Tumor Cell-Derived Periostin Regulates Cytokines That Maintain Breast Cancer Stem Cells

This article is featured in Highlights of This Issue, p. 1

Cancer stem cells (CSC) represent a highly malignant subpopulation of tumor cells that have been proposed to drive tumorigenesis, metastasis, disease recurrence and resistance to chemotherapy (1, 2). Breast CSCs (CD44^high/CD24^low/−) were first identified based on their ability to initiate tumor growth and differentiate into non-CSCs when injected at limiting dilution (3). Since then, various other phenotypic characteristics have been associated with breast CSCs including high aldehyde dehydrogenase (ALDH) activity (4), an enhanced capacity to generate mammospheres, a surrogate marker for self-renewal and stem cells (5, 6), and activation of an epithelial–mesenchymal transition (EMT) program (7, 8). The relevance of CSCs to breast cancer is supported by clinical findings that link markers of their presence (i.e., expression of stem cell–associated genes) to a poor disease prognosis (9) and chemotherapeutic resistance (10).

Several signaling pathways have emerged as critical regulators of the CSC state. Many of these signals constitute important developmental programs that, in the case of breast cancer, may act to sustain a population of CSCs (11, 12). Recently, cytokines, namely IL6 and IL8, have been implicated in the maintenance of breast CSCs (13–16). And conditions such as hypoxia have also been shown to promote the appearance of breast CSCs (17). Therefore, in addition to defined genetic mutations, the surrounding tumor microenvironment may play a major role in the promotion of the CSC state.

Basal-like breast cancer (BLBC) is an aggressive molecular subtype of the disease that is enriched for cells with stem-like features such as a CD44^high/CD24^low surface marker profile (18), activation of a mesenchymal transcriptional program (19) and resistance to therapy (20). The relevance and underlying basis for this enrichment are not well understood but it raises the possibility that BLBC pathogenesis could be linked to the creation of a tumor microenvironment conducive to the formation or maintenance of CSCs.

Here, we examined the hypothesis that BLBC cells secrete periostin (encoded by the POSTN gene, also known as OSF-2), a mesenchymal extracellular matrix (ECM) protein, to support a population of CSCs. These studies highlight a role for tumor cell-derived POSTN, which signals through integrin α_vβ₃ to regulate the expression of key cytokines and mediates stemness in BLBC cells.

All M cell lines (MI, MII, MIII) were cultured in DMEM F/12 media containing 5% horse serum, 10 μg/mL insulin, 20 ng/mL EGF, 100 ng/mL cholera toxin, and 0.5 μg/mL hydrocortisone, as previously described (21). SUM159 cell lines were cultured in Ham's F/12 supplemented with 5% FBS, 10 μg/mL insulin, and 0.5 μg/mL hydrocortisone while Hs578T cells were maintained in DMEM containing 10% FBS. All other cell lines were cultured as described previously (22). Mammospheres were grown in MammoCult media (Stem Cell Technologies) containing 0.48 μg/mL hydrocortisone and 1% methylcellulose. All cells were grown in the presence of penicillin/streptomycin (1%) and CO₂ (5%) at 37°C in a humidified incubator.

A detailed list of all antibodies and reagents used is provided in the Supplementary Data.

Stable knockdown of periostin or integrin β3 was achieved by transduction of lentiviral vectors (pLKO.1) followed by selection with puromycin (2 μg/mL). shRNA sequences can be found in the Supplementary Data.

Total RNA was isolated with TRIzol (Life Technologies) and cDNA synthesized using random primers and SuperScript II Reverse Transcriptase (Life Technologies). qRT-PCR was performed with Power SYBR Green PCR Master Mix (Life Technologies) and run on an Applied Biosystems 7900HT Sequence Detection System. Relative RNA expression was calculated using the ΔΔC_t method and normalized to β-actin. Primer sequences are listed in the Supplementary Data.

Secreted periostin was measured with an ELISA kit (AdipoGen) according to the manufacturer's protocol. Protein was concentrated using Amicon centrifugation filters (30K membrane, Millipore) before analysis. To measure cytokine production, cells were starved in serum-reduced media (0.1% serum) and conditioned media collected 48 hours later. IL8 protein was quantified using a Quantikine ELISA Kit (R&D Systems), whereas IL-6 quantification was performed with a CyPlex (CyVek, Inc.).

Protein was isolated from cells on ice with RIPA buffer containing protease and phosphatase inhibitor cocktails (Roche). For examination of signaling events, cells were grown in serum-reduced media (0.1% serum) for 24 or 48 hours before protein isolation. Proteins were separated by SDS-PAGE and immobilized on PVDF membranes. Secondary antibodies were conjugated to peroxidase and detected by chemiluminescence with ECL solution (PerkinElmer).

For examination of cell surface proteins, cells were trypsinized, washed in FACS buffer (PBS with 0.2% BSA, 0.09% sodium azide and 1 mmol/L EDTA) and stained with antibody for 30 minutes at 4°C. To quantify the percentage of ALDH-positive cells, the ALDEFLUOR assay was used (Stem Cell Technologies). Briefly, cells were trypsinized, suspended in assay buffer, and treated for 30 minutes with the ALDEFLUOR reagent (BODIPY-aminoacetaldehyde, BAAA; 3 μmol/L), in the presence or absence of the ALDH inhibitor diethylaminobenzaldehyde (DEAB; 37.5 μmol/L). Samples were analyzed on FACScan, FACSCalibur, LSR II or Accuri C6 (BD) cytometers. Cells were sorted using a different buffer (PBS, 2% horse serum, 1 mmol/L ETDA) with a MoFlo (Beckman Coulter).

Tumor xenograft experiments were conducted under the approval of the Institutional Animal Care and Use Committee at Boston University School of Medicine (Boston, MA). Bioluminescent cells were generated using a pMSCV-Luc-PGK-hygro plasmid (Addgene-8782). Fifty thousand cells were injected subcutaneously, in the absence of Matrigel (in 0.1 mL of serum-free F/12 media), into 6- to 8 week-old NOD/SCID mice (The Jackson Laboratory). Animals were monitored on a weekly basis using an IVIS Spectrum Imaging System (Caliper) for a period of 6 months, unless tumor growth necessitated euthanasia.

Total RNA was isolated using TRizol (Life Technologies) and cleaned up with the RNeasy Mini kit (Qiagen). Biotin labeling was performed using the Ambion WT Expression Kit (Life Technologies) and the GeneChip WT Terminal Labeling and Controls Kit (Affymetrix), followed by hybridization to GeneChip Human Gene 2.0 ST arrays (Affymetrix). Raw CEL files were normalized using the Robust Multiarray Average [PubMed ID 14960456]. Differential gene expression between shPN and shGFP was assessed using the moderated t test implemented in the “limma” package. Raw CEL files from GEO Series GSE21653 were normalized in the same manner, and differential expression between molecular subtypes was assessed using the Welch t test. All microarray analyses were performed using the R environment for statistical computing.

Cells were cotransfected with the 3× KB-L NF-κB luciferase reporter plasmid (Addgene #26699) and a Renilla-luciferase plasmid using X-tremeGENE HP DNA transfection reagent (Roche). Twenty-four hours after transfection, cells were starved in reduced serum medium (0.1% FBS) for 16 hours. TNFα treatment (20 ng/mL, R&D Systems) served as a positive control. For experiments that involved inhibition of the ERK pathway, inhibitors were added 24 hours after transfection and luminescence was measured 24 hours later. Luminescence was quantified with the Dual-Luciferase Reporter Assay (Promega) using a GloMax microplate reader. Firefly luciferase signal was normalized to the Renilla luciferase signal to determine relative luminescence units (RLU).

Statistical analysis was performed using a two-tailed paired Student t test. A P value less than 0.05 was considered statistically significant. Error bars represent ± SEM.

To identify tumor cell-derived factors that could support the maintenance of breast CSCs, we examined a well-established cell line model of breast cancer progression (23, 24). This model system consists of three cell lines: MCF10A (MI) and two derivatives of this line, MCF10AT1k.cl2 (MII) and MCF10CA1h (MIII). These cell lines exhibit increasing degrees of malignancy (24). Previous characterization by our laboratory has indicated that while MI and MII cells display numerous epithelial features, MIII cells clearly exhibit a phenotype associated with activation of an EMT program (21).

Given the connection between EMT and CSCs (7, 8), we wondered whether MIII cells also exhibit features associated with the stem cell state. In support of this notion, MIII cells predominantly displayed the CD44^high/CD24^low surface phenotype (Fig. 1A) and preferentially formed mammospheres, doing so with nearly a 9-fold greater efficiency than MII cells (Fig. 1B). Because the MIII cell line displayed various traits reminiscent of breast CSCs, we reasoned that it could serve as a good model to identify regulators of the CSC state.

Because all of the M model cell lines share the same genetic lineage, we hypothesized that altered expression of secreted microenvironmental factors may contribute to the CSC-like features of MIII cells. Our previous gene expression profiling studies (21) revealed that periostin (POSTN) was overexpressed in the MIII cell line. This was particularly interesting because POSTN is regulated by Twist (25) and TGFβ (26), two factors that promote passage through an EMT and acquisition of stem cell phenotypes (7). We validated that POSTN is a transcriptional target of the TGFβ pathway in this model of breast cancer, as treatment of MII cells with TGFβ1 resulted in a strong increase in POSTN transcription (Supplementary Fig. S1A). Conversely, active TGFβ signaling was necessary to sustain POSTN expression in MIII cells (Supplementary Fig. S1B).

Using qRT-PCR, we confirmed that the level of POSTN mRNA was significantly increased in MIII cells compared with MI or MII cells (Fig. 1C). As expected, this increase in transcription was correlated with enhanced secretion of POSTN into the surrounding media (Fig. 1D). Moreover, we found that POSTN was highly expressed in populations enriched for CSCs. For instance, MIII cells grown as mammospheres, which are predominately formed by CSCs (6), expressed nearly 10 times more POSTN than their adherent counterparts (Fig. 1E). In a second approach, when MII cells were fractionated based on the surface expression of CD44 and CD24, we found that cells within the CD44^high/CD24^low (CSC) population expressed significantly more POSTN than cells within the CD44^high/CD24^high (non-CSC) fraction (Fig. 1F). Overall, in the MCF10A breast cancer model, high POSTN expression directly correlated with multiple phenotypes attributed to CSCs.

In examining gene expression profiles of the M cell lines, we noted that, in addition to POSTN, MIII cells also expressed high levels of integrin β3 (ITGB3), a subunit of the integrin α_vβ₃ complex that is a known receptor for POSTN (27). Indeed, we found that the surface levels of the functional integrin α_vβ₃ complex were abundant on MIII cells but nearly absent from MI and MII cells (Fig. 1G). The coordinated expression of the receptor-ligand combination required for the activity of an intact signaling axis suggested that, in MIII cells, POSTN could potentially signal in an autocrine fashion.

Wishing to expand on this observation, we examined whether such an axis was present in other breast cancer cells. To this end, we measured the mRNA levels of both POSTN and ITGB3 in a panel of nine breast cancer cell lines. A high level of expression of both genes was detected in four cell lines: BT549, SUM1315, SUM159, and Hs578T cells (Fig. 1H). Notably, and similar to MIII cells (21), all four of these cell lines fall under the molecular classification of BLBC cells (22). In a subset of these lines, we verified that the increase in POSTN and ITGB3 expression correlated with enhanced POSTN secretion (Supplementary Fig. S2A) and increased surface levels of the integrin α_vβ₃ receptor complex (Supplementary Fig. S2B). These data indicated that a periostin–integrin β3 signaling axis is preserved, and potentially operative, in a subset of BLBC cells.

To test the relevance of POSTN expression to CSCs, we opted to focus on the SUM159 line because it has a well-defined CSC population (28). We generated SUM159 cells with stable knockdown of POSTN (shPN) or ITGB3 (shBeta3) through expression of corresponding shRNA constructs (Supplementary Fig. S3A and S3B). Knockdown of POSTN or ITGB3 in SUM159 cells did not have an effect on cell proliferation or apoptosis (Supplementary Fig. S3C and S3D) and did not lead to any morphologic changes or reversion of the EMT program. (Supplementary Fig. S3E and S3F). However, POSTN knockdown resulted in a significant reduction in the ability of SUM159 cells to generate mammospheres, with shPN cells forming 72% fewer mammospheres than control (shGFP) cells (Fig. 2A). Similarly, SUM159 shBeta3 cells also exhibited a significant reduction in mammosphere formation potential (Fig. 2A). Notably, the most dramatic effect of POSTN knockdown was observed in the well-established (28), ALDH-positive CSC population of SUM159 cells. Under conditions of reduced serum (0.1% serum), which are often used to analyze the effect of ECM proteins and integrin signaling (29), close to 15% of SUM159 shGFP were ALDH-positive. On the other hand, only 2% of the cells were ALDH positive in SUM159 shPN cells, representing a 7.4-fold reduction in this highly tumorigenic cell population (Fig. 2B). Furthermore, SUM159 shBeta3 cells had a 2.9-fold reduction in the ALDH-positive CSC population (Fig. 2B). Importantly, we verified these findings with an additional shRNA hairpin directed at either POSTN or ITGB3 (Supplementary Fig. S4A–S4C). These results suggest that BLBC cells rely on POSTN and ITGB3 to sustain a population of CSCs.

Additional evidence for this notion was obtained from experiments with the MIII cell line. Here too, knockdown of POSTN (Supplementary Fig. S5A) led to a significant reduction in mammosphere-forming potential; MIII shPN cells generated 52% fewer mammospheres than control cells (Supplementary Fig. S5B). Notably, when MIII cells were sorted based on the surface level of integrin α_vβ₃, there was a dramatic difference in the capacity to initiate mammosphere growth, where only MIII cells with high surface levels of α_vβ₃ could robustly form mammospheres (Fig. 2C). The overlapping phenotypes observed following perturbation of POSTN or ITGB3 imply that these proteins function in a closely related or overlapping pathway to regulate the CSC state in BLBC.

Consistent with the above findings, when shGFP and shPN cells were seeded as mammospheres at limiting dilution we detected a clear reduction in the estimated stem cell frequency, with the sphere formation efficiency of 100 shGFP control cells being roughly equal to that of 1000 shPN cells (Fig. 2D). Importantly, knockdown of POSTN impaired the ability of SUM159 cells to initiate tumors when injected at limiting numbers (5 × 10⁴ cells). Six months after injection, 50% of mice had developed tumors in the control group while no tumors were detected in animals injected with SUM159 shPN cells (Fig. 2E). This result is consistent with our in vitro findings and lend support to the idea that tumor cell-derived periostin is crucial for tumor initiation, the salient characteristic of CSCs.

To understand the molecular events downstream of POSTN, we performed genome-wide expression profiling of SUM159 shGFP and shPN cells and used Gene Set Enrichment Analysis (GSEA) to identify the major pathways that were altered upon POSTN knockdown in SUM159 cells. This analysis revealed six gene sets, containing a preponderance of cytokines, that were significantly repressed in SUM159 shPN cells (Supplementary Fig. S6A). Given the connection between cytokines and breast CSCs (13), we wished to investigate this further. Close examination of the twenty genes that comprise the cytokine pathway gene set highlighted five cytokines that were clearly repressed in SUM159 shPN cells compared with shGFP cells: IL1A, IL8, IL6, IL18, and IL16 (Fig. 3A). Of these five cytokines, IL1A, IL8, and IL6 were found at the leading edge of other downregulated gene sets (Supplementary Fig. S6B), implying that they may represent particularly important mediators downstream of POSTN.

We used qRT-PCR to confirm the repression of IL1A, IL6, and IL8 in SUM159 shPN cells; importantly, the expression of all three genes was also significantly reduced in SUM159 shBeta3 cells (Fig. 3B). Further, conditioned media from shPN and shBeta3 cells contained significantly less IL8 and IL6 protein than media conditioned from shGFP control cells (Fig. 3C and D). Of these three cytokines, IL6 expression was most closely correlated with the expression of POSTN in a panel of six BLBC cell lines (Fig. 3E), suggesting that the regulation of IL6 expression by POSTN may be common in BLBC cells.

In light of the reduced cytokine levels, and because STAT3 is a transducer of cytokine signaling that has been linked to the maintenance of breast CSCs (15), we were prompted to investigate the STAT3 pathway in cells with knockdown of POSTN or ITGB3. Indeed, although SUM159 shGFP cells exhibited relatively high levels of active, phosphorylated STAT3, both SUM159 shPN and shBeta3 cells had markedly reduced levels of phosphorylated STAT3 (Fig. 3F). The impaired transcription of IL6 and IL8, along with reduced activation of STAT3, was also verified using an additional hairpin directed at POSTN or ITGB3 (Supplementary Fig. S4D and S4E).

To test whether impaired production of secreted factors was responsible for this effect, we collected conditioned media from both SUM159 shGFP and shPN cells, and then treated the three cell lines with this or serum-reduced media (mock) for 60 minutes. The diminished activation of STAT3 in the knockdown cells could be rescued by transient exposure to media conditioned from shGFP cells, but not media conditioned from shPN cells (Fig. 3G). These data indicate that impaired production of secreted factors from SUM159 shPN and shBeta3 cells, ostensibly the cytokines mentioned above, is responsible for the observed reduction in STAT3 signaling, rather than any inherent intracellular defects. In addition, we found that treatment of shPN cells with media conditioned from control shGFP cells led to a 1.5-fold increase in the ALDH-positive subpopulation (Supplementary Fig. S7A). This effect was not observed in shBeta3 cells, suggesting that both POSTN and cytokine signaling were required to maintain this population.

Next, we wished to determine the proximal signals downstream of the POSTN–ITGB3 pathway that were responsible for controlling cytokine production. Our approach to achieve this aim was to examine the expression profiles of SUM159 shGFP and shPN cells using Ingenuity Pathway Analysis (IPA) software to look for upstream regulatory networks that might help to explain the differential gene expression profiles between these two cell lines. Interestingly, we noted that, in comparison with the control shGFP cells, SUM159 shPN cells exhibited significant activation of gene expression profiles characteristic of cells treated with two ERK pathway inhibitors (Supplementary Fig. S7B and S7C). This suggested that cells with disrupted POSTN signaling might show impaired activation of the ERK pathway. Consistent with this, cells with knockdown of POSTN or ITGB3 were found to have dramatically reduced levels of phosphorylated, active ERK1/2 (Fig. 4A). This defect in ERK signaling appears to be upstream of cytokine transcription, as pharmacologic inhibition of the ERK pathway led to significantly reduced expression of IL6, IL8, and IL1A (Fig. 4B and Supplementary Fig. S7D). In line with a recent report that linked POSTN signaling to activation of the NF-κB pathway (30), we found that knockdown of POSTN or ITGB3 impaired the transcriptional activity of NF-κB as assessed by a luciferase reporter assay (Fig. 4C). This result was verified using an independent shRNA hairpin directed at POSTN or ITGB3. (Supplementary Fig. S4F). Furthermore, pharmacologic disruption of ERK signaling also caused a decrease in NF-κB transcriptional activity (Fig. 4D).

To test whether the connection between POSTN-controlled cytokine expression and CSCs is applicable outside of the SUM159 line, we also knocked down POSTN in Hs578T cells (Supplementary Fig. S8A), another BLBC cell line (22) that coexpresses high levels of both POSTN and integrin α_vβ₃ (Fig. 1H and Supplementary Fig. S2A and S2B). Consistent with our previous observations, Hs578T cells with knockdown of POSTN formed 57.9% fewer mammospheres (Supplementary Fig. S8B) and showed a 3-fold reduction in the ALDH-positive subpopulation (Supplementary Fig. S8C). The loss of these CSC characteristics was also associated with reduced IL6 expression (Supplementary Fig. S8D) and impaired activation of the STAT3 pathway (Supplementary Fig. S8E). This suggests that the functional relevance of POSTN-regulated cytokine pathways may be more broadly applicable to other basal-like CSCs.

We reasoned that cancers with an operative POSTN-ITGB3 signaling axis might be enriched with CSCs and therefore associated with a worse clinical prognosis. On the basis of the fact that ITGB3 marks a population of luminal progenitor cells, the presumed cell-of-origin for basal-like cancers (31), one would predict that this would be limited to a subset of POSTN-expressing BLBCs, many of which are likely to be positive for ITGB3. In accordance with this notion, in silico Kaplan–Meier analysis of a public breast cancer dataset (32) revealed that high POSTN expression was associated with reduced relapse-free survival in basal-like (HR = 1.42, P = 0.0084), but not luminal A (HR = 1.08, P = 0.4), breast cancers (Fig. 5A). As POSTN has been suggested to promote angiogenesis through activation of integrin α_vβ₃ on endothelial cells (33), it is possible that high POSTN expression could be related to neoangiogenesis, which is also a prognostic factor in breast cancer (34). However, the effect of angiogenesis would seem to be independent of molecular subtype, suggesting that the association with relapse-free survival could be based on an alternative explanation. Indeed, our findings are in line with a recent study that found a correlation between POSTN expression, breast CSC content (as assessed by CD44^high/CD24^low cells), and a worse clinical prognosis (35).

In addition, we examined the expression of POSTN, IL6, and IL8 in another independent microarray dataset of 226 primary breast cancers (GEO Series GSE21653; ref. 36). In this dataset, POSTN was highly expressed (median percentile = 99.6%) across all molecular subtypes (Supplementary Fig. S9), but IL6 and IL8 levels were significantly elevated in BLBC (Fig. 5B), in agreement with previous findings that have implicated these cytokines in the growth and tumorigenicity of BLBC stem cells (15, 16, 37). Overall, these clinical data are consistent with a model wherein POSTN can control the production of cytokines in the context of BLBC.

EMT-induced cancer cell lines have been successfully used to screen for selective inhibitors of CSCs (38) and we reasoned that MIII cells—which clearly underwent EMT and display numerous other CSC traits—could represent a similar cell line model in which to study regulators of the CSC state. We hypothesized that proteins secreted into the local microenvironment would play a particularly important role in the maintenance of breast CSCs. Consistent with this notion, our previous gene expression profiling study (21) found a correlation between TGFβ-mediated EMT and POSTN expression, and here we focused our efforts on the function of this protein in regulation of the CSC state.

As a matricellular protein, POSTN can potentially serve as a crucial mediator between CSCs and their surrounding microenvironment (39). Yet there is limited mechanistic detail as to how POSTN may function in BLBC cells. We found a POSTN–ITGB3 signaling axis is operational in a number of breast cancer cell lines, and in each case, the cells could be classified molecularly under the basal-like subtype. This is notable for a number of reasons. First, numerous lines of evidence support the idea that BLBC arises from transformation of luminal progenitor cells (40–42), which are specifically marked by expression of ITGB3 (43). Second, BLBC cells share numerous phenotypic similarities with CSCs (18–20), which may be especially true of a related disease subtype, claudin-low breast cancers (44, 45). In this context, we suggest that tumor cell-derived POSTN may be particularly important for the maintenance of a population of CSCs in basal-like disease.

In support of this hypothesis, our results from knockdown experiments, carried out in three different BLBC cell lines, showed that POSTN expression was required to maintain CSC phenotypes. These findings complement work from others, which showed that ectopic expression of POSTN promotes stem-like properties in human mammary epithelial cells (46). Our data also support a major role for the integrin α_vβ₃ receptor in sustaining mammosphere growth and an ALDH-positive subpopulation. Another study has addressed the relationship between integrin β3 and CSCs and reported that, while ITGB3 is highly expressed in breast CSCs, knockdown of this gene resulted in minimal effects on mammosphere growth (47). Interestingly, these experiments were performed in derivatives of SK-BR-3 cells, a luminal cell line and, as argued above, this specific integrin may be more relevant in the context of BLBC.

We also sought to identify the pathways downstream of the POSTN-ITGB3 signaling axis that are relevant in the context of CSC regulation. Here, we focused on a defined cytokine network, consisting of IL6, IL8, and IL1A, which was repressed following POSTN knockdown. Interestingly, other gene sets significantly repressed in shPN cells included the hematopoietic lineage and stem cell gene sets. As POSTN has been found to localize to stem cell niches in the bone marrow (48), this suggests that POSTN-mediated regulation of cytokine production might also have physiologic relevance to hematopoietic stem cells. Furthermore, our observations are also consistent with a report that demonstrated that POSTN was required for the production of multiple proinflammatory cytokines during chronic skin inflammation (30). Thus, in the context of cancer, breast CSCs may activate a similar program downstream of POSTN. We also found that cells with defective POSTN signaling showed reduced activation of ERK signaling and, consequently, impaired cytokine transcription. This result is in accordance with a recent study, which reported that activation of the MAPK pathway due to DUSP4 loss promoted the formation of CSCs and increased the expression of IL6 and IL8, in this case through the transcription factors ETS-1 and c-JUN (49). Our data suggest that, at least downstream of POSTN, the NF-κB pathway is likely to be responsible for driving cytokine expression, similar to what has been observed during skin inflammation (30). Nevertheless, NF-κB transcriptional activity appears to lie downstream of the ERK pathway, raising the possibility that POSTN-mediated regulation of the CSC state may be susceptible to interference by MAPK inhibitors.

Overall, this leads to a model in which the POSTN-ITGB3 signaling axis regulates the production of cytokines, which in turn activate STAT3 signaling. This model is compelling because of the rapidly accumulating evidence that links cytokines, especially IL6 and IL8, to the regulation of breast CSCs (15, 16, 28, 50). These alterations may be further understood when considering recent studies that have reported on the cellular plasticity that exists between non-CSCs and CSCs (14, 51). For instance, it has been shown that the IL6-STAT3 pathway can shift this equilibrium in favor of CSCs (14). Thus, one possibility may be that POSTN, at least in part through control of cytokine production, is able to tilt this equilibrium to promote the formation and/or maintenance of CSCs.

In addition to the experiments presented here, others have recently found that POSTN plays a role in the regulation of breast CSCs during the process of metastatic colonization (48). It has also been shown to act as a factor derived from neovascular tip cells in the perivascular niche, where it can serve to break tumor cell dormancy (52). Our work, while consistent with the central idea that POSTN is required for the maintenance of CSCs at sites of tumor cell dissemination, also differs in important aspects. First, we found that, in a subset of basal-like cancer lines, POSTN can be tumor cell derived, potentially allowing these cells to generate their own functional niche and rendering them more independent from their surrounding microenvironment. Similarly, POSTN has recently been reported to be secreted by glioblastoma stem cells (53). Second, our data highlight a major role for integrin α_vβ₃ in facilitating POSTN signaling, which represents an important regulatory pathway for the maintenance of breast CSCs. Indeed, a recent study found integrin β3 to be a marker of CSCs and a mediator of resistance to targeted therapy, not only in breast cancer, but also in lung and pancreatic cancers (54). Therefore, it is conceivable that periostin could contribute to CSC maintenance and resistance to therapy in breast as well as other carcinomas. Finally, our study implies that a POSTN signaling network can, in principle, act in the primary tumor as well, helping to maintain a fraction of CSCs at earlier stages of tumor progression (Fig. 6). Therefore, we suggest that a subset of BLBCs can establish their own microenvironmental niche supportive of breast CSCs through the production of tumor cell-derived POSTN, which might be clinically relevant as a biomarker or therapeutic target.

No potential conflicts of interest were disclosed.

Conception and design: A.W. Lambert, S. Ozturk, C.K. Wong, P. Papageorgis, S. Thiagalingam

Development of methodology: A.W. Lambert, C.K. Wong, S. Ozturk, R. Raghunathan, B.M. Reinhard, H.M. Abdolmaleky, S. Thiagalingam

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.W. Lambert, C.K. Wong, S. Ozturk, P. Papageorgis, R. Raghunathan, Y. Alekseyev, S. Thiagalingam

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.W. Lambert, C.K. Wong, S. Ozturk, P. Papageorgis, A.C. Gower, B.M. Reinhard, S. Thiagalingam

Writing, review, and/or revision of the manuscript: A.W. Lambert, C.K. Wong, S. Ozturk, P. Papageorgis, B.M. Reinhard, H.M. Abdolmaleky, S. Thiagalingam

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.C. Gower

Study supervision: S. Thiagalingam

The authors thank Drs. David Sherr and Ramon Parsons for generously providing reagents.

This work was supported by grants from Susan G. Komen for the Cure (KG081435) and NIH/NCI (CA165707) (to S. Thiagalingam) and NIH/NCI (CA138509) (to B.M. Reinhard and S. Thiagalingam). A.W. Lambert is a recipient of a predoctoral traineeship award from the Department of Defense, Breast Cancer Research Program (W81XWH-11-1-006). P. Papageorgis was supported with a postdoctoral fellowship from the Research Promotion Foundation, Cyprus (DIDAKTOR/0609/24). This work was also supported by a seed grant from the Boston University Genome Science Institute as well as the Boston University Flow Cytometry Core Facility, and the core facilities of the Boston University Clinical and Translational Science Institute (CTSA award UL1-TR000157).

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.