Inhibitors of STAT3, β‐catenin, and IGF‐1R sensitize mouse PIK3CA‐mutant breast cancer to PI3K inhibitors

Although mutations in the phosphoinositide 3‐kinase catalytic subunit (PIK3CA) are common in breast cancer, PI3K inhibitors alone have shown modest efficacy. We sought to identify additional pathways altered in PIK3CA‐mutant tumors that might be targeted in combination with PI3K inhibitors. We generated two transgenic mouse models expressing the human PIK3CA‐H1047R‐ and the ‐E545K hotspot‐mutant genes in the mammary gland and evaluated their effects on development and tumor formation. Molecular analysis identified pathways altered in these mutant tumors, which were also targeted in multiple cell lines derived from the PIK3CA tumors. Finally, public databases were analyzed to determine whether novel pathways identified in the mouse tumors were altered in human tumors harboring mutant PIK3CA. Mutant mice showed increased branching and delayed involution of the mammary gland compared to parental FVB/N mice. Mammary tumors arose in 30% of the MMTV‐PIK3CA‐H1047R and in 13% of ‐E545K mice. Compared to MMTV‐Her‐2 transgenic mouse mammary tumors, H1047R tumors showed increased upregulation of Wnt/β‐catenin/Axin2, hepatocyte growth factor (Hgf)/Stat3, insulin‐like growth factor 2 (Igf‐2), and Igf‐1R pathways. Inhibitors of STAT3, β‐catenin, and IGF‐1R sensitized H1047R‐derived mouse tumor cells and PIK3CA‐H1047R overexpressing human HS578T breast cancer cells to the cytotoxic effects of PI3K inhibitors. Analysis of The Cancer Genome Atlas database showed that, unlike primary PIK3CA‐wild‐type and HER‐2+ breast carcinomas, PIK3CA‐mutant tumors display increased expression of AXIN2, HGF, STAT3, IGF‐1, and IGF‐2 mRNA and activation of AKT, IGF1‐MTOR, and WNT canonical signaling pathways. Drugs targeting additional pathways that are altered in PIK3CA‐mutant tumors may improve treatment regimens using PI3K inhibitors alone.


Introduction
The phosphatidylinositol 3-kinase (PI3K) pathway is a key regulator of growth, survival, and metabolism in both normal and malignant cells (Zhao and Vogt, 2008). Over 70% of breast cancers show activation of the PI3K pathway through mechanisms such as HER2 amplification, deletion of the tumor suppressor PTEN, or oncogenic mutations in PIK3CA (Zhao and Vogt, 2008). The PIK3CA gene encodes the catalytic subunit p110a, and its amplification and/or mutation is associated with several types of solid tumors (Bachman et al., 2004;Kadota et al., 2009;Samuels et al., 2004;Wu et al., 2005). Activating somatic mutations in PIK3CA Abbreviations muPIK3CA, mutant PIK3CA; PI3Ki, PI3K inhibitor; PIK3CA, phosphoinositide 3-kinase catalytic subunit. are present in~30% of human breast cancers of all stages (Bachman et al., 2004;Barbareschi et al., 2007;Miller, 2012;Saal et al., 2005). In 47% of these cases, mutations occur in the kinase domain, most frequently the H1047R substitution in exon 20. In 33% of these cases, mutations occur in the helical domain; here, the most frequent ones are the E545K and E542K substitutions in exon 9 (Bachman et al., 2004;Bader et al., 2005). PIK3CA mutations cause a gain of enzymatic function, stimulate AKT signaling, induce growth factor, and anchorage independent growth in culture (Bader et al., 2005;Ikenoue et al., 2005;Isakoff et al., 2005). Expression of H1047R-mutant PIK3CA in the mammary gland induces tumors in transgenic mice (Adams et al., 2011;Meyer et al., 2011;Tikoo et al., 2012;Yuan et al., 2013). Therefore, inhibition of PI3K represents a potentially attractive strategy for the treatment of breast cancer. In fact, a number of PI3K inhibitors (PI3Ki) have entered clinical trials (Zardavas et al., 2014).
Laboratory studies and early clinical trials indicate that several of the PI3Ki demonstrate preferential inhibition of tumors with PIK3CA mutations (Mayer et al., 2017;O'Brien et al., 2010). However, while long-term stabilization and partial tumor responses are observed in PIK3CA-positive breast cancers treated with PI3Ki (NCT01219699) (Brana and Siu, 2012), the majority of mutant cancers do not experience substantial regression (Bendell et al., 2012). Strategies to overcome both de novo and adaptive resistance to PI3Ki have employed combination therapy. Combined inhibition of PI3K and mammalian target of rapamycin complex (mTORC) is associated with stable disease in phase 1 clinical trials (Markman et al., 2012;Papadopoulos et al., 2014). In addition, clinical trials using a combination of PI3Ki with fulvestrant, IGF-1R antibody (AMG479), or lapatinib have been initiated (Zardavas et al., 2014).
This study sought to characterize the molecular mechanisms of action of mutant p110a in tumors derived from genetically engineered mouse models to further define targets for treating PIK3CA-driven tumors and their metastatic progression, with the goal of developing mechanism-based combination therapies.

Cell lines and reagents
The HS578T cell line was obtained from the American Type Culture Collection (Manassas, VA, USA). Low passages of the cell line were used to avoid loss of fidelity to the original cells. Stable cell lines were derived from PIK3CA-H1047R mouse mammary tumors. Cell lines derived from recurrent H1047R mouse tumors (RDR-C234 and RDR-A677) were kindly provided by Jean Zhao from Dana-Farber Cancer Institute . Sources of other reagents were as follows: XAV939 (Tocris, Minneapolis, MN, USA); S31-201 (SelleckChem, Houston, TX, USA); linsitinib, OSI-906 (LC Labs, Woburn, MA, USA); LY294002 (Sigma-Aldrich, St. Louis, MO, USA), and GDC-0941 (Genentech, San Francisco, CA, USA). Neu-N (c-erbB2, ERBB2, Her-2) mice were derived from the colony of Guy et al. (1992), bred to homozygosity as verified by Southern blot analysis (Reilly et al., 2000), and maintained at Johns Hopkins University.

Site-directed mutagenesis and cloning of human PIK3CA-mutant transgenes into MKbpAII vector
The plasmid containing the PIK3CA-mutant H1047R and E545K sequences were generated by the use of site-directed mutagenesis kit (Quick Change; Stratagene, La Jolla, CA, USA). Automated design of mutagenic primers for site-directed mutagenesis was performed by PRIMERX software (Bioinformatics.org).
Mammary gland-specific expression of PIK3CA was driven using wild-type or E545K-or H1047R PIK3CA-mutant genes cloned into the MKbpAII vector containing the mouse mammary tumor virus MMTV-LTR promoter/enhancer and KCR intron to increase transgene expression (gift from Jeffrey Rosen, Baylor College of Medicine).

Generation, handling, and genotyping of transgenic mice
All animal studies were performed following approval of the Institutional Animal Care and Use Committee of the Johns Hopkins School of Medicine. Transgenic mice were generated by microinjecting the 9.4-kb fragment of pMMTV-PIK3CA into the pronucleus of single-cell embryos isolated from superovulated FVB/N mice by the Transgenic Mouse Core Facility at National Cancer Institute. The transgenic founders from the different PIK3CA-mutant strains were transferred to Johns Hopkins animal facility. The transgenic F1 generation was crossed with FVB/N mice and the progeny maintained as hemizygous for the PIK3CA-mutant allele. The genotype of transgenic mice was assessed in tail clip DNA by PCR using MMTV and PIK3CA-specific primers (forward primer, 5 0 -CTGGTCATCATCCTGCCTTT-3 0 ; reverse primer, 5 0 -CCTCACGGAGGCATTCTAAA-3 0 ). The presence of the respective PIK3CA mutations in the transgenic mouse DNA was confirmed by sequencing.

Determination of transgene copy number and expression
Southern blot analysis and real-time quantitative PCR (D'Haene et al., 2010) were employed to determine the transgene copy number inserted into the founder genome. Briefly, mouse genomic DNA was digested with Bam H1, hybridized with the E545K or H1047R PIK3CA probes, and the membranes were developed using the Odyssey reagent (Licor Biosciences, Lincoln, NE, USA), according to the manufacturer's instructions. For gene expression, real-time quantitative PCR was conducted using the Maxima SYBR Green/ROX Master Mix (Fermentas, Waltham, MA, USA) per the manufacturer protocol as previously described (Merino et al., 2016). Primer sequences will be provided upon request.

Histology
5-lm thick sections of FFPE mouse tumor fragments were stained with hematoxylin and eosin (H&E) and assessed by the pathologist blinded to group of origin.
2.9. Cell proliferation assay 5 9 10 3 cells per well were plated in 96-well plates in triplicate and treated with the indicated drugs, alone and in combination. After 48 h, cell proliferation assay was performed as previously described (Merino et al., 2016).

Analysis of TCGA data
The processed RNA-seq dataset and clinical information of the The Cancer Genome Atlas (TCGA) breast carcinoma (BRCA) database hosted on the Firehose server managed by the Broad Institute (Center, 2015) were used for analysis of correlations of gene expression with disease subtype. PIK3CA mutation and HER2 amplification information was obtained from the cBioPortal interactive analysis platform (Gao et al., 2013). Single sample gene set analysis (ssGSEA) (Barbie et al., 2009) implemented by the GSVA package (Hanzelmann et al., 2013) was used to estimate activation scores of PIK3CA-related pathways curated by the Molecular Signatures Database (Liberzon et al., 2011). Pairwise Welch's corrected ttest with Bonferroni correction for multiple testing was used to calculate significance in differences across groups.

Statistical analysis
mRNA expression, assessed by Q-PCR, was expressed as mean AE standard errors of mean. Mann-Whitney U-test was performed to assess median of mRNA and protein levels for tumors. Student's t-test was performed to compare cell viability, mRNA, and protein levels in cell lines and expressed as the mean (AE SEM). Statistical significance was defined as P < 0.05 and calculated using GRAPHPAD PRISM software (GraphPad Software, La Jolla, CA, USA).

Generation of transgenic mice with mammary gland-specific overexpression of PIK3CA-H1047R-and -E545K-mutant genes
To examine the role of PIK3CA-H1047R and -E545K mutations during mammary development and their ability to induce mammary tumors, we generated transgenic mice with expression of the PIK3CA-mutant genes under the control of the mammary epithelialspecific MMTV promoter (Fig. 1A). The presence of mutant PIK3CA in the transgenic mouse DNA was confirmed by sequencing. Southern blot analysis (Fig. S1A) and quantitative PCR (Fig. S1B) were employed to determine copy numbers of the transgene inserted into the founder genome. As PIK3CA mutations occur on single alleles in the genome, we selected the founders that had a small number of transgene insertions in their genome. In comparison with FVB/N control mice, levels of PIK3CA mRNA (Fig. S1C), protein ( Fig. S1D), as well as its downstream effector protein phosphorylated Akt-Ser ( Fig. S1D), were found to be higher in the mammary glands of pregnant transgenic mice. These findings are as predicted because the MMTV promoter is hormonally regulated and is activated in mammary gland during pregnancy and peaks at lactation (Wagner et al., 2001).

Mutants of PIK3CA increase side branching, delay involution, and induce tumorigenesis
To examine the effects of PIK3CA mutations on mammary gland development, we performed whole-mount staining analysis of mammary glands from newly generated PIK3CA-H1047R and -E545K low copy number transgenic mice at different stages of mammary gland development (virgin, pregnancy, and involution). At 8 weeks, the ductal tree of the PIK3CA-H1047R and -E545K transgenic mice had a significantly greater level of branching in comparison with control FVB/N mice (Fig. 1B). On day 13 of pregnancy, there were no significant phenotypic differences in the morphology of the mammary gland among these mice, except a slight increase in branching observed in the H1047R mutants ( Fig. S1E). At day 29 of involution, most of the alveolar structures of FVB/N and PIK3CA-E545K mammary glands collapsed into clusters of epithelial cells and more adipocytes filled the spaces (Fig. 1C). On the other hand, PIK3CA-H1047R mice showed delayed involution and the alveolar structures were largely intact and filled with milk ( Fig. 1C). Using Wntknockout mice, Brisken et al. (2000) showed that Wnt proteins function as paracrine factors that operate downstream of progesterone and the progesterone receptor to mediate the process of side branching (Brennan and Brown, 2004). As we observed altered side branching in the mammary glands of PIK3CAmutant mice, we investigated whether the Wnt/b-catenin pathway is induced in the PIK3CA-H1047R-and PIK3CA-E545K-mutant mammary glands. We found that mammary glands of PIK3CA-mutant mice showed an increase in activated b-catenin (Fig. S1F) and its downstream gene, cyclin D1 (Fig. S1D).

Wnt/b-catenin pathway is activated in PIK3CA-H1047R-mutant tumors
Higher levels of b-catenin expression in the normal glands of PIK3CA-mutant mice (Fig. S1F), and reports that PIK3CA-H1047R caused an expansion of the luminal progenitor population (Tikoo et al., 2012), led us to hypothesize that the Wnt signaling pathway is likely to be activated in the PIK3CA-H1047R human breast tumors. We also compared signaling responses between Her2-and PIK3CA-H1047R mouse tumors.
Although PI3K is downstream of Her-2 signaling, we saw many differences between mammary tumors arising in transgenic mice overexpressing Her-2 versus PIK3CA-H1047R, reflecting differences observed between wild-type and mutant PI3K. PIK3CA-H1047R mammary tumors showed strong activation of phosphorylated AKT, phosphorylated Gsk3-b, cyclin D1, and activated b-catenin compared to tumors from Her-2 transgenic mice ( Fig. 2A). The mRNA expression levels of Axin 2, a transcriptional target of the Wnt/b-catenin pathway (Jho et al., 2002), were higher in PIK3CA-H1047R-mutant tumors compared to Her-2 tumors (Fig. 2B). Axin 2 was also induced in tumor cells derived from recurrent tumors (RDR-C234 and RDR-A677) from the H1047R-doxycyclineinducible mouse model   (Fig. 2B). Consistent with predicted consequences of activation of the Wnt/b-catenin pathway, single cells isolated from PIK3CA-H1047R tumors contained a larger CD24med/CD49fhi stem cell population (63%) compared to normal mammary glands (12% at 2 months and 20% at pregnancy) (Fig. S2). Although it was reported that PIK3CA-H1047R caused an expansion of the luminal progenitor population (Tikoo et al., 2012), we did not observe a difference in the CD24med/CD49fhi stem cell population between FVBN and PIK3CA-H1047R mammary glands.
To extend these findings in mouse mammary tumors to human tumor cells, we overexpressed PIK3CA-H1047R-and PIK3CA-E545K-mutant genes in the human breast cancer cell line, HS578T. Human breast cancer cells expressing mutant PIK3CA also showed activation of the WNT/b-catenin pathway, which was detected as an increase in activated b-catenin (ABC) protein (Fig. 2C), the ligand WNT3A mRNA (Fig. 2D, left panel), and a decrease in mRNA expression of the negative modulator of the pathway, the secreted frizzled-related protein 1 (sFRP-1) (Fig. 2D, right panel).

STAT3 is activated in PIK3CA-H1047R tumors
An IL-6-STAT3 loop was shown to mediate resistance to PI3K inhibitors by inducing epithelial-mesenchymal  transition (EMT) and cancer stem cell expansion in human breast cancer cells (Yang et al., 2014). Therefore, we investigated whether Il6-Stat3 signaling is induced in H1047R-mouse tumors and could be utilized as a therapeutic target in combination with PI3K inhibitors. Phosphorylated Stat3 expression was higher in the mouse mammary tumors from PIK3CA-H1047R mice in comparison with tumors from Her-2/neu transgenic mice, as determined by IHC (Fig. 3A,B) and western blot analysis (Fig. 3C). Upstream of Stat3, Il6 tended to be highly expressed (Fig. 3D) and hepatocyte growth factor (Hgf) is significantly highly expressed (Fig. 3E) in mammary tumors arising in PIK3CA-H1047R in comparison with Her-2 transgenic mice.
There was no increase in the cytotoxic effect of the combination therapies in comparison with PI3K inhibition alone on parental HS-578T breast cancer cells (Fig. 5B, a). In contrast, HS578T cells that overexpressed PIK3CA-H1047R responded more efficiently to combinations of PI3K and STAT3 inhibitors compared to single agents (Fig. 5B, b). Similar to cell lines derived from our PIK3CA-H1047R transgenic mice tumors (Fig. 5A, a) the combination of PI3K and Wnt/b-catenin inhibitors did not improve the cytotoxicity achieved by the PI3Ki alone (Fig. 5B, b). These results suggest that combination of PI3Ki with STAT3 inhibitor is more effective than combination of PI3Ki with the Wnt/b-catenin inhibitor. In addition, H1047R mouse tumor cells, RDR-C234 and RDR-A677, derived from recurrent tumors, responded more efficiently to a combination of inhibitors of PI3K (both LY294002 and GDC-0941) and IGF-1R (OSI-906) in comparison with single treatments (Fig. 5C). These results with mouse and human cells containing the PIK3CA hotspot mutation H1047R suggest that targeting specific pathways that are additionally activated in these tumors potentiates the killing effect of PI3Ki and may overcome drug resistance.
3.7. STAT3, b-catenin, and IGF pathways are induced in human PIK3CA tumors We next evaluated whether the pathways activated in the mutated PIK3CA mouse tumors involved in the cell killing effect in combination with the PI3Ki are also regulated in luminal human tumors containing the PIK3CA hotspot mutations. Analysis of 965 primary breast tumors from TCGA RNA-seq dataset showed that breast tumors harboring H1047R and other PIK3CA mutations (including E545K and E542K) have significant upregulation of AXIN2, HGF, STAT3, IGF-1, and IGF-2 mRNA compared with PIK3CA-wild-type and HER-2-amplified tumors (Fig. 6A). Estimated using the ssGSEA (single sample gene set analysis) method (Barbie et al., 2009), PIK3CA-mutant breast tumors showed activated AKT, IGF1-mTOR, and WNT canonical signaling pathways compared to PIK3CA-wild-type and HER2overexpressing tumors (Fig. 6B).
Thus, our comprehensive molecular analysis of PIK3CA-H1047R mouse and in silico analysis of human tumors revealed activation of Wnt/b-catenin, STAT3, and IGF-1R pathways. Therefore, combined targeting of pathways altered in PIK3CA-mutant tumors offers the potential to improve treatment regimens with PI3K inhibitors.

Discussion
PIK3CA, the gene encoding the catalytic subunit p110a of PI3K, is an excellent target for therapy. Since their discovery, PI3K inhibitors have been developed but have experienced limited success in the clinic. One of the underlying reasons is the possible reliance of the tumors on activated compensatory pathways occurring upon PI3K blockade. To decipher other pathways that are activated in PIK3CA-mutant tumors and with a goal of developing improved treatment regimens for mutant tumors, we generated transgenic mouse models with mammary gland-specific overexpression of the PIK3CA hot spot mutations, H1047R and E545K. We have identified key alternate pathways active in mutant tumors and have demonstrated effective therapy regimens that use combinations of PI3K-i and other pathway-specific drugs.
Characterization of the transgenic models showed that mutant mice have increased mammary ductal branching compared to control FVB/N, and H1047R mutants displayed delayed involution (Fig. 1B,C). Mouse models of PIK3CA-H1047R transgenic mice showed abnormalities in mammary gland development such as an increase in mammary ductal side branching (Adams et al., 2011;Tikoo et al., 2012). Further, transgenic mice with alteration in PI3K pathway, such as AKT activation (Schwertfeger et al., 2001), PIK3CA membrane localization (Renner et al., 2008), and PIK3CA-H1047R (Meyer et al., 2011), showed delayed involution, attributed to a reduced number of apoptotic cells. Transgenic mice with loss of PTEN exhibited excessive ductal branching, delayed involution, and severely reduced apoptosis (Jho et al., 2002). In PIK3CA-mutant mice generated by us, we observed that the PIK3CA-mutant glands have an increase in activated b-catenin and the downstream cyclin D1. The phenotypes observed in the PIK3CA-mutant mice were in line with the known role of Wnt signaling in ductal branching (Brennan and Brown, 2004). We observed mammary tumors at the lower end of the frequencies (range 25-100%) as reported by others in the PIK3CA-H1047R mice (Adams et al., 2011;Meyer et al., 2011;Tikoo et al., 2012;Yuan et al., 2013). Tumor incidence was possibly proportional to the expression levels of the mutant protein. We also generated PI3KCA-E545K transgenic mice. Here again, the incidence was low -4 (13%) of 30 PIK3CA-E545K female mice developed mammary tumors. As previously reported in PIK3CA-H1047R-knock-in mouse models (Tikoo et al., 2012;Yuan et al., 2013) and PIK3CA-E545K (Meyer et al., 2011), we observed that PIK3CA-H1047R and -E545K tumors arose with long latency (> 12 months) (Fig. 1D). It was previously reported that expression of PIK3CA-mutant H1047R induces heterogeneous tumors (Meyer et al., 2011), an observation in line with our own findings of differing histologies in the PIK3CA-H1047 tumors. Expression of PIK3CA-mutant E545K in the mouse mammary gland was also shown to induce heterogeneous tumors although it is less potently tumorigenic compared to mutant H1047R (Meyer et al., 2013). Also, similar to others (Koren and Bentires-Alj, 2013), we observed that PIK3CA-H1047R tumors are ER alpha positive (Fig. S1G).
Another interesting observation was that the expression of insulin-like growth factor 2 (Igf2) and Igf-1R was significantly higher, while expression of Igf-2R, which degrades IGF-2 (Brown et al., 2009), was lower in H1047R tumors (Fig. 4). This finding is consistent with previous observations that Igf-2R promoted endocytosis and lysosomal degradation of Igf-2, thereby antagonizing its action and acting as a tumor suppressor protein (Brown et al., 2009). These findings are also in accordance with previous reports, which showed that activation of PI3K pathway, through reduction in PTEN in breast tumor cells, is associated with increased proliferation in response to IGF-2 (Church et al., 2012). In addition, a correlation between IGF-1R and downstream PI3K pathway activation was reported in drug resistance Gallardo et al., 2012;Leroy et al., 2016), and IGF-1R and PI3K dual treatments act synergistically in targeting breast cancer cells (Ayub et al., 2015).
Here, we have demonstrated that the use of specific inhibitors to the pathways which we identified as induced in PIK3CA-mutant tumors, such as STAT3, b-catenin, and IGF-1R, potentiates the cytotoxic effect of PI3K inhibitor on mutant PIK3CA tumor cells (Fig. 5). In addition, we showed that H1047R and other PIK3CA mutants, including E545K and E542K containing BRCAs, have significant upregulation of several genes/pathways identified in the mouse PIK3CA-mutant tumors, when compared to human PIK3CA-wild-type and HER-2-positive tumors (Fig. 6), suggesting the translational potential of our findings in this paper using model systems.

Conclusions
In summary, by generating transgenic mice expressing two hotspot mutants of PIK3CA in the mammary glands and through detailed characterization of their tumors, we have identified activation of Wnt/b-catenin, Stat3, and Igf-1R pathways. We have shown that the combined targeting of these additional pathways altered in PIK3CA-mutant tumors may improve the success of treatment regimens that include PI3K inhibitors (Fig. S4).

Supporting information
Additional Supporting Information may be found online in the supporting information tab for this article: Fig. S1. Determination of PIK3CA mutant transgene copy number and expression, and characterization of mutant normal mammary gland. Fig. S2. Stem cell population in PIK3CA-H1047R normal mammary glands and tumors. Fig. S3. PIK3CA mutant tumor cells are targeted with Wnt, STAT3 and IGF-1R drugs in combination with PI3K inhibitors. Fig. S4. Schematic representation of signaling pathways activated in PIK3CA mutant breast tumor cells.