Polyclonal PIPKIγ anti-serum was generated from rabbit (Covance, Princeton, NJ, USA) using purified His-tagged PIPKIγ. Anti-serum was purified on an affinity column generated by coupling recombinant C-terminus of PIPKIγ to cyanogen bromide-activated Sepharose 4B (Sigma-Aldrich, Saint Louis, MO, USA) as described 172527. The affinity-purified antibody recognizes both PIPKIγi1 and PIPKIγi2.

The siRNA sequence targeting PIPKIγ is 5'-GGACCUGGACUUCAUGCAG-3'. The sequence of control scrambled siRNA is 5'-GUACCUGUACUUCAUGCAG-3'. Oligonucleotide sequences used for generation of short hairpin RNA (shRNA) specific for PIPKIγ were: GCCACCTTCTTTCGAAGAA (PIPKIγ shRNA) and GCCTTCTTCGCTAAACGAA (Control shRNA). Generation of replication-defective infectious viral particles and the transduction of the cells were carried out following the protocol provided by Addgene (Addgene Inc., Cambridge, MA, USA). In brief, synthesized oligonucleotides were annealed and cloned into HpaI and XhoI sites of pLL3.7 vector (Addgene Inc., Cambridge, MA, USA). Stabl3 competent cells (Invitrogen, Carlsbad, CA, USA) were used for transformation and DNA purification to minimize the mutagenesis. The integrity of lentiviral vector-containing cloned shRNA sequences were validated by DNA sequencing.

MDA-MB-231 and MDA-MB-435S cells were cultured using DMEM supplemented with 10% FBS. SKBR3 cells were cultured in DMEM/F12 with 10% FBS. For siRNA transfection, cells were transfected with Oligofectamine (Invitrogen, Carlsbad, CA, USA) following the manufacturer's instructions.

A tissue microarray was constructed out of 438 archival invasive breast carcinoma samples at Vancouver General Hospital. These cases were collected between 1974 and 1995. Ethics board approval was obtained for all cases. Patients' demographics, pathological features of the tumors and expression of various biomarkers have been reported previously 28. In brief, median age of the patients at the moment of diagnosis was 61.5 years; median survival time was 11.9 years. Histological distribution included 379 infiltrating ductal carcinoma, 41 infiltrating lobular carcinoma and 8 special types.

Three tissue microarray blocks were assembled using a manual tissue microarrayer (Beecher Instruments, Inc., Silver Springs, MD, USA) from formalin fixed paraffin-embedded tissue as described previously 29. Tissue sections (4 μm thick) were cut from the donor blocks and stained with hematoxylin and eosin for tissue review. Representative areas of tumor were circled on the slides and corresponding donor blocks; duplicate 0.6 mm cores were taken from these blocks and inserted into three recipient blocks. Sections (4 μm thick) were cut from the recipient blocks and deparaffinized with CitriSolve and dehydrated through three alcohol changes. Antigen retrieval was performed using a steamer for 30 minutes in 0.1 M citrate buffer (pH 6.0). After that, sections were rinsed with PBS three times for five minutes each time. Hydrogen peroxide and serum free protein block (Dako, Carpenteria, CA, USA) were used to block endogenous peroxidases and prevent non-specific protein binding. Sections were then incubated with anti-PIPKIγ antibody (0.5 μg/ml) in a sealed immunochamber overnight at 4°C and Dako Envision anti-rabbit secondary antibody was applied at room temperature for 30 minutes. The NovaRed Substrate Kit (Vector Labs, Burlingame, CA, USA) was used to visualize the protein. Slides were then counterstained with hematoxylin and mounted.

The hematoxylin and eosin and immunohistochemistry images of all cores used in this study are publicly available at the companion site 30. The site was constructed using a GPEC database and a Java applet provided by Bacus Laboratories, Inc (Bacus Laboratories, Inc., Lombard, IL, USA). All the slides were scanned with a BLISS scanner (Bacus Laboratories, Inc., Lombard, IL, USA), and posted on the site. WebSlide Browser for Windows (Bacus Laboratories, Inc., Lombard, IL, USA) can be used for viewing preview images of the arrays and images of individual cores.

The assays were performed in modified Boyden chamber transwell (Neuroprobe, Gaithersburg, MD, USA) as described 3132. The membrane was pre-coated with type I collagen (10 μg/ml). Per well, 50,000 cells were applied. Chemotaxis assays were performed at 37°C in humidified air with 5% CO₂ for four hours. Cells migrated through to the underside of the membrane were counted in five high power fields, in a blinded fashion. The migration index for each experiment was calculated as the mean number of cells that migrated toward medium-containing 1% FBS divided by the mean number of cells that migrated toward medium-containing BSA only.

Matrigel-coated Transwells (BD Bioscience, San Jose, CA, USA) were incubated with DMEM for four hours and 5 × 10⁴ cells were plated in the upper chambers. The lower chambers contained 1% FBS conditioned DMEM. The inserts were incubated at 37°C in humidified air with 5% CO₂ for 24 hours. The cells that had invaded the lower surface of the membrane were fixed with 4% polyformaldehyde and stained with 0.2% crystal violet. The number of cells that had invaded was quantified by counting random fields using a light microscope.

MDA-MB-231, MDA-MB-435S, or SKBR3 cells infected with lentivirus containing either control shRNA or PIPKIγ shRNA were seeded into 12-well culture plates at a density of 1000 cells/well. Then, manual cell counting was performed every two days for eight days.

We performed Kaplan-Meier survival analysis using log-rank test to determine differences in survival of the patients with different levels of PIPKIγ protein expression. Breslow test was also used to emphasize survival differences in the first 5 to 10 years of the follow up. The alpha level was determined as 5%. All the tests were two-sided. Spearman's correlation was utilized to estimate correlations between PIPKIγ protein content with that of other biomarkers and clinical prognostic factors.

The cellular functions of PIPKIγ suggest the potential for roles in epithelial cancer progression. To explore the potential that PIPKIγ may correlate with disease progression and outcome we began to assess the content of PIPKIγ in a well-characterized tissue microarray of breast cancer patients. For this approach we used well-characterized antibody that specifically detect all splice variants of the PIPKIγ as described in Materials and Methods. As a control, PIPKIγ expression was knocked down with siRNA (Small interfering RNA) in breast epithelial (MCF10a) or carcinomas cell lines and this resulted in a loss of protein by western blotting and a loss of the immunostaining in these cells 1625.

Normal breast tissues show strong staining for E-cadherin epithelia cells lining the ducts and PIPKIγ also shows strong staining of these cells (Figure 1a). To assess how the PIPKIγ expression changes within breast cancers we have screened a breast cancer tissue microarray. The expression of PIPKIγ in breast carcinomas was demonstrated via tissue microarray that was constructed out of 438 archival invasive breast carcinoma samples at the Vancouver General Hospital 33. Of the 438 breast carcinoma represented on the tissue microarray, 330 specimens (75.3%) were considered interpretable. As shown in the lower panels of Figure 1a, in tumors with a loss of E-cadherin there is also a loss of PIPKIγ staining. The bottom panel of Figure 1a shows that in a fraction of tumors E-cadherin was expressed but not targeted to the plasma membrane and these tumors also lost PIPKIγ. These combined data are shown statistically in Table 1.

In Figure 1b, examples of the different PIPKIγ staining of breast tumors are shown. The slides were scored as 'negative' if no PIPKIγ staining was detected, 'weak staining' if there was any amount of weak membranous staining and/or strong membranous staining in less than 50% of tumor cells, and 'strong staining' if there was strong membranous staining in 50% or more carcinoma cells. Among the 358 specimens analyzed, 149 specimens were PIPKIγ staining negative (41.6%), 144 specimens were PIPKIγ staining weakly positive (40.2%), and 65 specimens were PIPKIγ staining strongly positive (18.2%) (Table 2).

To investigate if PIPKIγ tissue content correlates to breast cancer prognosis, Kaplan-Meier survival curves were generated using PIPKIγ antibody staining. As shown in Figure 2a, these curves show that strong positive PIPKIγ expression was correlated to poor outcome (Log Rank chi-squared = 6.078, P = 0.014; Breslow chi-squared = 7.454, P = 0.006). It indicates that PIPKIγ expression is inversely correlated to the survival of breast cancer patients. To further define the correlation of PIPKIγ expression with breast cancer prognosis, patients were stratified based on the lymph node status. Kaplan-Meier survival curves were generated for lymph node negative and lymph node positive. As shown in Figure 2b, there was no significant difference in survival of the patients with PIPKIγ strong positive and weak positive or negative tumor in lymph node negative subset (Log Rank chi-squared = 1.01, P = 0.315; Breslow chi-squared = 0.945, P = 0.331). In the lymph node positive subset, the longer term survival difference was not significant in the log-rank test (Log Rank chi-squared = 2.154, P = 0.142, Figure 2c); however, the 5- and 10-year survival rate of PIPKIγ strong positive patients is significantly lower than PIPKIγ negative or weakly positive patients and had trend towards significance in Breslow test (Breslow chi-squared = 3.267, P = 0.071, Figure 2c). Considering that the median age of the patients at the moment of diagnosis is 61.5 years, the longer-term survival as 20 or 25 years may not clearly reflect the survival of breast cancer because it is very close to the normal average human life limit. The 5- and 10-year survival rate difference supports that PIPKIγ expression is inversely correlated to the survival of breast cancer patients.

To further characterize PIPKIγ expression in breast cancer, Spearman's correlation was used to demonstrate the correlation of PIPKIγ expression with clinical characteristics such as lymph nodal status, tumor size and Nottingham tumor grade. As shown in Table 1, PIPKIγ expression does not correlate significantly with lymph nodal status and tumor size. There is a weak significant correlation with the Nottingham grade.

PIPKIγ expression was correlated with the expression of some established breast cancer biomarkers on a tissue microarray. From this analysis there was a correlation between the loss of PIPKIγ with loss of E-cadherin expression and plasma membrane targeting. It is consistent with the finding that PIPKIγ directly binds to E-cadherin and modulates E-cadherin trafficking 16. Epidermal growth factor receptor (EGFR)/human epidermal growth factor receptor (HER) family receptors are often overexpressed in breast cancers and are used as breast cancer biomarkers 34. Interestingly, our results show that both HER1 (EGFR) and HER2/neu expression are correlated with PIPKIγ expression (Table 1). EGFR/HER family receptors play key roles in breast cancer metastasis and their expression are correlated with worse prognostic patient survival rates 3536. The correlation of PIPKIγ with EGFR/HER family receptors is consistent with the finding that PIPKIγ expression is inversely correlated to the survival of breast cancer patients. In addition, the results of tissue microarrays show that the expression of PIPKIγ is inversely correlated to estrogen receptor (ER) and progestin receptor (PR). ER and PR status are important for distinguishing different breast cancer subtypes and they are critical for ER⁺ or PR⁺ breast cancer cell growth 3738. Our finding is the first time to discover a correlation between PIPKIγ with ovarian hormone receptors ER and PR.

Previously, it has been shown that PIPKIγ is required for growth factor-induced HeLa (Helen Lane) cell migration 25. As PIPKIγ expression is inversely correlated to the survival of breast cancer patients, to regulate cancer cell metastasis might be one of the mechanisms for PIPKIγ function in breast cancer. In order to verify this possibility, the effect of PIPKIγ knockdown on two breast cancer cell lines, MDA-MB-231 and MDA-MB-435S, migration and invasion was demonstrated. As shown in Figure 3a, PIPKIγ-specific siRNA could specifically knockdown expression of PIPKIγ in MDA-MB-231 or MDA-MB-435S cells, compared with the actin control. The effect of PIPKIγ-knockdown on breast cancer cell migration was quantified using a modified Boyden chamber transwell assay. As shown in Figure 3b, the knockdown of PIPKIγ blocked migration of both MDA-MB-231 and MDA-MB-435S. To investigate the effect of PIPKIγ-knockdown on breast cancer cell invasion, a Matrigel invasion assay was used. As shown in Figure 3c, the knockdown of PIPKIγ also blocked invasion of both MDA-MB-231 and MDA-MB-435S.

To further investigate the role of PIPKIγ in the pathogenesis of breast cancer progression, lenti-virus-vector-based PIPKIγ shRNA was used to establish PIPKIγ-knockdown stable cell lines in MDA-MB-231, MDA-MB-435S, and SKBR3 cells. The knockdown of PIPKIγ expression in these stable cell lines was confirmed by western blot as shown in Figure 4. The effect of PIPKIγ-knockdown on breast cancer cell growth was then determined. As shown in Figure 4, PIPKIγ-knockdown decreased cell growth of MDA-MB-231, MDA-MB-435S, and SKBR3 cells (Figures 4a to c). This result indicates an important role of PIPKIγ in breast cancer cell proliferation and is consistent with the finding that PIPKIγ expression is inversely correlated to the survival of breast cancer patients.