Novel determinants of NOTCH1 trafficking and signaling in breast epithelial cells

The evolutionarily conserved Notch pathway controls cell–cell communication during development and in adult metazoans. It influences cell fate decisions, cell proliferation and cell differentiation, and contributes to the maintenance of normal tissue homeostasis. Consequently, misregulation of the Notch pathway is associated with a wide range of diseases, including congenital disorders and cancers with little to no cure. Signaling by Notch receptors is regulated by a complex set of cellular processes that include maturation and trafficking to the plasma membrane, endocytic uptake and sorting, lysosomal and proteasomal degradation, and ligand-dependent and independent proteolytic cleavages. We devised assays to follow quantitively the lifetime of endogenous human NOTCH1 receptor in breast epithelial cells in culture. Based on such analysis, we executed a high-content screen of 2749 human genes for which modulatory compounds exist, to identify new regulators of Notch signaling activation that might be amenable to pharmacologic intervention. We uncovered 39 new NOTCH1 genetic modulators that affect different steps of NOTCH1 cellular dynamics. In particular, we find that PTPN23 and HCN2 act as positive NOTCH1 regulators by promoting endocytic trafficking and NOTCH1 maturation in the Golgi apparatus, respectively, while SGK3 serves as a negative regulator that can be modulated by pharmacologic inhibition. Our findings might be relevant in the search of new strategies to counteract pathologic Notch signaling.


Introduction 63
The evolutionarily conserved Notch pathway is a form of direct cell-cell communication 64 that is extensively deployed in the regulation of multiple cellular functions during development and 65 in adults. It influences cell fate decisions, cell proliferation and cell differentiation, and contributes 66 to the maintenance of normal tissue homeostasis in both invertebrate and vertebrate metazoans 67  Upon synthesis in the endoplasmic reticulum (ER), Notch proteins are processed at site 70 S1 by Furin in the TGN (Trans Golgi Network), the distal portion of Golgi apparatus (GA). Such 71 cleavage generates the mature receptors exposed on the cell surface that are held together by 72 Ca 2+ coordination within the heterodimerization domain (HD; see Fig. 1A for a schematic of human 73 NOTCH1) [5,6]. Once at the cell surface, Notch receptor activation can be initiated by binding the 74 Delta/Serrate/Lag2 (DSL) family of Notch ligands that are present on the surfaces of adjacent 75 signal-sending cells [7]. Following ligand binding, the receptors undergo two activating proteolytic 76 cleavages; the first one is extracellular and is executed by ADAM10 at site S2 [8]. The second is 77 intracellular and is mediated at site S3 by the γ-secretase complex [9-11] (see Fig. 1A for a 78 schematic of human NOTCH1). These cleavages generate the transcriptionally competent NICD 79 (Notch Intracellular Domain) which translocates to the nucleus and activates the expression of 80 Notch target genes [12]. In addition, Notch signaling activation may occur through non-canonical 81 means on the endosomal surface, with or without stimulation by ligands [13,14]. 82 As a membrane-tethered transcription factor, the Notch signaling output is not only 83 controlled by ligands, ADAM metalloproteases, the γ-secretase complex and the CSL 84 transcriptional complex, but also through intracellular trafficking [2,12,15-18]. Indeed, loss or 85 modulation of various activities occurring at membranes positively or negatively impact signaling 86 in both model organisms and humans. These activities include glycosylation enzymes [19][20][21], 87 internalization regulators such as Dynamin [22], and ubiquitin ligases such as Dx (Deltex) [23-88 27], Su(Dx) (suppressor of Deltex)/AIP4/ITCH in mammals [28,29], and NEDD4 [17,30]. In 89 addition, components of the endo-lysosomal system, such as Endosomal Sorting Required for 90 Transport (ESCRT) proteins [31][32][33] or the vacuolar-ATPase (V-ATPase) pump are required for 91 endocytic Notch trafficking and signaling [34][35][36][37][38][39][40]. It is plausible that numerous other factors, yet 92 to be identified and characterized, contribute to fine tune Notch liberation from membranes 93 signaling by harnessing the intracellular trafficking machinery. Thus, aiming to improve knowledge 94 of Notch receptor trafficking and to identify novel genes that control Notch trafficking and 95 signaling, here we combine quantitative immunofluorescence, RNA interference, and automated 96 microscopic imaging to describe the kinetics of endogenous NOTCH1 export to the plasma 97

EGTA stimulation of Notch cleavage, nuclear translocation and signaling 122
Notch signaling activation was stimulated by Ca 2+ depletion [6]. For short-term Ca 2+ 123 removal, EGTA was added at a final concentration of 10mM for up to 30 minutes, while for 124 extended periods, EGTA was used at a final concentration of 2.5mM for 2-4 hours, followed by 125 downstream assays. a Leica TCS SL confocal system. Where noted, digital images were processed using Photoshop 145 or ImageJ software without biased manipulation. 146 ImageJ was used for quantitative analysis of IF images following spatial image calibration 147 based on imaging scale bars. To assess colocalization in co-stained samples, imaging channels 148 were first separated using the split channel function of ImageJ. To quantify NOTCH1 149 accumulation in the Golgi in HCN2-and SGK3-silenced conditions, and the accumulation of 150 NOTCH1 and EGFR in the Golgi and endoplasmic reticulum (ER) upon EGTA treatment, 151 ImageJ's freehand selection tool was used to trace regions of interest (ROIs) based on Giantin, 152 TGN46, or RTN3 signals. The ROIs were then saved and transferred onto the NOTCH1 and 153 EGFR channels using the ROI manager tool. Next, the areas (µm 2 ) and integrated densities 154 (IntDens) of the ROIs were measured to determine organelle (GA) size and the signal intensities 155 of NOTCH1 and EGFR (in the GA and ER), respectively. To quantify the levels of NOTCH1 and 156 EGFR at the PM, ImageJ's freehand tool was used to mark ROIs covering the area between the 157 inner and outer sides of the PM based on the NOTCH1 and EGFR cell surface signal. The IntDens 158 within the ROIs was then measured to determine NOTCH1 and EGFR levels at the PM. To 159 measure whole cell levels of NOTCH1/EGFR, whole-cell ROIs were created by tracing the outer 160 side of the PM based on the NOTCH1 and EGFR signals, followed by IntDens measurement to 161 determine the levels of NOTCH1 and EGFR in whole cells. The proportions (%) of NOTCH1 and 162 EGFR in each compartment were determined by applying the following formula in their respective 163 channels: % IntDens = (ROI IntDens ÷ whole cell ROI IntDens) × 100. At least 5 ROIs, in at least 164 five images were analyzed. The number of endolysosomal puncta, NOTCH1-and EGFR-positive 165 puncta, and the rate of NOTCH1 colocalization with EEA1 (early endosomes) and LAMP1 (late 166 endosomes/lysosomes) in PTPN23-and SGK3-silenced samples, as well as the rate of NOTCH1 167 and EGFR colocalization with LAMP1 in Bafilomycin (BafA1)-treated cells were determined as 168 follows. First, ImageJ's freehand tool was used to trace the cytoplasmic area between the edge 169 of the nucleus and the inner side of the PM on merged (NOTCH1+DAPI+EEA1, 170 NOTCH1+DAPI+LAMP1, or EGFR+DAPI+LAMP1) composite images. The ROIs were then 171 saved and transferred onto the respective single channels using the ROI manager tool. The 172 number of NOTCH1-, EGFR-, EEA1-, or LAMP1-positive puncta was then determined using 173 ImageJ's ComDet v.0.5.5 plugin (DOI: https://doi.org/10.5281/zenodo.6546038) after empirically 174 determining the spot detection thresholds for each channel. For spot (puncta) detection, the 175 intensity threshold was set at 4 and colocalization determined using a maximum distance of 2 176 pixels between colocalized spots. To determine the rate of colocalization with the organelle 177 markers, the NOTCH1 or EGFR channels were merged the EEA1 or LAMP1 channels to generate 178 composite images and the ComDet plugin used to determine the rate of puncta colocalization. At 179 least 5 ROIs, in at least 5 images were analyzed. 180

Western blot analyses 181
For western blot analyses, cells were scraped into 1 ml of the media they had been 182 cultured in, transferred into prechilled 1.5 ml Eppendorf tubes, and placed on ice. They were then 183 centrifuged at full speed (14,000 rpm) for 5 minutes at 4°C and the pellets rinsed once with ice-184 cold PBS 1X. Unless indicated otherwise, cell pellets were then lysed by resuspension in 60 µl 185 Rad, Cat #: 1705062). They were then imaged on a ChemiDoc MP imaging system (Bio-Rad). 215 Band intensities were quantified using Image Lab software (Bio-Rad) followed by statistical 216 analysis and graph visualization using GraphPad Prism. 217

Lambda phosphatase (λ-PPA) treatment 218
For the λ-PPA assay, two types of RIPA lysis buffer were used -RIPA supplemented with 219 protease (phenylmethylsulfonyl fluoride and 1 tablet of 1X cOmplete mini protease inhibitor 220 cocktail, EDTA-free) and phosphatase (sodium fluoride and sulfur monoxide) inhibitors, and RIPA 221 supplemented with protease inhibitors only (phenylmethylsulfonyl fluoride and 1 tablet of 1X 222 cOmplete mini protease inhibitor cocktail, EDTA-free). The first lysis buffer (containing 223 phosphatase inhibitors) was used as a negative control for phosphatase treatment. MCF10A cells 224 were harvested, lysed in the appropriate ice-cold RIPA buffer (with or without phosphatase 225 inhibitor), and protein quantified as described in section 2.5. Next, for each sample (protein 226 concentration: 1 µg/μl), 25 μg of protein (25 µl) were subjected to λ-PPA treatment following 227 manufacturer instructions. Briefly, 0.5 µl of λ-PPA (20,000 U) were added into each tube 228 (containing 25 µl of the sample), along with 3.2 µl of 10X λ-PPA buffer and 3.2 µl of 10X MnCl2 229 (final volume: 32 µl). The samples were then mixed by gently flicking the tube and incubated for 230 30 minutes at 30°C. Next, 10.75 µl of 4X Laemmli loading buffer were added into each sample, 231 followed by western blotting using the antibody against the γ-secretase-cleaved (Val1744) 232

NOTCH1 intracellular domain (N1ICD). 233
Cycloheximide treatment and assessment of NOTCH1 protein stability 234 Where indicated, after gene knockdown for 72 hours, MCF10A cells were treated with 10 235 μM cycloheximide (CHX) to block protein synthesis or an equal volume of DMSO (vehicle) for 4 236 hours. Cells were then harvested, lysed, and protein concentration quantified as described in 237 section 2.5. To assess NOTCH1 protein stability, 20μg of the protein lysate per sample were 238 subjected to western blot analysis using the antibody against full length NOTCH1 (N1FL) and 239 transmembrane NOTCH1 (N1TM), as well as the antibody against the γ-secretase-cleaved 240 N1ICD. Band intensities were then analyzed on GraphPad Prism to determine the differences 241 between the levels of N1FL, N1TM, and N1ICD in the CHX-treated vs untreated cells. 242

Gene silencing 243
Gene knockdowns were performed in 384-well plate format. Each gene/well was targeted 244 with a pool of four distinct siRNAs against different sequences of the respective target transcript.  To stimulate NOTCH1 activation, at the end of the 72 hours of gene knockdown, 3 plates 263 from each library were treated with EGTA for 2 hours to stimulate Notch cleavage. To do this, 10 264 µl of fresh media containing 12.5 mM EGTA was directly added into the wells to a final 265 concentration of 2.5 mM EGTA per well, and a final volume of 50 µl/well. The NoEGTA stimulation 266 plates received 10 µl of sterile water. The plates were then incubated for 2 hours at 37°C and 267 then fixed with 2% PFA for 15 minutes at room temperature. Fixation was done by adding 50 µl 268 of 4% PFA directly into the wells. Solutions were dispensed into the wells using a Multidrop™ 269 Combi Reagent Dispenser (ThermoFisher Scientific). 270

Automated immunofluorescence analyses 271
Automated immunostaining for high content screening (HCS) was performed using a 272 Biotek EL406 washer/dispenser equipped with a 192-tube aspiration manifold. The solution was 273 removed from the wells and the cells rinsed once with 1X PBS. Cells were permeabilized with 274 0.05% triton in 1% BSA blocking solution for 1 hour at room temperature followed by a single 275 wash with 1X PBS. Cells were then incubated for 1 hour at room temperature with the anti- Cat #: P5282), and Alexa Fluor 488 anti-rat secondary antibody at 1:400 (ThermoFisher Scientific, 281 Cat #: A21208). They were then washed thrice with 1X PBS before imaging. Where imaging could 282 not be performed immediately, the immunostained cells were stored at 4°C for not more than 48 283 hours prior to image acquisition. 284

Automated image acquisition 285
The 384-well plates were scanned using an automated Olympus Scan^R (Tokyo, Japan) 286 microscope equipped with a Hamilton arm for plate handling. Eight fields of view and three 287 emission fluorescent channels (DAPI, phalloidin, and Alexa 488), were acquired for each well 288 using a 20X objective. Imaging data were annotated and transferred to the Isilon infrastructure 289 and network software and indexed by plate barcode for storage. In the annotation each well was 290 assigned information regarding the date of the experiment, the gene knocked down, the sub-291 genomic library the knocked down gene belongs to, and the treatment (EGTA vs NoEGTA). 292 Images were then uploaded to the Columbus server (PerkinElmer), where they could be accessed 293 for visual examination and automated analysis. 294

High content screen image analysis 295
An in-house Acapella (PerkinElmer) image analysis script was developed and used for 296 batch image analysis and to quantitatively describe a set of phenotypic features. All analyzed 297 images were first subjected to background correction and exclusion of unevenly illuminated 298 images. Background correction was done separately for each channel. DAPI, which was used to 299 mask the nuclei and phalloidin, which labeled cell surfaces, were used for cell segmentation. 300 Segmentation was performed using a modified version of the watershed algorithm, which allowed 301 the inversion of phalloidin channel images so as to display high pixel intensities in the cell and 302 low intensities along the cell membrane, allowing application of the watershed approach to identify 303 cell boundaries. This algorithm also detects and excludes regions of the image fields not covered 304 by cells. Once the cells, their membranes, and corresponding nuclei were detected, each cell was 305 segmented into the nucleus based on DAPI staining, the membrane based on phalloidin signal, 306 and the cytosol (region between the nuclei and cell surface membrane). Cells that were adjacent 307 to field of view borders, those deemed to be too small or too big, those with saturated pixel 308 intensities or those that were improperly segmented, were excluded from downstream image 309 analysis. To exclude out-of-focus images, we used boxplot statistics on the distribution of intensity 310 contrast values (on the DAPI channel) of all nuclei detected in the entire well. Using the first and 311 the third quantiles of this distribution, we estimated the lower inferior fence (LIF) using the 95% 312 confidence interval. For each field of view, we next established the number of nuclei presenting a 313 contrast lower than the LIF and classified fields in which >50% of the nuclei failed to cross the LIF 314 threshold as being out of focus. Data analysis on the EGTA stimulated plates was performed 315 independently of the corresponding non EGTA-treated plates. For each gene (well), the 316 parameters of interest were quantified and reported as z-score values [41]. 317

Candidate gene selection for validation 318
For each library plate, candidate genes from the EGTA and the NoEGTA conditions were 319 considered to affect respective parameters if their knockdown shifted the z-score positively or 320 negatively, the further the shift, the stronger the phenotype. The main parameters quantified were, plates, grown while observing the growth characteristics, and then transferred into 6-well plates 337 after reaching 70%-80% confluence. Eventually each single cell clone population was split into 338 two wells of a 6-well plate and frozen after reaching 70%-80% confluence, followed by long term 339 storage in liquid nitrogen. Each clone was then tested using a luciferase assay for Notch signaling 340 inducibility using EGTA. The two best clones with the widest Notch signaling induction assay 341 window, measured as the ratio between basal luciferase activity (in the absence of EGTA) to hours followed by fluorescence reading on a Promega Glomax multimode plate reader. 360

Secondary screen for signaling activity 361
Relevant genes were knocked down in MCF10A-RbpJK-Luc Notch signaling reporter 362 cells, on white opaque 384-well plates (Corning, Cat #: 3712) followed by luciferase/proliferation 363 assays using 500 µM D-luciferin and 2.5% resazurin after 72 hours. Where indicated EGTA was 364 used at a final concentration of 2.5 mM. Briefly, a mix of D-luciferin+resazurin was prepared 365 immediately before the experiment by dissolving D-luciferin and resazurin in normal MCF10A 366 cells media at a concentration 3.6X higher than the desired final concentration. Where EGTA-367 stimulated Notch activation was desired, a mix of D-luciferin+resazurin+EGTA was prepared by 368 including EGTA at a concentration 3.6X higher than its desired final concentration. Next, 15 µl of 369 the appropriate solution of Notch luciferase assay mix was added into each well of the 384-well 370 plate and the cells incubated for 4 hours in normal conditions. To assess Notch signaling output 371 (luminescence) and cell viability (resazurin fluorescence), luminescence and fluorescence were 372 read on a PHERAstar BMG LABTECH microplate reader. 373

Statistical analyses 374
All experimental data were analyzed on GraphPad Prism. All data represent at least 3 375 independent replicates and are shown in graphs as individual values, mean ± SD, or normalized 376 values. Statistical differences between two groups were evaluated using unpaired two-tailed 377 Student's t-tests, with *, **, ***, ****, and ns indicating p <0.05, p <0.01, p <0.001, p <0.0001, and 378 not significant, respectively. Data and statistical analyses for the screen were done as described 379 in section 2.11. 380

Acknowledgements and funding information 381
We Immuno-localization with an antibody that recognizes the intracellular domain of NOTCH1, shows 395 that endogenous NOTCH1 is localized on the cell surface of confluent MCF10A cells (Fig. 1B). stably expressing the Notch activation reporter RBPj-luc (Fig. 1D). 403 We previously observed that the S3-cleaved NOTCH1 intracellular domain (N1ICD) is 404 visible in western blots as a band doublet around the 120 kDa region [34] (Fig. 1E). We suspected 405 that the slightly higher molecular weight band of the doublet may represent the N1ICD that is 406 phosphorylated and destined for proteasomal degradation, while the lower molecular weight band 407 represents newly generated N1ICD that has not been phosphorylated [43][44][45]. Indeed, in extracts 408 of cells treated with the γ-secretase inhibitor DAPT to block the production of N1ICD, we observed 409 the presence of the upper band only (Fig. 1E). To test if the upper band of the N1ICD doublet was 410 indeed phosphorylated, we harvested proteins from unstimulated MCF10A cells, or from cells 411 stimulated with EGTA for 10 minutes to induce N1ICD generation, or from cells stimulated with 412 EGTA for 10 minutes and then returned to normal medium for 1.5 hours. We then treated the 413 lysates with or without λ-phosphatase (λ-PPA) [46]. λ-PPA treatment resulted in a minor downshift 414 of the N1ICD in extracts stimulated with EGTA for 10 minutes and in a large downshift in those in 415 which the EGTA had been washed out for 1.5 hours after stimulation, indicating that N1ICD is 416 progressively phosphorylated shortly after S3 cleavage (Fig. 1F). Together, these data indicate 417 that MCF10A cells are endowed with an endogenous pool of plasma membrane-localized 418 NOTCH1 that can be canonically activated in a S2 and S3-dependent fashion to promote target 419 gene transcription before being phosphorylated. 420

Dynamics of NOTCH1 receptor trafficking in MCF10A cells 421
To study the dynamics of endogenous NOTCH1 trafficking, we first treated MCF10A cells 422 with the V-ATPase inhibitor BAfA1 for 4 hours to block lysosomal protein degradation. As  and controls in unstimulated cells (Fig. 3A). To identify genes that are also important for NOTCH1 467 nuclear localization upon signaling activation, we subjected duplicate plates to EGTA stimulation. 468 We then stained cells with an antibody that recognizes the cytoplasmic portion of NOTCH1. The 469 cells were counterstained with DAPI and phalloidin to visualize nuclei and cell cortices, 470 respectively (Fig. 3B). We devised an automated high content image analysis pipeline to identify 471 factors that when silenced affected NOTCH1 subcellular localization by altering receptor amounts 472 on the cell surface, in the cytoplasm, or in the nucleus (Fig 3B). In such primary screen, we 473 identified 231 potential modulators of NOTCH1 localization (  Fig. 2A). We reasoned that PTPN23 depletion might interfere with 492 NOTCH1 endosomal trafficking or sorting. To identify the compartment in which the NOTCH1 493 accumulated, we co-stained cells with antibodies against NOTCH1 and the endo-lysosomal 494 markers EEA1 or LAMP1, which identify early endosomes and lysosomes, respectively. Efficient 495 depletion of PTPN23 (Supplemental Fig. 2B) led to significant expansion of the cytoplasmic pool 496 of puncta positive for EEA1, LAMP1, and NOTCH1 (Fig. 5A, quantified in A'). Quantitative 497 analysis also revealed that the NOTCH1 that accumulated upon PTPN23 depletion mainly resided 498 in puncta positive for EEA1 (Fig. 5A'). Taken together, these data indicated that PTPN23 depletion 499 might interfere with early to late endosomal trafficking, thereby trapping NOTCH1 mostly in early 500

endosomes. 501
We next evaluated how loss of PTPN23 might affect Notch processing and signaling. 502 Western blot analysis revealed a marked reduction of both N1FL and N1TM upon PTPN23 KD 503 ( Fig. 5B) but surprisingly, the levels of N1ICD were not significantly altered, both in unstimulated 504 and stimulated conditions (Fig. 5C). To assess the effect of PTPN23 depletion on Notch signaling, 505 we measured Notch-mediated RBPj-luc expression and found that when compared with control 506 cells, PTPN23 silencing significantly suppressed Notch signaling in unstimulated conditions but 507 not in the presence of EGTA (Fig. 5D). Similar observations were obtained when measuring HES1 508 expression by quantitative reverse transcription PCR (RT-qPCR) (Fig 5E). Because reduced 509 biosynthetic N1FL levels could also result from low NOTCH1 transcription, we also measured 510 NOTCH1 mRNA levels and found that upon PTPN23 depletion, NOTCH1 expression levels were 511 mildly but significantly reduced (Fig 5F). These data indicate that PTPN23 is required for NOTCH1 512 sorting into MVBs and for sustaining Notch expression and basal signaling and illustrate the 513 screen's ability to identify novel modulators of NOTCH1 trafficking and signaling. 514

HCN2 silencing traps NOTCH1 in an enlarged GA and suppresses Notch signaling 515
Automated analysis of HCS IF images identified 4 genes encoding channels (CACNB4, 516 TRPM7, HCN2, and MLC1) that upon KD in unstimulated conditions, caused cytosolic NOTCH1 517 accumulation (out of a total of 44 hits, a 9.1% enrichment; Supplemental file 3). Of these, HCN2 518 KD led to the most striking phenotype, with strong NOTCH1 accumulation in the perinuclear space 519 when compared with mock-silenced (non-targeting/NT) controls (Supplemental Fig. 3A). To test 520 if NOTCH1 accumulated in the perinuclear GA, we knocked down HCN2 (Supplemental Fig. 3B) 521 and co-stained for NOTCH1 and the GA protein GIANTIN. Confocal analysis confirmed the 522 perinuclear NOTCH1 localization observed in the HCS images and revealed an expansion of the 523 Giantin-positive GA and increased localization of NOTCH1 in such compartment, when compared 524 to control cells (Fig. 6A, quantified in A'). Additionally, the analysis revealed that when compared 525 with the control, HCN2 silencing markedly reduced cell surface NOTCH1 levels (Fig 6A, quantified  526 in A'), suggesting that HCN2 silencing might interfere with trafficking of newly made NOTCH1 to 527 the cell surface. To assess this, we compared NOTCH1 and EGFR localization upon HCN2 KD 528 and 4-hour EGTA treatment. Both NOTCH1 and EGFR mostly failed to return to the PM after 1 529 or 4-hour w/o (Supplemental Fig. 4A; quantified in A'), indicating that both receptors do not traffic 530 efficiently to the PM in the absence of HCN2. Relative to mock silencing, HCN2 depletion also led 531 to alteration of the TGN with loss of TGN46 and slight elevation of Golgin-97 (Fig 6B, quantified  532 in B'). 533 Next, we extended the analysis of HCN2 KD to Notch signaling output. HCN2 depletion 534 significantly reduced expression of a Notch signaling reporter in unstimulated cells and limited the 535 expression of the NOTCH1 target gene, HES1, relative to controls (Fig. 6C-D). Consistent with 536 these observations, Western blot analysis revealed that when compared with the control, HCN2 537 silencing markedly reduced the levels of N1ICD both in the absence and presence of EGTA (Fig.  538   6E). Intriguingly, Western blot analysis of the levels of the S1 substrate, N1FL, revealed the 539 opposite trend, with HCN2 KD elevating N1FL levels relative to the control ( Fig. 6F; Supplemental 540 Fig. 3D). In contrast, the levels of N1TM, the S2-processed form of NOTCH1, were reduced in 541 HCN2 KD cells, when compared with controls (Fig. 6F). Together, these findings suggest that 542 HCN2 is required to support a TGN organization that allows NOTCH1 S1 processing, PM delivery 543 and signaling. 544

SGK3 silencing elevates cytosolic NOTCH1 levels 545
Automated image analysis identified 5 kinases that when silenced, interfered with 546  Fig. 5). 553 To validate this observation, we silenced SGK3 and analyzed the cells by confocal microscopy 554 after co-staining with antibodies against GOLGIN-97, EEA1 and LAMP1, to mark the early TGN, 555 early endosomes, and lysosomes, respectively. This analysis confirmed the presence of elevated 556 intracellular NOTCH1 levels but revealed no significant difference in its colocalization with the 557 Golgin-97-, EEA1-and LAMP1-positive compartments (Fig. 7A-B, quantified in A'-B'). We also 558 observed a significant elevation of EEA1 and LAMP1 puncta, suggesting that SGK3 might 559 regulate the endocytic compartment. 560 Given the diffused accumulation of NOTCH1 observed upon SGK3 silencing, we 561 wondered whether this correlated with elevated NOTCH1 processing and signaling levels. We 562 first tested which NOTCH1 form accumulated upon SGK3 KD and found that SGK3 depletion 563 leads to elevated N1FL, N1TM, and N1ICD levels in both unstimulated and stimulated conditions 564 ( Fig. 7C-D). We next used the Notch signaling reporter cell line MCF10A-RbpJk-Luc and 565 observed that when compared with the control, silencing SGK3 caused significantly higher Notch 566 signaling output upon EGTA stimulation (Fig. 7E). Although signaling did not differ significantly in 567 unstimulated SGK3-silenced MCF10A-RbpJk-Luc cells when compared to mock silencing, RT-568 qPCR analysis revealed that SGK3 KD in MCF10A cells significantly upregulated the expression 569 of the Notch signaling targets HEY1, but not HES1. Depletion lead also to a slight transcriptional 570 elevation of NOTCH1 (Fig. 7F-H). 571 The observation that depleting SGK3 in MCF10A cells leads to elevated N1ICD levels 572 made us wonder whether SGK3 reduction might impair N1ICD degradation. To test this 573 possibility, we silenced SGK3 and then treated the cells with cycloheximide (CHX) to block protein 574 synthesis or with DMSO as a negative control. Because blocking the synthesis of new proteins 575 would allow cells to turn over proteins that had been present before CHX addition, we reasoned 576 that if SGK3 is required for efficient N1ICD degradation, upon CHX treatment SGK3-depleted 577 cells would exhibit elevated N1ICD levels when compared with control cells. Western blot analysis 578 using antibodies against N1FL and N1ICD revealed that SGK3 silencing caused marked increase 579 in the levels of N1FL and N1ICD when compared with control. However, N1FL and N1ICD levels 580 were markedly decreased in both control and SGK3 KD cells upon CHX treatment (Fig. 7I), 581 indicating that SGK3 is not required for N1ICD destabilization. Finally, to assess whether SGK3 582 contributed directly or indirectly to N1ICD phosphorylation, we silenced SGK3 in MCF10A cells 583 and then performed the λ-PPA assay. This analysis revealed that in SGK3 KD cells, accumulated 584 N1ICD was still phosphorylated, and that phosphorylation could be erased by treating the cell 585 extracts with λ-PPA (Fig. 7J), indicating that SGK3 is not required for N1ICD phosphorylation. 586 Taken together, these data indicate that SGK3 negatively regulates NOTCH1 levels and signaling 587 indirectly by a mechanism other than control of N1ICD stability and/or degradation. 588

Pharmacologic SGK3 inhibition elevates N1ICD levels and Notch signaling 589
To explore whether SGK3 modulation of NOTCH1 could be controlled pharmacologically, 590 Strikingly, EGF failed to elicit NDRG1 phosphorylation in VPS34-IN1-treated cells and erased the 597 basal low level of NDRG1 phosphorylation in unstimulated cells (Fig 8A), indicating that treatment 598 with VPS34-IN1 potently blocks SGK3 activity in cultured MCF10A cells. 599 Next, to determine if pharmacologic VPS34-IN1 treatment reproduces the effects of SGK3 600 silencing on NOTCH1, we treated MCF10A cells with VPS34-IN1 and measured N1ICD levels. 601 Western blot analysis revealed that relative to mock treatment, VPS34-IN1 significantly elevated 602 the levels of N1ICD (Fig 8B). To establish whether VPS34-IN1 treatment also elevates Notch 603 targets, we examined the expression levels of HEY1 by RT-qPCR. This analysis revealed that 604 when compared with control cells, HEY1 levels were significantly higher in VPS34-IN1-treated 605 cells (Fig. 8C). Taken together, these data indicate that as with SGK3 silencing, pharmacologic 606 VPS34-IN1 treatment elevates NOTCH1 levels and promotes Notch signaling. Among genes that we have identified, some have previously been associated with 672 regulation of the mTOR pathway (CIB2, MASTL, MC1R, SGK3, ZNRF1, ZNRF2). ZNRF1 appears 673 to be required for EGFR sorting to late endosomes, a prerequisite for lysosomal degradation and 674 signaling termination [78]. In contrast, ZNRF2 has been found to regulate mTOR signaling at 675 lysosomes [79]. Also, the calcium and integrin binding protein 2 encoded by the gene CIB2, which 676 we found here as a negative regulator of NOTCH1, is a negative regulator of mTOR signaling in We have characterized the activity of SGK3 as a negative regulator of NOTCH1 684 expression and signaling. SGK3 encodes a PI3K-dependent endosome-localized 685 serine/threonine kinase with similar substrate specificity to AKT, a kinase acting upstream of 686 mTOR widely involved in tumorigenesis [83]. All forms of NOTCH1 accumulate in SGK3 depleted 687 cells and we have excluded that SGK3 directly phosphorylates N1ICD. Thus, SGK3 is likely to 688 act indirectly on NOTCH1. Because SGK3 has been proposed to promote mTOR activation in 689 endo-lysosomal compartments, an interesting possibility is that mTOR and NOTCH1 might be 690 alternatively regulated. Interestingly, resistance to mTOR inhibition in breast cancer cells is 691 mediated by hVPS34 and SGK3, defining an AKT independent mTOR activation pathway [84]. 692

Considering that inhibition of Notch activity on TNBCC abrogates SGK3 expression [85], a 693
Notch/SGK3 regulatory loop might exist in breast cells, that will be further investigated in future. 694 Finally, it will be important to test whether SGK3 acts as a negative regulator in contexts in which 695 NOTCH1 acts as a tumor suppressor, such as during skin tumorigenesis [86]. 696 The relocalization of NOTCH1 to early endosomes upon PTPN23 KD matches earlier 697 observations in human cells that indicate that PTPN23 resides on early endosomes and that its 698 Finally, MET, which encodes the hepatocyte growth factor receptor (HGFR), is a proto-oncogenic 720 receptor tyrosine kinase found to be coregulated with Notch in physiologic as well as several 721 pathologic conditions, including in gastric and breast cancer cells [101][102][103][104][105]. These combined 722 studies warrant further characterization of the identified candidate genes in breast cancer cells. 723 A few genes that we have isolated in the screen have been previously involved in 724 regulation of Notch signaling (ADORA1, CALM2, JAK3, MAP3K12, PLAU, TRPM7)  . The N1ECD contains several EGF-like repeats, some of which are binding sites for Notch ligands as well as the negative regulatory region (NRR), which includes the Lin12/Notch Repeats (LNR) and the heterodimerization domain (HD). The N1ICD and N1ECD portions of Notch receptors are produced by Furin cleavage occurring at the S1 site during trafficking to the trans-Golgi network (TGN). N1ICD and N1ECD are non-covalently held together by Ca 2+ at the HD. Ca 2+ depletion causes ADAM10 and γ-secretase to sequentially cleave the receptor at S2 and S3 sites. This leads to the release in the cytoplasm of the N1 intracellular domain (N1ICD), which translocates to the nucleus. N1ICD contains the RBP-J-kappa-associated module (RAM), ankyrin repeats (ANK), nuclear localization signals (NLS), and the proline (P), glutamic acid (E), serine (S) and threonine (T) -PEST domain, which limits NICD half-life. (B) Immunofluorescence reveals that EGTA treatment relocates most endogenous NOTCH1 from the cell surface to the nucleus and this is inhibited by silencing ADAM10 or PSENEN. (C) siRNA against ADAM10 or PSENEN effectively deplete their mRNA. (D) The Notch reporter cell line, MCF10A-RbpJk-Luc, reports strong EGTA-induced Notch signaling, and this effect is markedly suppressed by silencing NOTCH1, PSENEN or ADAM10. (E) Western blot analysis of MCF10A cell extracts using an antibody against S3-cleaved N1 shows that unstimulated cells express low levels of N1ICD, which appears as a doublet. γ-secretase inhibition for 3 hours (DAPT) markedly reduces basal levels of the N1ICD lower band. (F) Western blot analysis of MCF10A lysates indicates that λ-phosphatase (λ-PPA) treatment leads to disappearance of the upper band of the N1ICD doublet, indicating that it corresponds to a phosphorylated N1CD form. *** and **** indicate p <0.001 and p <0.0001, respectively. . Compared with non-EGTA treated cells, 4-hour EGTA treatment causes diffused NOTCH1 and EGFR accumulation in the cytosol, with some EGFR remaining on the PM (green arrows). The Giantin signal is also diffused consistent with an expected fragmentation of the GA. EGTA washout (w/o) for 1 hour causes all NOTCH1 and most EGFR signal to localize in the GA (red arrows). EGTA w/o for 4 hours or overnight (ON) restores normal intracellular distribution of NOTCH1 and EGFR (quantified in C'; *, **, ***, ****, ns indicate p <0.05, p <0.01, p <0.001, p <0.0001, and not significant, respectively). (2) The cell surface was segmented using the phalloidin signal and overlaid on the NOTCH1 channel (3) to quantify levels of cell surface NOTCH1 (4) and the area between the cell cortex and the nuclei was used to determine levels of cytoplasmic NOTCH1 (5). (6) The phalloidin mask was also used to count cells and establish cell-to-cell boundaries. (7) Nuclear size was used as a readout of cell viability as compact, pycnotic nuclei identify dead cells. Using such a pipeline, each gene KD was defined by its effects on the amount of NOTCH1 at the cell cortex, in the cytoplasm or in the nucleus. Candidates that when compared with controls led to marked cytotoxicity, did not affect intracellular NOTCH1 (N1) localization, or caused general loss of NOTCH1 signal, were excluded from further analysis (orange, blue, and light blue circles, respectively). A total of 231 genes that altered intracellular NOTCH1 localization were identified (green). 117 did so in unstimulated condition, 63 upon EGTA stimulation and 51 in both conditions. (B) The 231 candidates underwent secondary screening. 73 of the 231 led to changes in NOTCH1 localization but only 51 genes reproduced the primary screen phenotypes. (C) Of these 51, 38 affected NOTCH1 trafficking in the unstimulated No EGTA condition, 13 altered NOTCH1 trafficking in the stimulated +EGTA condition. PTPN23 silencing causes marked increase in the number of EEA1, LAMP1, and NOTCH1 (N1) puncta, as well as accumulation of intracellular NOTCH1 positive puncta, which mostly colocalize with EEA1 (arrows; quantified in A'). (B-C) Western blot analyses indicate that PTPN23 depletion reduces the levels of N1FL and N1TM but does not alter N1ICD levels with or without EGTA. Note that levels of N1ICD in unstimulated and EGTA stimulated cells are not comparable in C, due to different detection methods (see materials and methods). (D-F) The Notch reporter cell line MCF10A-RbpJk-Luc and RT-qPCR analysis reveal that PTPN23 depletion suppresses basal Notch signaling. *, ***, ****, and ns indicates p <0.05, p <0.001, p <0.0001, and not significant, respectively. The Notch reporter cell line, MCF10A-RbpJk-Luc and RT-qPCR analysis reveal that HCN2 silencing suppresses basal Notch signaling. (E) Western blot analyses indicate that HCN2 depletion reduces both basal and EGTAstimulated N1ICD levels. Note that levels of N1ICD in unstimulated and EGTA stimulated cells are not comparable in C, due to different detection methods (see materials and methods). (F) Western blot analysis indicates that HCN2 silencing results in marked accumulation of N1FL. *, **, ****, and ns indicate p <0.05, p <0.01, p <0.0001, and not significant, respectively. Western blot analyses indicate that SGK3 does not destabilize N1ICD or N1FL. (J) Western blot analyses of lysates treated with λ-phosphatase (λ-PPA) showed that similar to controls, most of the N1ICD that accumulates upon SGK3 silencing is phosphorylated. *, **, ***, ****, and ns indicate p <0.05, p <0.01, p <0.001, p <0.0001, and not significant, respectively.