CDK1 and CDK2 regulate phosphorylation-dependent NICD1 turnover and the periodicity of the segmentation clock

All vertebrates share a segmented body axis. Segments form periodically from the rostral end of the presomitic mesoderm (PSM) and this periodicity is regulated by the segmentation clock, a molecular oscillator that drives dynamic clock gene expression across the PSM with a periodicity that matches somite formation. Notch signalling is crucial to this process. Altering Notch intracellular domain (NICD) stability affects both the clock period and somite size. However, the mechanistic details of how NICD stability is regulated are unclear. We identified a highly conserved site crucial for NICD recognition by the SCF E3 ligase, which targets NICD for degradation. We demonstrate both CDK1 and CDK2 can phosphorylate NICD in the domain where this crucial residue lies and that NICD levels vary in a cell cycle-dependent manner. Inhibiting CDK1 or CDK2 activity increases NICD levels both in vitro and in vivo, leading to a delay of clock gene oscillations.

However, the mechanistic details of how NICD stability is regulated are unclear. 23 We identified a highly conserved site crucial for NICD recognition by the SCF E3 ligase, which 24 targets NICD for degradation. We demonstrate both CDK1 and CDK2 can phosphorylate 25 NICD in the domain where this crucial residue lies and that NICD levels vary in a cell cycle-26 dependent manner. Inhibiting CDK1 or CDK2 activity increases NICD levels both in vitro and 27 in vivo, leading to a delay of clock gene oscillations. 28

INTRODUCTION 29
Segmentation, a process which occurs early during vertebrate body plan formation, generates 30 repeated segments (or somites) that later give rise to the vertebral column, most skeletal 31 musculature and dermis [1,2]. 32 During somitogenesis, pairs of somites bud off the rostral end of the unsegmented presomitic 33 mesoderm (PSM) with a periodicity that is species specific. The periodicity of segment 34 formation is regulated by a molecular oscillator, known as the somitogenesis clock, which 35 drives oscillatory gene expression within the PSM tissue from which somites are derived [2-36 4]. 37 These clock genes are targets of the Notch, Wnt and FGF pathways [5,6]. Aberrant 38 somitogenesis leads to severe segmentation and skeletal defects [7]. In humans, defects in 39 segmentation lead to congenital scoliosis (CS), with an infant mortality rate of 50% that 40 comprises many vertebral skeletal and muscular pathologies, including the family of 41 spondylocostal dysostoses (SCD). For CS, whilst the aetiology is unclear, linkage analyses 42 have shown mutations in four genes lead to familial forms of SCD [8]. Significantly these are 43 components of the Notch pathway, which plays multiple roles during segmentation. Notch is 44 crucial to the segmentation process in mice, since in the absence of Notch signalling, the 45 segmentation clock stops and no somites form [9]. 46 On a single cell level in the PSM, oscillatory clock gene expression is established through 47 positive and negative feedback loops of unstable clock gene products which potentiate or 48 inhibit the pathway that activates them. Synchronisation of clock gene oscillations between 49 neighbouring cells is reliant on Notch signalling [10][11][12][13]. Mathematical models predict the 50 period of clock gene oscillations can be approximated as a sum of the delays involved in 51 transcription, splicing, translation and transport of clock gene products, and in particular 52 through the regulation of the half-lives of both mRNA and protein of unstable regulators [14-53 17]. Whilst great progress has been made in demonstrating the role of transcription and 54 splicing delays in setting the clock period, little experimental work investigating whether 55 stability of clock components affects clock period has been performed. 56 clock. However, the specific kinases/molecular mechanism of action remain ill-defined and 84 leave open the question of whether this coupling is a general or conserved mechanism. 85 In this manuscript we identify the phosphorylated residues within human NICD. We 86 demonstrate that purified recombinant Cyclin-dependent kinase 1 (CDK1) and Cyclin-87 dependent kinase 2 (CDK2) phosphorylate NICD within the PEST domain. A point mutation 88 affecting a conserved serine residue within this CDK substrate domain of the NICD PEST 89 motif prevents NICD interaction with endogenous FBXW7. Strikingly, we show that NICD 90 levels fluctuate in a cell cycle dependent manner anti-correlating with high levels of CDK1/2 91 activity. Lastly, highly specific inhibitors of CDK1 or CDK2 lead to increased levels of NICD in 92 vitro and in vivo and delay the mouse somitogenesis clock and somite formation. Quantification of the density of western blot bands in at least three independent experiments 120 confirmed that the increase in NICD levels was statistically significant after treatment with 121 Roscovitine, DRB and XAV939, as shown in Figures 1B, 1D and Supplementary Figure 1B. 122 Two other inhibitors were used as positive and negative controls for the assay. LY411575 is 123 a γ-secretase inhibitor that prevents Notch1 cleavage and thus inhibits activation of target 124 gene expression [59,60]. As expected, LY411575 treatment significantly reduced NICD levels 125 (Figures 1A-D, Supplementary Figure 1). Phosphorylation of the C-Terminal PEST domain 126 of NICD leads to recruitment of FBXW7 and thus to NICD ubiquitylation and proteasomal 127 degradation [34-36, 43, 44, 46-48]. When E3 ligase activity is reduced with the NEDDylation 128 inhibitor MLN4924 [61], NICD levels increase, since NICD degradation is stopped in the 129 presence of this compound (Figures 1A-D, Supplementary Figure 1). 130 Interestingly, we were able to detect two distinct bands by western blot with the NICD antibody, 131 particularly when cells were treated with MLN4924. We hypothesised this reflected the 132 presence of non-phospho and phospho-NICD species. To test this hypothesis, we treated 133 MLN4924-treated lysates with λ phosphatase which abrogated the appearance of the higher 134 band by western blot with the NICD antibody ( Figures 1A and 1C). These data demonstrate 135 that the higher band detected corresponds to a phosphorylated isoform of NICD. 136 In order to determine if the increased levels of NICD were due to increased NICD production 137 and/or increased NICD stability, we exposed HEK293 cells to LY411575 treatment for the last 138 indicating that the increase in NICD levels is not due to increased NICD production, but to an 143 increased stability (Supplementary Figure 2). 144 Taken together, these results show that exposure to this group of inhibitors leads to increased 145 levels of NICD in a variety of cell lines, in the same way that they do in the mouse and chicken 146 PSM tissue, suggesting they regulate a conserved mechanism leading to increased NICD 147 levels and reduced NICD turnover. In order to investigate whether this selection of small molecule inhibitors have different effects 152 on NICD phosphorylation, we treated HEK293 cells with the inhibitors and performed a Phos-153 tag assay [62]. Following MLN4924 treatment, a variety of bands indicative of different 154 phospho-species of NICD was observed ( Figure 1E). Given that very few bands are present 155 in the control sample (DMSO, Figure 1E), the bands detected after MLN4924 treatment are 156 likely to be very labile isoforms of the NICD peptide, which are rapidly degraded in the DMSO 157 sample. In contrast, when whole cell lysate was treated with both MLN4924 and λ 158 phosphatase only one band, of the lowest molecular weight, was detectable, further supporting 159 the notion that the ladder of bands obtained upon MLN4924 treatment reflects a variety of 160 unstable phosphorylated NICD isoforms, which is completely depleted in the presence of λ 161 phosphatase ( Figure 1E). As expected from data showing NICD levels are increased after 162 inhibitor treatment (Figures 1A-D), all three kinase inhibitors Roscovitine, DRB and XAV939 163 cause a noticeable increase in the number and intensity of bands compared to control cells. 164 However, compared to the effect seen with MLN4924, phos-tag technology reveals 165 Roscovitine, DRB and XAV939 have a reduced number of phospho-bands as compared to 166 MLN4924, indicative of the fact these kinase inhibitors act to reduce NICD phosphorylation. 167 Moreover, each of the inhibitors presents a distinct profile of NICD phospho-species. These 168 data suggest that NICD is targeted by several kinases and/or phosphorylation events which 169 are differentially sensitive to these inhibitors ( Figure 1E). 170 NICD-FBXW7 interact at the endogenous levels in HEK293 cells, in a phosphorylation-171 dependent manner. The involvement of the F-box protein component of the SCF E3 ligase 172 complex, FBXW7, in NICD degradation has been previously reported [46][47][48]. However, to 173 date, the NICD-FBXW7 interaction has only been shown in overexpressed systems [35, 44, 174 46, 63-66] Thus, we examined the binding of NICD to FBXW7 using co-immunoprecipitation 175 analysis at the endogenous level. 176 FBXW7 was immunoprecipitated from HEK293 cells treated with DMSO or MLN4924 for 3 177 hours and extracts were probed with NICD antibody. NICD directly binds to FBXW7 ( Figure  178 2A). After MLN4924 treatment, the amount of NICD bound to FBXW7 was significantly higher 179 compared to control cells and this was particularly evident with the higher molecular weight 180 isoform of NICD (Figures 2A-B), confirming again NICD-FBXW7 interaction is 181 In order to determine if the change in the NICD phosphorylation profile observed after 183 treatment with the CDK inhibitors reduced the NICD-FBXW7 interaction, we performed the 184 same co-immunoprecipitation assay after CDK inhibitor treatment. In order to maximise the 185 amount of NICD immunoprecipitated with FBXW7, cells were treated with MLN4924 (to 186 prevent NICD degradation) in the presence or absence of the CDK inhibitors. A significantly 187 reduced interaction between NICD and FBXW7 was observed after treating HEK293 cells with 188 Roscovitine or DRB for 3 hours (Figures 2C and 2E). Statistical analyses, carried out on the 189 density of western blot bands after immunoprecipitation, confirmed a significant reduction in 190 the interaction between NICD and FBXW7 following either Roscovitine or DRB treatment 191 ( Figures 2D and 2F). 192 Taken together, these data demonstrate, for the first time, that NICD interacts with FBXW7 at 193 endogenous levels in HEK293 cells, and this interaction is dependent on phosphorylation. 194 To further validate the involvement of FBXW7 in endogenous NICD turnover, we conducted a 195 siRNA-mediated depletion of FBXW7 in HEK293 cells ( Figure 2G). siRNA treatment efficiently 196 depleted FBXW7 protein levels and led to an increase in levels of the FBXW7 target protein 197 Cyclin E. FBXW7 depletion also resulted in increased levels of NICD,and,in particular,an 198 accumulation of the phosphorylated form of NICD ( Figure 2G). 199

Serine 2513 is essential for the NICD-FBXW7 interaction 201
We utilised Mass Spectrometry as an unbiased approach to identify which NICD residues are 202 phosphorylated in HEK293 cells transiently transfected with human NICD-GFP, followed by 203 immunoprecipitation of NICD-GFP. Gel slices were processed and submitted to MS analysis. 204 We identified 15 phospho-sites on exogenous hNICD, highlighted in green in Figure 3A. To 205 investigate the relevance of those phosphorylation sites in NICD turnover, we screened those 206 located within the PEST domain (such as S2527), and others based on the FBXW7 phospho-207 degron motif (such as S2205, S2513, S2516, S2538) which is known to be [RK] S/T P [RK] X 208 S/T/E/D, where X is any amino acid and RK is any amino acid except arginine (R) or lysine 209 (K) [67]. 210 Thus, we generated 5 peptides each carrying a serine to alanine point mutation in an identified 211 site. Following transient transfection of HEK293 cells with wild-type or mutated peptides we 212 performed immunoprecipitation using GFP-conjugated beads, to evaluate peptide binding 213 efficiency with endogenous FBXW7. 214 In 2004, Fryer and colleagues proposed that CDK8 phosphorylates serine residues 2481, 215 2484, 2506 (2513, 2516, 2538 in our annotation). They reported that when all of these residues 216 were mutated from serine to alanine, in vitro phosphorylation by recombinant CyClinC:CDK8 217 was dramatically reduced [35], suggesting these sites could be implicated in NICD stability. 218 However, we find individual mutations on serine residues 2205, 2516, 2527 and 2538 did not 219 affect the NICD-FBXW7 interaction ( Figure 3B, Supplementary Figure 3). However, 220 mutating serine 2513 to alanine, to render this residue non-phosphorylatable, completely 221 abolished the NICD-FBXW7 interaction ( Figure 3B). Cells transfected with the double mutant 222 S2513A/S2516A also showed a dramatic loss of the NICD-FBXW7 interaction. This did not 223 reflect a reduction in the level of immunoprecipitated GFP ( Figure 3B). 224 Thus, our data suggest that only serine 2513, of those we have tested, is the key NICD 225 phosphorylation site required for interaction with FBXW7, and thus potentially crucial for NICD 226 stability and turnover. screen.mrc.ac.uk/kinase-inhibitors) (University of Dundee) indicates that, at the 233 concentrations used in our assays, Roscovitine, in particular, is a potent inhibitor of CDK2, but 234 a very weak inhibitor of CK1, GSK3β and more than 50 other kinases tested. In order to 235 confirm results derived from the Kinase Profiling Inhibitor Database, we performed a kinase 236 assay in collaboration with the MRC-PPU International Centre for Kinase Profiling. We tested 237 three NICD phospho-peptides against the activity of a panel of different kinases including 238 CDK1 and CDK2 ( Table 1). Of seven kinases tested, 5 had no specific activity against any of 239 the peptides. In contrast, CDK1 and CDK2 elicited a very high activity against Peptide 1, which 240 contained serine residues 2513 and 2516, previously identified by Mass spectrometry analysis 241 to be phosphorylated in NICD and in particular 2513 we have shown to be crucial for the NICD-242

FBXW7 interaction. 243
These results demonstrate that CDK1 and CDK2 phosphorylate the C-terminal region of NICD 244 in vitro. Therefore, we decided to evaluate the contribution of these kinases to endogenous 245 NICD turnover and FBXW7 interaction in the cell lines. CDK2 siRNA treatment in HEK293 246 cells efficiently depleted CDK2 protein levels, with no effect on levels of CDK4. Under these 247 conditions NICD levels were significantly increased compared to control scrambled siRNA 248 treated cells, indicative of reduced NICD turnover (Figures 4A-B). 249 We repeated this assay and depleted CDK1 by a siRNA-mediated approach in HEK293 cells. 250 Under these conditions we also observed an increase in NICD protein levels compared to the 251 control ( Figure 4C), and this increase was statistically significant ( Figure 4D).Interestingly, 252 we also detected elevated levels of CDK2 and CDK8 following CDK1 depletion, suggesting a 253 possible compensation effect was occurring upon CDK1 knockdown. Nevertheless, this did 254 not prevent the effect of loss of CDK1 upon NICD turnover. 255 CDK8 has previously been proposed as a potential kinase involved in NICD phosphorylation 256 and turnover [35,65]. We monitored NICD levels following efficient siRNA-mediated depletion 257 of CDK8 in HEK293 cells. This resulted in very slightly elevated NICD levels (Figures 4E-F). 258 These data suggest CDK8 may also phosphorylate NICD and regulate its turnover as 259 previously proposed [35,65]. 260 Taken together, these data provide further validation of CDK1 and CDK2 involvement in NICD 261 phosphorylation and turnover. 262 263

Pharmacological inhibition of CDK2 and CDK1 activity increases levels of NICD in vitro 264
As a complementary approach to siRNA-mediated loss of function, we selected two small 265 molecule inhibitors that have a highly selective inhibitory activity against CDK2 kinase: 266 (Purvalanol B and GSK650394A). The specificity of each of these kinase inhibitors has been 267 tested (at two different concentrations) by the International Centre for Kinase profiling within 268 the MRC Protein Phosphorylation Unit at the University of Dundee. At 1 and 10 µM, 269 Roscovitine specifically inhibits more than 96% of CDK2 activity. Purvalanol B inhibits more 270 than 95% CDK2 activity, at 0.1 and 1 µM. GSK650394A inhibits 99% of CDK2 activity at both 271  Figures 4A-B). 279 Given our observation that both CDK1 and CDK2 can phosphorylate NICD peptides in vitro, 280 we similarly treated HEK293 cells with a CDK1 specific inhibitor, . Following 281 exposure to 10 µM of RO-3306 for 3 hours HEK293 cells showed elevated NICD levels 282 compared to control DMSO treated cells (Figures 5D-E). These data further support the 283 hypothesis that both CDK1 and CDK2 are likely to be involved in phosphorylation-mediated 284 regulation of NICD turnover. 285

NICD levels fluctuate over the cell cycle 287
It is well known that CDK1 and CDK2 share several substrates, with a consequent functional 288 redundancy [69]. Our data demonstrate that these two kinases can phosphorylate NICD in 289 vitro and in the absence of either kinase NICD levels increase in HEK293 cells suggesting 290 they are not acting redundantly in this context. In order to further test this, we analysed whether 291 NICD levels fluctuate during the cell cycle where the role of both CDK1 and CDK2 has been 292 extensively reported in regulating transition to distinct cell cycle phases [70]. 293 To that end, we synchronised HEK293 cells by using a double thymidine block assay. After 294 releasing from the second thymidine block, cells were collected at indicated time points and 295 cell cycle characterization was performed by Fluorescence Activated Cell Sorting (FACS) 296 ( Figure 6A). Figure 6A shows the distribution though the cell cycle of HEK293 cells after 297 synchronization at early S-Phase (0 hour), as previously reported [71]. Two and four hours 298 post release, the majority of cells were in S-phase. At six hours post release, the majority of 299 cells were in G2 phase, while eight hours after release, the majority of cells were in late 300 G2/early M phase. At ten hours post release, the majority of cells had exited mitotic phase and 301 already entered G1. After twelve hours from release, cells were in mid/late G1 of the new cell 302 cycle ( Figure 6A). The graph in Figure 6B represents the cell cycle distribution of HEK293 303 cells at distinct time points post double thymidine block and release from three independent 304 experiments analysed by FACS. Western blotting analysis of synchronized HEK293 cell 305 extracts showed the expression of several cell cycle regulatory proteins at the same distinct 306 time points reflecting distinct cell cycle phases as described above ( Figure 6C). Interestingly, 307 we found that NICD levels fluctuated in a striking manner during the cell cycle, whereas we 308 saw no change to levels of FBXW7. At 2, 4, 8 and 12 hours post release, we observed a 309 dramatic decrease of NICD expression corresponding to CDK2-dependent G1/S phase, 310 whereas the decrease 8 hours after double thymidine block release occurred in the CDK1-311 dependent G2/M phase transition. These data suggest that NICD levels fluctuate during the 312 cell cycle in a CDK1 and CDK2-dependent manner. 313 In addition, we analysed the cell cycle distribution by FACS in HEK293 cells after CDK2-siRNA 314 mediated depletion, which we have shown leads to a significant increase in NICD levels 315 ( Figure 4A) and we observed a statistically significant accumulation of cells in G1 phase 316 compared to the control, as expected for cells deprived of CDK2 activity and therefore unable 317 to pass the G1/S checkpoint (Figures 6D-E). These data suggest that the drop in NICD levels 318 occurring in G1/S phase is due to CDK2 phosphorylation of NICD. 319 320

Pharmacological inhibition of CDK2 increases NICD levels and delays the pace of the 321 segmentation clock in mouse PSM explants 322
In order to address whether CDK2 phosphorylation of NICD is involved in driving the NICD-323 FBXW7 interaction, we performed a co-immunoprecipitation assay with FBXW7 antibody and 324 analysed NICD by western blot after CDK2 inhibitor treatment. As above, in order to maximise 325 the amount of NICD immunoprecipitated HEK293 cells were treated with MLN4924 (to prevent 326 NICD degradation) +/-Purvalanol B (0.1 µM). A significantly reduced interaction between 327 NICD and FBXW7 was observed after treating HEK293 cells with Purvalanol B for 3 hours. 328 This did not reflect a reduction in the level of immunoprecipitated FBXW7 ( Figure 7A). 329 Statistical analysis on the density of western blot bands after immunoprecipitation, confirmed 330 an extremely significant reduction in the NICD-FBXW7 interaction following Purvalanol B 331 treatment ( Figure 7B). 332 In order to address the potential in vivo role of CDK-mediated NICD phosphorylation during 333 somitogenesis, we cultured E10.5 mouse PSM explants for 4 hours in the presence of 1µM of 334 Purvalanol B. Initially we analysed NICD levels by western Blot and just as in the in vitro 335 context, CDK2 inhibition resulted in increased NICD levels as compared to control embryos 336 ( Figure 7C). This provides the first in vivo evidence CDK2 is likely to be involved in NICD 337

turnover. 338
Previous reports have suggested perturbations to NICD turnover leading to increased NICD 339 levels/stability are closely linked to an increase in the period of segmentation clock oscillations 340 in the PSM [18]. To further explore whether clock gene oscillations were delayed following 341 CDK2 inhibition, we used the half-embryo assay, where the PSM from one half of an E10.5 342 mouse embryo is cultured in control media, while the contralateral half from the same embryo 343 is cultured in the presence of 1µM of Purvalanol B for 4 hours. In 55.5% of cases examined 344 (n=10/18), exposure of PSM explants to Purvalanol B caused a delay in the pace of oscillatory 345 mLfng expression across the PSM as compared to the control explant ( Figure 7D). Also, in 346 some cases the treated explant ("+") developed one somite less compared to the control 347 ( Figure 7D). 348 Additionally, we cultured E10.5 mouse PSM explants in the presence or absence of the CDK1 349 inhibitor, RO-3306, for 4 hours at 10µM. Analysis of NICD levels by western blot revealed that 350 inhibition of CDK1 leads to elevated levels of NICD compared to control embryos ( Figure 7E). 351 E10.5 mouse half embryo explants exposed to RO-3306 treatment for 4 hours showed 352 delayed clock oscillations of mLfng as compared to DMSO treated contralateral half embryo 353 explants ( Figure 7F). These data, not only support our in vitro findings, but importantly 354 provide, for the first time, the in vivo evidence that CDK1 and CDK2 are involved in NICD 355 stability and turnover and that this molecular regulation of NICD turnover is extricably linked 356 to the pace of the segmentation clock. 357 358

Mathematical model links NICD regulation and cell cycle 359
To understand how our findings on the molecular details of NICD regulation in individual cells 360 give rise to tissue-scale delay of the segmentation clock, we first developed a mathematical  We assume a -compartment model that describes sequential progression of a cell through 366 the cell cycle ( Figure 8A). It is assumed that CDKs that phosphorylate NICD, resulting in its 367 interaction with FBXW7 and subsequent degradation, are active in some of the cell cycle 368 states. We assume that: (i) NICD exists in non-phosphorylated and phosphorylated forms; (ii) 369 NICD is produced at constant rate 1 ; (iii) both forms of NICD get degraded at background 370 rate 7 ; (iv) NICD gets phosphorylated by CDKs at rate 3 ; (v) pNICD degrades at rate 6 with 371 6 > 7 ; and (vi) dephosphorylation of NICD occurs at rate 2 . We consider the 372 pharmacological perturbations as follows: LY411575 treatment experiments, in which NICD 373 production is inhibited, are simulated by setting 1 = 0; Purvalanol B/Roscovitine treatment 374 experiments, which target only CDK2, reduce the number of compartments where CDKs are 375 active by a factor of two; and MLN4924 treatment, which inhibits FBXW7 mediated 376 degradation of NICD, are simulated by removing the fast mode of decay ( 6 = 0). [18]. Here, we impose, without explicitly describing the molecular circuitry, this assumption. 398 We assume that each cellular oscillator has a natural frequency ( ) that is a function of cell 399 cycle position; when CDKs are active the baseline clock frequency, 0 , is increased by an 400 acceleration factor ( Figure 8D). As the cell cycle is asynchronous in the PSM, it is therefore 401 comprised of a population of oscillators with frequencies 0 or 0 . Given sufficiently strong 402 coupling a population of such oscillators yields synchronous oscillations with a period that is 403 an average of the individual oscillator periods (Figures 8E-8F). Upon CDK inhibition, the 404 relative number of faster oscillators is reduced hence the average period decreases (Figure 405  (Figure 7). 408

DISCUSSION 409
The Notch pathway plays multiple critical roles in somitogenesis. Notch signalling is critical for 410 dynamic clock gene expression and for somite formation in mice [9]. Notch-Delta signalling is 411 essential to synchronise clock gene oscillations among neighbouring cells of the PSM [10][11][12][13]. 412 Moreover, a recent publication using a pharmacological approach, has shown that modulating 413 the half-life of NICD affects the clock oscillation period and somite size [18]. It is important to 414 note that NICD degradation and turnover still occurred under these conditions but less 415 efficiently. In this study, we have significantly extended those findings to determine the 416 mechanistic details of regulation of NICD turnover, and we report for the first time that cell 417 cycle-dependent CDK1 and CDK2 activity is involved in NICD turnover, which has broad 418 implications across all developmental and disease contexts where Notch plays a role. 419 We demonstrate that kinase inhibitors (Roscovitine, DRB and XAV939), previously shown to 420 prolong NICD half-life and delay the segmentation clock pace in mouse and chick PSM in vivo, 421 [18] also increased endogenous levels of human NICD when a range of primary human cell 422 lines were treated for 3 hours. This highlights the conserved effect of these inhibitors on 423 regulating NICD stability, which is perhaps not surprising given the high degree of sequence 424 similarity between mouse, human and chicken NICD. 425 We demonstrate that, following exposure to these kinase inhibitors, the increased stability of 426 NICD accompanies a change in the array of phosphorylated isoforms of NICD observed. The 427 different profiles of phosphorylated isoforms of NICD observed, following exposure to each of 428 the inhibitors, suggest they each inhibit distinct kinases. In each case, however, the highest 429 molecular bands were no longer visible. It is possible that the multiple phosphorylation bands 430 are indicative of unique events, some of which may reflect priming phosphorylation events that 431 facilitate or increase the efficiency of secondary phosphorylation events, which then act as 432 phospho-degron signals to recruit E3 ligases that target NICD for degradation. It is not possible 433 to determine whether the lower molecular weight bands that remain, following inhibitor 434 treatment, are indicative of the loss of primary or secondary phosphorylation events. 435 436 A number of reports have highlighted the fact that the SCF FBXW7 E3 ligase plays an important 437 role in NICD degradation [35,44,[46][47][48]66]. SCFs (Skp1, Cullin-1, F-box protein) are a class 438 of E3 ligases that use Cullin-1 as a scaffold and F-box proteins as substrate receptors. FBXW7 439 is an evolutionary conserved F-box protein. Substrate phosphorylation instigates FBXW7 440 binding to a conserved CDC4 phospho-degron motif which then recruits the rest of the E3 441 ligase complex, including Cullin1, thereby targeting the substrate for ubiquitination and 442 subsequent degradation by the proteasome [45,49]. The role of FBXW7 in NICD regulation 443 is also supported by data showing that, when mammalian Sel-10 (homologue of FBXW7 in C. 444 elegans) is mutated, NICD is much more stable [46]. To date, the interaction between NICD 445 and FBXW7 by co-immunoprecipitation has only been shown using overexpressed proteins, 446 due to the fact this interaction is very transient and leads to efficient degradation of NICD [35, 447 44, 47, 48, 66]. In order to stabilise the interacting complex, we conducted experiments in the 448 presence of MLN4924, a Cullin1 neddylation inhibitor which prevents activation of Cullin1. 449 Under these conditions, we demonstrate for the first time that NICD and FBXW7 can interact 450 at endogenous levels in HEK293 cells. Moreover, this allowed us to demonstrate that, when 451 phosphorylation is disrupted by small molecule CDK inhibitors (Roscovitine or DRB), the 452 NICD-FBXW7 interaction is reduced. It is important to note that neither inhibitor abolished the 453 NICD-FBXW7 interaction which suggests again that they are each inhibiting only some of the 454 kinase activity involved in NICD phosphorylation and subsequent recruitment of FBXW7. This 455 aligns with the observation that in the chick/mouse PSM both of these inhibitors increase NICD 456 stability and the period of the segmentation clock but that NICD turnover still occurs in this 457 tissue, and thus is dependent on a number of different kinases/phosphorylation events 458 differentially targeted by these two inhibitors [18]. 459 By Mass Spectrometry analysis, we identified 15 phospho-sites within human exogenous 460 NICD in HEK293 cells, some of which have been previously identified and reported to be 461 involved with NICD turnover [35,65]. Among those phospho-sites identified, we found that 462 Serine 2513, when mutated to alanine, thereby rendering the site non phosphorylatable, was 463 essential for the interaction between NICD and FBXW7. Point mutations in a number of other 464 phosphorylated residues showed these are non-essential for this NICD-FBXW7 interaction, 465 including two residues that have previously been reported to potentially be required for the 466 interaction; S2484 and S2506 (S2516 and S2538 in our annotation) [35,66]. However, Fryer 467 and colleagues used exogenous proteins and mutated all three sites simultaneously so the 468 individual contribution/requirement of each residue was not addressed in that study. O'Neil 469 and colleagues have also reported, using exogenous proteins, that phosphorylation of the 470 Threonine residue at T2487 in mouse NICD (T2511 in our annotation) is key to driving the 471 NICD-FBXW7 interaction and subsequent ubiquitination of NICD [44]. We did not observe 472 phosphorylation at this residue in human NICD. However it would be interesting to repeat the 473 mass spec analysis using different enzyme digestions to determine if this reveals additional 474 phosphorylated sites that play a role in this interaction. 475 SCF substrates typically have many phospho-degrons that can be widely dispersed. It will be 476 important to test the requirement of all the phosphorylated residues we have identified in NICD 477 for driving and regulating the NICD-FBXW7 interaction. 478 We have identified CDK1 and CDK2 as two kinases that can phosphorylate NICD in the PEST 479 region that harbours residue serine 2513, which we have demonstrated to be crucial for the 480 interaction between NICD and FBXW7. Through two loss of function approaches, namely 481 siRNA and a pharmacological approach, we further demonstrated the role of both CDK1 and 482 CDK2 kinases in the regulation of NICD turnover. Indeed, by transfecting HEK293 cells with 483 CDK1, or CDK2 specific siRNA, or treating HEK293 cells or iPS cells with highly selective 484 small molecule inhibitors against CDK1 and CDK2, we could appreciate an increase in NICD 485 levels, indicating that inhibiting phosphorylation by CDK1 or CDK2, renders NICD more stable. 486 As described above, the serine 2513 residue has previously been reported to potentially be 487 involved in NICD turnover in other systems [35]. Using gain and loss of function experiments 488 with exogenous proteins, the authors reported phosphorylation occurred through 489 CyclinC:CDK8. Indeed, by transfecting HEK293 cells with CDK8 specific siRNA, we could also 490 appreciate an increase in NICD levels, indicating that this kinase could also play a role in NICD 491 turnover, although the effect appears less dramatic compared to that observed with CDK1 or 492 CDK2 specific siRNA. 493 CDK1 and CDK2 activity and thus phosphorylation of their substrates changes in a cell cycle 494 dependent manner [72]. Strikingly, we find that NICD levels vary in HEK293 cells in a cell 495 cycle-dependent way such that we observe lowest levels of NICD in phases of the cell cycle 496 where CDK1/Cyclin B1 and CDK2/Cyclin E levels are highest and therefore when these 497 complexes are reported to be most active [73]. This finding suggests that NICD activity, signal 498 duration and signal strength is likely to vary in a cell cycle dependent manner, potentially 499 leading to differential transcriptional output in different phases of the cell cycle. 500 Given our findings, it is striking that CDK2 homozygous null mice are viable [74]. However, it 501 has been reported that CDK1 and CDK2 share more than 50% of their targets [75] which is 502 likely to allow for a lot of redundancy in this knock-out line, as CDK1 can compensate for loss 503 of CDK2 by forming active complexes with A-, B-, E-and D-type cyclins. The CDK1-null 504 homozygous mice however are embryonic lethal and a CDK2 knock-in to the CDK1 locus is 505 unable to rescue this phenotype [76,77]. It would be interesting to examine the role of the 506 CDK1 conditional mutant in the context of Notch signalling and somitogenesis [78]. 507 Interestingly, Cyclin C-null mice, which are embryonic lethal at 10.5, show dramatically 508 reduced NICD1 phosphorylation in vivo, and elevated NICD1 levels. The authors show that 509 Cyclin C can complex with CDK3, CDK19, CDK8, CDK1 and CDK2 to phosphorylate NICD1 510 and promote NICD1 degradation [65]. Thus, it would be very interesting to examine a PSM 511 conditional Cyclin C loss of function to determine whether this cyclin is involved in regulating 512 NICD turnover in this tissue. It is noteworthy Cyclin C levels did not appear to vary in a cell 513 cycle dependent manner in HEK293 cells, although it is possible the activity of Cyclin C/CDK 514 complexes varies in a cell cycle-dependent manner through post-translational modifications 515 rather than protein levels per se. 516 The finding that CDK1 inhibition or CDK2 inhibition leads to increased levels of NICD in E10.5 517 mouse embryo PSM lysates and to a delay in clock gene oscillations and somite formation 518 provides the first in vivo evidence that both CDK1 and CDK2 are involved in NICD turnover. 519 It is noteworthy that inhibition of CDK2 activity with a highly selective inhibitor reduces the 520 NICD-FBXW7 interaction in HEK293 cells but does not block it completely, as we saw with 521 Roscovitine and DRB. Indeed, inhibition of CDK1 or CDK2 in the mouse PSM with highly 522 selective inhibitors raises NICD levels and also slows, but does not block, dynamic Notch 523 target clock gene expression, and somitogenesis. This suggests that some NICD turnover 524 persists under these conditions, possibly through redundancy between CDK1 and 2 and/or 525 through CDK8-mediated phosphorylation, the subsequent recruitment of E3 ligases and 526 degradation of NICD. Moreover, a report by Chiang et al. has also identified a region 527 downstream of the PEST sequence, termed S4, that is involved in NICD degradation, but that 528 is independent of FBXW7 activity [79]. These data indicate there are several mechanisms 529 regulating NICD turnover which are partially redundant. 530 We also developed a mathematical model that coupled cell cycle dynamics to NICD 531 degradation. Using HEK293 cells, model parameters were identified that recapitulated the 532 it was assumed that levels of NICD are anti-correlated with clock somitogenesis frequency. 539 Hence, the posterior PSM is represented by a population of phase coupled oscillators whose 540 frequency is cell cycle dependent. Simulating CDK2 inhibition removes a pool of faster 541 oscillators thus reducing the tissue period. 542 Notch plays a key role as a gatekeeper protecting progenitor and/or stem cells in multiple 543 developmental contexts, in part through preventing differentiation and in part through 544 regulating components of the cell cycle [80][81][82]. Our novel finding of a reciprocal auto-545 regulatory role between the cell cycle regulated CDKs, CDK1 and CDK2, and NICD turnover 546 has potentially great relevance to the developmental biology community and may provide 547 additional insight into disease/cancer contexts where this autoregulation may have gone awry. 548  cell lysates treated with DMSO or MLN4924 were subjected to immunoprecipitation using 587 FBXW7 antibody, or IgG antibody as negative control, and precipitated material was analysed 588 by western blot using NICD antibody. Western blot with FBXW7 antibody served as loading 589 control for immunoprecipitation efficiency. 10% of cell lysate before immunoprecipitation was 590 used as input control and β-Actin served as loading control. 591 (B) Quantification of the density of western blot bands in (A) performed by ImageJ software. 592 Data are expressed as fold changes compared to DMSO. All data represent the mean ± SEM 593 from three independent experiments. Student t-test was used to determine p values, with 594 **p≤0.01. 595 (C) Roscovitine treatment reduced the NICD-FBXW7 interaction. 500 µg of HEK293 cell 596 lysates treated with MLN4924 or MLN4924 together with Roscovitine were subjected to 597 immunoprecipitation using FBXW7 antibody, or IgG antibody as negative control, and 598 precipitated material was analysed by western blot using NICD antibody. Western blot with 599 FBXW7 antibody served as loading control for immunoprecipitation efficiency. 10% of cell 600 lysate before immunoprecipitation was used as input control and β-Actin served as loading 601 control. 602

(D) Quantification of the density of western blot bands in (C) performed by ImageJ software. 603
Data are expressed as fold changes compared to MLN4924 treated samples. All data 604 represent the mean ± SEM from three independent experiments. Student's t-test analysis was 605 performed, with ***p≤0.001. 606 (E) Interaction between NICD and FBXW7 is reduced following DRB treatment. 500 µg of 607 HEK293 cell lysates treated with MLN4924 or MLN4924 together with DRB were subjected to 608 immunoprecipitation using FBXW7 antibody, or IgG antibody as negative control, and 609 precipitated material was analysed by western blot using NICD antibody. Western blot with 610 FBXW7 antibody served as loading control for immunoprecipitation efficiency. 10% of cell 611 lysate before immunoprecipitation was used as input control and β-Actin served as loading Cyclin E were determined by western blot. β-Actin served as loading control. 620 interaction. hNICD-GFP phospho-mutant peptides encoding non-phosphorylatable residues 629 at S2513 and/or 2516 (serine to alanine) were expressed in HEK293 cells. The exogenously 630 expressed protein was subsequently immunoprecipitated with anti-GFP antibody and 631 precipitated material was analysed by western blot using FBXW7 antibody. Wild-type hNICD-632 GFP and GFP only vectors were included as positive and negative controls, respectively. 633 Western blot using GFP antibody served as immunoprecipitation efficiency control. β-Actin 634 has been used as loading control for the input lanes. 635   is able to inhibit more than 95% of CDK2 activity, but is far less effective against other kinases. 676 Purvalanol B (0.1 and 1µM) inhibits more than 94% of CDK2 activity. At both 1 and 10 µM, 677 GSK650394A is able to inhibit more than 98% of CDK2 activity. Source: The Kinase Profiling 678 Inhibitor Database (http://www.kinase-screen.mrc.ac.uk/kinase-inhibitors).    Dundee. All NOTCH1 mutants were generated and obtained from MRC-PPU reagents, 852

University of Dundee. 853
Briefly, the fragment NOTCH1 1754-2555 (end) was synthesized by GeneArt with flanking 854 BamHI and NotI restriction sites to facilitate cloning, the sequence was codon-optimized for 855 mammalian expression. This was then digested and ligated into expression vector pCMV5D 856 GFP to make the wild type clone pCMV5D GFP NOTCH1 1754-end. 857 Site-directed mutagenesis was carried out using the QuikChange Lightning Site-Directed 858 Mutagenesis Kit (Agilent Technologies) but substituting the Taq with KOD Hot Start DNA 859 polymerase (Novagen). All mutations were confirmed by sequencing.  To synchronize HEK293 cells at G1/S, a double-thymidine block assay was performed. Cells 984 were treated with 2.5 mM of thymidine (Sigma) for 18 hrs, washed twice with PBS, released 985 into fresh media for 9 hrs, treated for a further 14 hrs with 2.5mM of thymidine, then released 986 again into fresh media following two washes in PBS. Cells were then collected for FACS and 987 Western Blot protein analyses at the indicated time points, following protocols already 988 To account for the experimental observations of NICD degradation in cell lines, we develop a cell cycle dependent model of NICD production and degradation (see Figure  8A). The position of the j th cell in the cell cycle is represented by a set of K discrete state variables, X ij (t), defined such that It is assumed that cells transition sequentially between states at rate r 1 and that mitosis is followed by G 1 entry. The averaged fraction of cells in state i, given by where n is the number of cells in the experiment, satisfy We note that (author?) (1) propose a similar model and show that relatively large number of compartments (K = 25) compartments can be needed to reproduce experimentally observed variances in cell cycle distribution times. Letting X(t) represent the population average fraction of cells where CDKs are active, then where M, the set of CDK active states, has m members. We make the following assumptions about NICD: (i) NICD exists in two states (nonphosphorylated and phosphorylated); (ii) NICD is produced and degraded at constant rates k 1 and k 7 , respectively; (iii) NICD is phosphorylated by CDK1 and/or CDK2 at 1 rate k 3 ; (iv) phosphorylated NICD has a larger degradation rate, k 6 , than the background rate, k 7 ; and (v) dephosphorylation occurs at rate k 2 . Letting N 1j (t) and N 2j (t) represent the amount of NICD and phospho-NICD in the j th cell at time t, the population averages are defined to be and satisfy As the cell cycle variables are uncoupled from the NICD dynamics, in a typical cell line experiment each of the X j 's will tend to the equilibrium value and the fraction of CDK active cells is Hence equations (8) reduce to To compute NICD levels equations (11) are solved to steady state (N * 1 , N * 2 ) and the total amount of NICD ( Figure 8B and 8C) is N * = N * 1 + N * 2 . LY treatment experiments, in which the production of NICD is reduced owing to gammasecretase inhibition, are simulated by setting k 1 = 0. As PurB/Roscovitine is assumed to inhibit CDK2, the effect in the cell cycle synchronised experiments is to reduce the parameter m. As MLN treatment inhibits FBXW7 mediated degradation, it is simulated by removing the fast decay mode (i.e. k 6 = 0). Now consider a situation in which cells are synchronised as a result of double thymidine block. At the point of release the cell cycle distribution is no longer at equilibrium, rather the initial conditions are where S is a set of states where cells get paused as a result of double thymidine block. After the cells are released from the cell cycle block there are peaks of CDK activity (see Figure 8B) and therefore dynamic levels of NICD.

Somitogenesis clock and the cell cycle
To explore the coupling between cell cycle mediated NICD degradation and the somitogenesis clock oscillator in PSM tissue, we introduce an additional set of variables, θ i (t), that represent the position of the i th cell in the somitogenesis clock cycle. As PSM oscillators are coupled via Delta-Notch signalling we consider a phase coupled oscillator model (e.g. 2; 3; 4) of the caudal PSM given where ω i , the natural frequency of the i th cell, is a functional of cell cycle position, the sum is taken over all oscillators and the coupling function represents the effect of Delta-Notch signalling in the caudal PSM. In a previous model (5) we demonstrated how competition between NICD and Hes7 could yield a phenotype in which increased levels of NICD are correlated with a longer clock period; here we impose this assumption in the phase coupled oscillator model by assuming that the clock frequency is inversely correlated with levels of NICD ( Figure  8E). Hence as the cell progresses through the cell cycle, basal levels of NICD fluctuate and modify the natural frequency. We assume that ω i (X 1 , X 2 , ..., X K ) = rω 0 CDK active ω 0 otherwise.
where the parameter r represents the degree of acceleration of the clock when CDK is active. Equations (13) -(14) represent a mathematical model of coupled PSM oscillators whose frequencies are modulated by the cell cycle position. The model allows us to explore whether there is an emergent population-scale period and how it is affected by perturbing the cell cycle. To do this we simulate a population of N cells and find that given adequate oscillator coupling the oscillators synchronise to an emergent frequency which is an average over the individual oscillator frequencies 3 Parameter inference

Cell cycle
The parameters used in the model are presented in Table 2. To recapitulate the cell cycle phase distributions presented in Figure 6 of the Main text we use a 35 state model. The transition rate between compartments, r 1 , is defined such that the mean time for a cell to progress from state 1 to is the cell cycle period, T C = 12h. Based upon the data in Figure 6 in the Main Text we assume the cell cycle times presented in Table 2 (see Figure 1).

NICD modelling
The parameter k 1 represents the number of molecules produced per cell per hour. The fast and slow degradation rates are values inferred from (author?) (5). To determine the parameters k 2 and k 3 we use the DMSO and MLN treatment Blot data presented in Figure 1 of the Main Text. The measured normalised levels of of non-phosphorylated and phosphorylated forms of NICD in different conditions are given in Table 1. A least squares minimisation algorithm is used to minimise the error between the model and the data in Table 1 for the control and MLN datasets. Hence values for the parameters k 2 and k 3 are inferred.

Somitogenesis clock
The somitogenesis clock period in the mouse PSM is approximately two hours. As we do not have direct measurements of coupling strength or acceleration factor we simulate model behaviour over a range the specified parameter ranges.