Doublecortin-like kinase 1 is a therapeutic target in squamous cell carcinoma

Doublecortin like kinase 1 (DCLK1) plays a crucial role in several cancers including colon and pancreatic adenocarcinomas. However, its role in squamous cell carcinoma (SCC) remains unknown. To this end, we examined DCLK1 expression in head and neck squamous cell carcinoma (HNSCC) and anal squamous cell carcinoma (ASCC). We found that DCLK1 is elevated in patient SCC tissue, which correlated with cancer progression and poorer overall survival. Furthermore, DCLK1 expression is significantly elevated in HPV negative cancer tissues, which are typically aggressive with poor responses to radiation therapy. To understand the role of DCLK1 in tumorigenesis, we used specific shRNA to suppress DCLK1 expression. This significantly reduced tumor growth, spheroid formation, and migration of HNSCC cancer cells. To further the translational relevance of our studies, we sought to identify a selective DCLK1 inhibitor. Current attempts to target DCLK1 using pharmacologic approaches have relied on non-specific suppression of DCLK1 kinase activity. Here, we demonstrate that DiFiD [3,5-bis (2,4-difluorobenzylidene)-4-piperidone] binds to DCLK1 with high selectivity. Moreover, DiFiD mediated suppression of DCLK1 led to G2/M arrest and apoptosis and significantly suppressed tumor growth of HNSCC xenografts and ASCC patient derived xenografts, supporting that DCLK1 is critical for SCC growth.


Introduction
9 levels on western blot were pictured by Bio-Rad ChemiDoc-XRS+ instrument and analyzed by 185 image lab software. Band intensity was calculated relative to the lowest dilution of pronase. 186 187 Immunoblotting 188 HNSCC cells (HN5, UMSCC1, OSC19, FaDu; 1x10 6 cells) were seeded in 100 mm tissue culture 189 dishes. Cells were treated with respective IC 50 µM concentrations of DiFiD or vehicle control for 190 24 and 48 hours. Fifty µg of the total protein lysate were subjected to polyacrylamide 191 electrophoresis and transferred to PVDF membranes with a wet-blot transfer apparatus  Rad, Hercules, CA). After blocking in 5% milk overnight and incubation with primary antibody 193 (using manufacturer recommended dilutions), the signal was developed with horseradish 194 peroxidase-conjugated secondary antibody (dilution 1:5000) and ECL Western blotting detection 195 reagents (Amersham-Pharmacia, Piscataway, NJ). Actin and GAPDH were used as loading Paraffin-embedded tissues were cut to 4 μm sections, deparaffinized and subjected to antigen 204 retrieval. The tissue sections were blocked with UltraVision Hydrogen Peroxide block for 10 205 minutes (Thermo Scientific). The slides were incubated with primary antibodies (1:200 dilution) 206 overnight at 4 º C. DCLK1 antibody was purchased from Abcam (#37994) (Cambridge, MA) and 207 used according to a previously published protocol (21). The following day, the primary antibody 208 was washed, and tissues were incubated with HRP Polymer Quanto for 10 minutes then 209 developed with a DAB Quanto Chromogen-Substrate mixture and counterstained with 210 hematoxylin. The slides were assessed using a Nikon Eclipse Ti microscope under a 200X 211 magnification. Staining intensity was determined by Aperio Imaging Software (Aperio 212 ImageScope V12.3.3). The slides were also independently assessed in a blinded manner by two 213 board-certified pathologists. 214 215 Paraffin-embedded spheroids were sectioned into 4 µm slices, deparaffinized and subjected to 216 antigen retrieval. Spheroids were blocked with 1% bovine serum albumin (BSA) in PBST for 30 217 min at room temp. Spheroids were stained with primary DCLK1 antibody (as above) overnight 218 at 4ºC. The next day, slides were washed with PBS, and spheroid sections were incubated at 219 room temperature in the dark with FITC tagged secondary antibody (Jackson Immuno) at 1:100 220 dilution for 1 hour. Slides were then rinsed with PBS and counterstained with DAPI and H&E. 221 222

Proliferation and colony formation assay 223
Cells (5000 cells/well) were plated in 96-well plates, allowed to grow for 24 hours in complete 224 DMEM media containing 10% FBS, and treated with increasing doses of respective vehicle 225 control (DMSO) or DiFiD. Cell viability was measured by enzymatic hexosaminidase assay as 226 described previously ( tools were used to analyze the docking. Total Kollman and Gasteiger charges were added to the 266 protein and the ligand prior to docking. We used Lamarckian GA to find the best conformations 267 and chose approximately 10 conformations for further analyses. The most stable compound 268 conformation was selected based on the scoring function and the lowest binding energy, and 269 visualized using Pymol (https://pymol.org/2/) (24). The ability of DiFiD to interact with and stabilize DCLK1 in cells was determined using CETSA 273 (25). Briefly, cells (8 × 10 6 ) were treated with media containing DMSO or DiFiD (5 μM) for 4 274 hours. After treatment, the cells were aliquoted into PCR tubes and exposed to a temperature 275 gradient. Subsequently, cells were lysed using three repeated freeze-thaw cycles in liquid progression, 4-nitro-quinoline-oxide (4-NQO,100 ppm in sterile drinking water ad libitum) was 283 administered for 16 weeks to C3H mice (n=20) using a previously reported protocol (26). Mice 284 were then given sterile drinking water for 3 weeks, at which time animals were sacrificed and 285 tongues excised. Immunohistochemistry was then used to assess DCLK1 expression. 286 287 Five-week-old female Foxn1/nude mice (Charles River Laboratory) were injected with 1×10 6 288 HN5 or FaDu cells in the flank. One week following implantation, mice were randomized into two 289 groups with 7 mice per group in the HN5 injected mice and 10 mice per group in the FaDu treated 290 mice. Animals were treated with either vehicle control (2.5% DMSO in water) or DiFiD (2 mg/kg 291 body weight), administered intraperitoneally daily for 15 days. Tumor growth was measured 292 every 2-3 days by a blinded observer measuring tumor diameters using vernier calipers and 293 volume was calculated (Tumor volume = longest dimension × shortest dimension 2 x 0.52) as 294 previously described (27). At the end of treatment, animals were euthanized, and the tumors 295 were collected, weighed, and processed for downstream analytical assays. 296 297 Additionally, we established a patient-derived xenograft from an anal SCC that metastasized to 298 the liver. The tumor was passaged twice through NOD SCID gamma (NSG) mice. ASCC tumors 299 from the third passage were implanted subcutaneously into flanks of 8-week-old mice. Mice were 300 randomized based on the tumor volume into two groups with 10 mice (5 female and 5 male) in 301 the control group and 14 mice (9 female and 5 male) in the DiFiD treatment group. Average 302 tumor volumes across both groups were 50 ± 8.9 mm 3 . Mice were treated daily with either vehicle 303 control (DMSO in water) or DiFiD (2 mg/kg body weight) intraperitoneally daily for 15 days. 304 Tumor growth was measured, and the volume calculated as mentioned before. Fractional tumor 305 volumes were calculated as per previously published protocols (28). 306 307

Statistical analysis 308
Data are reported as mean ± SEM. Parametric, one-tailed t-test with Welch correction was used 309 to assess significance in all experiments unless stated. Outliers were detected using Graphpad 310 software. Significant outliers were identified using the Grubb's Test with Alpha=0.05. Significant 311 outliers were removed from further statistical analysis. For in vivo studies, repeated measures 312 ANOVA test was employed to assess the level of significance in tumor volumes between 313 treatment arms. For TCGA survivorship comparison, log-rank (Mantel-Cox) test assessed 314 differences between curves. To generate a best expression cut off, patients were stratified into 315 two groups and association between survival and RPKM was examined. The RPKM value that 316 yields the maximum difference between survival of the two groups at the lowest log-rank P-value 317 determined best expression cut off. From this, a high/low expression cut off (0.37) was applied. 318 All statistical calculations were performed on GraphPad Prism software (version 6.03), with 319 significance determined by p<0.05. 320 321

DCLK1 is upregulated in Squamous Cell Cancers 323
DCLK1 is associated with pro-survival signaling in various cancers, including colorectal and 324 pancreatic cancers (8,18,19). Its expression is elevated in HNSCC tumor samples compared 325 to normal oral epithelial tissue ( Fig. 1A  immortalized esophageal epithelial line. DCLK1 was differentially expressed with the highest 345 expression observed in HN5 and FaDu cells, and lowest expression in Het-1A cells (Fig. 1G). 346 Immunofluorescence revealed DCLK1 to be expressed in HN5 spheroids ( Supplementary Fig.  347 S1C). Therefore, HN5 and FaDu were used for subsequent studies. 348 . These data suggest that DCLK1 is elevated in HNSCC compared to normal oral mucosa and 349 is associated with tumor progression. Further, anal squamous cell carcinoma (ASCC) patient 350 samples (n=17) were stained for DCLK1 expression. DCLK1 was more significantly expressed 351 in ASCC tumor compared to normal anal mucosa (Supplementary Fig.S1D and E). We observed 352 increased nuclear staining along the tumor invasive front ( Supplementary Fig. S1D). 353

DCLK1 suppression inhibits HNSCC growth 355
To evaluate the antitumor efficacy of targeting DCLK1 in HNSCC, we attenuated DCLK1 levels 356 using two shRNA constructs, sh3 and sh4. DCLK1 mRNA and protein were significantly reduced 357 Cell cycle regulator, cyclin D1 expression was significantly reduced, while pro-apoptotic, Bax 370 expression was greatly increased in DCLK1 knockdown tumors. Collectively, these data indicate 371 DCLK1 may help drive cancer progression, and DCLK1 suppression may lead to cell cycle arrest 372 and apoptosis. 373 374

DiFiD binds to DCLK1 and inhibits its activity 375
The small molecule, 3,5-bis (2,4-difluorobenzylidene)-4-piperidone (DiFiD) was previously 376 shown to suppress the growth of pancreatic cancer cells in vitro and in vivo (14). However, the 377 precise target for DiFiD remained unknown. DCLK1 is implicated in pancreatic and colorectal 378 cancer progression. Therefore, we first determined compound-protein interaction. Molecular 379 docking predicted a DiFiD interacts with DCLK1 by forming hydrogen bonds with aspartic acid 380 533, with a binding energy of -7.9 kcal/mol, as shown in ribbon and surface views (Fig. 3C). 381 Additionally, the binding kinetics of DiFiD and DCLK1 were studied via surface plasmon 382 resonance (SPR). We observed dissociation constants of K D1 = 71 nM and K D2 =902 nM (Fig.  383   3E). Taken together this data shows that DiFiD binds DCLK1 with high affinity. To confirm that 384 DCLK1 is a binding target of DiFiD in the cells, we performed cellular thermal shift assay 385 (CETSA) to assess protein stability following thermal denaturation. FaDu cells were treated with 386 5 µM DiFiD at 37°C for 4 hours. Cells were aliquoted into equal volumes, subjected to a thermal 387 gradient and DCLK1 expression was then evaluated by western blot. Thermal denaturation of 388 DCLK1 occurred at 54°C in the DMSO treated control group, which increased to 58°C in the 389 presence of DiFiD (Fig. 3F). To further validate DiFiD binding with DCLK1, we performed the 390 drug affinity responsive target stability (DARTS) assay. Briefly, FaDu cell lysates were incubated 391 with DMSO or DiFiD (5 µM) for 30 minutes followed by treatment with increasing concentration 392 of pronase. Starting with 0 mg/mL stock solution of pronase, we incubated cells with increasing 393 protein:pronase ratio. We observed that DiFiD protected DCLK1 from protease-mediated 394 degradation, as DCLK1 expression was extended to 1: 1/800 protein: pronase ratio compared 395 to 1: 1/1600 in the DMSO control arm (Fig. 3G). Altogether, these data suggest that DiFiD binds 396 to DCLK1.  Fig. S3C and S3D). This suggests that DiFiD does not stabilize DCLK1 lacking 406 the kinase domain. Therefore, we conclude that DiFiD preferentially binds to the kinase domain 407 of DCLK1, further confirming molecular docking predictions. 408 409 DCLK1 belongs to the family of kinases with homology to calmodulin kinases, but it does not 410 depend on Ca2+/calmodulin for its kinase activity (29). Since DiFiD interacts with the kinase 411 domain, we next assessed the effect of DiFiD on DCLK1 kinase activity by performing an in vitro 412 kinase assay. Here, we incubated recombinant DCLK1 with GS peptide as substrate to assess 413 the ability of DCLK1 to phosphorylate the peptide (18,19). Along with DiFiD, we also tested the 414 ability of EF24 (IKK inhibitor) and KN62, a pan-CaMKII inhibitor. We observed greater inhibition 415 of DCLK1-mediated ATP consumption following incubation with DiFiD, than with either EF24 or 416 KN-62 (Fig. 3H). To determine the specificity of DiFiD to DCLK1, we also performed studies with 417 CaMKIIα, CaMKIIβ, and CaMKIV. We observed that DiFiD did not affect the activities of these 418 CaMK proteins. These data demonstrate a higher selectivity and specificity of DiFiD for DCLK1. 419 420 Since DiFiD has favorable binding to DCLK1, inhibits its kinase activity, and is relatively 421 ineffective at reducing Het1A proliferation ( Supplementary Fig. 4B), we hypothesized that cells 422 expressing lower levels of DCLK1 would be resistant to DiFiD. Consequently, we performed 423 hexosaminidase assays with cells where DCLK1 was knocked down using specific shRNA. 424 There was an increase in their respective half maximal inhibitory concentration (IC 50 ) values 425 from 1.3 μM to greater than 11 μM compared to control cells (Fig. 3I). These data demonstrate 426 DiFiD activity is dependent upon the presence of DCLK1, and DiFiD has poor activity against 427 cells with low DCLK1 expression. These data further suggest that DCLK1 is a highly specific 428 direct target of DiFiD. 429 430

DiFiD demonstrates potent cytotoxicity in HNSCC in vitro 431
Since DiFiD inhibits DCLK1 activity, we sought to determine the effect of DiFiD on HNSCC 432 cancer cell viability. We observed that DiFiD inhibits HNSCC viability in a dose and time-433 dependent manner ( Fig. 4A and Supplementary Fig. S3A). We identified an effective dose to 434 assess the mechanism of action. The IC 50 was measured by hexosaminidase assay and 435 observed within 48 hours at concentrations of 750 nM and 1.5 µM in HN5 and FaDu cell lines, 436 respectively. In addition, we tested toxicity of DiFiD on an immortalized non-cancerous cell line, 20 Het1A (30) (Fig. 1G). We observed that the IC 50 for Het1A at 48 hours was 9 µM, a 6-12-fold 438 increase compared to FaDu and HN5 cells, respectively ( Supplementary Fig. S4B). In addition, 439 we observed that DiFiD attenuates spheroid growth in both HN5 and FaDu cell lines when 440 treated at IC 50 concentrations, suggesting that it affects anchorage independent growth ( in CDC2 expression were observed in the treatment arm ( Fig. 5B and 5C). We also performed 459 western blot analysis for pro-and cleaved-PARP protein. DiFiD treatment induced PARP 460 cleavage in both HN5 and FaDu cell lines ( Fig. 5B and 5C). This suggests that DiFiD further 461 induces apoptosis in HNSCC cancer cell lines associated with a G 2 /M arrest. This was confirmed 462 by Annexin V/PI staining that demonstrated an increased percentage of DiFiD treated cells were 463 in the apoptotic and dead fractions compared to vehicle control (Fig. 5D). To assess the role of 464 caspases in the induction of apoptosis by DiFiD, we performed caspase 3/7 assay to assess 465 effector caspase activity. DiFiD induced a significant upregulation in the activity of the effector 466 caspases (Fig. 5E, p<0.001). Taken together, these data confirmed that DiFiD mediated 467 suppression of DCLK1 activity lead to mitotic catastrophe of HNSCC. 468 469

DiFiD has antitumor effects in vivo 470
To determine the in vivo antitumor activity of DiFiD, we treated FaDu and HN5 subcutaneous 471 tumors in Foxn1 nu/nu mice. Briefly, HN5 or FaDu cells were injected subcutaneously into the 472 flanks of nude mice and were subsequently treated with DiFiD at 2 mg/kg/day for 15 days ( To elucidate the molecular mechanism whereby DiFiD exerts its antitumor effects on HNSCC, 484 we analyzed xenograft tumors using western blotting. Tumor samples were subjected to 485 electrophoresis, and subsequently, expression of Bax, Cyclin D1, and DCLK1 were determined. 486 DiFiD significantly induced the expression of Bax in HN5 (Fig. 6F and Supplementary Fig. 6B) 487 and FaDu (Fig. 6G and Supplementary Fig. 6F) tumor samples relative to vehicle control. 488 Additionally, DiFiD significantly reduced expression of Cyclin D1 (Fig. 6F and Supplementary 489 SCCs from the head and neck and anorectal regions are highly aggressive malignancies with 498 high rates of local recurrence, distant metastasis, and poor clinical outcomes, including reduced 499 survival. Current standard of care include surgery followed by chemoradiotherapy. However, 500 despite aggressive treatment, the survival rate remains low highlighting a significant unmet 501 medical need in SCC patients. In this article, we demonstrate that DCLK1 is a clinically relevant 502 target for HNSCC and ASCC. DCLK1 is well characterized as a reserve, stress induced stem 503 cell marker in pancreatic and colorectal cancers (5, 6). Using multiple platforms, we demonstrate 504 DCLK1 expression is significantly upregulated in HNSCC compared to normal oral mucosa of 505 patient tissues. Furthermore, high expression of DCLK1 correlates with poor clinical outcomes. 506 These findings agree with a recent report associating high DCLK1 levels with poor HNSCC 507 patient survival (4). 508

509
Genetic knockdown of DCLK1 has demonstrated promising findings in neuroblastoma, 510 colorectal, and pancreatic tumors (31, 32). DCLK1 knockdown triggers apoptosis and inhibits 511 proliferation, mitochondrial function, and ATP synthesis in neuroblastoma cells (32). 512 Furthermore, DCLK1 siRNA nanoparticle delivery to colorectal and pancreatic tumor xenografts 513 resulted in the significant inhibition of tumor growth with seemingly high tolerance (31). Our data 514 presented herein supports these previous findings, as we observed decreased spheroid growth, 515 colony formation, migration, and invasion following knockdown of DCLK1. As such, due to the 516 high expression of DCLK1 in HNSCC tumor tissues and cell lines, we postulated that DCLK1 517 may be a potential therapeutic target for HNSCC. 518

519
There are no known DCLK1 inhibitors in development, in fact, few compounds have been 520 reported to inhibit DCLK1, and none with high specificity. Yet, in a single study, Weygant et al.,521 in targeting leucine-rich repeat kinase 2 (LRRK2), with the small molecule inhibitor LRRK2-IN-522 1, reported inhibition of DCLK1, and subsequent attenuation of HCT116 (colon) and AsPC-1 523 (pancreatic) growth, and invasiveness (33). However, direct binding of DCLK1 by LRRK-IN-1 524 was not examined in this study and the effects are most likely due to the indirect effects of LRRK 525 inhibition on DCLK1 activity (34). Furthermore, several kinase inhibitors with anti-tumor activity, 526 such as XMD8-92 (MAPK7 inhibitor), BI-2536 (PLK1 inhibitor), and TAE-684 (ALK inhibitor), 527 demonstrate non-specific activity, with a comparable affinity towards DCLK1 as much as their 528 target kinases (35). Therefore, while inhibition of DCLK1 may play a role in the therapeutic 529 activity of these compounds, it is highly valuable to identify a specific DCLK1 inhibitor for future 530 clinical applications. 531 532 Previously, we reported that DiFiD shows antitumor activity towards pancreatic cancer cells [11]. 533 However, the direct target for DiFiD was not elucidated. We used an in vitro kinase assay to 534 demonstrate that DiFiD effectively inhibits DCLK1 kinase activity, but not related CaMK family 535 members, suggesting that the compound is a specific competitive inhibitor of DCLK1. We further 536 confirmed a direct interaction through in vitro binding assays and SPR analysis. The CETSA and 537 DARTs binding assays involved the uptake of the compound by cells before thermal or 538 enzymatic denaturation, respectively (25). We showed that DCLK1 was robustly stabilized by 539 DiFiD. Additionally, cells stably transfected with expression plasmids containing only the DCX 540 domain did not exhibit thermal stabilization as was observed with full length DCLK1 containing 541 the kinase domain. Interestingly, we found that DCX domain required higher temperatures to 542 denature compared to full length protein. This is likely due to enhanced microtubule association 543 and stabilization of the domain, as this domain in DCLK1 is critical for its microtubule binding 544 and polymerization activity (7, 36). SPR analysis, which is the gold standard for target 545 interactions, identified strong DiFiD:DCLK1 binding with low nM concentration equilibrium 546 dissociation constants K D . Taken together, these data demonstrate that DiFiD has high 547 selectivity for DCLK1 with binding sites located in the c-terminal kinase domain. In our studies, we observed significant antitumor effects in SCC preclinical models following 577 treatment with DiFiD. DCLK1 is expressed in various normal cell types, including neurons, 578 osteoblasts, and colon stem cells (48), and is involved in physiological processes, including 579 retrograde transport, neuronal migration, and neurogenesis. The data presented in this article 580 suggest that DiFiD is well tolerated at doses that demonstrated antitumor activity, as mice 581 maintained normal weight gain and ambulation. DiFiD tolerance was observed in an earlier study 582 (14), further supporting the notion that it is well tolerated at the doses used to inhibit cancer 583 growth. Representative images of colony formation assays for DCLK1 control, sh3 and sh4 FaDu clones. 804