, which encodes a portion of the ZU5 C 280 domain common to both AnkB220 and AnkB440 ( Figure 8A). AnkB440 KO cortical neurons expressing these ASD-linked 281 AnkB440 variants failed to rescue GC collapse in response to Sema 3A ( Figure 8E,F). These variants do not appear to affect 282 protein stability, given that they expressed normal levels of full-length AnkB440 protein when transfected in HEK293T 283 cells, which lack endogenous AnkB440 expression ( Figure  ankyrin-G (AnkG), associated with bipolar disorder (Baum et al., 2008;Ferreira et al., 2008) and schizophrenia (Cruz et al.,

12
have been identified in individuals with ASD and intellectual disability (De Rubeis et al., 2014;Iossifov et al., 2014;Iossifov 295 et al., 2015) and it is ranked as a top high confidence ASD gene with one of the highest mutability scores (Ruzzo et al., 296 2019). Despite their structural similarities and degree of sequence conservation, ankyrins diverge in cell type expression, 297 subcellular localization, and protein partners in the brain (Lorenzo, 2020). For instance, AnkG preferentially localizes to 298 the axon initial segment (AIS), where it acts as the master regulator of AIS organization, while AnkB is widely distributed 299 throughout the axon (Lorenzo et al., 2014;Yang et al., 2019;Lorenzo, 2020). Ankyrins achieve a second order of functional 300 specialization through alternative splicing, which yields giant isoforms with unique inserted sequences in both AnkG and 301 AnkB in neurons (Bennet and Lorenzo, 2016;Lorenzo, 2020). ASD-linked AnkB variants fall both within the inserted region 302 unique to AnkB440 and in domains shared by AnkB440 and AnkB220 isoforms. 303 The present study sheds new light into the isoform-specific functions of AnkB in modulating axonal architecture and 304 guidance in the developing brain and the potential contribution of AnkB deficits to ASD pathology. We previously showed 305 that expression of the truncated product of the frameshift mutation p.P2589fs that models the de novo p.(R2608fs) ASD 306 variant in AnkB440 causes stochastic increases in structural cortical connectivity in mouse brains and exuberant collateral 307 axon branching in cortical neuron cultures (Yang et al., 2019). We confirmed the development of axon hyperbranching in 308 cultured neurons selectively lacking AnkB440 (Chan et al., 2020), but not in neurons from AnkB220 KO mice. Using Golgi 309 staining, we determined that AnkB440 KO, but not AnkB220 KO, cortical neurons grow more collateral branches, 310 consistent with the in vitro results. We also found thickening of the corpus callosum in AnkB440 KO mice, which could 311 result from volumetric gains caused by hyperbranching of callosal axons. These findings support critical and specialized 312 roles of AnkB440 in modulating axon collateral branch formation and pruning. 313 Previous results implicated AnkB440 in suppressing collateral branch formation via the stabilization of microtubule 314 bundles near the plasma membrane (Yang et al., 2019;Chan et al., 2020). Loss of AnkB440 or expression of the 290-kDa 315 truncated AnkB440 product promotes microtubule unbundling, which facilitates microtubule invasion of the nascent 316 filopodia, a necessary step in the commitment to forming a new branch (Yu et al., 1994). The formation of transient actin 317 nucleation sites marks the point of formation and precedes the emergence of the precursor filopodia (Gallo, 2011). In this 318 study, we show that selective loss of AnkB440 KO in cultured cortical neurons results in larger number of actin patches 319 relative to control and AnkB220 KO neurons, which correlated with higher number of axonal filopodia and collateral branches. Thus, AnkB440 modulates collateral branch formation through a combined mechanism that suppresses the 321 stochastic formation of collateral filopodia by lowering the density of F-actin-rich branch initiation points, as well as the 322 invasion of microtubules into the maturing filopodia. Although F-actin patches serving as precursor to filopodia have been 323 observed in vitro (Loudon et al., 2006;Spillane et al., 2011) andin vivo (Spillane et al., 2011;Hand et al., 2015), little is 324 known about the factors that regulate their formation and dynamics (Armijo-Weingart and Gallo, 2017). How AnkB440 325 modulate actin patches is not clear. The crosstalk between actin and microtubule networks have been proposed to 326 orchestrate the emergence of branches (Dent and Kalil, 2001;Pacheco and Gallo, 2016), and splaying of microtubule 327 bundles have been observed to correlate with F-actin accumulation at branch points (Dent and Kalil, 2001). Thus, it is 328 possible that microtubule unbundling resulting from AnkB440 deficits promotes the seeding and growth of F-actin patches 329 at sites of axon branch formation. Alternatively, AnkB440 may associate with actin through their common partner βII-330 spectrin, although loss of βII-spectrin does not lead to axonal hyperbranching in vitro and may not affect the number or 331 dynamics of actin patches (Lorenzo, et al., 2019). While loss of AnkB220 does not affect axonal branching in vitro and in 332 mouse brains, its expression in AnkB440 null neurons may be necessary to sustain hyperbranching. This is possibly due to 333 the role of AnkB220 in axonal organelle and vesicle transport and growth, which are not affected by exclusive loss of 334 AnkB440. 335 While cell-autonomous factors that modulate axonal branching in vitro, such as the formation of actin patches, correlate 336 with the number of axon branches (Gallo, 2011), live imaging of actin dynamics during in vivo axonal development found 337 that that correlation does not hold in projection neurons in the mouse cortex (Hand et al., 2015). Instead, the number of 338 collateral axon branches along the axon is determined by the cortical layer they transverse, indicating that extrinsic factors 339 that modulate branch formation and pruning may be involved in determining the pattern of axonal innervation (Hand et 340 al., 2015). Several reports support the activity of Sema 3A as an extrinsic repellent cue that inhibits axon branching in vitro 341 (Dent et al., 2004) and prunes cortical axons and collateral axon branches in vivo (Polleux et al., 1998;Bagri et al., 2003). 342 We show that AnkB440 is required to transduce repellent Sema 3A signals in vitro to facilitate the collapse of GCs from 343 the axon and collateral branches, which offers a plausible cellular mechanism underlying the ectopic neuronal connectivity composed of L1CAM-Nrp1 at the cell surface (Castellani et al., 2000;Castellani et al., 2002;Castellani et al., 2004). 347 Interestingly, GCs of neurons lacking the F-actin and AnkB partner βII-spectrin respond normally to Sema 3A, which 348 indicates that βII-spectrin is not required for Sema 3A-induced GC collapse. However, given βII-spectrin's role in the 349 development and wiring of axons in mouse brains (Galiano et al., 2012;Lorenzo et al., 2019) and the recent identification 350 that pathogenic variants in SPTBN1, which encodes βII-spectrin, cause a neurodevelopmental syndrome associated with 351 deficits in cortical connectivity (Cousin et al., 2020), we cannot rule out its involvement in axonal guidance through 352 alternative mechanisms. Interestingly, although AnkB440 binding to microtubules suppresses branch initiation, loss of this 353 interaction does not affect the response to Sema 3A during GC collapse. Instead, transduction of Sema 3A signaling via 354 the AnkB440-L1CAM-Nrp1 complex modulates F-actin dynamics through LIMK phosphorylation of cofilin. 355 Our structure-function studies in AnkB440 KO cortical neurons found that de novo ASD variants p. ( and it may affect AnkB's ability to bind PI3P lipids, given its proximity to the PI3P lipids binding region (Lorenzo et al., important to determine whether AnkB440 binds PI3P lipids and the significance of AnkB440-PI3P lipid binding activity for 373 axonal development and axonal connectivity in the developing brain. 374 Besides the significance of ANK2 in ASD, neuronal pathways involving the AnkB440-L1CAM complex are also relevant to 375 other neurological diseases. For example, a few hundred variants in L1CAM have been described in individuals with CC 376 hypoplasia, retardation, adducted thumbs, apasticity and hydrocephalus (CRASH) syndrome (Rosenthal et al., 1992;Jouet 377 et al., 1994;Weller and Gärtner, 2001;Vos et al., 2010). Pathological L1CAM variants can affect both its extra-and 378 intracellular domains and disrupt binding to molecular partners including AnkB and Nrp1 (Schäfer and Altevogt, 2010). 379 Studies in mouse models that constitutively lack L1CAM have reported CC hypoplasia, cerebellar and other brain 380 malformations, and axon guidance defects in the corticospinal tract (Dahme et al., 1997;Cohen et al., 1998;Fransen et 381 al., 1998). Interestingly, neurons differentiated from human embryonic stem (ES) cells in which expression of endogenous 382 LICAM was knockdown through homologous recombination showed reduced axonal length and deficient axonal branching 383 relative to matching control neurons (Patzke et al., 2016). These results are in contrast with the cellular phenotypes we 384 observe in AnkB440 KO neurons, even though ES cell-derived L1CAM KO neurons showed noticeable downregulation of 385 AnkB, which appear to have been largely driven by significant loss of AnkB440. We found that loss of AnkB440 reduces 386 L1CAM abundance at the cell surface but does not significantly change total levels of L1CAM in the cortex and in cortical 387 neuron cultures. L1CAM expression is also normal in brains of AnkB440 R2589fs mice (Yang et al., 2019). While it is plausible 388 that normal AnkB expression requires L1CAM, and not the reverse, it would be important to confirm whether these 389 changes in protein expression are specific to human ES cell-derived neurons. The phenotypic differences between 390 AnkB440-and L1CAM-deficient mice point to additional, independent pathways involving AnkB440 and L1CAM, including 391 the functional relationship of L1CAM with other ankyrins, which collectively underscore the importance of these proteins 392 in neuronal structure and signaling. For instance, a recent report implicates an L1CAM-AnkG association at the AIS of 393 neocortical pyramidal neurons in their innervation by GABAergic chandelier cells (Tai et al., 2019). AnkB440 it is widely 394 distributed through all axonal domains, including the AIS. Further work will be needed to determine whether AnkB440-395 L1CAM complexes localize at the AIS and if they contribute to this specific type of pyramidal neuron innervation. 396 In summary, our findings unveil a critical role of AnkB440 and its binding to L1CAM in promoting the clustering of L1CAM-complex enables the chemorepellent action of Sema 3A to induce GC collapse in axons and collateral branches, thereby 399 providing novel insight into the mechanisms of axonal guidance and branch pruning. As we show, ANK2 variants cause 400 disruption of this signaling axis, which might lead to cortical miswiring and contribute to the neuropathology of ASD. previously reported (Scotland et al., 1998;Lorenzo et al., 2014;Yang et al., 2019;Chan et al., 2020). βII-spectrin floxed 570 mice (Sptbn1 flox/flox , a gift from Dr. Mathew Rasband) have been previously reported (Galiano et al., 2012;Lorenzo et al., 571 2019). AnkB440 KO and AnkB220 KO (described below) mice respectively lacking AnkB440 and AnkB220 in neural 572 progenitors were generated by crossing AnkB440 flox/flox or AnkB220 flox/flox animals to the Nestin-Cre line [B6.Cg-Tg(Nes-573 cre)1Kln/J, stock number 003771] from The Jackson Laboratory. A similar breeding strategy was used to generate mice 574 with loss of βII-spectrin in neural progenitor. All mice were housed at 22°C ± 2°C on a 12-hour-light/12-hour-dark cycle 575 and fed ad libitum regular chow and water. 576

Generation of conditional AnkB220 knockout mice 577
Mice carrying a floxed allele that selectively targets the AnkB220 isoform (AnkB220 flox/flox ) were generated by the Animal 578 Model Core at the University of North Carolina at Chapel Hill using CRISPR/Cas9-mediated integration of a targeting vector 579 into mouse embryonic stem (ES) cells, followed by ES cell injection into blastocytes and production of chimeric progeny. 580 In brief, the CCTop website (https://crispr.cos.uni-heidelberg.de) was used to identify potential Cas9 guide RNAs targeting 581 Ank2 intron 35 and the 5' end of exon 37. Selected guide RNAs were cloned into a T7 promoter vector followed by in vitro 582 transcription and spin column purification. Functional testing was performed by transfecting Cas9 protein/guide RNA 583 ribonucleoprotein complexes into a mouse embryonic fibroblast cell line. The guide RNA target regions were amplified 584 from transfected cells and analyzed by T7endo1 assay (NEB) to detect genome editing activity at the target site. Guide 585 RNAs selected for genome editing in mouse embryonic stem cells were Ank2-i35-sg73T (protospacer sequence 5'-586 GGTTCTAGTCTTCCCGA -3') and Ank2-E37-sg79B (protospacer sequence 5'-GTCCGGACTTGCTAAGAC -3'). A donor vector 587 was constructed for homologous recombination that included a 1002 bp 5' homology arm corresponding to the sequence 588 immediately 5' of the cut site of Cas9/Ank2-i35-sg73T; a LoxP site; 368 bp 3' end of Ank2 intron 35 including splice acceptor 589 sequence; 3707 bp cDNA encompassing exons 36 and 38-46 (isoform lacking exon 37); a 814 bp stop cassette comprised 590 of 3 tandem copies of SV40 polyadenylation sequence; a FRT-flanked selection cassette with PGK mammalian promoter, 591 a EM7 bacterial promoter, a neomycin resistance gene and PGK polyadenylation cassette, all in reverse orientation relative 592 to the Ank2 elements; a LoxP site; a second 368 bp 3' end of Ank2 intron 35 including splice acceptor sequence; a 86 bp 593 segment including 22 bp exon 36 fused to 64 bp 5' end of exon 37; silent point mutations designed to disrupt the Ank2-594 E37-sg79B target site. The sequence GTCTTA was mutated to GTGCTC, corresponding to mutation of a valine codon from 595 GTC to GTG and a leucine codon from TTA to CTC; and a 996 bp 3' homology arm corresponding to sequences immediately 596 3' of the silent point mutations. 597 The donor vector was incorporated into C57BL/6N ES cells by nucleofection with 3 µM Cas9 protein (Thermo Scientific), 598 1.6 25 µM each Ank2-i35-sg73T and Ank2-E37-sg79B guide RNAs and 200 ng/µl (20 µg total) circular donor vector DNA. 599 Cells were selected on G418 and resistant clones were screened for homologous integration of the donor vector at the 600 Ank2 locus. PCR-positive clones were analyzed by Southern blot with 5' and 3' external probes and neomycin cassette 601 internal probe. Two clones, F6 and H10, were identified with homologous integration of the donor. Targeted ES cell clones 602 F6 and H10 were injected in Albino C57BL/6N blastocysts for chimera production. Chimeras were mated to transgenic 603 animals expressing Flp recombinase on Albino C57BL/6N genetic background. Germline transmission of the targeted allele 604 was obtained from both clones, although clone H10 gave more germline transmission pups. Clone H10 had an apparent 605 random integration event in addition to the homologous event. Therefore, pups were screened to identify clones with the 606 homologous integration event in absence of the random integration event. Selected founders were bred for five 607 generations to C57BL/6J mice after which heterozygous carriers of the Ank2 targeting allele (AnkB220 flox/+ ) were bred to 608 each other to generate homozygous carriers (AnkB220 flox/flox ). The WT Ank2 allele was identified by PCR using primers produce a 328 bp DNA fragment. The AnkB220 flox allele was detected by PCR using primers ABCS-RE-F2 (5'-611 GCTTGGCTGTGTTCACAAACA-3') and ABCS-RE-R2 (5'-GACTTGCGAGCACAGGAACTT-3'), which produce a 639 bp DNA 612 fragment. 613

Plasmid transfection for biochemistry analysis 651
Transfection of Halo-tagged AnkB440 plasmids were conducted in HEK293T cells grown in 10 cm culture plates using the 652 calcium phosphate transfection kit (Takara) and 8 µg of plasmid. Cell pellets were collected 48 hours after transfection. 653 654

Labeling and detection of biotinylated surface proteins 669
Cortical neuronal cultures were washed three times with ice-cold PBSCM (PBS + 1mM MgCl2 + 0.1 mM CaCl2) and incubated 670 with 0.5 mg/ml Sulfo-NHS-SS-biotin (Life Technologies) for 1 hour at 4°C. Reactive biotin was quenched by two consecutive 671 7-minute incubations with 20 mM glycine in PBSCM on ice. Cell lysates were prepared in TBS containing 150 mM NaCl, 672 0.32 M sucrose, 2 mM EDTA, 1% Triton X-100, 0.5% NP40, 0.1% SDS, and complete protease inhibitor cocktail (Sigma). Cell 673 lysates were incubated with rotation for 1 hour at 4°C and centrifuged at 100,000 x g for 30 min. Soluble fractions were 674 collected and incubated with high capacity NeutrAvidin™ agarose beads (Pierce) overnight at 4°C to capture biotinylated 675 surface proteins. Beads were washed three times with TBST. Proteins were eluted in 5x-PAGE buffer and resolved by SDS-676 PAGE and western blot. 677

Histology and immunohistochemistry 678
Brains were fixed by transcardial perfusion with 4% PFA in PBS before overnight immersion in the same fixative at 4ºC. 679 After fixation, brains were stored in PBS at 4ºC until use. Brains were then transferred to 70% ethanol for 24 hours and 680 paraffin embedded. 10 µm coronal brain sections were cut using a microtome (Leica RM2135) and mounted on glass 681 slides. Sections were deparaffinized and rehydrated using a standard protocol of washes: 3 x 3 xylene washes, 3 x 2 min 682 100% ethanol washes, and 1 x 2 min 95%, 80%, 70%, ethanol each, followed by ≥5 min in PBS. Sections were processed 683 for antigen retrieval using 10 mM sodium citrate with 0.5% Tween-20, pH 6 in a pressure cooker for 3 minutes at maximum 684 pressure. Sections were cooled, washed in PBS and blocked in antibody buffer (4% BSA, 0.1% Tween-20 in PBS) for 90 685 minutes at room temperature. Tissue sections were then incubated with primary antibody in antibody buffer overnight at

Golgi stain 689
Golgi staining of PND25 brains was conducted using the FD Rapid GolgiStain Kit (FD Neurotechnologies Inc.) In brief, brains 690 were immersed in Solutions A+B for 2-3 weeks, before being transferred to solution C for 3-6 days. 100 µm coronal 691 cryosections from areas of the somatosensory cortex were collected on gelatin-coated microscope slides, counter-stained 692 following manufacturer recommendations, and mounted in Permount for imaging. 693 Neuronal cultures and HEK293T cells were washed with cold PBS, fixed with 4% PFA/4% sucrose for 15 min, and 729 permeabilized with 0.2% Triton-X100 in PBS for 10 min at room temperature. Cells were blocked in antibody buffer for 730 one hour at room temperature and processed for fluorescent staining as tissue sections. For F-actin labeling, Alexa Fluor 731 488-, Alexa Fluor 568-, or Alexa Fluor 633-conjugated phalloidin (1:100) was added to the secondary antibody mix. To label 732 surface Nrp1, fixed but non-permeabilized DIV3 cortical neurons, untreated and treated with Sema 3A, were blocked as 733 described above and incubated overnight at 44°C with a rabbit anti-Nrp1 antibody that recognizes an extracellular epitope, 734 followed by overnight incubation with donkey anti-rabbit IgG conjugated to Alexa Fluor 568. Neurons were fixed again for 735 10 min with 4% PFA in PBS, permeabilized for 8 minutes with 0.2% Triton X-100 in PBS, reblocked with antibody buffer for 736 30 minutes at room temperature and incubated with primary antibodies overnight at 4°C. Secondary antibodies and 737 phalloidin were applied in antibody buffer overnight at °4C. Cells were mounted for imaging with Prolong Gold Antifade Proximity ligation assay (PLA) 740 PLA was performed using the commercial Duolink kit (Sigma-Aldrich) following the manufacturer's recommendations. 741 Fixed and permeabilized neurons were incubated overnight with a pair of primary antibodies specific for the putative 742 interacting partners, each produced in different species. Duolink minus-and plus-probes were used to detect antibody-743 labeled proteins. In the case of the detection of PLA signal between surface Nrp1 and L1CAM, fixed, but non-permeabilized 744 neurons were first incubated overnight at 4 °C with a rabbit anti-Nrp1 antibody that recognizes an extracellular epitope, 745 followed by cell permeabilization and overnight incubation with mouse anti-L1CAM. 746

Image acquisition and image analysis 747
Brain sections stained with antibodies were imaged using a Zeiss 780 laser scanning confocal microscope (Zeiss) and 405-748 , 488-, 561-, and 633-nm lasers. Single images and Z-stacks with optical sections of 1 μm intervals and tile scans were 749 collected using the 20x (0.8 NA) and Plan Apochromat 40x oil (1.3 NA) and 63x oil (1.4 NA) objective lenses. Images were 750 processed, and measurements taken and analyzed using NIH ImageJ software. Three-dimensional rendering of confocal 751 Z-stacks was performed using Imaris (Bitplane). Golgi-stained brains were imaged using a Nikon Ti2 Eclipse scope running 752 NIS-Elements. Widefield Z-stacks were taken with 2 µm optical section in the primary somatosensory cortex with a 40x/NA 753 objective. 754

Time-lapse video microscopy and movie analyses 755
Live microscopy of neuronal cultures was carried out using a Zeiss 780 laser scanning confocal microscope (Zeiss) equipped 756 with a GaAsP detector and a temperature-and CO2-controlled incubation chamber as previously reported (Snouwaert et 757 al., 2018). Movies were taken in the distal axon and captured at a rate of 1 frame/second for time intervals ranging from 758 60-300 seconds with a 40x oil objective (1.4NA) using the zoom and definite focus functions. Movies were processed and 759 analyzed using ImageJ (http://rsb.info.nih.gov/ij). Kymographs were obtained using the KymoToolBox plugin 760 for ImageJ (https://github.com/fabricecordelieres/IJ_KymoToolBox). In details, space (x axis in µm) and time 761 (y axis in sec) calibrated kymographs were generated from video files. In addition, the KymoToolBox plugin was used to 762 manually follow a subset of particles from each kymograph and report the tracked particles on the original kymograph 763 and video files using a color code for movement directionality (red for anterograde, green for retrograde and blue for 764 stationary particles). Quantitative analyses were performed manually by following the trajectories of individual particles to calculate dynamic parameters including, net and directional velocities and net and directional run length, as well as 766 time of pause or movement in a direction of transport. Anterograde and retrograde motile vesicles were defined as 767 particles showing a net displacement >3 μm in one direction. Stationary vesicles were defined as particles with a net 768 displacement <2 μm. 769

Statistical analysis 770
Sample size (n) for evaluations of growth cone collapse was estimated using power analyses and expected effect sizes 771 based on preliminary data in which we used similar methodologies, specimens, and reagents. We assumed a moderate 772 effect size (f=0.25-0.4), an error probability of 0.05, and sufficient power (1-β=0.8). GraphPad Prism (GraphPad Software) 773 was used for statistical analysis. Two groups of measurements were compared by unpaired, two tailed students t-test. 774 Multiple groups were compared by one-way ANOVA followed by Tukey or Dunnett's multiple comparisons tests. 775 experiments. The box and whisker plots in E, F, and H represent all data points collected arranged from minimum to 807 maximum. One-way ANOVA with Dunnett's post hoc analysis test for multiple comparisons. ****p < 0.0001. 808  neurons expressing GFP-tagged AnkB220 untreated and treated with Sema 3A and stained for GFP to visualize AnkB220. 863 Scale bar, 1 μm. (G) Percent of GC collapsed before and after Sema 3A treatment of AnkB440 KO neurons rescued with 864 AnkB440 or AnkB220 cDNAs. Data represent mean ± SEM collected from an average of n=150 GCs/treatment AnkB440-GFP are targeted to the plasma membrane or found diffused throughout the cytoplasm for each experiment. 884 Data represent mean ± SEM collected from an average of n=15 cells/transfection from three independent experiments. 885 One-way ANOVA with Dunnett's post hoc analysis test for multiple comparisons. ****p < 0.0001, ns p > 0.05. 886 Images of the distal portion of the main axon of DIV3 cortical neurons untreated and treated with Sema 3A for 1 and 5 918 minutes and stained with phalloidin, βIII-tubulin, and pCofilin Ser3 . Dotted lines indicate GCs. Scale bar, 2 μm. (E) 919 Quantification of pCofilin S3 signal at GCs relative to GC area at the basal state and upon Sema 3A treatment for 1 and 5 920 minutes. Data represent mean ± SEM collected from an average of n=25-40 GCs/treatment condition/genotype from three 921 independent experiments. One-way ANOVA with Dunnett's post hoc analysis test for multiple comparisons. ****p < 922 0.0001, ***p < 0.001, **p < 0.01. 923