PDLIM2 Is a Marker of Adhesion and β-Catenin Activity in Triple-Negative Breast Cancer Free

Triple-negative breast cancers (TNBC) lack estrogen receptor, progesterone receptor (ER⁻, PR⁻), and amplification of human epidermal growth factor receptor 2 (HER2⁻), and constitute 15% to 20% of all breast cancers. TNBC displays considerable genetic heterogeneity, a high risk for distant metastasis, is refractory to many therapies, and often has a poor prognosis (1–5). Although many studies have helped define subtypes within TNBC, the underlying drivers of this cancer are still unclear, and there is an urgent need for better biomarkers and therapeutic targets (5–7).

Growth factor and adhesion signaling have been strongly implicated in TNBC (8–11). Our studies on insulin-like growth factor-1 (IGF1) signaling in cancer led us to identify the PDLIM2 (Mystique) protein (12), as a feedback regulator of IGF1 and adhesion signaling (12–14). PDLIM2 is a member of the PDZ-LIM domain family encoded by a locus on chromosome 8p21 (12, 15, 16), a region that is associated with metastasis and often disrupted in cancer (17). Located at the cytoskeleton and nucleus of epithelial cells and hemopoietic cells, PDLIM2 regulates the stability and activity of important transcription factors including STATs, NF-κB, and IRFs (12, 18–21). PDLIM2 can suppress cellular transformation (12, 13, 16), and it may be repressed by methylation in breast cell lines (22, 23). PDLIM2 inactivation has also been reported in classical Hodgkin and anaplastic large cell lymphoma (24). However, PDLIM2 is robustly expressed in breast and prostate cancer cells that exhibit a migratory phenotype (13, 16), and sustains a transcription programme for epithelial to mesenchymal transition (EMT; ref. 21). In line with this, PDLIM2 expression has been linked to highly aggressive ovarian cancer (25), lymph node metastasis in low-grade breast cancers (26), and was identified as a moderate dependency gene in basal breast cancers (27). It has also been linked to the growth of neurofibromatosis 2 (NF2)-associated menginioma and schwannoma (28).

In the nontumorigenic MCF10A breast epithelial cell line, PDLIM2 is required for maintaining cell polarization and 3D acinar formation (14), and PDLIM2 expression increases upon retinoic-acid induced differentiation of breast cancer cells (23). However, although PDLIM2 expression has been linked with tumor suppression and with sustaining a migratory phenotype, it is still unclear what it contributes to the progression or phenotype of any subgroup of breast cancer. Here, we investigated this by assessing PDLIM2 expression in different cohorts of breast tumors and manipulating its expression in cell lines. We found that PDLIM2 is expressed in a subset of TNBCs (50%–60%), and in approximately 20% of other breast cancer subtypes. In TN tumors and cell lines, cytoplasmic PDLIM2 promotes the accumulation of active β-catenin. Our study suggests that PDLIM2 mediates adhesion and growth factor signals derived from the tumor microenvironment to enhance β-catenin activation in a large proportion of TNBCs.

Cell lines were purchased from ATCC, except CAL-51 (DSMZ) and authentication established by PCR-single-locus-technology (Eurofins Medigenomix, Forensik GmbH, Ebersberg, Germany), up to November 2018. MDA-MB-231-LUC2 cells were purchased for in vivo experiments (Caliper Life Sciences). All cells were tested monthly for mycoplasma by specific DNA staining and maintained as previously described (14, 29, 30), up to 6 to 8 weeks for use in experiments. Further details can be found in Supplementary Materials and Methods: Tables 1, 3.

For serum starvation and growth factor stimulation, cells were incubated in serum-free medium for 4 hours prior to stimulation, or not, with 10 ng/mL TGFβ1 or IGF1. For non-adherent conditions, cells were trypsinized, centrifuged, and washed prior to re-suspension in complete medium into 50 mL tubes. Cells were incubated for 4 hours at 37°C, with gentle rotation. Alternatively, cells were seeded on petri dishes coated with 100 μL/cm² of Polyhema (6 mg/mL; Sigma).

For wound healing assays, confluent monolayer cultures were wounded by scoring with a sterile pipette tip, and imaged at 0 and 24 hours post-wounding. Wound widths were measured as described in Supplementary Materials and Methods.

Colony formation was measured by assessing plating efficiency. MDA-MB-231-LUC2 cells were seeded at 5 × 10² cells/well in six-well plates, (minimum triplicates per clone). After 9 to 10 days, cells were stained with 0.01% crystal violet and colonies of more than 50 cells were counted.

For 3D on Top assays, cells were cultured on matrigel with matrigel-supplemented medium, as described by Kenny and colleagues (31) and monitored over 4 days. Entire 3D cultures were extracted for Western blot analysis of protein expression as described (31).

Two separate clones of MDA-MB-231-LUC2 cells stably expressing shRNA scramble control or shRNA targeting PDLIM2 (1 × 10⁶ cells) were injected into the tail vein of 4- to 5-week-old female HsdOla:MF1-Foxn1^nu mice (Harlan; n = 6–7 per group, randomized; three separate cohorts). The Bioluminescent signal (luminescence counts) per animal per cell clone and images were collected after 45 to 49 days by using the IVIS Spectrum in vivo imaging system (PerkinElmer). These experiments were performed under licence to R. O’Connor from the Irish Department of Health and protocol approval by University College Cork Animal Experimentation Ethics Committee (#2012/008). Further details are in Supplementary Materials and Methods.

Total cell protein was extracted using RIPA buffer and subcellular fractionations were performed as described previously (20, 32). Western blot analyses were performed using the Odyssey Image scanner system and protein expression levels were normalized to protein loading control by densitometry using Licor Image Studio software as previously described (14, 20, 21, 32).

All tissues were stained using a PDLIM2 mouse monoclonal antibody (Abcam). Details of IHC staining procedures used for each sample collection are described in supplementary methods.

Breast Cancer Tissue Microarrays were generated from Formalin-fixed paraffin-embedded primary tumor blocks as previously described (33). Each tumor was represented by three independent cores. Breast Cancer subtypes were determined from biomarker expression, and classified according to St Gallen International Expert Consensus (34) as: luminal A (ER and/or PR positive, HER2 negative); luminal B [ER and/or PR positive and HER2 negative (HER2−) or HER2 overexpressed/amplified (HER2+)]; HER2 enriched (nonluminal, ER and PR negative and HER2 overexpressed/amplified); basal-like/triple negative (ER, PR, and HER2−).

Slides from the Netherlands Cancer Institute (NKI) and Cambridge University (CAM) contained tissue cores from individual patients within the cohorts (35), and we analyzed IHC staining of tumor samples from 128 patients with TNBC. Further details on RATHER cohorts analyses are in Supplementary Materials and Methods.

Breast Cancer Tissue sections were prepared from TNBC biopsies at the University Medical Centre Mannheim as described previously (36).

Further details are in Supplementary Materials and Methods. Written informed consent and ethical approval was obtained at each Institution (33, 35, 36).

RNA extraction, cDNA synthesis, and quantitative PCRs were performed as previously described (21), with details and primer sequences in Supplementary Materials and Methods and Supplementary Table S2.

MDA-MB-231-LUC2 cells were transfected with pSUPER vectors encoding shRNAs targeting PDLIM2 (shPDLIM2) or control shRNA (shScramble), and BT549 cells with pcDNA3-HA-Empty Vector (EV) or HA-PDLIM2 using Lipofectamine 2000. Stable clones were generated by selection in G418 (12, 21). HEK293T cells were transiently transfected with HA-EV or HA-PDLIM2 using calcium phosphate protocol (32). HCC1806 were transiently transfected with siRNA negative control and 2 different siRNAs targeting PDLIM2 using oligofectamine (12). Further details are in Supplementary Materials and Methods.

HEK293T cells grown on 96-well OptiPlate microplates (PerkinElmer) were transfected with the TOPflash plasmid, which contains three copies of the Tcf/Lef sites upstream of a thymidine kinase (TK) promoter and the firefly luciferase gene, and the FOPflash, which contains mutated copies of Tcf/Lef sites using Lipofectamine 2000. Cells were cotransfected with 0.2 μg of internal control reporter Renilla reniformis luciferase construct (pRL-TK; Promega) to normalize transfection efficiency, and luciferase activity was measured using a dual luciferase Assay System Kit (Promega).

Cells were fixed with 4% paraformaldehyde, probed with primary antibodies followed by Alexa 488- or Cy3-conjugated secondary antibodies and/or TRITC-phalloidin. Nuclei were visualized with Hoechst dye and imaged as described previously (12, 14), and in Supplementary Materials and Methods.

RNA was purified, amplified, labeled, and hybridized to the Agendia custom-designed whole genome microarrays (Agilent Technologies) and raw fluorescence intensities quantified, as previously described (35). GEO accessions are GSE66647 (RPPA) and GSE68057 (RNA microarray expression). Further details of analyses of the RATHER cohorts are in Supplementary Materials and Methods.

Kaplan–Meier survival analysis was performed using the SPSS statistics software (IBM). Univariate analysis was performed using the Cox regression model to illustrate the relationship between PDLIM2 protein expression and breast cancer-specific survival (BCSS), recurrence-free survival (RFS), and distant recurrence-free survival (DRFS), respectively. Hazard ratios (HR) and 95% confidence intervals (95% CI) were evaluated for each survival outcome using the Cox regression model.

Graphs were prepared using GraphPad Prism. Data were analyzed for statistical significance using Fisher exact test, one-way ANOVA or Student t test as indicated in legends. A P value of <0.05 was deemed significant and graded P values are denoted as follows: *, P ≤ 0.05; **, P ≤ 0.005; ***, P ≤ 0.0005, unless otherwise noted.

To gain insight into how PDLIM2 contributes to human breast cancer, we first assessed its expression in tissue microarrays (TMA) using IHC scored as negative (0), low (1), moderate (2), or high (3). Images representing the relative intensities of these IHC staining scores are shown in Supplementary Fig. S1A. In a TMA representing 248 breast tumors (Northern Ireland Biobank; NIB), PDLIM2 was expressed in 25.8% of all tumors (Fig. 1A). By separating these tumors using the Gallen scores of tumor classification (34), it was clear that the majority of Luminal A (HER2-negative), Luminal B-HER2-negative or -positive and HER2-enriched tumors did not express PDLIM2 (Fig. 1B and C). However, within the basal/TNBC samples more than half (53%) of tumors expressed PDLIM2 (Fig. 1B and C). Interestingly, PDLIM2 expression in tumor-associated stroma or tumor-infiltrating leukocytes was evident in 30% to 40% of all non-TNBC types in the TMA, and again, expression was proportionately higher in TNBC (61.8%; Fig. 1D and E; Supplementary Fig. S1B and S1C). Overall, 70% of PDLIM2-positive tumors (all subtypes) had PDLIM2-positive stroma, whereas within PDLIM2-negative tumors, 50% of TNBC and 30% of non-TNBC had PDLIM2-positive stroma (Supplementary Fig. S1C). PDLIM2 staining in TNBC and non-TNBC TMAs are shown in Fig. 1F and Supplementary Fig. S1D, respectively.

In summary, PDLIM2 is expressed in more than 50% of TNBC tumor cells and 60% have PDLIM2-positive stroma, whereas in other breast cancer subtypes, it is expressed in approximately 20% of tumors and 40% of stroma.

PDLIM2 association with TNBC was further tested in additional TNBC cohorts. These were from the RATHER consortium (www.ratherproject.com), which includes cohorts from the Netherlands Cancer Institute (NKI, n = 71) and the University of Cambridge, UK (Cam; n = 57). We also tested a smaller cohort (n = 15) from Heidelberg University. As can be seen in Fig. 2A, PDLIM2 was expressed in 53% of TNBC from the NIB cohort (data from Fig. 1), 72% of the NKI cohort, 47% of the CAM cohort and 70% of the Heidelberg cohort (Fig. 2A and B). The distribution of weighted IHC scores for PDLIM2 staining is shown in Supplementary Fig. S2A. Overall, PDLIM2 was expressed in 60% of TNBCs across the four cohorts (Fig. 2C), and was low or absent in neighboring normal tissue, as shown in Supplementary Fig. S2B (Heidelberg cohort). Interestingly, PDLIM2 is present in immune cells (mostly lymphocytes) surrounding epithelial ducts in this normal tissue (Supplementary Fig. S2B).

We next asked whether PDLIM2 expression could be correlated with survival and/or clinicopathological features. No significant correlation was found with PDLIM2 expression and overall survival in patients with TNBC from the NIB cohorts, and in the RATHER cohorts, PDLIM2 was not significantly correlated with outcome (breast cancer specific-, recurrence free-, or distant recurrence-free survival; Supplementary Fig. S2C). Similar trends were observed when data were segregated into weighted scores, and there was no significant association with histological grade, number of positive lymph nodes, or tumor size (Supplementary Fig. S2C).

To test whether PDLIM2 is associated with specific TNBC subtypes, we applied the 80 gene signature from Burstein and colleagues (6) to the RATHER cohorts (Fig. 2D). The majority of tumors were classified as mesenchymal (MES), but there was no significant correlation of PDLIM2 with any subtype (Fig. 2E; Supplementary Fig. S3A and S3B). However, trends indicated higher PDLIM2 expression in the Basal-like Immune Activated (BLIA) and Suppressed (BLIS) subtypes (the latter of which is associated with poorest outcomes; ref. 6), compared with MES or luminal androgen receptor (LAR) subtypes (Supplementary Fig. S3B and S3C). Interestingly, in these analyses, we noted that PDLIM2 mRNA levels were not strongly correlated with PDLIM2 protein expression (Supplementary Fig. S3D).

Taken together, the data demonstrate that PDLIM2 is expressed in approximately 60% of tumor cells derived from four cohorts of TNBC, but is not definitively linked to TNBC subtypes or clinical outcomes.

Because PDLIM2 subcellular location has previously been associated with integrating adhesion signals with transcription factor stability/activity (12, 20, 21), we next analyzed its location in TNBC tissue. It was clear that PDLIM2 is predominantly expressed at the cytoplasm/membrane and excluded from the nucleus in tumor cells (Fig. 3A). By quantifying PDLIM2 staining at the cytoplasm/membrane or also in the nucleus (nuclear-only expression was not observed in tumor cells), it was evident that PDLIM2 was restricted to cytoplasmic/membranous areas (Fig. 3B and C). This is in contrast to high levels of nuclear PDLIM2 in stromal cells (Fig. 1F insets; Supplementary Figs. S2B and S3E).

To determine functional significance for cytoplasmic PDLIM2 in TNBC, we used available reverse phase protein array (RPPA) and RNA profiling data for the RATHER cohorts, and asked whether PDLIM2 expression was associated with signaling pathways, or with transcription factor activity. Included in these analyses were 63 proteins and 139 genes with known or anticipated links to PDLIM2 function (Supplementary Tables SI and SII; refs. 12, 14, 18, 20, 21). The Wilcoxon rank sum test was used to determine significance. RPPA data showed that high levels of phospho-β-catenin (serine 675; active) were significantly correlated with PDLIM2-positive tumors (Fig. 3D; Supplementary Table SI). Phospho-MET (Tyrosine 1349) and the adhesion-associated proteins fibronectin and FAK were also high in PDLIM2-positive tumors. In contrast, mTOR and Ki67 were low in PDLIM2-positive tumors (Fig. 3D). Interestingly, although the expression levels of several genes previously shown to be regulated by PDLIM2 (21), were significantly different in PDLIM2- negative and -positive tumors, mRNA levels corresponding to the differentially expressed proteins within the RPPA data were not significantly different (Supplementary Table SII and SIIb).

Overall, these data demonstrate that PDLIM2 expression in TNBC is restricted to the cytoplasm and this positively correlates with β-catenin activity and adhesion signaling.

To further test PDLIM2 association with specific signaling pathways in TNBC, we interrogated RNAseq and RPPA datasets available for approximately 80 breast cell lines (30). PDLIM2 is not represented in the RPPA analysis but is present in the RNAseq data. Given the observed lack of correlation between PDLIM2 mRNA and protein expression in the tumor samples (Supplementary Fig. S3D), we first tested whether PDLIM2 mRNA could be correlated with protein expression in cell lines, using eight TNBC cell lines from the Marcotte study (30), and the TN, nontumorigenic breast cell line, MCF10A (classifications in Supplementary Fig. S4A). In agreement with observations in tissues, we found that a subset of breast cell lines expressed high levels of PDLIM2 protein, whereas others did not (Fig. 4A and B).

Analysis of the RNAseq dataset from the TN cell lines (30) demonstrated that PDLIM2 RNA levels do not significantly correlate with protein expression (Supplementary Fig. S4B). The PDLIM2 gene encodes several RNA variants including PDLIM2 RNA variants 1, 2, and 3 (NM_198042, NM_021630, NM_176871; also referred to as Mystique 1, 2, 3). We previously reported that PDLIM2 protein is encoded by PDLIM2 variant 2 (12, 16), and in the nine TN cell lines, levels of PDLIM2 variant 2 mRNA correlated well with protein (Fig. 4C), whereas variant 3 mRNA was generally inversely correlated with protein (Fig. 4D). PDLIM2 variant 1 was barely detectable. Thus, because RNAseq data may include all PDLIM2 RNA transcripts, it cannot be used to infer PDLIM2 protein expression. Moreover, because PDLIM2 mRNA levels did not correlate with differentially expressed proteins in the RATHER cohorts (Fig. 3; Supplementary Tables SI and SII, SIIb), our analyses indicate that PDLIM2 mRNA expression profiles in cancer cohorts do not reflect protein expression or function. This disparity may also explain why total PDLIM2 mRNA does not correlate strongly with clinical outcomes.

Using the nine TN breast cell lines, we next asked whether PDLIM2 protein correlates with adhesion signals and β-catenin, as was observed in the RPPA data (Fig. 3D). β1-integrin was tested as a marker of extracellular matrix (ECM) adhesion, and we previously reported that PDLIM2 is essential for feedback regulation of β1-integrin signals (14). E-Cadherin and β-catenin are markers of cell–cell adhesion. β1-integrin was observed to be generally higher in PDLIM2-positive than PDLIM2-negative breast cell lines (Fig. 4E and G; Supplementary Fig. S4C). Furthermore, E-cadherin and β-catenin were also expressed at higher levels in PDLIM2-positive than PDLIM2-negative cells (Fig. 4E–G), with the exception of MDA-MB-231 cells where E-cadherin is suppressed by methylation (37). In addition, β-catenin phosphorylated on Serine 675 was higher in PDLIM2-positive TN cells (Fig. 4E). Furthermore, levels of phospho-EGFR and phospho-IGF1R were high in PDLIM2-positive TN cells (Supplementary Fig. S4D and S4E), although there was no clear correlation between phospho-c-MET and PDLIM2 expression (Supplementary Fig. S4F).

Overall, we conclude that PDLIM2 expression is associated with high levels of cell adhesion markers, active growth factor receptors, and β-catenin.

Although PDLIM2 is restricted to the cytoplasm of TN tumors (Figs. 1–3), it could be observed in the cytoplasm and nucleus of cultured cell lines by both immunofluorescence and subcellular fractionation (Fig. 5A and B; Supplementary Fig. S5A and S5B). These observations suggest that cell adhesion and growth factors promote PDLIM2 shuttling between the cytoplasm and nucleus, so we tested this further. We found that under nonadherent culture conditions, PDLIM2 accumulates in the nucleus (Supplementary Fig. S5B and S5C). De-adhesion of MCF10A cells for 24 hours was sufficient to reduce PDLIM2 in the cytoplasm and induce accumulation in the nucleus, even under serum-starved conditions (Fig. 5C and D; Supplementary Fig. S5B and S5C). Re-adhesion of cells for up to 24 hours was sufficient to restore cytoplasmic PDLIM2 with a concomitant decrease in nuclear levels (Fig. 5C). Thus, cell adhesion is necessary and sufficient for nuclear exclusion of PDLIM2.

Serum starvation caused markedly decreased nuclear and increased cytoplasmic PDLIM2 (Fig. 5E–H; Supplementary Fig. S5C). Stimulation of MCF10A with either TGFβ (Fig. 5E, G, and H) or IGF1 (Fig. 5F and G) promoted translocation of PDLIM2 into the nucleus over time. Interestingly, the substantial retardation of cytoplasmic PDLIM2 protein mobility, which we previously confirmed as serine phosphorylation (20), was less evident in the nuclear fractions (arrows; Fig. 5E and F), indicating that PDLIM2 serine phosphorylation facilitates it cytoplasmic sequestration. Overall, these data show that adhesion and growth factor signaling in TN breast cells control PDLIM2 phosphorylation and subcellular localization.

Because PDLIM2 expression correlates with active β-catenin (Fig. 3D, Fig. 4E), we next asked whether PDLIM2 enhances β-catenin activity by using two approaches. First, we assessed active β-catenin in cell lines using an antibody that detects β-catenin only when it is not phosphorylated on serine 45 and therefore active (38). This showed that PDLIM2-expressing cells generally express active β-catenin (Fig. 6A). Exceptions were MDA-MB-231 cells, which express PDLIM2 in the presence of low β-catenin, and HS578T, which do not express PDLIM2 but have high active β-catenin expression. Similar results were observed using the antibody that detects active β-catenin phosphorylated on serine 675 (Fig. 4E; Supplementary Fig. S5D; refs. 38, 39). Immunofluorescence staining illustrated that active β-catenin is present at the plasma membrane, throughout the cytoplasm and in the nucleus of PDLIM2-positive cells (Fig. 6B). Interestingly, in cells with low PDLIM2, active β-catenin is predominantly at the plasma membrane (Fig. 6B).

The second approach was to test the effects of ectopic HA-PDLIM2 expression on β-catenin activity. In BT549 cells, HA-PDLIM2 was mostly evident in the cytoplasm and at the actin cytoskeleton (Supplementary Fig. S5E). Higher total and active β-catenin levels (Fig. 6C; Supplementary Fig. S6A) were observed in the cytoplasm and nucleus (Supplementary Fig. S6A and S6B), and the migratory potential was enhanced compared with controls (Supplementary Fig. S6C). Furthermore, ectopic expression of PDLIM2 was sufficient to activate β-catenin in HEK293T cells in luciferase reporter assays (Fig. 6D and E). PDLIM2 suppression had minor effects on the levels of β-catenin and adhesion proteins, in HCC1806 cells and MDA-MB-231-LUC2 cells (Supplementary Fig. S6E and S6F).

Because active β-catenin is associated with PDLIM2 in TNBC tissues and cell lines (Figs. 3 and 6A and B), and PDLIM2 is restricted to the cytoplasm of TNBC tumor cells (Fig. 3), we next asked whether the subcellular localization of PDLIM2 is important for β-catenin expression and activation. To test this, we used cell detachment to induce nuclear accumulation of PDLIM2 and serum starvation to induce nuclear exclusion, and then measured β-catenin levels and phosphorylation (activity) status in PDLIM2-positive and -negative cells. As expected (Fig. 5; Supplementary Fig. S5; refs. 12, 20), cells in suspension exhibited less cytoplasmic PDLIM2 than adherent cells, with a concomitant accumulation of nuclear PDLIM2 (Fig. 6F; Supplementary Fig. S6G and S6H). These cells showed a clear reduction in active β-catenin levels in both the cytoplasm and nucleus that was not evident in PDLIM2-negative cells. Serum starvation had little effect on levels of active β-catenin, indicating that nuclear PDLIM2 is not directly involved in regulating β-catenin levels or activity (Supplementary Fig. S6H and S6I). We also confirmed that overexpressed HA-PDLIM2 accumulated more in the nucleus than in the cytoplasm in non-adherent BT549 cells, and this was accompanied by a marked reduction in active β-catenin, especially in the nucleus (Fig. 6G).

Overall, we conclude that PDLIM2 accumulation in the cytoplasm (and out of the nucleus) enhances nuclear levels of active β-catenin.

In a previous study, PDLIM2 suppression in DU145 prostate cells was shown to reverse the EMT phenotype characterized by re-expression of epithelial markers and loss of directional migration (21). In that study we also showed MDA-MB-231-LUC2 cells with stably suppressed PDLIM2 exhibited enhanced proliferative rates and reduced growth in soft agarose compared with controls. As can be seen in Fig. 7A and B, two clones of MDA-MB-231-LUC2 cells stably expressing shRNA targeting PDLIM2 (shPDLIM2) exhibited greatly reduced colony formation (plating efficiency) compared with controls (shScr). To test the in vivo significance of PDLIM2 suppression in TNBC, we assessed in vivo tumor spread of these MDA-MB-231-LUC2 clones in nude mice. Each of the shPDLIM2 clones exhibited little if any tumor burden after 45 to 49 days compared with controls, noting that control shScr 1 cells exhibited better growth in vivo than shScr 2 (Fig. 7C). This result is consistent with reduced clonogenic growth and migratory potential previously observed with PDLIM2 suppression (21), and suggests that high PDLIM2 may facilitate breast cancer progression.

To further test PDLIM2 function, we assessed the effects of PDLIM2 overexpression on BT549 3D cell growth using 3D “on top” cell culture assays (31). In control BT549 cultures, cells formed stellate 3D structures that have previously been associated with an invasive phenotype (31). However, BT549 cells stably expressing HA-PDLIM2 with higher active β-catenin, initially formed more rounded, small clustered structures than controls at day 2, and failed to fully adopt the stellate morphology by day 4 (Fig. 7D and E). These results indicate that PDLIM2 regulates the pathway controlling the stellate phenotype and also indicates that β-catenin activity is regulated during this reversible process.

Although several studies have characterized distinct types of TNBC (1–3, 6, 7), there remains an urgent need for biomarkers with prognostic value or to predict TNBC subgroups that would be amenable to novel therapies. Here, we found that the PDLIM2 protein is enriched in TNBCs (overall approximately 60%), whereas it is less frequently expressed in other breast cancers. Furthermore, the correlation of PDLIM2 expression with β-catenin activity and adhesion signaling suggests it defines a large cohort of TNBC with similar features. A direct link between PDLIM2 expression in TNBC and clinicopathological parameters was not established. However, this may be due to TNBC heterogeneity and the fact that PDLIM2 mRNA profiles in available TNBC databases actually represent several PDLIM2 mRNA variants, whereas only one mRNA variant (2) encodes protein. Our conclusion that the PDLIM2 protein is a marker for a large subset of TNBC is further supported by published genomic studies, including a gene dependency screen, that identified PDLIM2 as a moderate dependency gene in basal TNBC and functionally linked this to cell adhesion (27). Interestingly, this study also reported that basal TNBC is addicted to proteasomal activity, whereas genes encoding components of the proteasome and the COP9 signalosome (CSN) have been described as essential in TNBC basal A cell lines (30). These observations are all consistent with the reported actions of PDLIM2 in shuttling between the cytoskeleton and nucleus to regulate transcription factor stability (19, 21, 40), via association with the COP9 signalosome (21).

Suppression of PDLIM2 in MDA-MB-231 (and DU145) cells is sufficient to impair migratory potential and clonogenic growth (21), and as shown here, to impair in vivo spread of MDA-MB-231-LUC2 cells, whereas re-introduction of PDLIM2 into BT549 TNBC cells alters their invasive growth in 3D cultures and enhances directional migration. The exclusion of PDLIM2 from the nucleus of tumor cells, and the effects of adhesion and growth factors on promoting PDLIM2 translocation from the cytoplasm to nucleus in cell cultures, indicate that adhesion signaling promotes cytoplasmic retention of PDLIM2 in TNBC to regulate its function. Moreover, TN cell lines expressing cytoplasmic PDLIM2 generally also expressed high levels of β1-integrin, E-cadherin, β-catenin, and phosphorylated IGF1 or EGF receptors. However, although PDLIM2 suppression enhanced expression of integrins and ECM proteins in a MCF10A 3D model (14), suppression of PDLIM2 in TNBC cell lines did not have a significant effect on the levels of β1-integrin, for example. This may be expected considering the high levels of adhesion proteins in PDLIM2-expressing TNBC cells, and that PDLIM2 is itself regulated by, and a component of adhesion signaling in these cells. Overall, we propose that PDLIM2 functions to maintain a polarized phenotype associated with cell–cell adhesion in normal epithelial cells, but in cancer cells, it is a component of adhesion signaling that promotes loss of cell polarization, tumor growth, and motility.

Our findings indicate that β-catenin activity is a key output of cytoplasmic PDLIM2 in TNBC. Levels of active β-catenin were significantly elevated in PDLIM2-positive tumors and ectopic expression of PDLIM2 was sufficient to induce β-catenin expression and activity in cell lines. This is consistent with suppression of PDLIM2 in EMT-like cells causing reduced β-catenin nuclear activity (21). Several studies have linked canonical and atypical WNT/β-catenin signaling to breast cancer, although the role of β-catenin is complex (41–44). Canonical WNT signaling in TNBC may promote stabilization of β-catenin in the cytoplasm and subsequent transcription of genes involved in EMT (44). However, elevated β-catenin at the plasma membrane has been linked with a good prognosis in TNBC (42), whereas low β-catenin at cell membranes in EGFR-positive TNBC has been associated with poor survival (45). Low levels of E-cadherin at cell membranes may also correlate with elevated β-catenin activity and poor survival in TNBC (46).

How PDLIM2 integrates adhesion and growth factor signaling to enhance β-catenin phosphorylation and activation is not fully established. It is likely that PDLIM2 mediates cell adhesion-dependent phosphorylation and relocation of β-catenin to facilitate its subsequent nuclear activity. This is supported by our observations that cell de-adhesion reduces cytoplasmic PDLIM2 levels and phosphorylation/nuclear accumulation of active β-catenin, and that ectopic expression of PDLIM2 enhances β-catenin phosphorylation on serine 675 and its nuclear accumulation. Importantly, we also observed that although active β-catenin is low in PDLIM2-negative cell lines, it was predominantly at the plasma membrane or cell–cell contacts, whereas, in PDLIM2-positive cells, it was localized in the cytoplasm and nucleus. These suggest that PDLIM2 is not implicated in basal β-catenin phosphorylation, but rather, it facilitates the release of β-catenin from the membrane, thereby enabling its stabilization, phosphorylation, and subsequent activation. Interestingly, growth factor and adhesion signals have recently been reported to act through cellular kinases such as SRC and PAK to enhance β-catenin phosphorylation on serine 675, and its subsequent activation (39, 47).

In summary, PDLIM2 is a marker for a large subgroup of TNBC that is driven by adhesion, growth factor signals and β-catenin activity. The identification of a potential new protein marker to classify functional subsets of TNBCs across previously defined subtypes is an important development. Although PDLIM2 per se may not have a prognostic role, it could be a predictive biomarker to stratify TNBC for therapies that target the signaling pathways regulated by PDLIM2 or that directly target active β-catenin.

B. Moran has Marie Curie Industry Fellowship at Oncomark Ltd. W.M. Gallagher is a Chief Scientific Officer at OncoMark Limited; reports receiving commercial research grant from Carrick Therapeutics; has ownership interest (including stock, patents, etc.) in OncoMark Limited; and is a member of consultant/advisory board at Carrick Therapeutics. R.D. Kennedy is a Medical Director at Almac Diagnostics. C. Caldas reports receiving commercial research grant from AstraZeneca, Genentech, Roche, Servier; and is a consultant/advisory board member at AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: O.T. Cox, R. O'Connor

Development of methodology: O.T. Cox, S.J. Edmunds, M. Bustamante-Garrido, R.D. Kennedy, A. Marx, R. O'Connor

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): O.T. Cox, S.J. Edmunds, K. Simon-Keller, N.E. Buckley, M. Bustamante-Garrido, N. Healy, C.H. O'Flanagan, W.M. Gallagher, R. Bernards, S.-F. Chin, A. Marx, R. O'Connor

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): O.T. Cox, S.J. Edmunds, K. Simon-Keller, B. Li, B. Moran, N.E. Buckley, M. Bustamante-Garrido, C. Caldas, A. Marx, R. O'Connor

Writing, review, and/or revision of the manuscript: O.T. Cox, S.J. Edmunds, K. Simon-Keller, B. Moran, N.E. Buckley, M. Bustamante-Garrido, N. Healy, C.H. O'Flanagan, W.M. Gallagher, R.D. Kennedy, R. Bernards, A. Marx, R. O'Connor

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): O.T. Cox, K. Simon-Keller, B. Li, C. Caldas, A. Marx, R. O'Connor

Study supervision: O.T. Cox, R. O'Connor

This work was funded by the Irish Cancer Society Collaborative Cancer Centre Breast Predict CCRC13GAL (to W.M. Gallagher, R. O’Connor, supported O.T. Cox, B. Li, B. Moran), Science Foundation Ireland Principal Investigator awards 11/PI/1139 and 16/IA/4505 (to R. O’Connor, supported O.T. Cox, M. Bustamente-Garrido), and the European Union FP7 Marie Curie Industry-Academia Partnerships and Pathways (IAPP) Programme 251480: BiomarkerIGF (to R. O’Connor, supported S. E. Edmunds, O. T. Cox, C. H. O’Flannagan). The Northern Ireland Biobank (NIB) received funding from the HSC Research and Development Division of the Public Health Agency in Northern Ireland, Cancer Research UK through the Belfast CR- UK Centre, the Northern Ireland Experimental Cancer Medicine Centre, and the Friends of the Cancer Centre. The Northern Ireland Molecular Pathology Laboratory, which creates resources for the NIB has received funding from Cancer Research UK, the Friends of the Cancer Centre and the Sean Crummey Foundation. The Rational Therapy for Breast Cancer (RATHER), a Collaborative Project is funded under the European Union 7th Framework Programme (grant agreement no. 258967). We thank Dr. Gary Loughran, University College Cork, for helpful discussions on PDLIM2 RNA isoforms and RNA-seq analysis.

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.