Cell Cycle–Dependent Tumor Engraftment and Migration Are Enabled by Aurora-A Free

Proliferation and migration are hallmarks of carcinoma cells (1). These critical processes are often perceived as independent, but molecular pathways that control microtubule nucleation at centrosomes may interconnect them. The nucleation of microtubules at centrosomes (2), and not at cell–cell contacts (3), is needed to prepare polarized epithelial cells for cell division. Similarly, the scattering of epithelial cells relies upon the nucleation of microtubules at centrosomes (4). Although epithelial-to-mesenchymal transition (EMT) may be dispensable for metastasis in animal models (5, 6), motile cancer cell populations that have undergone EMT gain gene expression profiles that are shared with proliferative, tumor-initiating populations (7, 8). The examination of single metastatic cells derived from patient-derived xenografts of triple-negative breast tumors also indicates convergence between the expression of EMT-related gene products and proliferation. Metastatic cells isolated from low metastatic burden tissues express gene profiles similar to normal basal cells (9), analogous to rare invasive “leader” cells at the periphery of primary tumors (10). In these xenograft models, metastatic progression requires a switch from dormancy to cell cycle, and high metastatic burden associates with increased proliferation (9). Thus, carcinoma cell division, EMT, and metastasis may be interconnected by a requirement for the acquisition of microtubule nucleation capacity at the centrosome. Indeed, increased microtubule nucleation at supernumerary centrosomes is sufficient to trigger cell invasion in nontransformed human mammary epithelial cells (11).

Aurora kinase A (Aurora-A) increases the microtubule nucleation capacity at centrosomes prior to and during mitosis (12). During mitosis, Aurora-A is optimally active when in a complex with targeting protein for Xklp2 (TPX2; ref. 13). A gradient of Ran-GTP near mitotic chromosomes releases TPX2 from importin-α, a nuclear import receptor, and enables the formation of an Aurora-A–TPX2 heterodimer (14, 15). The mitotic localization of TPX2, and in turn its access to the kinase, is promoted by hyaluronan-mediated motility receptor (HMMR; refs. 16, 17), a nonmotor adaptor protein (18). However, less is known about Aurora-A activation and function within nonmitotic cells.

During interphase, Aurora-A and HMMR localize to centrosomes and microtubules, while TPX2 is largely nuclear (19); despite their occupation of distinct subcellular locations, Aurora-A activity at centrosomes, as determined by a fluorescence resonance energy transfer (FRET)–based reporter, is dampened in TPX2-silenced cells (20). However, it remains largely unexplored whether molecular pathways that control microtubule organization during mitosis also regulate these processes in nonmitotic cells and, if so, whether the augmented microtubule nucleation at centrosomes that occurs as a cell progresses through the cell cycle enhances that cell's migratory capacity.

Here, we find that cell-cycle progression dramatically impacts the engraftment kinetics of breast cancer cells. We use multiparameter imaging of wound closure assays to track cell-cycle progression, cell migration, and associated phenotypes in epithelial cells or carcinoma cells expressing a fluorescence ubiquitin cell-cycle indicator (FUCCI). We find cell migration occurs with an elevated velocity and directionality during S–G₂-phase, and cells in this phase possess front-polarized centrosomes with augmented microtubule nucleation capacity. Aurora-A and HMMR promote wound closure in G₂-phase cells and, in these cells, the silencing of HMMR dampens Aurora-A activity and impedes the nuclear transport of TPX2. Mechanistically, TPX2 nuclear transport relies upon HMMR-T703, which releases TPX2 from a complex with importin-α at the nuclear envelope. Finally, our analysis of 3,922 clinically annotated mammary carcinoma tissues finds phosphorylated HMMR-T703 (pHMMR) abundance to be a significant predictor of breast cancer–specific survival (BCSS) and relapse-free survival (RFS) in patients with estrogen receptor (ER)–negative breast cancer. We propose that epithelial cell migration and cell-cycle progression share common regulation of the microtubule cytoskeleton through the Aurora-A–TPX2–HMMR axis, which could represent an effective therapeutic target in ER-negative breast cancer.

All mice were maintained in the specific pathogen-free animal facility at BC Children's Hospital on a 12-hour light cycle, 20°C ± 2°C, with 50% ± 5% relative humidity, and with food and water ad libitum. All procedures involving animals were in accordance with the Canadian Council on Animal Care and UBC Animal Care Committee (protocol no. A15-0187).

A total of 10,000 4T1-luc2 Bioware Ultra cells (PerkinElmer, 124087) single cells suspended in 0.2 mL PBS were injected through the tail veins of BALB/c or nonobese diabetic (NOD)-severe combined immunodeficient (scid) IL2 receptor gamma chain gene null (gamma) mice (NSG mice). Successful injection was confirmed by bioluminescent detection of cells in lungs 2 to 4 hours after injection. Recipient mice were subsequently imaged at indicated time points after engraftment. All imaging was performed on the Spectral Instruments Imaging Ami-X platform 5 minutes after intraperitoneal injection of 1 mg d-luciferin potassium salt (dissolved in 0.1 mL PBS). Images were analyzed using the Amiview software program version 1.7.06 (Spectral Instruments Imaging).

4T1-luc2 Bioware Ultra cells (PerkinElmer, 124087) were grown in RPMI media supplemented with 10% FBS and penicillin/streptomycin. For synchronization in G₁-phase, cells were grown in media without serum supplementation for 48 hours. Normal murine mammary gland cells expressing FUCCI (nMuMG-FUCCI; Riken Institute) were grown in DMEM (high glucose) supplemented with 10% FBS, penicillin/streptomycin, and 10 μg/mL insulin. MCF-10A cells were grown as described previously (19). HeLa (ATCC) and HeLa-FUCCI cells (Riken Institute) were grown in DMEM (high glucose) supplemented with 10% FBS and penicillin/streptomycin. Synchronization of cells in G₂-phase (2 mmol/L thymidine block and release for 6 hours) or G₂–M-phases (100 ng/mL nocodazole block) and microtubule nucleation assays were performed as described previously (16).

Cells were collected, suspended in PBS, and fixed in ice-cold ethanol at −20°C. Ethanol-fixed samples were washed with cold 1% BSA in PBS and then suspended in 1% BSA in PBS containing 30 μg/mL propidium iodide and 40 μg/mL RNase A, in the dark at room temperature for 30 minutes. Cells were analyzed using the Accuri C6 flow cytometer (BD Biosciences).

RNA was extracted using TRIzol (Invitrogen, 15596026), quantified with NanoDrop, and converted to cDNA using the High-Capacity cDNA Reverse Transcriptase Kit (Applied Biosystems, 4368813) as per the manufacturer's protocols. Primers are listed in Supplementary Table S1 [generously provided by Brad Hoffman (University of British Columbia, British Columbia, Canada)]. Real-time PCR reactions were run in triplicate with an Applied Biosystems 7000 series machine (Invitrogen). Results were analyzed using the ΔΔC_t method. Expression of each transcript was normalized to TATA box binding protein (TBP) levels; expression levels of each transcript/gene measured in G₁-phase enriched 4T1-luc2 cells were normalized to levels measured in asynchronously growing cells.

Cells were grown in standard growth medium until 100% confluence and then serum starved for 72 hours. Prior to imaging, a wound mark was made and the media were replaced with standard growth media. Cells were imaged using an ImageXpress Micro High Content Screening System (Molecular Devices). Image analysis was performed using MetaXpress and ImageJ.

shRNA sequences targeting HMMR (5′-CGTCTCCTCTATGAAGAACTA-3′ and 5′-GCCAACTCAAATCGGAAGTAT-3′; Sigma-Aldrich) were packaged in lentivirus using psPAX2 and pMD2.G (psPAX2 and pMD2.G were gifts from Didier Trono; Addgene plasmids #12260 and 12259). Production and collection of lentiviral particles was performed as described previously (19). Virus was added to media for 24 hours, the media were replaced, and cells were analyzed 72 to 96 hours later.

On-target plus siRNA (Dharmacon) and scrambled siRNA were used as described previously (16). Plasmids used were: GFP-HMMR^WT and GFP (16); GFP-HMMR^T703A was created from GFP-HMMR^WT cDNA using a QuikChange Site-Directed Mutagenesis Kit (Agilent, 200515). Transfection of DNA and siRNA used JetPrime (Polyplus Transfection, 114-07) following the manufacturer's protocols. Cells were harvested or analyzed 96 hours posttransfection with siRNA (24 hours after rescue construct transfection).

Cells were fixed with ice-cold methanol at −20°C for 15 minutes and blocked with 3% BSA in PBST. Primary antibodies were incubated for 2 hours at room temperature, or overnight at 4°C, and secondary antibodies were incubated for 1 hour at room temperature. The following primary antibodies were used: Alexa-647-β-tubulin (TUBB; Thermo Fisher Scientific, MA5-16308-A647, 1:100), phosphorylated Aurora-A (pAurora-A; Cell Signaling Technology, 3079, 1:1,000), DDK (Origene, TA50011-100, 1:1,000), phosphorylated HMMR (pHMMR)(1:7,500; ref. 19), Ran-binding protein 2 (RANBP2; ms, Abcam, ab111811, 1:200), RANBP2 [rb, 1:2,000; generously provided by Dr. Mary Dasso (National Institute for Child Health and Human Development, Rockville, MD) and Dr. Frauke Melchior (Zentrum für Molekulare Biologie der Universität Heidelberg, Heidelberg, Germany)], TPX2 (Novus, mb500179, 1:1,000), and γ-tubulin (TUBG1; Sigma, T6557, 1:1,000). Antibodies conjugated with Alexa-488, Alexa-549, and Alexa-647 (Life Technologies) were used as secondary antibodies. Coverslips were mounted with Prolong Gold antifade reagent with DAPI (Life Technologies, P36935), and images were acquired with a confocal microscope (FluoView Fv10i, Olympus or Leica SP8, Leica). For the nuclear import assay, cells were visualized using a Fluoview FV1000 confocal laser-scanning microscope (Olympus). The quantitation of the ratio of intensity in the nucleus and cytoplasm was measured as described previously (21). Image analysis was performed using ImageJ.

HeLa cells were synchronized at G₂-phase or at G₂–M-phase and homogenized with lysis buffer with a phosphatase inhibitor (PhosSTOP; Roche, 04906845001) and cOmplete protease inhibitor cocktail (Roche, 11697498001). Immunoprecipitation with IgG control antibodies or antibodies targeting pHMMR, or importin-α, was carried out overnight at 4°C, followed by incubation with protein A/G beads at 4°C (Santa Cruz Biotechnology, sc-2003). Protein A/G beads were washed with lysis buffer three times. Bound proteins were separated by SDS-PAGE and analyzed by Western blot analysis with the following antibodies: actin (Sigma, a5060), Aurora-A (Abcam, ab13824), pAurora-A (Cell Signaling Technology), GAPDH (Proteintech, 60004-1-IG), histone H3 (Cell Signaling Technology, 9715), p-histone H3 (Cell Signaling Technology, 9701s), HMMR (Abcam, ab124729), importin (Novus, NB100-79807), and TPX2 (Novus). Secondary antibodies were HRP conjugated (GE Healthcare).

To remove HMMR from rabbit reticulocyte lysate (RRL, Promega, L4960), 50 μL of Dynabeads Protein G suspension (Life Technologies 1003D) was washed three times with 200 μL of PBS, followed by incubation with 2 μg of anti-HMMR/CD168 antibody (Abcam ab124729) in 200 μL PBS with 0.02% Tween-20 (PBST) for 2 hours at 4°C. Dynabead complexes were washed twice with 200 μL of PBST and then crosslinked with 400 μL of RRL for 2 hours at 4°C. After the incubation, the RRL was removed from the beads, analyzed by Western blot for successful depletion of HMMR, and used to perform nuclear import assay.

BSA covalently attached to the NLS of SV40T antigen (CGGGPKKKRKVED; NLS-BSA) was custom made (GenScript). NLS-BSA was labeled with Cy3 (Amersham Biosciences, PA33000) according to the manufacturer's protocol.

Adherent HeLa cells grown as monolayers on coverslips were permeabilized with 40 μg/mL digitonin (Sigma-Aldrich, D141) in transport buffer (TB: 20 mmol/L HEPES, pH 7.4, 110 mmol/L potassium acetate, 1 mmol/L EGTA, 5 mmol/L sodium acetate, 2 mmol/L magnesium acetate, and 2 mmol/L dithiothreitol) for 4 minutes. Permeabilized cells were washed with TB and incubated with TB containing 70-kDa Texas Red-labeled dextran (Invitrogen, D1864), Cy3-labeled NLS-BSA, or DDK-tagged TPX2 (TPX2-DDK; OriGene, MR221968) for 30 minutes at 37°C in the presence or absence of 20% RRL (Promega) or HMMR-depleted RRL, an energy-regenerating system (0.4 mmol/L ATP, 0.45 mmol/L GTP, 4.5 mmol/L phosphocreatine, and 18 U/mL phosphocreatine kinase; Sigma-Aldrich), cOmplete-Mini/EDTA-free Protease Inhibitor Cocktail (Roche) at 10 μg/mL, and 1.6 mg/mL of BSA. After incubation, cells were washed with TB three times and fixed with 4% paraformaldehyde for 10 minutes. Cells incubated with Texas Red dextran or Cy3-labeled NLS-BSA were washed with TB three times and mounted with Prolong Gold antifade reagent with DAPI. Cells incubated with TPX2-DDK were permeabilized with 0.2% Triton X-100 for 5 minutes and prepared for immunofluorescent microscopy as indicated above.

The cohort comprises breast cancer patients newly diagnosed between 1986 and 1992 referred to the British Columbia Cancer Agency and has been described previously (22–24). Detailed description of the demographic, pathologic, and treatment characteristics are described in Supplementary Table S2. Clinicopathologic information including staging, treatment, and follow-up information was available with a median follow-up of 12.3 years. Additional information includes histology, grade, tumor size, lymphovascular invasion (LVI), ER status, axillary lymph node involvement, type of systemic therapy, dates of diagnosis, recurrences (local, regional, or distant), or death. The Clinical Research Ethics Board of the British Columbia Cancer Agency approved this study. Two pathologists reviewed hematoxylin and eosin slide preparations from these blocks to identify areas of invasive carcinoma for inclusion into 0.6-mm core tissue microarrays.

Abundance of pHMMR was analyzed using a rabbit antihuman antibody (described in ref. 19) on 4-μm sections of the 17 tissue microarray blocks using previously described IHC methods (25). Tissue microarrays were scored visually by three observers, with the third observer being the arbitrator when the first two scores were discordant, using a scoring system that captured intensity of staining ranging from 0 (no/low staining), 1 (moderate), and 2 (strong). IHC methods for ER, progesterone receptor (PR), HER2, and Ki67 were as described previously (26).

Statistical analysis was performed using GraphPad Prism v5.01 for Windows (Graphpad Software). Pairwise comparisons were made using unpaired Student t test. Comparisons of multiple groups were made using one-way ANOVA with a Bonferroni posttest. Colocalization was measured using the M1 colocalization coefficient (TPX2/RANBP2).

For the tissue microarray, data distribution in groups and significance between different conditions was analyzed by using an unpaired t test (95% confidence interval, two tailed) or other comparison tests as indicated in GraphPad Prism software. P < 0.05 was considered statistically significant. The entire cohort was randomly divided into a training set and a validation set. Complete data were available for 1,583 and 1,592 cases from the training and validation sets, respectively (N = 3,175). Prespecified analyses for pHMMR association with clinicopathologic variables and outcome were initially conducted on the training set, confirmed on the validation set, and then performed on the entire cohort as presented in this study. Exploratory analyses in the different treatment subsets were done to further assess the association of pHMMR abundance with survival.

All statistical analyses were performed using SPSS 18.0 and R 2.15.0. χ² analysis was used to test associations of pHMMR abundance (as a categorical variable) with age, menstrual status, nodal status, local treatment and systemic treatment, tumor size, grade, histologic type, LVI, and expression of ER, PR, HER2, Ki67 (<14% vs. ≥14%), and molecular subtypes by IHC (27). Luminal A tumors were defined as ER or PR positive, negative for HER2, and low Ki67. Luminal B subtypes were all tumors positive either for ER or PR as well as high Ki67 but negative for HER2. Luminal/HER2 positive subtype was HER2 positive as well as positive for either ER or PR. HER2 subtypes were all those exclusively positive for HER2 with hormone receptor negativity. Tumors negative for all three receptors, ER, PR, and HER2, but positive for either of CK5/6 or EGFR were defined as basal-like subtype. Systemic therapy was categorized as no adjuvant systemic therapy (no AST), tamoxifen only, chemotherapy only, and tamoxifen with chemotherapy.

Univariate survival analyses were performed using the Kaplan–Meier method, and survival differences were estimated using the log-rank test. For multivariate analysis, a Cox proportional hazards ratio model was used to estimate the adjusted HR significance. The primary endpoint for survival analyses was BCSS, defined as the time from date of diagnosis of primary breast cancer to date of death due to breast cancer as the primary or underlying cause. RFS, defined as time from date of diagnosis of primary breast cancer to date of first local, regional, or distant recurrence, was used as a secondary endpoint.

Triple-negative breast cancer cells engaged in the cell cycle metastasize better than quiescent cells in patient-derived xenograft models (9). We tested whether a particular phase of the cell cycle allows for better engraftment of murine breast cancer cells in syngeneic animals. A total of 10,000 4T1-luc2 cells that were either growing asynchronously or G₁-phase enriched (Supplementary Fig. S1A) were injected into the tail vein of BALB/c mice. Using luciferase activity, we monitored engraftment and tumor growth over 3 weeks. 4T1-luc2 cell populations enriched for cells in G₁-phase were significantly impaired in their ability to engraft BALB/c mice (Fig. 1A and B). We performed similar experiments by injecting asynchronously or G₁-phase–enriched 4T1-luc2 cells into the tail vein of NSG mice (Supplementary Fig. S1B). Again, G₁-phase enriched cells injected in NSG mice were significantly impaired in the kinetics of engraftment and exhibited an approximately 1-week delay in tumor growth (Fig. 1C). We examined whether the reduction in engraftment potential observed in G₁-phase enriched 4T1-luc2 cells correlated with changes to the expression of defined EMT markers, including Cdh1, Snai1, Snai2, vimentin, or Cdh2. The expression level for Hmmr, which is elevated during G₂–M-phase (28), was included as a positive control. Relative to the expression observed in asynchronously growing cells, G₁-phase enriched cells express elevated levels of the epithelial cell marker Cdh1 and 4- to 10-fold lower levels of Snai1, Snai2, or vimentin, markers for mesenchymal cells, and Hmmr, a proliferation marker (Fig. 1D). These data indicate that the enrichment of 4T1-luc2 cells in G₁-phase dampens their expression of EMT markers and severely diminishes their engraftment potential.

To determine whether G₁-phase cells are impaired in migratory capacity measured in vitro, we utilized live cell imaging to follow the kinetics of wound closure for nonmalignant, murine mammary epithelial cells expressing a cell-cycle indicator (nMuMG-FUCCI; Fig. 2A). nMuMG-FUCCI cells were seeded in 96-well plates, grown to confluence, and serum starved to synchronize at G₁-phase. Wounds were introduced and complete closure was observed 48 hours following the scratch (Fig. 2B). We noted at 24 hours postwounding that the cells occupying the leading edge and wound area were largely in S–G₂-phase, while cells more distal from the scratch were largely in G₁-phase (Fig. 2C and D), suggesting that either G₁-phase cells at the edge of the wound transition into S–G₂-phase more quickly than more centrally located cells or that S–G₂-phase cells have a migratory advantage. To distinguish between these possibilities, individual cells at the wound edge were tracked for the first 24 hours postwounding using a wide-field, high-content cell imaging system with multiparameter assessment of cell-cycle phase, wound closure, and migration velocity (Fig. 2E). The region of the wound was defined, and only cells (indicated by masked nuclei) within this region of interest were measured. Based upon the FUCCI signal, cells were classified as those that remained within one phase (cells 1 and 4, Fig. 2E) and those that transitioned through cell-cycle phases during the imaging window (cells 2 and 3, Fig. 2E). We found that cells in S–G₂-phase migrated with approximately 1.7-fold greater velocity than neighboring cells that remained in G₁-phase (14.1 ± 4.8 μm/hour vs. 8.8 ± 4.1 μm/hour, P < 0.0001; Fig. 2F). When we restricted the analysis to individual cells that transitioned from G₁-phase to S–G₂-phase during imaging, we confirmed a significant increase in migration velocity during S–G₂-phase (13.2 ± 3.7 μm/hour vs. 9.9 ± 3.9 μm/hr, P < 0.001; Fig. 2G). We repeated these studies using malignant, cervical carcinoma HeLa-FUCCI cells and found a congruent correlation between progression into S–G₂-phase and elevated migration velocity (Supplementary Fig. S2).

For mechanistic insight into why S–G2-phase cells may migrate more efficiently than cells in G₁-phase, we incubated nMuMG-FUCCI cells with thymidine, or equivalent DMSO control, to prevent their entry into G₂-phase during wound closure. Relative to control-treated cells, wound closure was impaired by nearly 45% for thymidine-synchronized cells (Fig. 3A and B). Next, we augmented the frequency of nMuMG-FUCCI cells in G₂-phase by incubation with nocodazole, which depolymerizes microtubules and prevents entry into mitosis. Although we observed a robust enrichment of G₂-phase cells following nocodazole treatment, wound closure was abolished (Fig. 3A and B), indicating a requirement for the microtubule cytoskeleton. Thus, cell-cycle progression into G₂-phase is strongly correlated with an increased velocity during wound closure assays with nMuMG-FUCCI cells, and the microtubule cytoskeleton functions as a critical mediator for migration.

By imaging wounds during the first 24 hours of closure, we were able to measure the velocity and directionality of migration for individual cells (Fig. 3C). Consistent with the kinetics of wound closure (Fig. 3B), thymidine-synchronized, G₁-phase cells and nocodazole-treated, G₂-phase cells migrated with a significantly reduced mean velocity relative to DMSO-treated, S–G₂-phase cells (Fig. 3D). Moreover, the migration of DMSO-treated cells was highly biased to be directed toward the wound as indicated by plotting the displacement of individual cells along the x-axis and y-axis (Fig. 3E). At the end of 24 hours, 89.4% ± 6.0% of DMSO-treated cells were displaced within a quadrant defined by ±45-degree angles from their origin. However, the displacement of thymidine-synchronized, G₁-phase cells and nocodazole-treated, G₂-phase cells was far less directional (Fig. 3E).

Movement of the centrosome, or microtubule-organizing center, toward the leading edge is a critical step in the front-rear polarization, scattering, and migration of certain cell types, including epithelial cells (4, 29). Moreover, augmented microtubule nucleation at supernumerary centrosomes is known to trigger invasive behavior in mammary epithelial cells (11). Therefore, we examined whether G₁-phase and S–G₂-phase nMuMG-FUCCI cells differ in centrosome polarity and microtubule nucleation capacity during wound closure. At 18 hours postwounding, nMuMG-FUCCI cells were fixed and immunostained for TUBG1. The frequency of front-polarized TUBG1-positive centrosomes (i.e., those positioned between the nucleus and wound, Fig. 3F) was augmented in S–G₂-phase cells (Fig. 3G). Next, we treated scratch wound assays with 15 micromol/L nocodazole for 60 minutes at 18 hours postwounding to depolymerize microtubules in nMuMG-FUCCI cells. Cells were washed and microtubules were allowed to polymerize for 2 minutes prior to fixation (Fig. 3H). The microtubule fibers regrown in nMuMG-FUCCI cells located at the leading edge were significantly longer in S–G₂-phase cells than in G₁-phase cells (Fig. 3I), indicating an augmented microtubule nucleation capacity in S–G₂-phase. Together, these data indicate fundamental differences in centrosome orientation and microtubule nucleation capacity between S–G₂-phase cells and G₁-phase cells during wound closure assays, which may alter their relative abilities to break epithelial cohesion and migrate into a wound.

Aurora-A promotes microtubule nucleation within mitotic cells (30). To assess whether active Aurora-A is required for the migration of nonmitotic cells during wound closure, we first established that 2.5 μmol/L MLN8237 (a specific small-molecule Aurora-A inhibitor) was a minimal dose needed to reduce either the levels of phosphorylated histone H3 (Supplementary Fig. S3A), a kinase substrate, or the levels of the autophosphorylated kinase (Supplementary Fig. S3B) in lysates prepared from nocodazole-synchronized nMuMG-FUCCI or MCF-10A cells, respectively. The treatment of nMuMG-FUCCI cells with 2.5 μmol/L MLN8237 for the first 24 hours following wounding was sufficient to significantly reduce the area of wound closure (Fig. 4A and B). The treatment of nMuMG-FUCCI cells with 2.5 μmol/L MLN8237, however, did not significantly change the proportion of S–G₂-phase cells at the leading edge (Fig. 4C, P = 0.13), indicating that inhibition of Aurora-A dampens migration without altering cell-cycle progression into G₂-phase. We next analyzed the kinetics and directionality of cell migration for the first 24 hours in the presence of DMSO or MLN8327. Aurora-A inhibition not only diminished the velocity (Fig. 4D) but also dampened the directionality of nMuMG-FUCCI cell migration (Fig. 4E). These altered migration phenotypes induced through Aurora kinase inhibition were accompanied by a significantly reduced frequency of front-polarized centrosomes (Fig. 4F and G) and diminished microtubule nucleation capacity at the centrosome (Fig. 4H and I) when the analyses were restricted to S–G₂-phase cells at the leading edge. Thus, Aurora-A activity enables microtubule nucleation at centrosomes, centrosome polarization, and directed migration in scratch wound closure assays.

HMMR localizes TPX2 during mitosis and acts as an upstream regulator of Aurora-A activity that enables microtubule nucleation at duplicated centrosomes and near chromosomes (16, 17). As TPX2 enhances the accumulation of Aurora-A during G₂-phase (31) and Aurora kinase activity at centrosomes (20), we postulated that HMMR may also promote Aurora-A activity and cell migration during G₂-phase. To test this postulate, we silenced HMMR by separately treating HeLa cells with scrambled siRNA control or siRNA targeting the 5′- or 3′- untranslated regions (UTR) in HMMR (Fig. 5A), which enabled rescue with cDNA lacking those UTRs. Alternatively, we treated HeLa cells with lentivirus encoding a nonhairpin shRNA control or shRNA targeting HMMR (Fig. 5B). Following these treatments, we determined that HMMR was dispensable for the centrosome localization of pAurora-A (Fig. 5C), but the levels of the active kinase were dampened in HMMR-silenced cells relative to control-treated cells (Fig. 5D). To determine the effect of this dampened activity on the cell's capacity to nucleate microtubules, microtubule regrowth assays were performed 3 days following transduction with lentivirus. We restricted our analysis of microtubule regrowth capacity to cyclin B1–positive, G₂-phase cells (Fig. 5E) and found that silencing HMMR resulted in a significant reduction in the length of microtubule fibers at the centrosome (Fig. 5F). Moreover, the fraction of cyclin B1–positive, G₂-phase cells with front-polarized centrosomes (Fig. 5G) were also reduced in HMMR-silenced cell populations (Fig. 5H), which correlated with significantly impaired wound closure measured at 24 hours (Fig. 5I and J) and dampened migration velocity (Fig. 5K). Together, these data indicate that the migration of G₂-phase epithelial cells requires Aurora-A activity, which can be reduced by treatment with a small-molecule inhibitor (MLN8237) or through the silencing of HMMR, an upstream molecular regulator.

In nonmitotic cells, little is known about the extent or mechanism by which TPX2 in the nucleus can activate Aurora-A other than the observation that the subcellular localization of TPX2 is altered in HMMR-silenced cells (19). Indeed, TPX2 colocalizes with the nucleoporin Nup153 in HMMR-silenced cells (Supplementary Fig. S4A), suggesting a role for HMMR in the nuclear transport of TPX2. HMMR is a microtubule-associated protein and is usually segregated from TPX2 in the nucleus (28); however, Aurora-A phosphorylates HMMR at threonine-703 (pHMMR) and pHMMR localizes to the nucleus (Supplementary Fig. S4B; ref. 19). To test whether this phosphorylation event in HMMR affects the nuclear localization of TPX2, we created a GFP-tagged phospho-dead mutant (GFP-HMMR^T703A) as well as wild-type HMMR (GFP-HMMR^WT; Supplementary Fig. S4C). We treated cells with siRNA targeting the 5′- or 3′-UTRs in HMMR, or scrambled siRNA control, and transiently transfected cells with cDNAs encoding GFP-HMMR variants, which are resistant to the HMMR-targeting siRNA. In scrambled siRNA-treated control cells, the expression of GFP, GFP-HMMR^WT, or GFP-HMMR^T703A did not alter the nuclear localization of TPX2 (Fig. 6A). In HMMR-silenced cells, only the expression of GFP-HMMR^WT, and not GFP alone or GFP-HMMR^T703A, rescued the nuclear localization of TPX2 (Fig. 6A and B). These data indicate that the nuclear localization of TPX2 is promoted by HMMR threonine-703.

To investigate a requirement for HMMR in the nuclear transport of TPX2, we used an in situ nuclear import assay wherein HeLa cells were transfected with siRNA targeting HMMR, or scrambled siRNA, and subsequently permeabilized with digitonin to release soluble cytosolic components (32). Recombinant flag-tagged TPX2 (TPX-DDK), an energy regeneration mixture, and rabbit reticulocyte lysate, which was immunodepleted of HMMR (Supplementary Fig. S5A), were then added to these cells to enable nuclear import. The addition of fluorescently labeled BSA conjugated with a nuclear localization signal (NLS-BSA) served as a positive control for nuclear import. To monitor nuclear integrity, we confirmed that fluorescently labeled dextran remained cytosolic and did not enter the nucleus during the experimental procedures (Supplementary Fig. S5B). Moreover, in the absence of energy and cytosol (demarked No Transport), nuclear transport did not occur for either TPX2-DDK (Fig. 6C) or NLS-BSA (Supplementary Fig. S5C). When scrambled siRNA control-treated cells were incubated cytosol and energy (demarked Transport), both NLS-BSA (Supplementary Fig. S5C) and TPX2-DDK (Fig. 6C) were efficiently transported into the nucleus. However, the levels of TPX2-DDK in the nucleus, measured as a ratio with the levels in the cytoplasm, were quantitatively reduced in HMMR-silenced cells (Fig. 6C and D). HMMR-silenced cells did not have alterations in the organization of the microtubule cytoskeleton or the shape and area of the nucleus (Supplementary Fig. S5D and S5E), which may explain the deficient nuclear transport of TPX2-DDK. Similar to the localization of endogenous TPX2 in HMMR-silenced cells, we also observed an augmented colocalization of TPX2-DDK with the nucleoporin RANBP2 (Fig. 6E and F) indicative of TPX2-DDK accumulation at the nuclear envelope. We next tested whether complexes of TPX2 and importin-α were augmented in HMMR-silenced cells. Cell lysates from control-treated or HMMR-silenced cells were immunoprecipitated with antibodies targeting importin-α. In HMMR-silenced cell lysates, we observed an overall reduction in the total levels of TPX2 (input, Fig. 6G); in spite of this, we also observed an increase in the levels of TPX2 coprecipitated with antibodies targeting importin-α from HMMR-silenced cell lysates (Fig. 6G). Taken together, our data uncover HMMR-T703 as a critical regulator of TPX2 nuclear transport that is needed to reduce TPX2–importin-α complexes at the nuclear envelope. These data support an axis of Aurora-A–TPX2–HMMR in the determination of microtubule nucleation and migration in nonmitotic cells.

To study the putative significance of the HMMR–TPX2–Aurora kinase A axis in primary breast tumor tissues, we evaluated pHMMR abundance in a large tumor tissue array. Tumor was present for evaluation in 3,175 of the 3,992 cases; 817 cases were excluded for no or insufficient tissue present. Phosphorylated HMMR abundance (intensity score ≥1) was seen in 58.7% (1863 of 3175) tumors. The intensity of staining in the tumors ranged from no or weak staining (41.3%), to moderate staining (44.1%), or strong staining (14.6%; Fig. 7A). A significant positive correlation between pHMMR and Ki67 was observed (P < 0.005).

In patients with ER-positive breast cancer, the intensity of pHMMR staining did not predict BCSS or RFS (Fig. 7B and F). However, pHMMR was a significant predictor of BCSS and RFS in patients with ER-negative breast cancer (Fig. 7C and G), triple-negative phenotype breast cancer (Fig. 7D and H), and basal-like subtype breast cancer (Fig. 7E and I). In the multivariate model, including age at diagnosis, grade, nodal status, tumor size, LVI, ER, and HER2 status, pHMMR was an independent predictor of RFS in patients with basal-like subtype (P = 0.02), but was not significant in multivariate analyses of BCSS (P = 0.07). Similarly, in multivariate analyses, pHMMR was not a significant independent predictor of BCSS or RFS in all ER-negative breast cancer (P = 0.09 and P = 0.15, respectively).

Within the entire cohort, patients were treated with either a combination of chemotherapy and tamoxifen (n = 237), chemotherapy alone (n = 588), tamoxifen alone (n = 968), or no AST (n = 1,362). We were specifically interested in the predictive value of pHMMR in the cohort of patients that received no AST. Phosphorylated HMMR abundance did not predict BCSS or RFS across all tumors that received no AST (Fig. 7J and L). However, when the analysis was restricted to the patients with ER-negative tumors that received no AST (n = 453), pHMMR was an independent predictor of both BCSS and RFS (Fig. 7K and M), and this association was retained in multivariate analyses (P = 0.01 and P = 0.002, respectively); in patients with ER-negative tumors that received no AST, there was also an association between pHMMR and distant nodal metastasis (P < 0.05).

Apicobasal-polarized epithelial cells organize microtubules at cell–cell contacts (3) but, prior to cell division, Aurora-A enables cilia dissolution (2) and promotes the microtubule nucleation capacity, or maturation, of the centrosome (33, 34). Similarly, microtubule nucleation at the centrosome, which must be repositioned toward the leading edge to establish front-rear polarity (29), is critical to motile or scattering epithelial cells (4). Thus, cell-cycle progression and cell migration necessitate comparable changes with the organization of microtubules in epithelial or carcinoma cells. Here, we find that cell-cycle progression augments both engraftment kinetics in vivo and the expression of mesenchymal markers in vitro in asynchronously growing 4T1-luc2 cells. Cells in S–G₂-phase possess an increased velocity and directionality of migration as well as front-polarized centrosomes with augmented microtubule nucleation capacity. Thus, we find that epithelial cell migration and cell-cycle progression share common regulation of the microtubule cytoskeleton through the Aurora-A–TPX2–HMMR axis.

The connection between cell-cycle progression and migration in epithelial cells likely explains the strong relationship between cell populations that express an EMT signature and those with an augmented tumor-initiating potential (7, 8). When EMT is induced in renal fibrosis, for instance, the cells become delayed in G₂–M, and this delay is only reversed with the removal of the EMT induction (35). This relationship has important implications. Conventional theories hypothesize that rare malignant cells in a primary tumor possess unique properties or transcriptional programs that enable them to seed metastasis (9); analysis of transcriptional profiles in single cells isolated from low burden metastatic sites suggest that these pioneer cells may express basal-like or stem-like gene expression patterns, including dormancy-related genes, and that metastatic progression requires a switch to the cell cycle (9). Here, we find that the simple process of enriching for G₁-phase cells has dramatic negative consequences on the expression of defined basal-like cell markers and the efficiency of tumor engraftment. Moreover, we identified significant changes in migration velocity, centrosome polarization, and microtubule nucleation capacity as individual G₁-phase cells progressed through the cell cycle. Thus, a carcinoma cell transitioning through the cell cycle demonstrates plasticity that may be interpreted by gene expression profiling, or through xenotransplantation assays, as an EMT or increasing cancer “stemness.” But, a connection between cell-cycle progression and migratory capacity is likely cell-type specific given that hematopoietic cells, for example, do not engraft well during S–G₂–M but optimally engraft during G₀–G₁-phase (36, 37). Our findings are also highly relevant to the study of tumor initiation and cancer stem cells (CSC). The tumor xenotransplantation assay is the current gold standard for the functional detection of cancer stem cells. Here, we find that simple adjustment in cell-cycle progression within a standard breast cancer cell line significantly impacts the efficiency of engraftment analogous to the influence of the immune system in the recipient animal (38) or the use of tumor fragments as opposed to disassociated cells (39). Thus, for carcinoma samples, the CSC readout obtained from a tumor xenotransplantation assay may be strongly impacted by the cell-cycle profile of the inputted tumor sample.

A FRET-based reporter for Aurora-A activity at centrosomes indicates the need for nuclear-localized TPX2 to activate the kinase and stabilize microtubules during G₁ phase (20), but the mechanism for this regulation is not well studied. We have previously reported that silencing HMMR reduces the nuclear localization of TPX2 and alters Aurora-A activity (19). Here, we show that HMMR, and specifically T703, is needed for the nuclear transport of TPX2. In HMMR-silenced, nonmitotic cells, TPX2–importin-α complexes are augmented and TPX2 accumulates at the nuclear envelope while the activity of Aurora-A at the centrosome, as measured by the levels of pAurora-A, is dampened, similar to the effect observed in TPX2-silenced, G₁-phase cells (20). This observation contrasts with our prior interpretation that pHMMR negatively regulates Aurora-A activity (19), which was based upon measurements of kinase activity in cell lysates as opposed to individual cells. Rather, our analysis of pAurora-A indicates that HMMR promotes kinase activity in nonmitotic cells through the nuclear transport of TPX2 and facilitating TPX2 release from a complex with importin-α, similar to the role HMMR plays in mitotic cells and extracts (16, 17). Aurora-A activity at the centrosome is known to control microtubule dynamics in migrating endothelial cells (40) and neurons (41). Thus, in these cells and tissues, it will be important to determine whether Aurora-A activity is also reliant upon the control of TPX2 location by pHMMR.

A limitation of our study of pHMMR in the large breast cancer tissue microarrays relates to IHC approaches, which have been outlined previously (27). Briefly, the study has limited technical reproducibility and the readout for pHMMR was subjective and qualitative, although we relied upon a consensus score from three blinded observers to offset the latter limitation. Moreover, the study was trained and validated on tissue microarrays, which may miss focally higher areas of immunostaining that are more easily appreciated in whole sections. Survival data were derived from a median follow-up period of 12.3 years and, as a consequence, the treatment recommendations at the time of tissue collection (1986–1992) differed from those in contemporary practice, which tend to be more aggressive. Putting these limitations aside, however, the abundance of pHMMR appears to be a significant negative prognostic factor specifically in ER-negative breast tumors.

Proliferation is one of the most important prognostic factors for invasive breast cancer (42). In ER-positive, HER2-negative subtypes, a proliferative molecular module, defined in part by Aurora-A expression, was the common denominator of most prognostic signatures and the strongest parameter predicting clinical outcome (43). Indeed, Aurora-A expression, used in combination with the expression of HER2 and ER, is sufficient to discriminate between low and high proliferative luminal A and luminal B tumors (44). Moreover, stable DNA amplification of TPX2 was identified in ER-positive breast cancer and was found to be enriched in prognostic gene sets for ER-positive breast tumors (45). However, we found that elevated pHMMR levels did not discriminate ER-positive tumors with a poor prognosis (Fig. 7B and F). When considering this discrepancy, it is important to note that elevated transcript expression does not absolutely correlate with protein expression nor with kinase activity (phosphoproteome) at steady state or following drug treatment (46), so ER-positive tumors with elevated Aurora-A or TPX2 gene expression may not be detected with high pHMMR by IHC. More importantly, in the TMAs analyzed here, the Ki67 index associated with poor BCSS and RFS in luminal breast tumors (27). Thus, it is likely that luminal breast tumors detected to express high pHMMR are not simply Ki67-positive tumors with elevated levels of proliferation or with consistent elevated expression of TPX2 and Aurora-A.

The selectivity of the prognostic power for elevated pHMMR in TNBC and basal-like breast cancer may relate to a specific nonproliferative role of Aurora-A signaling in promoting the progression of these subtypes. Genomic and transcriptomic analyses of 2,000 breast tumors implicated elevated Aurora kinase signaling in a basal-like cancer–enriched subgroup (47). In this large cohort of breast tumors, basal-like breast tumors harbor chromosome 5q deletions associated in trans with elevated Aurora kinase signaling and cell division genes (47). In other tumor cells, genomic imbalance in chromosome 5q32-qter correlated with markers for Aurora kinase activity and elevated sensitivity to aurora kinase inhibitors (48). Chromosome 5q33-34 and specifically polymorphisms of HMMR and CCNG1 in this region (19, 49), modifies breast cancer risk in BRCA1 mutation carriers, which tend to develop ER-negative or basal-like breast tumors. Thus, heightened Aurora-A activity, as indicated by elevated pHMMR levels, may promote the progression of ER-negative breast tumors and consequently indicate a poor prognosis. If so, these tumors may possess heightened sensitivity to small-molecule Aurora kinase inhibitors.

Taken together, our data indicate that cell-cycle progression and the acquisition of migratory phenotypes are interconnected in carcinoma cells by a requirement for the acquisition of microtubule nucleation capacity at the centrosome that is enabled by the Aurora-A/TPX2/HMMR axis. Specifically, this molecular axis promotes the establishment of directional migration. The precise reason why this pathway is more strongly responsible for the progression of ER-negative breast tumors, and whether this pathway enables rare invasive “leader” cells or epithelial cell scattering, which is known to rely upon centrosome repositioning (4), remains unclear and warrants further study. Finally, our findings may also have implications for therapeutic targeting strategies in currently difficult-to-treat ER-negative breast cancer.

No potential conflicts of interest were disclosed.

Conception and design: T.L.H. Chu

Development of methodology: T.L.H. Chu, M. Connell, L. Zhou, S.M.R. Rahavi, A. Fotovati, N. Pante, C.A. Maxwell

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.L.H. Chu, M. Connell, L. Zhou, Z. He, H. Chen, S.M.R. Rahavi, P. Mohan, O. Nemirovsky, G.S.D. Reid, T.O. Nielsen, C.A. Maxwell

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.L.H. Chu, M. Connell, L. Zhou, Z. He, J. Won, H. Chen, S.M.R. Rahavi, P. Mohan, O. Nemirovsky, M.A. Pujana, G.S.D. Reid, T.O. Nielsen, N. Pante, C.A. Maxwell

Writing, review, and/or revision of the manuscript: T.L.H. Chu, M. Connell, S.M.R. Rahavi, P. Mohan, G.S.D. Reid, N. Pante, C.A. Maxwell

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.L.H. Chu, L. Zhou, Z. He, S.M.R. Rahavi, C.A. Maxwell

Study supervision: T.L.H. Chu, N. Pante, C.A. Maxwell

Other (animal study and bioluminescent imaging): S.M. Rahavi

This study was funded by the Canadian Institutes of Health Research (OBC-134038, awarded to C.A. Maxwell and N. Pante; MSH-136647 salary award to C.A. Maxwell) and the Canadian Breast Cancer Foundation (PhD fellowship, awarded to T.L.H. Chu). C.A. Maxwell and G.S.D. Reid are supported by salary awards from the Michael Cuccione Childhood Cancer Research Program, BCCH.

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.