Profiling of Residual Tumor Cells in Chemotherapy-treated MMTV-PyMT Mice
To determine the response of single tumor foci to cytotoxic chemotherapy, we utilized the MMTV-PyMT hyperplasia transplant system 
. Focal hyperplasias were resected from 3-week-old MMTV-PyMT;actin-GFP mice and orthotopically transplanted into the cleared mammary fat pads of FVB/n host mice (). Tumor outgrowths displayed stereotyped tumor growth and histologic progression as compared to MMTV-PyMT mice, with progression to early carcinoma and late carcinoma occurring at 8 and 15 weeks post-transplantation, respectively (). The hyperplasia transplant model is amenable for studying the individual steps in malignant progression, such as angiogenesis, tumor dissemination, and metastasis formation ().
Modeling breast cancer residual disease in the MMTV-PyMT hyperplasia transplant model.
Tumor-bearing mice at the early carcinoma stage (8 weeks post-transplant) were treated with cytotoxic chemotherapy to identify residual tumor cells. We modeled the clinical setting by administering the TAC chemotherapy regimen (taxol, adriamycin, cyclophosphamide), which is given to patients with poorly-differentiated breast adenocarcinoma 
. Tumor-bearing mice were treated with 25 mg/kg docetaxel, 5 mg/kg doxorubicin, and 120 mg/kg cyclophosphamide every 21 days, similar to patients. Mice lost 8–10% body weight after each dose but recovered within 21 days of dosage (data not shown). Tumors underwent significant regression within seven days after the first dose of chemotherapy. By day 10, tumors stabilized and started to grow once again (). Within 21 days after the first dose, the tumors relapsed and grew nearly to the size of the original tumor. Subsequent administration of chemotherapy every 21 days led to similar patterns of tumor regression and relapse (). We collected the untreated tumors, residual tumors at day 10, and relapsed tumors at day 80 for further analysis. GFP-positive tumor cells from these samples were FACS-sorted, and RNA was harvested for mRNA microarray profiling.
Enrichment of Jak/Stat, Notch, and Epigenetics Genes in Residual Tumors
Microarray profiling of untreated, residual, and relapsed tumors revealed a number of signaling pathways and biomarkers enriched in residual tumors (GEO Microarray Omnibus, accession number GSE43566). These include the Jak/Stat pathway, DNA damage response/repair pathways (Atm/Atr and Brca1-mediated), and Akt signaling pathways (). Ingenuity pathway analysis of microarray data identified Stat1 as an important signaling node in residual tumors (). Interestingly, several genes in the Ifn-γ/Jak/Stat pathway, such as Gbp1, Gbp3, Gbp4, Ifi47, Tgtp, and Stat1, were up-regulated in residual tumors relative to untreated and/or relapsed tumors (). We validated the enrichment of these genes in residual tumors with Fluidigm RT-PCR. GFP-positive tumor cells were sorted from adenomas (5-week outgrowths), carcinomas (18-week outgrowths), residual tumors, disseminated tumor cells in lungs, and lung metastases for this analysis. In agreement with microarray data, several members of the Ifn-γ/Jak/Stat pathway, including Stat1, Ifngr1, Tgtp1, Ifit1, Gbp3, and Gbp4, were significantly up-regulated in residual tumors relative to the other groups (). A subset of the genes, such as Stat1 and Gbp4, were also up-regulated in adenoma tumors, suggesting that elements of the pathway were active in early breast tumors.
Gene families enriched in residual MMTV-PyMT tumors after chemotherapy.
Ifn-γ/Jak/Stat signaling in MMTV-PyMT residual tumor cells after chemotherapy.
We further validated the enrichment of Jak/Stat pathway genes in residual tumors at the protein level. Stat1 immunohistochemistry of untreated MMTV-PyMT adenocarcinomas identified sub-populations of Stat1-positive tumor cells (). The number of Stat1 positive cells and the ratio of Stat1 positive-to-negative cells increased significantly in residual tumors after chemotherapy (′). Relapsed tumors contained only small Stat1 sub-populations, much like untreated tumors, suggesting that enrichment of Jak/Stat genes was specific to the residual disease state (Figure 3A″). Co-immunofluorescence with a macrophage marker showed that Stat1 was expressed in tumor cells, as was reported in breast cancers () 
. Further, p-Stat1-positive cells were detected in untreated and residual tumors. Interestingly, p-Stat1-positive tumors cells were also found in clustered sub-populations in tumors (′).
Enrichment of a Stat1-positive tumor sub-population in MMTV-PyMT residual tumor cells after chemotherapy.
We determined the kinetics and mechanism of Stat1 activity in residual tumors after chemotherapy. We collected tumor lysates at days 1, 3, 5, and 7 after chemotherapy and blotted for Stat1 and p-Stat1, -3, and -5 proteins (). Stat1 and p-Stat1 levels were significantly increased by day 7 after chemotherapy, corresponding to development of residual disease. Interestingly, Stat1-positive tumor cells in residual tumors displayed significantly less DNA damage than Stat1-negative tumor cells (). This suggested that Stat1 positive tumor cells, which pre-existed in untreated tumors, were inherently resistant to DNA-damaging agents, as reported 
. This may result from more active DNA repair mechanisms in Stat1 positive cells 
. The increased DNA repair activity of Stat1-positive tumor cells may have primed tumor cells for drug tolerance and tumor relapse after chemotherapy in this model.
In addition to Jak/Stat family genes, we identified Notch family members as biomarkers of residual tumor disease in this model. We first determined the expression levels of Notch family members in whole tumor lysates, which included both tumor and stromal populations. Taqman RT-PCR analysis confirmed the up-regulation of Notch-1, Notch-2, Notch-3, and Dll-1 in residual tumors relative to untreated or relapsed tumors (). We then performed Fluidigm RT-PCR analysis on sorted tumor populations to determine the tumor cell specificity of Notch gene expression. Sorted residual tumor cells were compared to sorted tumor cells from adenomas, carcinomas, disseminated tumor cells, and lung metastases. We found that Notch-1 and Dll1 were specifically up-regulated in residual tumors compared to primary tumors, disseminated cells, or metastases. Dll1 demonstrated elevated expression in both residual and untreated adenoma tumors, similar to Stat1 expression ().
Up-regulation of Notch family members in MMTV-PyMT residual tumors.
A number of chromatin-modifying genes, in particular histone methyltransferases, were significantly up-regulated in residual tumors. We validated the enrichment of these genes in sorted residual tumor cells by Fluidigm RT-PCR analysis. A number of methyltransferases specific to H3K4 (Mll1, Mll3, Setd1a), H3K9 (Ehmt2, Suv39h1, Setdb1) and H3K27 (Ezh1) were up-regulated in residual tumors compared to untreated primary tumors, disseminated cells, or metastases (). Dyrk3 (a histone-modifying kinase) and Ash1l (a methyltransferase) were also enriched in residual tumors (). In some cases, these genes were uniquely up-regulated in residual tumors (e.g., Setd1a, Dyrk3), while in other cases the genes were up-regulated in both residual and untreated adenoma tumors (e.g., Suv39h1, Mll1) ().
Up-regulation of chromatin modifying gene in MMTV-PyMT residual tumors.
Profiling of Disseminated Tumor Cells and Metastases in MMTV-PyMT Mice
The MMTV-PyMT hyperplasia transplant model allows the collection of disseminated tumor cells in various organs, including lung, spleen, brain, and liver (′). The vast majority of disseminated cells are single cells in the microvasculature of distant organs () 
. The number of disseminated tumor cells in distant organs rises in proportion to tumor size. In 18-week adenocarcinoma outgrowths, >5000 disseminated tumor cells can be detected in the lungs of tumor-bearing mice. However, primary tumor removal experiments showed that less than 0.1 percent of these disseminated tumor cells in lung are capable of forming lung metastases. Further, although disseminated cells can be detected in many organs, metastases only arise in the lungs () 
. This model can be used to determine the mechanisms of tumor dissemination and metastasis formation in an orthotopic, immunocompetent setting.
Profiling of disseminated cells and metastases in MMTV-PyMT hyperplasia transplant model.
We FACS-sorted GFP-positive tumor cells from adenomas (5 week outgrowths), carcinomas (18 week outgrowths), disseminated tumor cells from lungs, and lung metastases for microarray expression profiling (GEO Microarray Omnibus, accession number GSE43566). When compared to primary tumors or disseminated cells, lung metastases showed substantial differences in gene expression, with 3500–4000 genes differentially expressed (2-fold level) relative to adenoma or carcinoma (). For comparison, chemotherapy-treated residual or relapsed tumors had ~600 differentially expressed genes (2-fold level) relative to untreated tumors. Interestingly, disseminated cells appeared more similar to carcinoma than to metastases (1000 versus 2400 differentially expressed genes, respectively) (). These data indicate that lung metastases in the MMTV-PyMT model are molecularly distinct at the mRNA expression level from primary tumors or disseminated tumor cells.
Enrichment of JAK/STAT Genes and Epigenetic Regulators in Disseminated Tumor Cells and Metastases
We performed hierarchical clustering and pathway analysis of genes differentially expressed in metastases relative to primary tumors. The gene families most represented in differentially expressed genes were stem-cell-associated markers, including markers of neural and embryonic stem cells (Figure S1
). A number of stem-cell-related genes, such as Fzd3, Xiap, Sox2, and several Smad members were up-regulated in metastases relative to primary tumors. Given the substantial up-regulation of stem cell-associated genes, including cell fate determination genes and transcription factors, in metastases relative to primary tumors, we hypothesized that metastases had undergone epigenetic reprogramming during cancer progression. To test this, we performed hierarchical clustering of chromatin-modifying gene expression in metastases, disseminated tumor cells and primary tumors. We identified a cluster of lysine and arginine methyltransferases that were highly expressed in metastases (). Quantitative analysis of microarray data confirmed that these methyltransferases, which included Setd3, Setd5, Suv39h2, Smyd3, Prmt3, Prmt6, Nsd1, and Nsd2 were up-regulated in metastases compared to adenomas, carcinomas, or disseminated cells (Table S1
). A number of these genes catalyze the methylation of H3K4 and H3K9 residues. The enrichment of these methyltransferases suggested that metastases had acquired specific epigenetic markers during cancer progression.
We assessed the methylation status of H3K4, H3K9, and H3K27 in metastases and primary tumors in the MMTV-PyMT transplant model. Lung metastases had higher H3K4 trimethylation relative to adenomas and carcinomas, although the majority of this increase could be accounted for in the adenoma-carcinoma transition (). H3K9 trimethylation was more heterogeneous across tumors and metastases. Metastases demonstrated significantly elevated H3K9 trimethylation relative to tumors. In contrast, metastases had similar or slightly elevated levels of H3K27 trimethylation relative to adenomas or carcinomas (). Further analysis of H3K4 methylation status revealed that H3K4 tri-methyl, H3K4 di-methyl and H3K4 mono-methyl marks were increased in metastases relative to primary tumors. In all cases, carcinomas had increased levels of these markers relative to adenomas. Interestingly, H3K4 tri-methyl and H3K4 mono-methyl marks appeared more represented than H3K4 di-methyl across the tumor types ().
Histone-3 methylation marks in cancer progression.
Mining of microarray data was performed to identify biomarkers of disseminated tumor cells in this model. The Il-6/Jak/Stat pathway members emerged as biomarkers for disseminated tumors cells in lungs (). Il-6 mRNA levels were significantly up-regulated in disseminated cells relative to metastases, primary tumors or residual tumors. Il6ra was also enriched in disseminated cells, though it was also elevated in in residual tumors. Prdm1, a transcriptional repressor and effector of Il-6 signaling, was also uniquely up-regulated in disseminated tumor cells, suggesting that active Il-6/Jak/Stat signaling occurs in this population ().