Cooperation between Pik3ca and P53 Mutations in Mouse Mammary Tumor Formation

Running title: Mutations in Pik3ca and p53 cooperate in breast cancer model Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. Abstract PIK3CA, which codes for the p110α catalytic subunit of phosphatidylinositol 3' kinase, is one of the most frequently mutated genes in human breast cancer. Here we describe a mouse model for PIK3CA-induced breast cancer using the ROSA26 (R26) knock-in system, where targeted Pik3ca alleles can be activated through transgenic expression of Cre recombinase. We mated


Introduction
The phosphoinositide 3-kinase (PI3K) signaling pathway is one of the most frequently mutated pathways in cancer.One common and direct mechanism by which the PI3K pathway is activated in breast, endometrial, colorectal, urinary tract, thyroid, and ovarian cancer is through gain-of-function mutations in PIK3CA (1-3).PIK3CA mutations are particularly important in human breast cancer, with activated alleles detected in 25% to 30% of tumors (4-7).These mutations have been found in ductal carcinoma in situ, suggesting that they play a role in breast tumor initiation (8)(9)(10).In addition, they are present at high frequency in estrogen receptor a (ERa)-positive, Her2/Neu þ and triple-negative breast tumors (TNT), suggesting an important role for mutant PIK3CA in etiology of multiple breast tumor subtypes (4).Metaplastic breast cancers have the highest percentage of PIK3CA mutations (11).Importantly, PIK3CA mutations can induce p53dependent growth inhibition (12) and mutations in both genes occur together in some human breast tumors (13,14).
PIK3CA point mutations occur most often within 2 hotspots: the helix and kinase domains (1).H1047R mutations in the kinase domain account for approximately 40% of breast cancer PIK3CA mutant alleles (4).This mutant shows elevated kinase activity (15)(16)(17) and is capable of transforming cells in culture (18), including mammary epithelial cells (19,20).Despite the high frequency, and hence importance of this allele in breast cancer, there is no animal model to test for its role in transformation of mammary epithelium.Here, we describe generation of a Cre-mediated system to conditionally express Pik3ca alleles in mouse tissues, and use of this system to test for induction of mammary tumor formation by Pik3ca H1047R .We report that ectopic expression of Pik3ca H1047R , but not Pik3ca wt , induced mammary tumors at high frequency.Furthermore, we have used the knock-in system to test for cooperation between Pik3ca and p53 (3), showing that deletion of p53 enhances tumor formation and alters the spectrum of mammary tumor subtypes that form in our Pik3ca H1047R model.

Southern blot analysis
A standard Southern blot protocol was used.Genomic DNA (10 mg) was digested with EcoRV/SstI.An approximately 940 bp probe on the 3 0 side of the targeted insertion site (CTTGAAAGTGGAGTAACTAC to TCAGAAGCTTTGAACTA-GAA in the R26 locus) was 32 P-labeled, using the Random Primers DNA Labeling System (Invitrogen 18187013).This probe binds an 11 kb fragment in nontargeted R26, and a 9 kb fragment if the R26 locus is correctly targeted.

RT-PCR on lin À mammary epithelium
Mammary glands from young virgin mice were minced and digested in collagenase/hyaluronidase (Stem Cell Technologies 07912) for 4 hours (37 C).Mammary epithelial (lineage-depleted; lin À ) cells were isolated by using the EasySep kit (StemCell Technologies 19757).RT-PCR was performed as outlined earlier in the text.

Tumor collection
Mice were humanely sacrificed at endpoint, and mammary tumors dissected.Humane endpoint was based on criteria outlined by the CCAC (tumor volume ¼ 1.7 cm 3 , tumor mass ¼ 5% of body weight, or ulcerated tumor).Part of each tumor was fixed in 10% formalin and embedded in paraffin.The remaining tumor was divided into small samples and snap frozen.

Image capture
Images were captured with an AxioCam HRm digital camera (Zeiss) by using AxioVision (release 4.6.3)software.

Generation of a Cre-inducible Pik3ca mammary tumor model
To test for the ability of Pik3ca H1047R or wild-type Pik3ca (Pik3ca wt ) to induce mammary tumors, we used the Creconditional ROSA26 (R26) knock-in system, whereby Cremediated deletion of loxP-flanked transcriptional stop sequences allows for tissue-specific expression of either allele (Fig. 1A; refs.21,22).mESC were targeted and screened to identify clones with Cre-inducible transgene expression (Supplementary Fig. 1).Cre-responsive mESC clones were then used to generate chimeric mice, and germline transmission of each knock-in allele confirmed.R26-Pik3ca H1047R and R26-Pik3ca wt mice were mated with MMTV-Cre mice, which express Cre-recombinase in mammary epithelium.Two different strains of MMTV-Cre were used: MMTV-Cre lineA , which expresses Cre recombinase in mammary epithelium at high efficiency and also in skin and lymphocytes (25); MMTV-Cre NLST , which is less efficient but more mammary restricted in its expression (Supplementary Fig. 2; refs.26,27).Mammary-inducible transgene expression was observed in R26-Pik3ca;MMTV-Cre mice but not in Cre-negative littermates (Fig. 1B).
Female R26-Pik3ca H1047R ;MMTV-Cre mice from both Cre lines showed reduced survival (Fig. 1C; Supplementary Table 1; note, males were not studied).Some animals died abruptly, whereas most were sacrificed at humane endpoint due to lethargy and impaired breathing or solid tumor growth.Decreased survival was not because of Pik3ca overexpression as R26-Pik3ca wt ;MMTV-Cre mice remained healthy (Fig. 1C).R26-Pik3ca H1047R ;MMTV-Cre mice started developing mammary tumors at 5 months of age.Mammary tumors were observed in both virgin and multiparous mice using either MMTV-Cre strain.R26-Pik3ca H1047R ;MMTV-Cre lineA mice reached endpoint more rapidly than R26-Pik3ca H1047R ; MMTV-Cre NLST mice (Fig. 1C).In addition, R26-Pik3ca H1047R ;MMTV-Cre lineA mice reached endpoint with a more diverse set of tumor types.69% of NLST mice and 42% of line A mice and had palpable mammary tumors at endpoint (Fig. 1D).For this reason, we used MMTV-Cre NLST transgenics for subsequent studies.Development of lymphoma/thymoma, skin, and other nonmammary tumors was attributed to Cre expression in other tissues (25).Interestingly, survival of R26-Pik3ca H1047R mice, without Cre, was reduced in comparison with MMTV-Cre controls (Fig. 1C; Supplementary Table 1; P < 1Â10 À6 ).This was due to development of blood vessel lesions in some R26-Pik3ca H1047R animals (data not shown), potentially through spontaneous activation of Pik3ca H1047R expression in endothelial cells (28).
Histologic analysis revealed that mammary tumors from R26-Pik3ca H1047R ;MMTV-Cre NLST mice were typically either adenosquamous carcinoma or adenomyoepithioma (Fig. 3A; Supplementary Fig. 4).Adenosquamous tumors contained cystic regions composed of laminar keratin lined with squamous epithelium and adjacent glandular elements.Some tumors had squamous nodules or necrosis, and several had invasive margins.Expansile tumors were also observed.Adenomyoepitheliomas contained a mixture of glandular epithelium and interstitial fusiform cells with abundant polar cytoplasm.These myoepithelial stromal cells represented 10% to 80% of the tumor.Isolated lung metastasis were observed in rare R26-Pik3ca H1047R ;MMTV-Cre NLST mice (data not shown).Mammary tumors from p53 f/þ ;MMTV-Cre NLST mice were either spindle/epithelial-mesenchymal transition (EMT) type or poorly differentiated adenocarcinoma as previously reported (Fig. 3B; Supplementary Fig. 4; refs.[29][30][31].Spindle tumor cells had fusiform nuclei and polar cytoplasm.Swirling patterns and necrotic areas were observed.Poorly differentiated adenocarcinomas were composed of solid sheets of cells with little tissue architecture and some necrosis.Cells had large, pleomorphic nuclei and dark staining cyto-plasm.Interestingly, mammary tumors that formed in R26-Pik3ca H1047R ;p53 f/þ ;MMTV-Cre NLST double-mutant mice were typically either spindle/EMT or adenosquamous carcinoma (Fig. 3C; Supplementary Fig. 4).Radial scar and poorly differentiated adenocarcinomas were also observed, although at a lower frequency (Fig. 3C; Supplementary Fig. 4).Radial scar tumors were small and had a stellate outline composed of dense connective tissue and a center with distorted neoplastic glands.Gross metastasic lesions were not observed in R26-Pik3ca H1047R ;p53 f/þ ;MMTV-Cre NLST mice (data not shown).

Akt activation in Pik3ca H1047R and p53 mutant mammary tumors
To test for PI3K pathway activation, lysates from representative tumors were analyzed for Akt phosphorylation, and for expression of PTEN and phospho-c-Jun (S73).Western blot signals were normalized to b-actin protein expression, and compared with levels in mammary lysates from a non-tumorbearing 74-week-old R26-Pik3ca H1047R virgin female.Adenosquamous carcinomas from R26-Pik3ca H1047R ;MMTV-Cre NLST mice showed elevated accumulation of phospho-Akt S473 , PTEN, and phospho-c-Jun S73 (Fig. 4).Adenosquamous carcinomas from R26-Pik3ca H1047R ;p53 f/þ ;MMTV-Cre NLST showed increased PTEN and phospho-c-Jun S73 expression.Surprisingly, however, no change in phospho-Akt was observed in these tumors.The reason for this is not clear, perhaps representing a change in PI3K signaling such as a reduced dependence on Akt activation (32).Alternatively, the wild-type sample used for normalization in this experiment was whole mammary gland lysate.This tissue contains a large amount of insulin-responsive fat, and may show higher basal levels of Akt activation than mammary epithelium.We therefore tested additional adenosquamous tumors from R26-Pik3ca H1047R ; MMTV-Cre mice for phospho-Akt and used lin À mammary epithelium as control.Indeed, phospho-Akt T308 and phospho-Akt S473 were significantly elevated in 12 adenosquamous carcinomas from R26-Pik3ca H1047R ;MMTV-Cre mice (Supplementary Fig. 5).
Elevated levels of phospho-Akt T308 and phospho-Akt S473 were observed in lysates from adenomyoepithelioma lesions in R26-Pik3ca H1047R ;MMTV-Cre NLST mice and in lysates from R26-Pik3ca H1047R ;p53 f/þ ;MMTV-Cre NLST spindle/EMT tumors.Interestingly, elevated levels of phospho-Akt T308 and phospho-Akt S473 were also seen in lysates from spindle/EMT and in poorly differentiated adenocarcinomas from p53 f/þ ;MMTV-Cre NLST mice.Elevated levels of PTEN expression and of phospho-c-Jun accumulation were observed in each mammary tumor type studied.In the case of PTEN, this is likely a result of PI3K pathway-induced phosphorylation and stabilization of PTEN, a negative feedback loop which functions to decrease PIP 3 levels (33).Also, increased phospho-c-Jun is associated with elevated PI3K pathway activation (34).Finally, it has been suggested that activation of other PDK1 substrate kinases, such as SGK3, may play an important role downstream of PIK3CA gene mutation in breast cancer (32).We therefore analyzed SGK3 phosphorylation in these tumors.Phospho-SGK3 S486 was not elevated in any of the tumor types analyzed (Supplementary Fig. 6).
To quantify cell proliferation and apoptosis in each tumor type, we performed Ki67 immunohistochemistry and TUNEL analysis, respectively (Table 1; Supplementary Table 3).Each tumor type examined had very low levels of TUNEL-positive nuclei (<1% in all cases).This result is not surprising given that PI3K pathway activation and p53 gene deletion are both associated with enhanced cell survival.In contrast, the percentage of Ki67-positive nuclei varied with genotype and tumor type.Mammary tumors from R26-Pik3ca H1047R ; MMTV-Cre NLST mice had the highest proportion of Ki67positive nuclei, whereas p53 f/þ ;MMTV-Cre NLST tumors had the lowest.In tumor type analysis, adenosquamous carcinomas had the highest Ki67 index whereas poorly differentiated carcinomas had the lowest.

Discussion
PIK3CA is commonly mutated in human breast cancer.The PIK3CA locus is also amplified in some breast cancers (6,40,41).To test for initiation of mammary tumor formation by PIK3CA alleles we generated syngeneic R26-Pik3ca H1047R and R26-Pik3ca wt knock-in mice.Each strain carried a Cre-inducible allele of mouse Pik3ca, and each was crossed to MMTV-Cre mice to activate expression within mammary epithelium.Starting at 5 months, R26-Pik3ca H1047R ;MMTV-Cre females developed mammary tumors at a high frequency.Importantly, R26-Pik3ca wt ;MMTV-Cre mice did not.Thus, low-level ectopic expression of p110a H1047R , but not p110a wt , is sufficient to initiate mammary tumor formation in mice.This result is consistent with identification of H1047R mutations in premalignant lesions (8)(9)(10).We did not observe any gross abnormalities in glands of young R26-Pik3ca H1047R ;MMTV-Cre NLST mice, likely because of inefficient Cre-deletion in this line.It will be important to define early responses to PIK3-CA H1047R gene activation in our model, perhaps using line A or through ex vivo, Cre-induced, gene activation and transplantation.
PIK3CA mutations occur in all major subtypes of human breast cancer (4)(5)(6)(7)11).Our mouse model of Pik3ca H1047Rinduced breast cancer develops adenosquamous carcinoma and adenomyoepithelioma of the mammary gland at high frequency.Both tumor types were ERa þ .Both also contained luminal and myoepithelial keratin-positive cells, suggesting that the cell of origin had bilineage potential.As PIK3CA and TP53 are the 2 most commonly mutated genes in human breast cancer, and mutations in both occur together in many tumors, we tested for cooperation between these genes in our model.Indeed, we observed reduced survival of R26-Pik3ca H1047R ;p53 f ;MMTV-Cre NLST double-mutant animals.Cooperation between Pik3ca H1047R and p53 loss-of-function mutation is in contrast to the situation observed in MMTV-myrPik3ca mice, in which mammary tumor formation is enhanced by expression of CDK4 R24C but not by deletion of one p53 allele (42).In our model, apoptosis was very low in all of the tumors analyzed.In contrast, the proliferation rate varied significantly from one tumor type to another, tracking more closely with tumor type than with genotype.Thus, the observed cooperation between pik3ca H1047R and p53 loss-offunction mutation is not at the level of apoptosis or proliferation.Indeed, data presented in Figures 2A and 3C indicate that pik3ca H1047R and p53 mutations cooperate in mammary tumor initiation and interact to control mammary tumor type.In total, on wild-type and p53 mutant backgrounds, we observed  5 distinct types of mammary tumors in our PIK3CA model: adenosquamous carcinoma, adenomyoepithelialomas, spindle/EMT tumors, poorly differentiated adenocarcinomas, and radial scar type lesions.Although some of these tumor types are quite rare in humans, poorly differentiated adenocarcinomas are common, as are spindle/EMT tumors (Basal B tumors in humans).In addition, adenosquamous carcinomas are metaplastic, a tumor type in humans with a high frequency of PIK3CA mutations (11).This result is consistent with the wide spectrum of human breast cancers found to have PIK3CA mutations.In addition, our diverse collection of tumors includes ERa þ tumors and TNT-type tumors.This result contrasts with tumor models using Neu, Wnt, Myc, or Polyoma middle T, each of which induce a very specific mammary tumor type in mice (43).We tested for PI3K pathway activation in major mammary tumor types that developed in our PIK3CA H1047R model mice.To this end we analyzed Akt phosphorylation, which was elevated in each tumor type.The modest level of Akt activa-tion noted was somewhat surprising, although this was related to the use of whole mammary gland control tissue rather than lin À mammary epithelium for normalization (Supplementary Fig. 5).In any case, Akt-independent transformation has been noted in some breast tumors with a PIK3CA mutation (32), and other PDK1 substrates are thought to play a role in this case.Indeed, transgenic mice expressing activated alleles of Akt1 from the MMTV LTR do not form mammary tumors, highlighting the importance of other PI3K pathways in breast cancer (44,45).Future studies will be required to define the importance of specific PIP 3 -responsive kinases and pathways in our Pik3ca H1047R model.
By using the Cre-conditional knock-in system, we can compare mammary tumor formation in response to distinct alleles of Pik3ca, each expressed at precisely the same level and in the same cell types.For example, we can now directly compare mammary tumors in our R26-Pik3ca H1047R model with tumor formation in mice expressing helical domain mutants of Pik3ca.As our system is Cre-inducible, we can also compare mammary tumor formation in R26-Pik3ca H1047R ; MMTV-Cre NLST mice and in Pten f/f ;MMTV-Cre NLST mice, in which PI3K pathway activation will occur in the same cell of origin but through a distinct mechanism.The importance of cell of origin can now be probed in our system as transgenic mice are available to activate Pik3ca alleles in luminal committed cells (WAP-Cre) and in bipotential cells expressing cytokeratin 14 (K14-Cre).Finally, this model can be used to identify genetic and cellular events associated with progression/metastasis, to screen for novel therapeutic targets in Pik3ca H1047R mammary tumor cells ex vivo and to develop preclinical data on PI3K pathway inhibitors for treatment of breast cancer.

Figure 4 .
Figure 4. Akt activation in mammary tumors from Pik3ca H1047R model mice.A, Western blot analysis of PI3Kpathway components and phospho-c-Jun.B, Western blot signals were quantified and normalized with respect to b-actin.Mean fold increase compared with mammary lysate from a control mouse was calculated after normalization.Bars, group mean values; error bars, AESE; asterisks, significant increase in expression (1-sided 1-sample t test; a ¼ 0.05; *, P < 0.05 and **, P < 0.1).Yellow dashed line, protein expression in a normal mammary gland.