Serine-Threonine Kinase TAO3-Mediated Trafficking of Endosomes Containing the Invadopodia Scaffold TKS5α Promotes Cancer Invasion and Tumor Growth

Much progress has been made in recent years in the development of novel cancer therapeutics. Among small molecules, kinase inhibitors have met with particular clinical success, with approvals of agents targeting mutated “driver” kinases, as well as nonmutated but essential enzymes (1). Despite these successes, both intrinsic and acquired resistance limit long-term efficacy. This has led to the development of inhibitors specifically targeting resistance mechanisms, as well as to combination therapies. In the case of mutant B-RAF–driven melanoma, combination therapy with the B-RAF inhibitor, dabrafenib, and the MEK inhibitor trametinib was recently approved (2–4). This regimen results in improved survival, although resistance emerges in most patients after approximately 1 year. There are currently no kinase inhibitor strategies for those melanomas with wild-type B-RAF. Exciting progress has also been made in immunotherapy approaches to treat cancer, with noted successes for antibodies targeting PD-1 and CTLA-4 in all subtypes of melanoma (5). Nevertheless, resistance to these agents also limits long-term efficacy. Thus, it is important to continue to identify new therapeutic approaches for eventual use in combination with existing agents. We reasoned that identifying targets in pathways regulating aspects of the cancer phenotype distinct from cell-cycle progression and immune evasion might be such an opportunity.

We decided to focus on the invasive behavior of cancer cells. Invasion is required for cancer cells to move into and out of the bloodstream (intra- and extravasation), and therefore underlies the metastasis that is responsible for most cancer deaths. These steps will have already occurred prior to diagnosis in patients for whom removal of the primary tumor is not curative, and therefore therapeutic intervention for intra- and extravasation may not be beneficial. However, there are many examples where invasive behavior has been linked to tumor growth in both primary and metastatic sites, as well as chemoresistance (6–10); intervention in these processes would be expected to have therapeutic benefit.

One prominent mechanism by which tumor cells exhibit invasive behavior is by the formation of membrane protrusions known as invadopodia (11–13). These complex structures coordinate the actin cytoskeleton with pericellular proteolysis, and metallo-, cysteine, and serine protease activities have all been described at invadopodia (14). It has long been known that the presence of invadopodia correlates with invasive behavior in a number of cancer cell types in vitro. More recently, invadopodia were observed in human cancer specimens ex vivo (15, 16) and were shown to be required for tumor cell intra- and extravasation in model systems (17, 18). Tumor expression analysis has revealed a correlation between high expression of the obligate invadopodia scaffold protein TKS5α and worse outcomes in lial-derived tumors, non–small cell lung cancer and breast cancer (6, 19, 20), as well as with increasing stage in prostate cancer and melanoma (7, 21). Most importantly, the use of three-dimensional (3D) culture systems and xenograft assays has revealed that invadopodia also promote the growth of tumor cells (6, 7, 22). Although the full mechanistic details of this phenotype remain to be established, it seems likely that the pericellular proteolytic activity associated with invadopodia both remodels the extracellular matrix (ECM) to the benefit of the tumor and processes and activates growth factors and cytokines required to create a growth-promoting microenvironment. For example, the angiogenesis-eliciting vascular endothelial growth factor (VEGF) is processed by metalloproteases into protumorigenic forms (23), and there is some evidence that loss of invadopodia results in reduced tumor-associated VEGF (6).

What might be good therapeutic targets for invadopodia inhibition? The TKS5α scaffold itself, as well as many other key invadopodia proteins, have no catalytic activity, and thus developing drugs against them, although not impossible, would be challenging (24). Other key players, such as small GTPases, have to date proven largely intractable to inhibitor strategies. And inhibitors of the matrix metalloproteases have also not met with success to date, perhaps because of redundancy among different protease classes, as well as lack of specificity and potency. Some years ago, we established a high-content screening assay to identify in an unbiased way regulators of invadopodia formation and function (25). Using a small compound library in a proof-of-principle screen, we identified several small molecules annotated as cyclin-dependent kinase (CDK) inhibitors. We deconvoluted these hits to identify and subsequently validate CDK5 as their target, and provided the mechanism by which CDK5 regulates invadopodia formation (25). Interestingly, at around the same time it was recognized that CDK5 is overexpressed and indeed frequently amplified in pancreatic cancer, which elaborates invadopodia (26). Knockdown of CDK5 had no effect on growth of pancreatic cells on tissue culture plastic, but did markedly inhibit the growth of the cells in 3D matrices as well as in vivo, in keeping with a role for CDK5 as an invadopodia regulator (27). Together, these studies support the conclusion that invadopodia inhibition is a viable strategy to reduce tumor growth, and suggest that kinases might be a valuable class of invadopodia targets. Despite the value of kinase inhibitors for cancer therapy generally, and the large target class (there are more than 500 members of the kinome; ref. 28), it is quite remarkable that most research, as well as most target identification and validation (29), and inhibitor development (30) have focused on just a few enzymes. We therefore set out to screen the entire kinome to identify invadopodia regulators.

The C8161.9 human melanoma cell line (RRID:CVCL_0196), the bladder cancer cell lines T24 (RRID:CVCL_0554) and UM-UC-3 (RRID:CVCL_1783) and the breast cancer cell line Hs578t (RRID:CVCL_0332) were purchased from ATCC. The luciferase-expressing human breast cancer cell line MDA-MB-231-Luc (RRID:CVCL_1D88) was obtained from Xenogen. For these cell lines, we relied on authentication performed by the provider. The WM793 human melanoma cell line was a gift from Dr. Gary G. Chiang (AbbVie; RRID:CVCL_8787) and authenticated by the OHSU DNA core facility using short tandem repeat analysis. The Src-transformed NIH-3T3 (Src3T3) mouse cell line was generated by the corresponding author (S.A. Courtneidge) and has been described previously (31). Melanoma, breast cancer cell lines and Src3T3 cells were grown in Dulbecco's Modified Eagle's Medium (10-013-CV, Corning) containing 10% fetal bovine serum (FBS, HyClone and Gibco). T24 was cultured by McCoy's 5a Medium Modified (ATCC) with 10% FBS. UM-UC-3 was grown in Eagle's Minimum Essential Medium (ATCC) with 10% FBS. All cell lines were routinely tested for Mycoplasma contamination and confirmed negative for Mycoplasma species before use. Cell lines were used within 3 months of initiating growth in tissue culture.

The high-content screening assay for invadopodia formation has been previously described (25). Briefly, we ran the screen twice, each time in quadruplicate, with a negative control (nontarget scrambled siRNA). We used pools of three different siRNAs for each gene. The results from the two runs had a low coefficient of variability; therefore, we did not repeat the assay a third time. The effects of siRNA knockdown on invadopodia number were determined by eye, counting at least 100 cells for each field. We considered cells as positive if they had more than two invadopodia. siRNAs that triggered invadopodia inhibition <50% (compared with control) in cells that retained viability and otherwise contained a normal actin cytoskeleton were considered hits. In order to validate the screening results, we used the three individual siRNAs (Mission siRNA; Sigma). The validation step was repeated three times, and we scored for phenotypic concordance. For each hit, we tested the effect of kinase knockdown on gelatin degradation (using the pool siRNA only). The siRNA oligo IDs are listed in Supplementary Table S1.

Cancer Outlier Profile Analysis (COPA) was performed on public gene-expression data sets curated by Oncomine (32). Gene-expression values were median-centered, setting each gene's median expression value to zero. The median absolute deviation (MAD) was calculated and scaled to 1 by dividing each gene-expression value by its MAD. The 75th, 90th, and 95th percentiles of the transformed expression values were calculated for each gene, and then genes rank-ordered by their percentile scores, providing a prioritized list of outlier profiles. A gene rank overexpression threshold of 10% was set for this analysis. For each gene identified as an outlier in at least one data set, we tabulated all data sets with outlier calls and the total number of data sets for that cancer type with expression of that gene.

Src3T3 cells were cultured on glass coverslips and prepared for imaging using the same methodology as for the invadopodia assay. Confocal images were collected using a laser-scanning confocal microscope LSM880 equipped with AiryScan (Carl Zeiss). Time-lapse imaging was performed by 200 ms per frame for 1 minute, total 300 frames. Images were transferred to Imaris (Bitplane), which is a multidimensional analysis program based on the fluorescence intensity data. For colocalization analysis, surfaces (spots) were created on the TKS5, TAO3, and RAB11 separately and colocalized signals were assessed by “ImarisColoc” tool. Detailed methodology is shown in Supplementary Fig. S5. For vesicle trafficking, surfaces (spots) were created on the TKS5α-mCherry or RAB11-DsRed to track the vesicles and extract the “displacement length (μm)” and “length (μm)” as statistics Excel files. For making movies, the time-lapse films were edited by Apple's iMovie software.

Primary melanomas were stained with TAO3 (ab150388, 1:300, Abcam, RRID:AB_150388) on a Ventana Discovery autostainer. Negative control slides were stained with Rabbit IgG. Subsequently, TAO3 IHC was scored by a board-certified pathologist (G.V. Thomas). Scores were either 0 for no staining in the tumor, 1+ for weak staining, 2+ for moderate to strong staining. Next, an immunoscore was calculated from the formula: (0 × % cells staining negative) + (1 × % cells staining weakly positive) + (2 × % cells staining moderately strongly positive), giving a range of 0–200.

All animal experiments were conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals, and were reviewed by an Institutional Animal Care and Use Committee. Subcutaneous implantation was performed with minor modifications as described previously (7). Briefly, cells were resuspended in PBS/Matrigel mixture (1:1 ratio, BD Biosciences) to a final concentration of 4 × 10⁶ cells/mL. Athymic nude mice (nude Nu/J, The Jackson Laboratory; RRID:IMSR_JAX:002019) were injected in the flank with 100 μL, and tumors were allowed to form for 22 days. For the tumor growth experiment with the SBI-581 inhibitor, tumors were allowed to form for 10 days and SBI-581 (final concentration 10 mg/kg in DMSO with 1:5 dilution of PBS) were injected every day for 10 days. Tumor growth was measured every 2–3 days using calipers. The longest (L) and shortest (S) measurements were recorded, and tumor volumes were calculated as volume = 0.5*(L × S2) and expressed as mean volume (mm²).

Extravasation efficiency assay was performed as described previously (18). GFP-expressing C8161.9 cells (RRID:CVCL_0196) with pLKO.1-scrambled, TAO3 knockdown (KD), or rescued (SR) were injected intravenously in tail veins of 8-week-old SCID/Beige female mice (C.B-17/IcrHsd-Prkdc scid Lyst bg-J, ENVIGO). Twenty-four hours after the injection of 2.5 × 105 of cells, mice were euthanized, and tumor cells in lung blood vessels were removed by perfusion, then lungs were harvested. Frozen serial sections of all lungs (50 μm thickness) were taken every five sections. For the efficiency analysis, slides were counterstained with Hoechst for 15 minutes before mounting. All sections were scanned by Zeiss AxiaScan system with ZEN Blue software. GFP-positive cells were counted from all scanned lung sections and then calculated as “Number of GFP⁺ cancer cells in lung.” For inhibitor experiment, DMSO control or SBI-581 (30 mg/kg) was injected intraperitoneally, and then 30 minutes later GFP-labeled C8161.9 cells were injected through tail vein.

C8161.9 cells with endogenous TAO3 replacement by shRNA-TAO3 and wild-type TAO3 (shRNA-resistant, SR) or kinase-dead TAO3 (shRNA-resistant, KR) on five 150-mm culture dishes were grown to between 70% and 80% confluence. Cells were washed with 1X cold-PBS before lysis to remove any media containing protein contaminants. The cells were lysed by 10 mL of Urea Lysis Buffer (at room temperature, 50 mmol/L HEPES pH 8.5, 8 M urea, 1 mmol/L NaF, 1 mmol/L sodium orthovanadate, 10 mmol/L sodium β-pyrophosphate, 1 mmol/L β-glycerophosphate buffer) and disrupted by sonication on ice. Cell lysates were stored at −80°C prior to digestion and enrichment. Protein concentrations were then determined using the Pierce BCA Protein Assay Kit (Thermo Scientific). Approximately 12 mg of protein from each lysate was used. Protein was reduced with dithiothreitol, alkylated with iodoacetamide, urea diluted to 2 M, then trypsin (TPCK treated, Worthington) was added at a 25:1 protein:trypsin ratio. The samples were incubated overnight at 37°C before being quenched with TFA at a final concentration of 1%. The peptides were then solid phase extracted using Waters Sep tC18 cartridges (Waters Corporation). Peptide concentrations were then determined using the Pierce Quantitative Colorimetric Peptide Assay (Thermo Scientific) and dried by vacuum concentration. The phosphopeptides were enriched using previously published methods (33, 34) using Titanosphere TiO₂ 5 μm particles (GL Biosciences). The enriched phosphopeptides were purified by solid phase extraction using UltraMicroSpin columns (The Nest Group, Inc.) and dried down in preparation for TMT labeling. The enriched phosphopeptides from the SR and KR cells were then labeled with TMT reagents as recommended by the manufacturer (Thermo Scientific), mixed, and analyzed by two-dimensional liquid chromatography/mass spectrometry using an Orbitrap Fusion Tribrid mass spectrometer (Thermo Scientific) as previously described (35). The full proteomic data set can be found in the PRIDE database (https://www.ebi.ac.uk/pride).

The protein (TAO3 kinase domain (1–319) fused with GST) was expressed in the baculovirus intracellular expression system. Linear DNA was used for the transfection to make virus. Protein was expressed using 8 L of SF9 cells. After 72 hours of infection, cells were collected by centrifugation, redissolved in a buffer containing 50 mmol/L NaCl, 20 mmol/L TRIS pH 8, 1 mmol/L EDTA, 2 mmol/L of β-mercaptoethanol and including a standard protease inhibitor cocktail. Cells were lysed in this buffer with a glass dounce homogenizer on ice. The supernatant was separated from pellet by centrifugation for 1 hour at 20K rpm (SS34 rotor) and then incubated with glutathione–agarose bead for 2 HRS at 4°C. Beads were washed with 50 mmol/L NaCl, 20 mmol/L TRIS pH 8, 1 mmol/L EDTA, 5 mmol/L β-ME, and then the TAO3 kinase domain was cleaved off the beads by incubation with 3C protease. Cleaved released kinase domain was separated from contaminating GST by immediately running the sample in this buffer onto a mono-Q on the AKTA FPLC, eluting protein with a gradient of the same buffer containing 50 to 1 M NaCl. The peak containing the TAO3 kinase domain was collected and then run on a Superdex S200 gel filtration column prequilibrated with the final buffer 100 mmol/L NaCl, 25 mmol/L Tris-Cl pH 8.0, 2 mmol/L β-ME, and 0.3 mmol/L EDTA.

A time-resolved fluorescence resonance energy transfer (TR-FRET) assay was developed to monitor the inhibition of TAO3 kinase activity. We used the kinase domain of TAO3, and an HTRF-KinEASE (Cisbio) assay along with the S3 substrate and a Eu³⁺ Cryptate-conjugated anti-pSer/Thr antibody. This assay was miniaturized to 1,536-well format with a final volume of 2 μL + 2 μL detection reagent. Compounds (including a known kinase inhibitor collection (800 compounds) and an SBP Institute-selected kinase inhibitor scaffold library (4,800 compounds)) were tested at 10 μmol/L. Assay performance was very robust with an average Z' of 0.89, and with no individual plate with a Z' of less than 0.70. Hit threshold at 30% inhibition, and because the hit rate was low (not unexpected based on the lack of reported TAO3 inhibitors in the scientific and patent literature), this resulted in 82 hits (0.12%). To further prioritize confirmed hits, apparent IC₅₀ values were determined. Twelve selected compounds with potencies <2 μmol/L were validated using Protein Thermal Shift technology, which demonstrated direct binding of the compounds to TAO3. These compounds were also tested in dose response against an unrelated kinase (MEKK3) using the same assay format, with a minimum selectivity threshold of 5×. From these efforts, a series of oxindoles were identified as the most promising hits. Preliminary chemistry led to SBI-581 (IC₅₀ = 42 nmol/L against TAO3, IC₅₀ = 237 nmol/L against MEKK3).

SBI-581 was dosed at 1 mg/kg i.v. (as a 1 mg/mL solution in 75%PEG300/25% water), 10 mg/kg po (as a 3 mg/mL solution in Pharmatek #6), and 10 mg/kg i.p. (as a 3 mg/mL solution in 5% DMSO/5% Tween 80/90% water) to fasted male C57BL/6 mice (3 mice per cohort, RRID:IMSR_JAX:000664). Plasma samples were taken at 0.083, 0.25, 0.5, 1, 2, 4, 6, and 24 hours after dose (i.v.) and 0.25, 0.5, 1, 2, 4, 6, and 24 hours after dose (p.o. and i.p.) and measured for the level of SBI-581 via LC/MS/MS analysis.

The numbers of samples (technical replicate) and times of experiments (biological replicate) are indicated in each figure or figure legend. All data points and error bars represent the means and SEMs. GraphPad Prism software was used to make graph and calculate significance by Student t test. P value of <0.05 was considered to be statistically significant and indicated in figure legends.

Additional information on reagents and assays can be found in Supplementary Materials and Methods.

We previously described a high-content image-based screening assay designed to identify regulators of invadopodia formation (25). Our strategy was to use Src3T3 cells (a mouse model of fibrosarcoma that elaborates abundant invadopodia) for the initial screen. Top hits were subsequently validated in human cancer cell lines. This study led to the identification and validation of CDK5. We used the same assay here. A library of pooled siRNAs (three for each murine kinase; ref. 36), targeting all protein kinases as well as select lipid and metabolic kinases, was kindly provided by Dr. Pam Carroll of Merck. Src3T3 cells, in 1,536-well format, were transfected with the pools, incubated for 48 hours, then stained with phalloidin and DAPI to label the F-actin in invadopodia and nuclear DNA, respectively. Following automatic well focusing and image capture, each well was analyzed by eye for cell number and viability. For those wells where no cytotoxicity or apoptosis was observed, invadopodia were then evaluated. From this, 12 kinases were prioritized. One of these was the previously validated CDK5, which we did not explore further here. For the others, we next tested each siRNA separately, requiring at least two of the three to inhibit to proceed. One kinase was deemed off-target by this criterion, leaving 10 in our list (Fig. 1A). Representative images for these are shown in Fig. 1B, and the results are summarized in Table 1. Next, we performed a functional assay (gelatin degradation). Knockdown of three of these kinases, while inhibiting invadopodia formation, did not affect matrix degradation (Fig. 1C; Table 1), and were not pursued further.

Our top hits were thus narrowed to CAMK1, HK2 (a metabolic enzyme), MAP2K1, PAK6, RAF1, TAO3, and TGFβR2. Of these, we chose to focus on those kinases we considered understudied, that is, with no or few reports to date on involvement in cancer progression and/or invasion. This narrowed our list down to CAMK1, PAK6, and TAO3, all serine-threonine kinases.

The next step was to determine the expression of these three kinases in cancer tissue and cell lines. We first performed COPA (37) on public gene-expression data sets curated by Oncomine (32). Genes scoring in the top 10% of COPA scores at any of three percentile cutoffs (75th, 90th, and 95th) were deemed outliers in their respective data sets. At the 10% gene rank, TAOK3 was an outlier in 28 studies in 17 cancer types, PAK6 was an outlier in 31 studies in 14 cancer types and CAMK1 was an outlier in 44 studies in 16 cancer types. Some of these tumor types (e.g., melanoma, breast, bladder) have been associated with invadopodia formation in several studies, whereas others (particularly leukemia and lymphoma) have not been well studied. Because it was important that we be able to assess the role of these kinases in human cancer cells in vitro, we chose to focus on melanoma, breast, and bladder cancers (Supplementary Tables S2–S4).

We first determined the expression of the three kinases in representative melanoma, bladder, and breast cancer cell lines by qPCR, and for TAO3 and CAMK1 also by immunoblotting (Fig. 1D and E). All kinases were expressed in the melanoma cell lines, so we chose to pursue them further in this cancer type. We first used transient RNA interference to reduce expression of each in C8161.9 cells (Supplementary Fig. S1A and S1B), and evaluated both invadopodia formation and function (Fig. 1F and G). The invadopodia formation assay was also performed in another human melanoma cell line, WM793, which confirmed reduction in invadopodia number (Supplementary Fig. S1C). In both assays, TAO3 knockdown had a more robust inhibitory effect than knockdown of the other kinases. In our screening procedure, we excluded kinases that appeared to reduce cell viability. By using siRNA to knockdown each member of the TAO family in C8161.9 cells, and evaluating the cells 3 days later, we found that cell number was compromised by loss of TAO1 or TAO2, but loss of TAO3 had no effect (Supplementary Fig. S1D), confirming our screening data.

Finally, we determined the expression of TAO3 at the protein level in clinical specimens of melanoma. An automated IHC staining protocol was developed and then used to stain 20 samples of primary melanoma. The slides were scored (using a 0–2 scale, where 0 = negative; 1 = weak; 2 = mod-strong), and intensity was evaluated across each tumor to derive an immunoscore with a range of 0–200. Nineteen of 20 evaluable samples showed staining in some portion of the tumor, with 7 with scores of 101–150 and 8 with scores greater than 151, suggesting increased expression in melanomas. In some cases, particularly strong staining could be seen at the tumor border, and in disseminating melanoma cells (Supplementary Fig. S1E). Together, these analyses led us to nominate TAO3 as our lead kinase.

The gene-expression analyses we described earlier suggested that TAO3 was also highly differentially expressed in breast and bladder carcinomas, and was present in cell lines derived from these tumors (Fig. 1D and E; Supplementary Table S2). To determine the generality of our findings, we therefore tested whether TAO3 was required for invadopodia formation and function in these cells. Both invadopodia formation and gelatin degradation were inhibited (Supplementary Fig. S1F and S1G—note that T24 could not be evaluated for gelatin degradation because of matrix tears caused by pulling). In addition, we determined that the TAO3 siRNA did not affect the expression of TAO1 or TAO2 (Supplementary Fig. S1H). Together, these data define TAO3 as a regulator of invadopodia formation and invasion, and further reveal its importance in multiple tumor cell types.

To further validate TAO3 in melanoma, we generated a lentivirus expressing shRNA specific for TAO3, as well as nontargetable expression constructs for wild-type and kinase-dead (via mutation of lysine 53 in the ATP binding site of the catalytic domain) TAO3. Quantitative PCR and immunoblotting demonstrated the specificity of the knockdown and successful reexpression of the constructs (Supplementary Fig. S2A and S2B). Furthermore, using a similar knockdown and reexpression strategy, we found that the catalytic activity of TAO1 was not required to rescue the cytotoxic effects seen upon TAO1 knockdown (Supplementary Fig. S2C), which may inform specificity requirements for TAO3 inhibitors (see later). For the TAO3 experiments, analysis of invadopodia formation (Fig. 2A) and function (Fig. 2B) as well as invasion through matrigel (Fig. 2C) in C8161.9 melanoma cells revealed that TAO3 knockdown has a significant inhibitory effect in each case, which could be rescued by reexpression of wild-type but not kinase-inactive TAO3. We have previously shown that reducing the expression of the obligate invadopodia scaffold protein TKS5 has an inhibitory effect on growth in 3D collagen matrices, as well as in vivo (6, 7). We next determined the effect of TAO3 knockdown in these same assays. When the cells were cultured on top of a thin layer of type I collagen (collagen-I), no growth differences were seen. However, TAO3 knockdown cells showed a profound growth defect in 3D collagen-I as compared with control cells. Growth was rescued by reexpression of wild-type, but not kinase-inactive, TAO3 (Fig. 2D). We also observed similar growth differences in a tumor spheroid assay (Fig. 2E), where spheroid size is directly correlated with cell number (38). Similar effects on invadopodia formation and 3D growth were seen in a second melanoma cell line, WM793 (Supplementary Fig. S2D and S2E).

We next evaluated the effect of TAO3 knockdown in vivo. For this, we used two distinct assays. First, it has been reported that extravasation of tumor cells into the lungs requires TKS5-dependent invadopodia formation (18). To test if this was also the case for TAO3, we introduced GFP into the control, knockdown, and rescue cells to mark them, and then injected them into the tail veins of immunocompromised mice. One day later, mice were sacrificed, lungs were removed and sectioned, and then fluorescence microscopy was used to enumerate tumor cells. TAO3 knockdown markedly inhibited extravasation, which was restored by reexpression of the wild-type protein (Fig. 2F). Next, the control and TAO3 knockdown cells, as well as knockdown cells reexpressing wild-type or kinase-inactive TAO3, were injected subcutaneously into immunocompromised mice, and tumor growth was evaluated over time. TAO3 knockdown caused a profound inhibition of tumor growth, which was rescued by the wild-type but not kinase-inactive TAO3 (Fig. 2G). These data are consistent with the growth-inhibitory effects we observed in 3D cultures.

Our rescue studies had shown that the kinase activity of TAO3 was required for its role in invadopodia formation in vitro and tumor progression in vivo. To test the therapeutic potential of TAO3, we initiated a high-throughput screening campaign to identify small-molecule inhibitors of the TAO3 kinase domain. Briefly, we screened a number of compound libraries including a set of 800 known kinase inhibitors, as well as 4,800 compounds with similar structural features to known kinase inhibitors. We identified a number of hits, of which, a series of oxindoles were the most promising, based on potency against the TAO3 kinase domain and moderate selectivity against a broad panel of kinases. A preliminary round of chemistry around the hits identified SBI-581 as a proof-of-concept compound (Supplementary Fig. S3A). It displays good potency against TAO3 (IC₅₀ = 42 nmol/L) and moderate selectivity against a broad panel of kinases (Supplementary Table S5). In addition, we measured the pharmacokinetics (PK) of SBI-581 in mice. Although oral bioavailability was poor (%F < 5), SBI-581 displayed reasonable PK after i.p. injection (t_1/2 = 1.5 hours; AUC = 1,202 hours × ng/mL; C_max = ∼2 μmol/L after a 10 mg/kg dose).

We first used SBI-581 in our short-term assays of invadopodia formation (Fig. 3A) and gelatin degradation (Fig. 3B). We observed a dose-dependent inhibition in both cases, with an EC₅₀ of <50 nmol/L, a dose that had no effect on cell viability. We also evaluated spheroid growth and observed an inhibitor-dependent decrease after 2 days in 3D culture (Fig. 3C). Then we determined the effect of SBI-581 in vivo, using both the extravasation and tumor growth assays described earlier. For the extravasation assay, mice were injected intraperitoneally with 30 mg/kg SBI-581 (a dose chosen to achieve a projected plasma concentration of >6 μmol/L, a concentration well in excess of the IC₅₀ for TAO3), and then 30 minutes later GFP-labeled melanoma cells were introduced via the tail vein. Extravasated tumor cells in the lung were quantified one day later as described earlier. SBI-581 pretreatment significantly inhibited extravasation, compared with the DMSO control (Fig. 3D). For the tumor growth assay, C8161.9 cells were implanted subcutaneously, and 10 days later, when tumors had reached approximately 80 mm³, daily intraperitoneal dosing at 10 mg/kg was begun. We noted a profound inhibition of tumor growth (Fig. 3E) with no significant effect on either body weight (Supplementary Fig. S3B) or cell viability, even after several days in culture (Supplementary Fig. S3C). Together, these data support the validation of TAO3 as an invadopodia, invasion, and growth inhibitor.

It is very important to determine how TAO3 regulates invadopodia formation, both to support its future clinical development and to identify biomarkers to track its activity. As a first step, we determined its subcellular localization. In Src3T3 cells (which have rosettes of invadopodia), we noticed that much of the TAO3 signal was not at invadopodia, but rather in a perinuclear location reminiscent of endosomes (Fig. 4A). Indeed, GFP-tagged TAO3 expressed in Src3T3 cells colocalized with RAB11, a marker of recycling endosomes (Supplementary Fig. S4A; refs. 39, 40). Endosomes are transported along a microtubule network (41), and costaining experiments revealed a network of microtubules feeding into the invadopodia (Fig. 4A) consistent with other studies (42); TAO3 was present on these microtubules. We used high-resolution microscopy to confirm that TAO3 and RAB11 colocalized on cytoplasmic microtubules (Fig. 4B). There are no reports in the literature on the involvement of RAB11 in invadopodia formation and function. Therefore, we next used RNA interference to reduce its expression, which resulted in significant reduction in invadopodia number and gelatin degradation (Fig. 4C; Supplementary Fig. S4B).

Our next task was to determine how RAB11 and TAO3 might affect invadopodia formation. We thought first of the invadopodia scaffold protein TKS5α (24, 43), because it has an amino-terminal PX domain with specificity for PI3P and PI3,4P₂ (44), and PI3P in particular is highly enriched in endosomes (45, 46). Indeed, we have recently found, using high-resolution microscopy, that TKS5α is present both at invadopodia and on microtubules (an example shown in Supplementary Fig. S4C; ref. 38). Here we show that on microtubules, TKS5α colocalized with both TAO3 and RAB11 (Fig. 4D and E). Furthermore, inhibition of invadopodia formation by treatment of the cells either with the Src family kinase inhibitor SU11333, or with the TAO3 inhibitor SBI-581, promoted TKS5α accumulation at RAB11-positive vesicles (Supplementary Fig. S4D). The methodology used is described in Supplementary Fig. S5A–S5G.

We next determined the effect of the TAO3 inhibitor on endosomal trafficking of TKS5α. First, cells were engineered to express mCherry-labeled TKS5α and YFP-labeled tubulin, and trafficking was confirmed by time-lapse confocal microscopy (Fig. 5A; Supplementary Movie 1). Then, either vehicle control (DMSO) or 100 nmol/L SBI-581 was added to the cells, and both time trajectory and vesicle displacement length were evaluated (Fig. 5B and C; Supplementary Movie 2). The TAO3 inhibitor had a rapid and profound effect on both properties, reducing the population of motile TKS5α-positive vesicles from 24% to 6% of total. To determine if SBI-581 affected all RAB11-positive vesicles, the experiment was repeated using cells expressing DsRed-tagged RAB11. In this case, approximately the same fraction of all vesicles was motile (23%), and more than half of these retained motile characteristics after inhibitor treatment (Fig. 5C; Supplementary Movie S3). This effect appeared specific to TAO3 inhibition, because inhibiting Src family kinases with SU11333 was not inhibitory to vesicle motility (Supplementary Fig. S6A; Supplementary Movie S4). Finally, SBI-581 shows reasonable selectivity at the very low concentrations we used in this assay (Supplementary Table S5), but we did note that one such coinhibited kinase was ROCK2, which has noted roles in control of the actin cytoskeleton (47), and is possibly involved in invadopodia activity in response to matrix (48). To determine if ROCK2 played any role in invadopodia formation and endocytic trafficking of TKS5α, we used the ROCK inhibitor Fasudil. ROCK inhibition affected neither the endocytic trafficking of TKS5α-positive vesicles (Supplementary Fig. S6B) nor invadopodia formation (Supplementary Fig. S6C).

To determine how TAO3 might affect endocytic trafficking of TKS5α, we next initiated a phosphoproteomics screen, using mass spectrometry to identify and compare phosphopeptides in C8161.9 cells in which the endogenous TAO3 was replaced with either wild-type or kinase-inactive TAO3. Supplementary Table S6 shows the nine candidates whose phosphorylation was significantly increased in the presence of wild-type TAO3 compared with kinase-dead. Of these, we were particularly interested in cytoplasmic dynein 1 light intermediate chain 2 (LIC2), because dynein complexes act as motors to oppose the actions of kinesin motors to traffic vesicles backward and forward along microtubules (49, 50). In particular, light intermediate chains can recruit the dynein complex to endosomes and lysosomes (51) and also contact adaptors and provide a link to cargo (52). In keeping with a possible role of LIC2 in TAO3 function, we observed that these two proteins frequently colocalize in the cytoplasm (50.6% colocalization; n = 239 TAO3⁺ signals in n = 7 cells), but not at invadopodia (Fig. 6A). We next used knockdown and reconstitution experiments to evaluate the role of LIC2 in invadopodia formation, using both C8161.9 (Fig. 6B) and WM793 cells (Supplementary Fig. S7A and S7B). Knockdown of LIC2 resulted in an increase in invadopodia formation, which was rescued by reintroduction of the wild-type protein. In contrast, introduction of a LIC2 molecule mutated at the identified phosphorylation site (S202A) markedly inhibited invadopodia formation, whereas introduction of a “phosphomimetic” form (S202D) promoted invadopodia form to the same extent as LIC2 knockdown. To test whether SBI-581 exerted its effects through preventing TAO3 phosphorylation of LIC2 on S202, we compared the effect of DMSO and SBI-581 on these same cells (Fig. 6C; Supplementary Fig. S7C). In cells expressing the wild-type protein, SBI-581 inhibited invadopodia formation as expected. However, the inhibitor was without effect in cells lacking LIC2 or expressing the S202A mutant. A small, but significant inhibitory effect was observed in cells expressing S202D LIC2, which may reflect a role for other TAO3 phosphorylation sites on LIC2. Together, these data are consistent with LIC2 acting in an inhibitory fashion in invadopodia formation and suggest that this inhibition is relieved by phosphorylation of LIC2 on serine 202 by TAO3.

Here we used a high-content screening assay to identify kinases required for invadopodia formation, a process that promotes tumor intravasation and extravasation, as well as tumor growth at both primary and metastatic sites. We identified several kinases already implicated in invasive behavior, but we ultimately chose to focus on an “understudied” kinase, TAO3, which we found from public databases to be robustly expressed in breast cancer, bladder cancer, and melanoma, each highly invasive tumors known to elaborate invadopodia. We validated a role for TAO3 in invadopodia formation and function in vitro, as well as in extravasation and tumor growth of melanoma in vivo. The phenotypes we observe are in keeping with our invadopodia studies using the obligate scaffold protein Tks5α, which we have previously shown to be involved in invasion and tumor growth (6, 7, 22), through mechanisms including protease localization and type I collagen signaling (38). We note that loss of TAO3 expression was not associated with diminished cell viability in our studies, whereas loss of either TAO1 or TAO2 led to rapid cell death in a kinase-independent fashion. It is important to note that while the N-terminal kinase domains of the three proteins are highly homologous, their C-terminal ∼600–800 amino acid regions are far more distinct from one another. This raises the possibility that the C-terminal regions but not the kinase activity of TAO1 and TAO2 are essential for cell viability. Lastly, in a previous siRNA study using Hela cells, TAOK3 was identified as a survival gene (53); it is possible that loss of TAO1 or TAO2 (perhaps by RNAi cross-reaction) underlay the observed phenotype in that case. In Drosophila there is a single TAO gene that is involved in both the JNK pathway and actin-microtubule dynamics (54), whereas in more complex organisms, these functions appear to be mediated by different members of the TAO family.

In the course of our experiments, we found that individual knockdown of the 10 kinases from our primary screen showed that they were necessary for invadopodia formation, but three of them (AKT3, PKCθ, and PYK2) were not required for ECM degradation. To our knowledge, this is the first description of the separation of invasive function from invadopodia formation, and in the future, it will be very interesting to determine the mechanism(s) by which these kinases are involved in invasive behavior.

In order to explore the potential of TAO3 as a drug target, we identified SBI-581 via high-throughput screening of a kinase inhibitor library and preliminary chemistry. SBI-581 shows high potency against TAO3 (IC₅₀ = 42 nmol/L), moderate selectivity (>5–10× against the majority of a broad panel of kinases), and reasonable pharmacokinetics (PK) in mice using i.p. injection. It therefore represents a good starting point for future optimization. We note, however, that when tested in cells or in vitro, SBI-581 inhibited TAO3 function without significant cellular toxicity. Although this may suggest that some degree of TAO family specificity has already been achieved with SBI-581, it is more likely that the kinase activities of TAO1 and TAO2 are not required for their survival functions, as suggested by the rescue data in Supplementary Fig. S2C.

We made the unexpected discovery that TAO3 localizes to RAB11⁺ endosomes situated on microtubules, as well as to invadopodia. Indeed, we also found that RAB11 was required for invadopodia formation, identifying a new link between recycling endocytosis and invasive behavior. Importantly, TAO3 inhibition reduced the movement of recycling endosomes, particularly of those vesicles containing TKS5α, thereby inhibiting invadopodia formation. The regulatory process of endocytosis involves the packaging of selected plasma membrane proteins, and subsequent sorting either for eventual destruction in the lysosome or for recycling back to the membrane (55), with GTPases of the Rab family regulating each discrete step (56). Receptors, adhesion proteins, and ECM proteins are all targets for endocytosis, and in turn, signaling from these entities can regulate endocytosis. A link between endocytosis and cancer has long been recognized, whereby, for example, oncogenes can promote recycling of key receptors, as well as decrease surface expression of junctional proteins, to promote cell motility (57–59). Our data expand these properties to include invasive behavior.

We have begun to identify key TAO3 substrates and validated the dynein protein LIC2 as a target of TAO3 and a negative regulator of invadopodia formation. This suggests a model in which phosphorylation by TAO3 inhibits retrograde vesicular transport, with the net effect of promoting the anterograde transport of TKS5α-containing vesicles to the plasma membrane where invadopodia formation takes place. The mechanism by which TKS5α is loaded onto and off recycling endosomes is obviously of interest, as is the interaction of TAO3 with endosomes. To this latter point, we note that TAO3, kinesin, and dynein complexes all contain coiled-coil domains, which are known mediators of both intra- and intermolecular interactions (60). In a second, unpublished phosphoproteomic analysis, we identified additional TAO3 phosphorylation sites in LIC2, as well as in components of the kinesin complex. In the future, it will be important to define all of the key TAO3 substrates responsible for the phenotype we observe. Nevertheless, the mechanistic underpinnings of TAO3 loss and the potency of the tumor growth and extravasation inhibition we have observed here, coupled with the lack of detrimental effects of TAO3 loss or inhibition in normal cell types, and the high degree of overexpression of TAO3 in several cancer types, support the preclinical development and eventual clinical testing of selective TAO3 inhibitors.

No disclosures were reported.

S. Iizuka: Conceptualization, data curation, validation, investigation, visualization, methodology, writing–review and editing. M. Quintavalle: Data curation, formal analysis, validation, and investigation. J.C. Navarro: Investigation. K.P. Gribbin: Investigation. R.J. Ardecky: Investigation. M.M. Abelman: Investigation. C.-T. Ma: Investigation. E. Sergienko: Investigation. F.Y. Zeng: Investigation. I. Pass: Investigation. G.V. Thomas: Data curation, formal analysis, validation, and investigation. S.K. McWeeney: Data curation, formal analysis, and investigation. C.A. Hassig: Investigation. A.B. Pinkerton: Data curation, formal analysis, supervision, investigation, and writing–review and editing. S.A. Courtneidge: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing.

The authors thank Devon Kelly for assistance in identification and acquisition of human tumor samples from the OHSU Biolibrary, the OHSU Histopathology and Proteomics Shared Resources for their support, Susanne Heynen-Genel (SBP) for her support in the establishment of the high content screening assay, and Pam Carroll for spearheading the generation of the mouse kinome siRNA library. We acknowledge the expert technical assistance by Stefanie Kaech Petrie and Crystal Chaw in the Advanced Multiscale Microscopy Shared Resource (NIH P30 CA069533), and Larry David, Jennifer Cunliffe, John Klimek, and Phillip Wilmarth in the Proteomics Shared Resource at the OHSU Knight Cancer Institute. They also thank the personnel and support from the Cancer Center Structural Biology Core at the Sanford Burnham Prebys Medical Discovery Institute (SBP) and at the Lentivirus Production Cores at SBP and OHSU. This research in the Courtneidge lab was/is supported by R01 CA154002 (S. Iizuka, M. Quintavalle, and S. Courtneidge) and R01 CA217625 (S. Iizuka, J. Navarro, K. Gribbin, and S. Courtneidge) from the NCI, NIH, as well as by funds from the Knight Cancer Institute. Research and Shared Resource involvement was supported by NIH grants R01 CA169172 (G. Thomas), P30 CA069533 (G. Thomas), and P30 CA069533 13S5 (G. Thomas) through the Oregon Health and Science University–Knight Cancer Institute. Proteomic analysis in the OHSU Proteomics Shared Resource was supported by NIH grants P30 CA069533, P30 EY10572, S10 OD012246, and the OHSU School of Medicine Faculty Innovation Fund. Research at Sanford Burnham Prebys Medical Discovery Institute was supported by R01 CA217625 (M. Abelman, C.-T. Ma, E. Sergienko, F.-Y. Zeng, I. Pass, C. Hassig, and A. Pinkerton).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.