GSK3 Controls Migration of the Neural Crest Lineage

Migration of the neural crest lineage is critical to its physiological function. Mechanisms controlling neural crest migration are comparatively unknown, due to difficulties accessing this cell population in vivo. Here, we uncover novel requirements of glycogen synthase kinase 3 (GSK3) in regulating the neural crest. We demonstrate that GSK3 is tyrosine phosphorylated (pY) in neural crest cells and that this activation depends on anaplastic lymphoma kinase (ALK), a protein associated with neuroblastoma. Consistent with this, neuroblastoma cells with pathologically increased ALK activity express high levels of pY-GSK3 and migration of these cells can be inhibited by GSK3 or ALK blockade. In normal neural crest cells, loss of GSK3 leads to increased pFAK and misregulation of Rac1 and lamellipodin, key regulators of cell migration. Genetic reduction of GSK-3 results in failure of migration. All together, this work identifies a role for GSK3 in cell migration during neural crest development and cancer.


Abstract 27
Migration of the neural crest lineage is critical to its physiological function. Mechanisms 28 controlling neural crest migration are comparatively unknown, due to difficulties accessing 29 this cell population in vivo. Here, we uncover novel requirements of glycogen synthase 30 kinase 3 (GSK3) in regulating the neural crest. We demonstrate that GSK3 is tyrosine 31 phosphorylated (pY) in neural crest cells and that this activation depends on anaplastic 32 lymphoma kinase (ALK), a protein associated with neuroblastoma. Consistent with this, 33 neuroblastoma cells with pathologically increased ALK activity express high levels of pY-34 GSK3 and migration of these cells can be inhibited by GSK3 or ALK blockade. In normal 35 neural crest cells, loss of GSK3 leads to increased pFAK and misregulation of Rac1 and 36 lamellipodin, key regulators of cell migration. Genetic reduction of GSK-3 results in failure 37 of migration. All together, this work identifies a role for GSK3 in cell migration during 38 neural crest development and cancer. 39

40
The neural crest is a vertebrate-specific, motile population of cells born at the junction 41 of the neural and non-neural ectoderm. This lineage has contributed to our understanding of 42 cellular behaviours, such as contact inhibition of locomotion 1 . It is the origin of many cell 43 types found throughout the organism, including melanocytes, peripheral neurons, cardiac 44 outflow tract, and the craniofacial skeleton. Recent reports have highlighted the importance 45 of neural crest cells: their stem-like capacity, their ability to reprogram, to become cancerous, 46 and to drive vertebrate evolution 2-5 . The highly migratory activity of these cells is critical to 47 their in vivo function, not only are their ultimate tissue descendants widespread in the 48 organism, but failure to regulate migration and differentiation in the correct locations is 49 associated with diseases like neuroblastoma 6-8 . Despite its importance, the specific 50 mechanisms underlying this migratory activity and its control are poorly understood. 51 In our previous work, we demonstrated a critical role for the pleiotropic kinase GSK3 52 in craniofacial development 9 ; therefore, we sought to understand the regulation of GSK3 in 53 neural crest cells, which are integral to most of the craniofacial structures. In vertebrates, the 54 serine/threonine kinase GSK3 is encoded by two paralogous genes, GSK3α and GSK3β, 55 which are nearly identical throughout their kinase domains 10,11 , and have over 100 predicted 56 substrates 11,12 . The effect of GSK3 phosphorylation is substrate dependent and variable, cancer and retinitis pigmentosa 19,20 . As a consequence it is thought that regulation of GSK3 91 target selection is very context dependent. 92 Even focusing on the neural crest lineage, GSK3 is thought to have multiple 93 sequential roles beginning with a requirement in patterning of the dorsal axis 21-23 . Based 94 primarily on data from chicken and frog, neural crest-specific GSK3 targets include the Wnt 95 effector β-catenin, the metalloprotease ADAM13 and transcription factors snail and twist [24][25][26] . 96 Wnt dependent inhibition of GSK3 is clearly necessary for β-catenin-mediated transcriptional 97 activation during neural crest induction 27 . GSK3 also has proposed roles in the 98 phosphorylation of Twist and Snail, proteins which can regulate the activity and stability of 99 these targets, thus controlling the onset of the epithelial-mesenchymal transition (EMT) 24 . 100 Concurrently, GSK3 interactions with ADAM13 are proposed to be crucial for delamination 101 and entry into the EMT 25 . Given the variety of substrates and the precise timing of 102 development, there is no doubt that GSK3 activity must to be dynamically regulated during 103 neural crest development. However, as noted, mice lacking the inhibitory phosphorylation 104 sites at S21/S9-GSK3α/β have normal craniofacial development and are viable 13 . Therefore, 105 we focus on positive regulation of GSK3 via activating tyrosine phosphorylations. 106 Here, our analysis uncovers a surprisingly specific activation of GSK3 in neural crest 107 cells as they depart from the neuroepithelium and become migratory mesenchymal cells. This 108 activation is controlled by anaplastic lymphoma kinase, which has been implicated in 109 neuroblastoma and other cancers. Genetic and pharmacological loss of GSK3 activity leads to 110 cytoskeletal changes in migratory neural crest cells as well as in neuroblastoma, raising the 111 possibility that control of GSK3 is a rapid and reversible mechanism for controlling cell 112 migration dynamics in the neural crest lineage. 113 114

115
GSK3 is expressed, and specifically tyrosine phosphorylated, in migrating neural crest cells 116 In the embryo, neural crest cells are induced at the neuroepithelial border, 117 subsequently delaminating and becoming migratory. To confirm whether GSK3 was 118 preferentially enriched during neural crest cell migration, we examined mRNA and reporter 119 gene expression for both paralogous genes in frog and mouse, and found that these genes 120 were indeed expressed in migratory neural crest ( Figure 1A-H). We noted in particular the 121 enrichment of GSK3 expression in the neural plate stages, at the border of the neural plate 122 (NPB) ( Figure 1E, E', G, G') and later on in the migratory neural crest cell population, carrying a single allele of functional GSK3 (Wnt1::cre; GSK3α fl/+ ;β fl/fl or Wnt1::cre; 157 GSK3α fl/fl ;β fl/+ ). In both cases, we noted a reduction in Sox10 positive cells en route to the first 158 branchial arch and an accumulation of positive staining in the neural tube, suggesting that 159 both GSK3 proteins contribute to migration of the neural crest (Supplemental Figure 1, red 160 bracket). 161 While the timing of the Wnt1::cre transgene misses the initiation of neural crest 162 induction, it was still possible that some of the effects seen were due to GSK3 requirements 163 in the pre-migratory neural crest. In order to bypass early effects on the neural crest we 164 turned to pharmacological inhibition of GSK3 using the specific inhibitor BIO (6-165 bromoindirubin-3′-oxime 29 ). Making use of Xenopus allowed us to precisely time our 166 manipulations in a well-defined in vivo system. Treatments of Xenopus embryos confirmed 167 that GSK3 inhibition led to loss of the migration marker twist1 ( Figure 2I). Treatment of 168 Xenopus embryos at stage 12.5 confirmed that GSK3 inhibition led to loss of twist1. When 169 the embryos were released from treatment at stage 19, we found some recovery of twist1 170 expression ( Figure 2K). Loss of GSK3 function during the critical stages led to significant 171 changes in face shape as well as a smaller neural crest derived tail fin (Supplemental Figure  172 2A-C), as well as loss of the neural crest derived facial cartilages ( Figure 2J, L). Although the 173 facial cartilages were lost leading to narrowing of the head, the mesodermal cranial muscles 174 are still formed (Supplemental Figure 2D). To confirm that this effect was specifically due to 175 impairment of migration, we transplanted fluorescently labeled neural crest cells into a st.17 176 embryo and then treated with BIO; these cells did not migrate ( Figure 2N

Perturbing GSK3 function prevents migration of cranial neural crest cells 181
To bypass the neural crest induction stage as well as the embryonic lethality, we 182 turned back to the neural crest explant cultures. We found that dispersion of Xenopus cells 183 was inhibited by BIO treatment (Figure 3A-C). Note that in the Xenopus explants, the 184 premigratory neural crest population can be specifically dissected independently from the 185 neural plate. We then turned back to mouse neural crest cultures in order to better compare 186 our pharmacological inhibitors to genetic mutants. Treatment with two different specific 187 inhibitors, BIO and CHIR99021 (CHIR 30 ) prevented neural crest cell migration similar to that 188 observed in Xenopus (Fig 3D-I, Supplemental Movies 1-2). We found that the area covered by the pre-migratory neural crest cells was expanded in treated samples (Supplemental Figure  190 3A, 3B) with a concurrent decrease in the migratory population (Supplemental Figure 3B Nevertheless, to confirm this, we observed that neural crest explants from these mice appear 200 normal and retain sensitivity to BIO inhibition, demonstrating that serine phosphorylation of 201 GSK3 is dispensable during neural crest migration (Supplemental Figure 4A BIO and CHIR treated cells losing filamentous actin localisation, which is normally at the 213 edge of the cell (see Figure 3M, O, white arrowheads) and generating spiky protrusions (Fig  214   3N, 3P, yellow arrowheads). Microtubule organization was also disrupted in BIO-treated 215 cells. In normal cells, stabilized microtubules (marked by acetylated tubulin) extend from the 216 centrosome toward the leading edge of the cell (see Figure 3Q). In BIO-treated cells 217 stabilized microtubules appeared to accumulate perinuclearly ( Figure 3R, Supplemental 218 Figure 5C). We also examined a marker for unstable microtubules (YL1-2 35 ) and found a 219 significant decrease in this population, which also accumulated primarily at the rear of the 220 cell, behind the nucleus ( Figure 3S-T, arrowheads and Supplemental Figure 5D). Finally, we 221 examined membrane dynamics in both mutants and BIO-treated explants from mice carrying a genetically labelled membrane GFP ( Figure 3U-W). In the control explants, a large 223 proportion of the GFP was internalised within the cell, suggesting recycling of the membrane 224 ( Figure 3U, closed arrowhead). Instead, both treated and mutant cells had very strong 225 expression of GFP at the cell membrane ( Figure 3V, 3W, open arrowheads). Consistent with 226 these findings, we also found an accumulation of β-catenin at the membrane in BIO-treated 227 explants ( Figure 3X-Y), suggesting that loss of GSK3 activity led to extremely stable 228 membrane dynamics compared to controls.  FAK is thought to control cytoskeletal dynamics by repressing the function of the 245 small GTPase Rac1 38 . Therefore, the inhibition of FAK activity appears necessary to allow 246 Rac1 activation. With the accumulation of active FAK in our treated cells, we found that 247 Rac1 was now excluded from the leading edge of the migratory neural crest cells (compare 248 Figure 5B to A, arrowheads). Interestingly, we see an increase in Rac1 in nuclei ( Figure 5B Scar/WAVE 41,42 . When GSK3 is inhibited, leading edge localisation of lamellipodin is lost, and surprisingly, lamellipodin also relocalizes to the nucleus ( Figure 5F, 5G). As a 256 consequence, treated neural crest cells fail to generate stable fan-shaped lamellipodia ( Figure  257 5H-K, green arrowheads), with approximately 50% of delaminated cells having unstable 258 lamellipodia ( Figure 5K). We then turned back to genetic mutants to confirm these 259 phenotypes (Fig 5L-M). In this case, to bypass any complications of GSK3 involvement in 260 neural crest induction, we turned to a tamoxifen inducible Cre (pCAAG::CreER TM ) 43 , 261 allowing temporal deletions upon drug addition. As predicted, tamoxifen induced knockout 262 of GSK3 led to a loss of the wavelike lamellipodial protrusions, leaving only spiky filopodial 263 movements in neural crest cells (stills shown in Figure 5L-M and Supplemental Movies 3-5). 264 All together, this demonstrates that in the absence of GSK3 activity, neural crest cells make 265 filopodial protrusions at the expense of lamellipodia. 266 267

Anaplastic lymphoma kinase (ALK) is expressed in migratory neural crest cells 268
Aberrant neural crest development is thought to underlie neuroblastoma, an 269 aggressive paediatric cancer. Activating mutations in ALK contribute to a subset of 270 neuroblastoma cases, correlating with poor prognosis 44-49 . Because we saw specific 271 expression of pY-GSK3 at the leading edge and in migratory neural crest cells, we wondered 272 whether ALK might be responsible for activating GSK3. First, we set out to check whether 273 ALK was expressed during the appropriate stages of neural crest development. Expression of 274 ALK has previously been reported at E10.5, including in the diencephalon and facial 275 ganglia 50 ; however, to our knowledge, there has been no analysis performed during key 276 neural crest migration stages. To test this, we performed mRNA in situ hybridization from 277 E8.0 onwards ( Figure 6A, D). We found that by E8.5, ALK appeared specific to the neural 278 plate border corresponding to active cranial crest migration ( Fig 6A). This expression was 279 enriched at the neural plate border consistent with a role for ALK in the delaminating neural 280 crest cells ( Figure 6A, 6D). Furthermore, ALK continues to be expressed at 9.5dpc at the 281 neural plate border, and in a migratory neural crest destined for branchial arch I and II and at 282 the frontonasal process ( Fig 6D). Additional expression was seen in the heart, trunk and 283 limbs. We also examined localization of the active form of ALK protein. Using an antibody 284 recognizing ALK carrying a phospho-tyrosine residue (pY1507-ALK), we again found 285 enrichment of ALK in the right place at the right time to be acting upon GSK3 during neural 286 crest migration (Fig 6B-F). This neural crest specific expression was recapitulated in explant 287 cultures, where we noted a lack of ALK protein in neural plates (total ALK, Fig 6G-G') followed by an onset of expression in migratory neural crest cells, which was somewhat 289 nuclear, with diffuse staining throughout the cell (compare Figure 6H-I'). 290 291

Inhibition of ALK activity leads to a loss of pY-GSK3 and phenocopies GSK3 inhibition 292
To test whether ALK activity is required for tyrosine phosphorylation of GSK3, we 293 challenged the neural crest explants with specific inhibitors for ALK. These included 294 crizotinib (CTB), which is currently used as a chemotherapeutic, and AZD3463 (AZD). 295 Because both of these are dual function inhibitors (CTB blocking ALK and the c-met 296 receptor 51,52 , and AZD blocking ALK and insulin-like growth factor receptors 53 ), we also 297 treated with the highly selective inhibitor NVP-TAE684 (NVP 54 ) (Fig 6L-M). we found that 298 blocking ALK led to a loss of pY-GSK3 expression in neural crest explants, suggesting that 299 GSK3 was indeed a target of ALK kinase activity in this context (Fig 6J- Furthermore, in all three cases, blocking ALK function phenocopied loss of GSK3 activity 301 leading to a substantial decrease in delamination of the neural crest and a loss of the 302 migratory cell population ( Figure 5N-Q). We noted that NVP treatment was the most 303 effective at blocking neural crest migration while CTB had a much milder effect (Fig 6R). 304 Finally, ALK inhibitors CTB and NVP phenocopied the disruption of the actin cytoskeleton 305 seen when GSK3 was blocked (Fig 6S-W). 306 307 Neuroblastoma lines with high levels of ALK also express high levels of pY-GSK3 308 We then wondered whether high levels of ALK activity could drive excessive 309 activation of GSK3. To test this, we examined nine neuroblastoma lines and found a clear 310 correlation between levels of total ALK, active (pY-1507) ALK and pY-GSK3 ( Figure 7A). 311 We then focused on the Kelly neuroblastoma line, which carries an activating mutation in 312 ALK (F1174L) 55 and, as a comparison, LS 56 neuroblastoma cells which had very low levels 313 of ALK. In western blots, we again found much higher levels of ALK in Kelly cells, with 314 nearly undetectable levels in LS cells, more similar to that of mouse embryonic fibroblasts 315 (MEFs) ( Figure 7B). 316 317 ALK activity is required for pY-GSK3 in neuroblastoma lines 318 All together, our data raised the possibility that ALK regulation of GSK3 is a neural 319 crest specific activity that may have been co-opted during cancer progression. Indeed, 320 inhibition of ALK in the neuroblastoma lines also decreased pY-GSK3 levels ( Figure 7C-D, Supplemental Figure 6). The Kelly neuroblastoma line carries an ALK-F1174L mutation 322 which renders it somewhat insensitive to the ALK inhibitor crizotinib (CTB) 55 . Therefore, as 323 with the neural crest explants, we confirmed these findings using the two other inhibitors 324 AZD and NVP ( Figure 7D-E). We found that treatment with ALK inhibitors was sufficient to 325 decrease phosphorylation of GSK3 (Fig 7E). Treatment with BIO or CHIR also affected pY-326 GSK3 levels, consistent with some auto-regulation by GSK3 itself (Fig 7D). 327 Finally, we set out to determine whether the excessive levels of pY-GSK3 could 328 underlie the aggressive nature of neuroblastomas. If GSK3 activity is downstream of ALK in 329 this context, we would predict that inhibition of GSK3 in Kelly cells would be sufficient to 330 limit cell migration. Indeed, using scratch assays where we measured the movement of cells, 331 we observed that Kelly cell migration was blocked by ALK inhibitors (NVP/AZD) similarly 332 to GSK3 inhibition (BIO/CHIR) ( Figure 8A, C,. As noted before, Kelly cells were resistant to 333 CTB ( Figure 8A, C,). 334 In contrast to the Kelly cells, LS cells, which do not carry the ALK-F1174L variant, 335 behaved very differently. LS cells had substantially less pY-GSK3, which correlated with 336 much lower levels of active ALK (pY1507, Figure 7A or pY1586, Supplemental Figure 6A). 337 LS cells were insensitive to crizotinib treatment ( Figure 8B, D and Supplemental Figure 6B-338 D). But, while the other ALK inhibitors led to a substantial decrease in the area covered, 339 examination of the cultures showed substantial cell death (see Figure 8B, bottom right panel). 340 More interesting, we found that GSK3 inhibition in LS cells elicited an unusual phenotype in 341 the scratch assays, with cells moving together in aggregates, rather than as single cells 342 ( Figure 8B, arrowheads). It is possible that LS cells have a more "epithelial" morphology 343 than Kelly cells and that GSK3 loss mimicked the stable cell-cell interactions similar to those 344 in pre-migratory neural crest cells ( Figure 3U-Y). Consistent with our hypotheses, 345 morphologically, the Kelly cells responded similarly to motile neural crest cells, with BIO 346 treated Kelly cells appearing compacted with dense actin staining (Fig 8E). Nevertheless, 347 taken together, our data suggest that ALK activity is closely linked to GSK3 phosphorylation 348 and activity in primary neural crest and neuroblastoma cells. Here, we studied the effect of the serine/threonine kinase GSK3 during mammalian 358 neural crest migration. Previously, we have found that GSK3β is required for the palate 359 formation at specific time points during development 6 . This structure depends upon neural 360 crest migration. However, due to functional redundancy between GSK3α and GSK3β, the in 361 vivo activity has been difficult to study 57 . Furthermore, early loss of GSK3 leads to global 362 Wnt activation, which can obscure later developmental roles 24,25,58-60 . However, our studies 363 bypass these early complications and provide a refined understanding of the regulation and 364 function of GSK3. This effect on neural crest migration appears to be β-catenin independent, 365 as neural crest specific expression of stabilized β-catenin does not disrupt migration 61 . 366 Instead, GSK3 appears to act directly on the actin cytoskeleton, changing the dynamics of 367 lamellipodial formation. Our data demonstrate that this regulation may be via regulation of 368 FAK localization, as well as key downstream factors including Rac1, cdc42 and lamellipodin. 369 Interestingly, neural crest specific deletion of FAK, Rac1, cdc42 and lpd all have a range of 370 craniofacial anomalies 62,63 . However, it is worth noting that in our assays, we predominantly 371 found that these proteins were relocalised, and thus it is difficult to directly compare our 372 observations with the published null phenotypes. Nevertheless, these observations are worth 373 further in-depth study. 374 GSK3 can also regulate the dynamics of the actin cytoskeleton, microtubules and 375 cell to matrix adhesions 64 . To date, polarized inhibition via phosphorylation of S9 of 376 GSK3β has been thought to be the main mechanism for establishment of cell polarity, 377 especially in astrocytes 32 , and is also critical for glioma cell invasion 65 . All of these 378 scenarios depend on negative regulation of GSK3. Importantly, contrary to the neuronal 379 cell scenario, we find GSK3 inhibition via serine phosphorylation is not necessary for 380 neural crest migration (Supplementary Figure 4). However, given the complexity of GSK3 381 regulation, it would be interesting to see whether combining phosphorylation mutations on 382 the activating tyrosines and inhibitory serines has an additive effect on neural crest 383

migration. 384
Most important, we find that neural crest cell migration depends on GSK3 activity, 385 and that this correlates with high levels of tyrosine phosphorylation via ALK. While there 386 are other kinases which may be regulating GSK3 phosphorylation, including GSK3 itself 16 , the association with ALK in the context of neuroblastoma is particularly compelling. 388 However, future studies should include other tyrosine kinases which may be 389 phosphorylating GSK3. For instance, it has been reported that PYK2 66 , a putative 390 mammalian homologue of ZAK1, a kinase found in Dictyostelium shown to phosphorylate 391 GSK3β at Y216. However, there is still no clear evidence on how this finding could relate 392 to mammals. In pathological conditions, pY216-GSK3β has been found in prostate cancer, 393 and Src was found to promote this phosphorylation, and with it, cancer progression and 394 invasion 67 . These other kinases are worth considering in the future, and may be necessary 395 to regulate sub-populations of GSK3. 396 Particularly intriguing was the prediction that GSK3α is a putative ALK substrate 397 in cancer cells 18 . ALK is negatively correlated to neuroblastoma prognosis, with 398 hyperactivating mutations of this kinase found in some of these aggressive tumors. 399 Therefore, the discovery that pY-GSK3 was specifically expressed in delaminating and  Stage 45+ embryos were fixed in MEMFA for 1 hour at room temperature before 512 washing into ethanol. For cartilage staining embryos were washed into a 0.15% alcian blue 513 solution (70% EtOH/30% acetic acid) at room temperature for 3 days. When suitably stained, 514 embryos were rinsed 3x 15 mins in 95% EtOH. Rehydration was done stepwise into 2% 515 KOH then washed from 2%KOH stepwise into 80% glycerol/20% 2%KOH, 1 hour per wash 516 before washing overnight into the final solution. Dissection of cartilages was then performed 517 to increase visibility of craniofacial cartilages. Image processing and all the analysis were made using the IncuCyte® ZOOM Software.

Microscopy and Image Analysis 556
Live imaging was obtained using Nikon A1R. Confocal z-stacks were obtained using a Leica 557 TCS SP5 DM16000. Image sequences were reconstructed using Fiji-ImageJ analysis 558 software. 559