AurkA controls self-renewal of breast cancer-initiating cells promoting wnt3a stabilization through suppression of miR-128

AurkA overexpression was previously found in breast cancer and associated to its ability in controlling chromosome segregation during mitosis, however whether it may affect breast cancer cells, endorsed with stem properties (BCICs), is still unclear. Surprisingly, a strong correlation between AurkA expression and β-catenin localization in breast cancer tissues suggested a link between AurkA and Wnt signaling. In our study, AurkA knock-down reduced wnt3a mRNA and suppressed metastatic signature of MDA-MB-231 cells. As a consequence, the amount of BCICs and their migratory capability dramatically decreased. Conversely, wnt3a mRNA stabilization and increased CD44+/CD24low/− subpopulation was found in AurkA-overexpressing MCF7 cells. In vivo, AurkA-overexpressing primary breast cancer cells showed higher tumorigenic properties. Interestingly, we found that AurkA suppressed the expression of miR-128, inhibitor of wnt3a mRNA stabilization. Namely, miR-128 suppression realized after AurkA binding to Snail. Remarkably, a strong correlation between AurkA and miR-128 expression in breast cancer tissues confirmed our findings. This study provides novel insights into an undisclosed role for the kinase AurkA in self-renewal and migration of BCICs affecting response to cancer therapies, metastatic spread and recurrence. In addition, it suggests a new therapeutic strategy taking advantage of miR-128 to suppress AurkA-Wnt3a signaling.

The Wnt pathway plays a fundamental role in proper mammary gland development, regulating self-renewal of stem-progenitors cells 9 . Moreover, nuclear accumulation of β -catenin is considered a trigger for transcription of genes implicated in self-renewal of CICs 10,11 .
However, contrasting evidences showed that overexpression of Wnt1, Wnt3a and Wnt7a promoted hyperplasia of mammary gland 12 , in contrast Wnt7b and Wnt5a failed to show a tumorigenic role in mice 13 , suggesting that each Wnt member may activate different signaling pathways depending on the cellular context.
Because of this complexity, the role of Wnt pathway in breast cancer and metastasis remains still uncleared.
Here, we show a post-transcriptional regulation of wnt3a by AurkA and reveal a novel role for the mitotic kinase in regulation of BCICs self-renewal and their metastatic properties.
Surprisingly, our findings suggest that AurkA regulates self-renewal, migratory activity and metastatic signature of BCICs through Wnt3a/β -catenin pathway. Indeed, we show that AurkA favors wnt3a mRNA stabilization in BCICs inhibiting miR-128.
Recently, miR-128 was found deleted in several human tumors. Previous data show that miR-128 may regulate wnt3a mRNA to promote differentiation of rat mesenchymal stem cells into neural cells 28 . In breast cancer, miR-128 inhibits expression of some stem gene as BMI1, CSF1, KLF4, LIN28A, NANOG 29 , however the molecular mechanisms still remain unknown.
Here we show that AurkA may suppress miR-128 expression through activation of Snail. Moreover, our study shows a strong correlation between AurkA and miR-128 expression in breast clinical isolates (N = 32) which further supports our findings.
Collectively our data reveal a new undisclosed role for the kinase AurkA in maintaining of BCICs. Moreover, our study suggests AurkA/miR-128/Wnt3a axys as a druggable target to inhibit chemoresistance and recurrence in breast cancer.

Results
AurkA overexpression is associated with β-catenin nuclear/cytoplasmic localization. AurkA overexpression/amplification was found in several human cancers. In breast cancer, some evidences sustain it is associated to basal-like phenotype 23,30 , others suggest it may be a marker for progression and outcome of luminal-like subtype 31 .
We analyzed the expression of AurkA kinase in 89 breast cancer patients by immunohistochemistry (IHC) (Fig. 1A). A significant positive staining of the kinase was found in 41 out 89 breast cancer patients (46.07%). However no correlation was found between AurkA overexpression and clinical and pathological features as grading (P value = 0.759), Ki67 (P value = 0.574), tumor size (P value = 0.553) or linfonodal status (P value = 0.107) in N = 89 breast cancer samples (Table 1).
Similar statistical analyses were performed to assess a correlation with breast cancer subtypes. On the basis of hormone receptors (ER and Her2), breast cancers were grouped in Luminal (ER-positive, 24  comparison with normal breast tissues (Table 1). However, no significant correlation was found between AurkA overexpression and any of the breast cancer subgroups (P value = 0.0568) ( Table 1). Surprisingly, a strong correlation was found after evaluation of cellular localization of β -catenin (P value = 0.00001) by IHC. Noteworthy, we found that breast cancer samples showing increased levels of AurkA, lost cortical β -catenin (Fig. 1A).
A β -catenin dislocation was previously associated to high aggressive breast cancer cells 32 . In addition, nuclear β -catenin suggests the activation of the Wnt signaling 33 .
Interestingly, we found that breast cancer samples, overexpressing AurkA (>1.5 fold-change in comparison with normal breast), similarly showed increased levels of CD44 as revealed by Q-PCR. Conversely, low levels of AurkA mRNA (<1.5 fold-change) corresponded to low levels of CD44. This significant correlation (P value = 0.00001) was found in 26 out 32 breast cancer samples (Fig. 1B).
AurkA controls BCICs through regulation of Wnt3a/β-catenin. The discovery that overexpression of AurkA correlated with nuclear/cytoplasmic localization of β -catenin and increased expression of CD44 suggested that AurkA may have a role in controlling BCICs.
To verify this hypothesis, we modulated AurkA expression in breast cancer cell lines and evaluated the effects on BCICs.
Surprisingly, we found that modulating AurkA expression we were able to change expression levels of wnt3a ( Fig. 2A,B). Accordingly, we found an increase in the amount of CD44 + /CD24 low/− subpopulation in MCF-AurkA+ , ( Fig. 2A,a), due to a marked increase in CD44-positive cells ( Fig. 2A,b) and a decrease of CD24-positive cells ( Fig. 2A,c). Moreover, sphere-forming assay revealed that MCF-AurkA+ , increased their ability to grow as mammospheres in comparison with control cells (Fig. 2C).
As well, AldeFluor assay showed a significant decrease of ALDH-positive cells in MDA-shAurkA cells suggesting that inhibition of AurkA affected stem-like subpopulation (Fig. 2B, a and b), as confirmed by a marked decrease in mammospheres formation when MDA-shAurkA cells were grown in low-adhesion conditions (Fig. 2C).
The involvement of AurkA in regulation of BCICs was further confirmed by western blot analysis showing increased wnt3a and β -catenin in MCF-AurkA cells (Fig. 3A). Conversely β -catenin degradation and undetectable wnt3a levels were found in MDA-shAurkA (Fig. 3B). Those data strongly suggested that the kinase may control BCICs through the Wnt3a/β -catenin axis.
In addition, western blot analyses revealed a link between AurkA and some proteins involved in cellular migration and metastasis as Mmp9 and Stat3. In particular, we found that AurkA overexpression promoted stabilization of Mmp9 and Stat3 (Fig. 3A), conversely AurkA silencing severely affected the amount of the same proteins (Fig. 3B).
Collectively, these data suggest that AurkA may take part in mechanisms underpinning self-renewal and metastatic properties of BCICs.
Actually, this hypothesis was further supported by Q-PCR revealing increased expression of EMT and migratory genes such as CD44v6, Snail, Twist and c-Met in MCF-AurkA+ cells (Fig. 3C), whereas a significant decrease (except for twist) in MDA-shAurkA cells (Fig. 3D). Moreover, AurkA overexpression promoted migration of MCF7 cells which usually exhibit a low aptitude to migrate in invasion assays (Fig. 3C). Conversely, the marked migratory activity of MDA-MB-231 cells was inhibited after AurkA knock-down (Fig. 3D).

Tumorigenic primary breast cancer cells show higher levels of AurkA.
To assess if AurkA may promote a more aggressive phenotype in vivo, we isolated primary breast cancer cells (named KBr1, KBr2, KBr3 and KBr4) from 4 different mastectomized breast cancer patients and injected them into immune-compromised mice.
We analyzed correlation between AurkA expression and tumorigenic properties of cancer cells. We found that only 3 samples (KBr2, KBr3 and KBr4), showing a high expression of the kinase (similar to MDA-MB-231 cells), were tumorigenic. In contrast, AurkA-low expressing KBr1 failed to generate tumors after injection in vivo, as MCF7 cells (Fig. 4A).
Moreover, analyses on xenografts confirmed a correlation between AurkA overexpression and a more aggressive behaviour in vivo as revelead by positive MMP9 staining and high Ki67 values (Fig. 4A).
Actually, we found that the CD44-positive subpopulation in primary breast cancer cells (KBr2, KBr3 and KBr4), showed higher levels of AurkA and wnt3a mRNAs as well as increased migratory ability, compared to CD44-negative cells (Fig. 4B).
AurkA promotes stabilization of wnt3a mRNA through repression of miR-128. Our findings suggest a new role for AurkA, which proves to be able to control BCICs and, likely, aggressiveness of breast cancer. Next we wanted to unravel the molecular mechanisms underpinning this new role of the kinase.
Some evidences showed that AurkA may control the expression of some genes through regulation of several microRNAs (miRNAs), namely miR-21 in hepatocellular carcinoma 34 . In breast cancer, it was found that AurkA control the miR17-92 cluster through regulation of E2F1 transcription factor 35  Here, we hypothesized that AurkA may control the expression of a miRNA which should act as a repressor for wnt3a. An in-depth analysis by TargetScan 7.0 Software allowed to identify three different miRNAs as specific regulators for wnt3a, unknown to be AurkA target: miR-15, miR-16 and miR-128.
By Q-PCR analysis, we were able to assess a correlation between miR-128 and AurkA. Basically, we found that AurkA knock-down in MDA-MB-231 was ineffective for stabilization of miR-15 and reduced miR-16 expression, whereas, induced a significant increase in miR-128 levels (Fig. 5A, left graph).
Accordingly to our previous hypothesis, if AurkA controls transcription of wnt3a through a repressive activity of a miRNA, a marked increased expression of this miRNA should be found after AurkA knock-down. As a consequence, we considered miR128 as a target of AurkA, excluding miR15 and miR16.
A miR-128 specific binding to wnt3a mRNA was proved by luciferase assay. A vector, carrying the 3′ UTR region of wnt3a downstream luciferase gene, was transfected in HEK-293T cells (Fig. 5C, pMIR-wt-wnt3a). We observed a decreased luciferase activity in presence of mimics, whereas luciferase activity was restored by an inhibitor specific for endogenous miR-128 (Fig. 5C). Conversely, neither mimics nor inhibitor were able to affect luciferase activity when HEK-293T cells were transfected with a vector carrying a mutated 3′ UTR region (Fig. 5C, pMIR-mut-wnt3a).
Snail mediates repression of miR-128 in response to AurkA overexpression. Given our findings proving that miR-128 is able to repress wnt3a at post-transcriptional level, we hypothesized whether a transcription factor may exist which is able to repress miR-128 activity on a fashion that is dependent on AurkA.
Previous findings suggested Snail as a repressor for miR-128 29 . Based upon this first evidence, we evaluated if snail may be a more likely candidate. Q-PCR confirmed a correlation between AurkA and Snail, showing increased snail mRNA levels in MCF-AurkA cells and, conversely, a marked decrease in both MDA-shAurkA and KBr2-shAurkA cells (Fig. 5D).
A luciferase assay was carried out to verify a site-specific binding of Snail to E-Box1 and E-Box2 of miR-128 promoter (Fig. 5E, pGL3-miR-128), as previously described in ref. 29.
Our data confirm that miR-128 is a target of Snail in our experimental model, in addition they suggest that the transcription factor may be regulated by the kinase AurkA.
Indeed, we found that MDA-shAurkA and KBr2-shAurkA showed decreased snail in comparison with control cells (Fig. 6A), surprisingly no significant change was found in MCF-AurkA+ cells which showed Snail protein levels similar to control cells (Fig. 6A).
This finding suggest that AurkA likely controls Snail through regulation of its nuclear localization. A co-immunoprecipitation assay confirmed that Snail is a direct target of the kinase in MCF-7 and MDA-MB-231 cells (Fig. 6B). Moreover, immunofluorescence analyses suggested that AurkA-Snail Interaction may affects subcellular localization of Snail. Indeed, we found that AurkA overexpression in MCF-7 cells induced nuclear translocation of Snail, in comparison with control cells showing a moderate staining for Snail in the cytoplasm (Fig. 6C left panel).
Consistent with this new role for AurkA in controlling the stem-like subpopulation through Wnt3a/β -catenin signaling, a direct consequence seems that AurkA may affect chemoresistance and recurrence in cancer patients.

Discussion
In the last decade, growing evidences have strongly supported that a subset of cells within the tumor bulk, referred as Cancer Initiating Cells (CICs) may account for tumor growth, resistance to anti-cancer treatments and metastatic spreads. Hence, the amount of CICs within the tumor bulk dramatically affects survival and outcome of cancer patients.
So far, many studies have been addressed to the identification of "druggable" pathways with key roles in self-renewal and chemoresistance of CICs.
Our study shows a novel role for AurkA in maintaining Breast Cancer Initiating Cells (BCICs) through a pathway involving a Snail-miR-128-wnt3a/β -catenin axis as signaling mechanism (Fig. 6C, bottom picture).
So far oncogenic properties of AurkA has been attributed to its pivotal role during mitosis, our data suggest that AurkA control self-renewal (Fig. 2), metastatic signature and migratory activity of BCICs (Fig. 3).
In this study, we induced ectopic expression of AurkA in MCF-7 cells, a low-aggressive breast cancer cell line with low metastatic potential. We found that AurkA overexpression increased the BCICs subpopulation, identified as CD44 + /CD24 low/− cells ( Fig. 2A), able to grow as mammospheres (Fig. 2C); moreover it was associated with increased wnt3a mRNA and protein levels (Figs 2A and 3A), and a metastatic signature as expression of Twist, Snail, c-Met, CD44v6, MMP-9 and Stat3 (Fig. 3A,C).
Those effects were severely impaired after AurkA knock-down in MDA-MB-231 cells, a very aggressive and metastatic breast cancer cell line. MDA-shAurkA cells lost their metastatic signature and migratory activity (Fig. 3B,D). The amount of BCICs, identified as ALDH-positive cells (Fig. 2B, a and b), dramatically decreased and, surprisingly, they showed low levels of wnt3a mRNA and protein (Figs 2B and 3B).
Altogether those findings strongly suggested that AurkA may control self-renewal and invasive capacity of BCICs, contributing to a worse outcome in breast cancer patients.
This role seems to involve the activation of canonical Wnt3a pathway as suggested by the association between AurkA overexpression and nuclear/cytoplasmic localization of β -catenin (P value = 0.00001) found in 89 breast cancer primary tumors (Fig. 1A) as well as the association between AurkA expression and β -catenin stabilization found in our experimental model (western blot in Fig. 3A,B).
Nuclear β -catenin, previously associated to high aggressive breast cancer cells 32 , its considered a marker for Wnt signaling activation and self-renewal of cancer cells. This hypothesis was further corroborated by the correlation between AurkA and CD44, a marker for breast cancer stem cells (P value = 0.00001) in 26 out 32 breast cancer samples (Fig. 1B).
Those evidences were confirmed also in primary breast cancer. Indeed, we found that tumorigenic KBr2, KBr3 and KBr4 showed higher levels of AurkA, similar to MDA-MB-231 cells. In addition a positive staining for MMP9 and increased Ki67 values suggested that they may have a more aggressive phenotype (Fig. 4A). In contrast, KBr1, displaying AurkA levels similar to MCF-7 cells, failed to be tumorigenic after injection (Fig. 4A).
Moreover, analysis of the CD44-positive cells in KBr2 cells, showed higher levels of Wnt3a and AurkA (Fig. 4B) as well as marked migratory activity and adhesion-independent growth, which were severely impaired after AurkA knock-down as shown in Fig. 4C.
Those findings further confirm the role of AurkA in sustaining self-renewal and aggressive behavior of BCICs through activation of Wnt3a/β -catenin signalling.
Actually, we show a post-transcriptional control of wnt3a by AurkA. Indeed, the kinase overexpression promoted a dramatic decrease of miR-128.
It is well known that miRNAs may influence carcinogenesis acting as oncogenes or tumor suppressors, although the molecular mechanisms are still unclear.
Here we found that miR-128 is an inhibitor of self-renewal and metastatic signature of BCICs, because it inhibits stabilization of wnt3a mRNA, in a fashion that is dependent on AurkA kinase.
However AurkA is not a transcription factor, so it is conceivable that the kinase may regulate miR-128 gene expression controlling a transcription factor. Previous findings suggested Snail as a repressor for miR-128 29 .
Indeed, a luciferase assay confirmed a repressive activity of Snail on miR-128 gene in our system, in addition showed that Snail activity changed accordingly to Aurka overexpression/knock-down (Fig. 6A).
Snail appears to be a direct target of AurkA Indeed, here we show for the first time that AurkA binds to Snail, as revealed by co-immunoprecipitation assay (Fig. 6B). Moreover, immunofluorescence analyses suggested that the kinase may promote nuclear translocation of snail and, as a consequence, its repressive activity on gene expression (Fig. 6C).
To be thorough, we notice that AurkA knock-down in MDA-MB-231 or in KBr2 cells reduced snail mRNA and protein levels (Figs 5C and 6A), however AurkA overexpression in MCF-7 cells was not followed by Snail protein stabilization (Fig. 6A), although they showed increased mRNA levels (Fig. 5D). Therefore, we consider that the effects we have found in Snail stabilization were a side-effect depending on a regulation loop between Snail and miR-128, as it was previously reported 29 .
In light of this study, we suggest that the oncogenic role of AurkA in breast cancers is not only depending by its key role during mitosis, but may include its capacity to control the "hard-core" of the tumor, that is the BCICs in the inner mass which account for response to cancer therapies, metastatic spread and recurrence.
This observation explains the association between AurkA overexpression and the worse clinical outcome previously reported for several human tumors as for breast cancer 21,23,31 .
In conclusion, we want to point out that the pivotal role of Aurka affecting BCICs self-renewal and invasive properties is depending on a Snail-miR-128-Wnt3a/β -catenin axys (Fig. 6C bottom picture). This molecular mechanism is strongly supported also by the strong correlation between AurkA expression and miR-128 levels in 81,25% (26 out 32) (P = 0.04295) of breast cancer patients (Fig. 6D).
Finally, we suggest that targeting this signaling pathway would be helpful as a therapeutic strategy to treat chemo-resistance and metastatic spread in breast cancer patients.

Materials and Methods
Ethics Statement. Investigation has been conducted in accordance with the ethical standards and approved national and international guidelines. The study involving human tissues has been approved by Central Ethic Committee (CEC) at IRCCS Fondazione "Salvatore Maugeri", Pavia. Clinical tumor specimens from post-surgery breast cancer patients were collected by the Institutional Biobank "Bruno Boerci", after that written informed consent was obtained from each patient.
In vivo study carried out with mouse models was performed according to the approved international guidelines, in addition the experimental procedures involving animals were approved by Italian Ministry of Health.
Clinical isolates. Formalin-fixed, paraffin-embedded (FFPE) tumor tissues from a cohort of 89 primary invasive ductal breast carcinomas were obtained. Histological typing, grading, staging and evaluation of estrogen receptor (ER) and progesterone receptor (PgR) status were performed as part of routine diagnostic protocol according to standard histopathologic techniques by the Unit of Pathology at IRCCS.
Cell culture. Culture of breast cancer cell lines, isolation/culture of primary cells, invasion assay, MACS sorting were performed as previously described in ref. [30].
Sphere-forming Assay. Breast  Sphere forming was monitored for 1 week. Mammospheres showing greater than 40 μ m were scored, using an inverted light microscope DM5000B (Leica) equipped with a CCD camera and LAS software (Leica) for picture capture. Experiments were performed as triplicate and MFE (Mammosphere Forming Efficiency) was calculated as reported in 36 . Standard Deviation was calculated for each sample. Representative images were showed at 200X magnification.
Samples were analyzed on Leica DM1000 Microscope (Leica) equipped with LAS Software for image capture and analysis. For each slide a least 50 fields were analyzed.
For Ki67 counting, at least ten randomly selected regions for slides were analyzed and a minimum of 500 nuclei was counted for each sample. Representative images at 200X magnification has been shown.
Samples were analyzed on Leica DM1000 Microscope (Leica) equipped with LAS Software for image capture and analysis. For each slide a least 50 fields were analyzed. Representative images at 200X magnification has been shown.
Western blot analysis. SDS-PAGE was performed as previously described in 37  Q-PCR. Total RNA isolation, preparation of cDNA, Q-PCR and data analyses were performed as previously described in ref. 37. Probes for all genes (AurkA, wnt3a, miR-128, twist, snail, c-Met, CD44, CD44v6 and GAPDH and U6 small nuclear RNA as controls) areTaqMan ® Gene Expression Assay (Thermo Fisher Scientific).

In vivo experiments.
In vivo experimentation carried out with mouse models were realized as previously described in ref. 37.