A published database was used for bioinformatic analysis. This database contains gene expression and copy number information for 51 breast cancer cell lines (23) (http://caarraydb.nci.nih.gov/caarray/publicEx-perimentDetailAction.do?expId=1015897590151581 at http://cancer.lbl.gov/breastcancer/data.php). Among these 51 cell lines, 13 cell lines contain PIK3CA mutations. The other 38 are considered PIK3CA-wild type. We then used the strategy as follows to identify amplified gene that could synergize with mutant PIK3CA in breast cancer. a) Threshold aCGH and gene expression data: copy number variation (CNV) amplification based on a cut-off ≥0.2. Gene overexpression based on a cut-off >143.767 (3-fold of the median of all samples). b) CNV markers and genes with highly increased amplification/overexpression frequency based on the following criteria: i) frequency difference between cell line w/mutations and w/o ≥0.25 or ii) Fisher exact test P-value of the difference <0.05. c) Pairs of amplified CNV markers and overexpressed genes which are close to each other (distance ≤2 Mb). d) Pairs of amplified CNV markers and overexpressed genes which are close to each other (distance ≤2 Mb) and positively correlated.

MCF 10A and HCC1954 cells were obtained from the American Tissue culture collection. MCF10A cell lines expressing LacZ (negative control), PI3KCA-H1047R, HER2, and both PI3KCA-H1074R and HER2 genes were created in our laboratory at the Barbara Ann karmanos cancer Institute (KCI). Briefly, full-length PIK3CA-H1047R and HER2 were subcloned into a pENTR vector and recombinated into the pLenti-6/V5-DeST vector. The lentiviruses for the full-length genes were generated using the pLenti-virus-expression system (Invitrogen). The generated virus was used to infect targeted model cells. Stable cells were generated after being selected with blasticidin (10 μg/ml, Invivogen) and were grown at 37°C in an incubator with 5% CO₂Most of the cell lines used were originally grown in standard SFIHE medium (serum-free medium with insulin, hydrocortisone, and epidermal growth factor) made by Ham's F12 media supplemented with 0.1% bovine serum albumin, 0.5 μg/ml fungizone, 5 μg/ml gentamycin, 5 mM ethanolamine, 10 mM HEPES, 5 μg/ml transferrin, 10 μM T3, 50 μM selenium, 1 μg/ml hydrocortisone, 10 ng/ml EGF, and 5 μg/ml insulin. MCF10A cells expressing HER2 were grown in SFHE medium (serum-free medium with hydrocortisone and EGF). For experimentation, each individual cell line with its own different overexpressed gene was first placed in SFIHE or SFHE medium for a few days, and then placed in either SFIHE medium, SFIH (lacking EGF or epidermal growth factor), SFHE (lacking insulin), or SFH (lacking both EGF and insulin) medium in order to observe cell growth and colony formation. For the invasion assay, cells were first plated in each well insert with SFIH medium, and then SFIHE with 5% FBS was placed outside in the well. Tests on inhibition were done in SFIHE medium.

McF10A/HER2, McF10A/PIK3CA-h1047r, McF10A/HER2/PIK3CA-H1047R and McF10A/LacZ were seeded in triplicate at a density of 2×10³ cells per well in 96-well plates on day 0. Cells were grown in SFIHE (growth medium with insulin, hydrocortisone and EGF), SFIH (growth medium lacking EGF) or SFHE (growth medium lacking insulin) or SFH (lacking both EGF and insulin) medium. Cell proliferation was measured with an MTT assay kit (Fisher Scientific, Pittsburgh, PA, USA), trypsinized and counted with a particle counter (Beckman Coulter) with an absorbance value of 540 nm on days 1, 3, 5, and 7 to make the growth curve. Media was changed at each of these time points.

Each cell line was seeded in two wells of a 6-well plate with a density of 2×10³ cells per well for 1-2 weeks in SFIHE, SFIH, SFHE, or SFH culture mediums. As soon as distinct clones were visible with a microscope and were almost starting to merge, all plates were stained with crystal violet on day 12 of growth for approximately 5 min and then exposed to water gently. The numbers of individual clones were then counted and recorded, and the plates were scanned for clear pictures of clones.

A cell migration and cell invasion assay was performed in BD control chamber and BioCoat Matrigel Invasion chambers (BD Bioscience, Corning, NY, USA) according to the manufacturer's instructions. Cells (5×10⁴) were plated in the upper chamber in 500 μl of SFIH growth medium without FBS, and the lower chamber was filled with 750 μl of SFIHE growth medium with 5% FBS. After 48-h incubation, the cells that had invaded to the underside of the membrane were fixed and stained with the HEMA 3 kit (Fisher Scientific). The migration and invasive cells were then determined by counting the stained cells by microscopy of ×100 and images were taken.

LY294002 [pan PI3K inhibitor (24,25)], and CP-724714 [selective HER2 inhibitor (26,27)] were diluted to the concentrations of 3 and 10 μM and 0.1 and 1 μM in SFIHE medium, respectively, according to previous data. SFIHE medium with either concentration of the PI3K inhibitor, the HER2 inhibitor, or both inhibitors at both concentrations were added to only the MCF10A cell line expressing both the mutant PIK3CA-H1047R and the HER2 gene in a 96-well plate for cell growth, and in 6-well plates for colony formation with 2.5×10⁴ cells/well. Medium with fresh inhibitors was replaced every other day. For the 96-well plate, cells were trypsinized and counted with a particle counter (Beckman coulter) with an absorbance value of 540 nm on the 9th day of growth. Clones in the 6-well plates were stained with crystal violet on the 12th day of experimentation, and clones were counted. The plates were also scanned for pictures of clones.

Gene expression levels for the four different cell lines were measured with Human Gene 2.1 ST Array Strip at the Genomics Core of University of Michigan. Microarray data reported herein have been deposited at the NCBI Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE99512) with the accession number GSE99512. Microarray analyses and pathway analysis were performed by the Biostatistics core of the KCI. The raw intensity probe-level values were background corrected, log2 transformed, quantile normalized, and summarized to probe set level expression values. The 'rma' function from R package 'oligo', based on robust multi-array average (RMA) method (28), was used for all above pre-processions. Control type of probe sets, such as 'Background probes', 'Poly-A controls' and 'Hybridization controls', were excluded from further analyses. Probe sets with log2 expression values less than 4 for any two cell lines compared were excluded from the analyses of differential expression (DE) probe sets. The k-mean clustering method was used for identifying DE probe sets based on each of three comparisons PIK3CA vs. MCF10A_ Control, HER2 vs. MCF10A_Control, and PIK3CA_HER2 vs. MCF10A_Control. The identified up-/down-regulated probe sets were subjected to ingenuity pathway analysis (IPA). IPA canonical pathway analysis identified the enriched pathways that were most significant to the data set. The significance of the association between the data set and the IPA canonical pathway was measured by the Fisher's exact test. The −Log (P-value) ≥3 was used as the cut-off for identifying the common or unique pathways across three comparisons.

Six distinct chromosomal regions were identified within chromosomes 3, 7, 12, 14, 16, and 17 as having increased frequency of amplification in PIK3CA-mutant versus PIK3CA-WT cell lines. The regions with the most amplification include chromosome 7, 14, 16, and 17 (P<0.05, Table I). In particular, the ERBB2/HER2 region was one of the most correlated regions. The percent difference of amplification of the HER2 region between the cell lines with and without the PIK3CA mutation was 37%, which indicated that there may be a synergistic relationship between the PIK3CA mutation and HER2 gene amplification. In addition to the chromosome 17 region, which contains HER2 and two other genes, multiple other chromosomal regions were also found to be correlated with PIK3CA mutation. Chromosome 7 region contains three genes. Chromosome 12 region contains ten genes. Chromosome 14 region contains seven genes and chromosome 16 contains 11 genes (Table II).

To specifically study the cooperative effect of mutant PIK3CA and HER2 genes in breast cancer cell growth, we chose to create MCF10A cell models expressing mutant PIK3CA-h1047r (MCF10A/PIK3CA-H1047R), HER2 (MCF10A/HER2) or both mutant PIK3CA-H1047R and HER2 genes (MCF10A/HER2/PIK3CA-H1047R). A MCF10A cell line expressing LacZ was used as a control (MCF10A/LacZ) (Fig. 1A). A cell proliferation assay was performed in SFIHE, SFIH, SFHE, and SFH media in order to address proliferation under different growth stimulating conditions. In SFIHE medium, all the cell models demonstrate similar proliferation rate (Fig. 1B). However, in media deprived of the growth factor EGF and insulin, the MCF10A/HER2/PIK3CA-H1047R cells exhibit the greatest growth advantage over MCF10A/PIK3CA-H1047R, MCF10A/HER2, and MCF10A/LacZ (negative control) (Fig. 1B). Epidermal growth factor (EGF) is an essential factor in the acceleration of cell growth in cancer cells (29-31). Insulin allows cells to use glucose for energy and maintains the glucose level in blood and is tightly associated with cancer cell growth and proliferation (32-34). This finding suggests a cooperative effect of HER2 and mutant PIK3CA that is insulin- and EGF-independent. A colony formation assay showed similar synergistic effect of the two genetic alterations on the promotion of colony formation (Fig. 1C and D).

We further investigated the effect of HER2, the mutant PIK3CA-H1047R, or both genetic alterations on the migration and invasion using a chamber assay. In the migration assay, all the oncogene expressing cells showed higher migration efficiency than control cell MCF10A/LacZ. MCF10A/HER2 and MCF10A/PIK3CA-H1047R showed a similar migration cell numbers, while MCF10A/HER2/PIK3CA-H1047R did not demonstrate higher migration capability than either oncogene alone (Fig. 1E). In the invasion assay, all the oncogene-expressing cells showed higher invasive efficiency than control cell MCF10A/LacZ. Specifically, MCF10A/HER2/PIK3CA-H1047R demonstrated a stronger invasion capability than MCF10A/HER2 and MCF10A/PIK3CA-H1047R cells (Fig. 1F and G).

To test whether the cell model (MCF10A/HER2/PIK3CA-H1047R) with both genetic alterations can be synergistically inhibited by simultaneous targeting of both genes, LY294002, a pan-PI3K inhibitor, was used for targeting PIK3CA-H1047R and CP-724714, a selective HER2 inhibitor, was used for targeting active HER2. The results showed that the cells treated with CP-724714 had little to no change in cellular proliferation. Low dose LY294002 (3 μM) alone also had a minimal efficacy as an inhibitor of cell growth. However, cells exposed to a combination of both inhibitors with 10 μM LY294002 had a drastic decrease in cell proliferation. The combined inhibition effects were more significant in SFIH, SFHE, and SFH medium than in SFIHE medium (Fig. 2A and B).

To validate the results obtained from the established MCF10A/HER2/PIK3CA-H1047R cell model, we repeated drug treatment using the HCC1954 cell line, which contains both endogenous PIK3CA mutation and HER2 amplification. The treatment with different doses of CP724714 (0.1, 1, 2, 5, and 10 μM) marginally changed cell growth. The treatment of HCC1954 cells with LY294002 only significantly inhibited cell proliferation at high doses (8 and 10 μM). The combination of these two drugs evinced a dose-dependent inhibition in cell proliferation. With a low dose of LY294002 (3 μM) treatment, increasing the dose of CP724714 from 0.1 to 10 μM resulted in a cell proliferation inhibition effect similar to the high dose of LY294002 (10 μM). However, when a high dose of LY294002 (10 μM) was used, further treatment with CP724714 at 0.1, 1, 10 μM obtained an even more significant reduction in cell growth (Fig. 2C). A similar result was obtained in colony formation assay when HCC1954 cells were treated with different combinations (Fig. 2D).

The study of the biological function of three different cell models showed that the MCF10A/HER2/PIK3CA-H1047R cells harbor unexpected characteristics in comparison to MCF10A/HER2 and MCF10A/PIK3CA-H1047R including EGF and insulin-independent growth, stronger invasion capability, and sensitive only to double inhibition. These results suggest that the co-occuring oncogenic HER2 and mutant PIK3CA could lead to activation of novel cell signaling pathways which drive unique oncogenic properties in cancer cells. To investigate the underlying mechanism of the functional differences between cells with mutant PIK3CA, HER2 or both mutant PIK3CA and HER2, we performed an Affymetrix microarray analysis. After normalization with MCF10A/LacZ control cells, MCF10A/PIK3CA-H1047R cells showed 1,175 upregulated, 1,963 downregulated and 3,902 no changed genes; MCF10A/HER2 cells showed 762 upregulated, 2,271 downregulated and 4,016 no changed genes. The MCF10A/mutant PIK3CA/-HER2 showed 619 upregulated, 2,719 downregulated and 4,131 no changed genes (Fig. 3A). Although there are 76 upregulated and 935 downregulated common genes shared in all three cell models, each cell model was shown to contain hundreds of unique genes (Fig. 3B–D).

Further ingenuity pathway analysis showed that 24, 20 and 18 cell signaling pathways were enriched in MCF10A/PIK3CA-H1047R, MCF10A/HER2 and MCF10A/HER2/PIK3CA-H1047R, respectively, with a cut-off of -log (P-value) ≥3 (Fig. 4A–C). Under this condition, four common cell signaling were identified in all three cell models, including aryl hydrocarbon receptor (AhR) signaling, AMP-activated protein kinase (AMPK) signaling, protein ubiquitination pathways and hereditary breast cancer signaling (Fig. 4D and E, and Table III). Additionally, 25% (7 out of 24) signaling pathways in cells expressing PIK3CA-H1047R overlapped with 35% (7 out of 20) signaling pathways in HER2 overexpression cells (Fig. 4D). There is 55% (10 of 18) signaling pathways enriched only in cells with both PIK3CA-H1047R and HER2 and do not overlap with cells expressing either mutant PIK3CA or HER2 alone (Fig. 4F). Importantly, 50% (5 out of 10) of these cell signaling pathways are related to DNA replication stress, including role of BRCA1 in DNA damage response, role of CHK1 in cell cycle checkpoint control, estrogen-mediated S-phase entry, cell cycle control of chromosomal replication and mismatch repair in eukaryotes (Fig. 4F and Table IV).