Myeloid-Derived Suppressor Cells Endow Stem-like Qualities to Breast Cancer Cells through IL6/STAT3 and NO/NOTCH Cross-talk Signaling

The capacity of immunity to control and shape cancer progression (1, 2) has been the subject of intensive investigation. Active protective immunity contributes to tumor dormancy during immunoediting (3). Inhibitory immune elements form immune-suppressive networks in the tumor microenvironment and are the main obstacles for developing effective cancer immunotherapies (4, 5). However, it is poorly understood whether the immune-suppressive networks reshape cancer and cancer dormancy in specific human cancer, and in turn affect tumor progression, metastasis, and therapeutic response in the immune competent host.

It is thought that dormant cancer cells may be stem-like cells with high metastatic- and therapeutic-resistant potential. The definition of cancer stem cells (CSC) remains largely operational. Nonetheless, CSCs are thought to be involved in tumor initiation, progression, metastasis, and therapy resistance (6–8). In vivo, the stemness of cancer cells is not exclusively intrinsic to CSCs. Extrinsic mechanisms (9) provided by the tumor microenvironment may be essential for forming the stem cell niche (10). For example, mesenchymal stem cells (10), myeloid cells (11–13), and T-cell subsets (14, 15) promote tumor metastasis. It is poorly understood whether breast CSCs are the primary targets of these cells (10, 14, 16, 17).

STAT3 activation is often observed in cancers, particularly metastatic sites (18, 19). Persistently, activated STAT3 in tumor cells acts as a crucial oncogenic mediator and promotes tumorigenesis (18, 20–23). Many factors can activate transiently STAT3. However, it is not well understood how STAT3 is persistently activated in the cancer cells and what is the functional and clinical consequence of persistent STAT3 activation in human breast cancer.

In this report, we outline extensive studies on the interaction between myeloid-derived suppressor cells (MDSC) and CSCs in animal models and patients with breast cancer, dissect cellular, and molecular mechanisms by which MDSCs reshape breast cancer stemness via STAT3 and NOTCH cross-talk, and reveal the pathologic, clinical, and therapeutic relevance of the interaction between MDSCs and CSCs in breast cancer.

Patients diagnosed with breast cancer were recruited in the studies. All use of human subjects in this study was approved by the local Institutional Review Boards. We collected fresh breast cancer tissues from the University of Michigan Surgery Clinic. Fresh tumors were processed into single cell suspension and immediately used for MDSC and tumor (stem) cell analysis (13, 15, 24, 25). Specifically, tumor cell suspensions were prepared from solid tumors by enzymatic digestion in 50 mL of Hank's Balanced Salt Solution (Life Technologies) containing 40 mg of collagenase, 4 mg of DNase I, and 100 units of hyaluronidase (Sigma Co.) for 2 hours. Cells were washed twice in HBSS. MDSCs were enriched by depleting tumor cells, B- and T-cell subsets (PE-selection kit; StemCell Technology) and 7-AAD exclusion, and sorted with high-speed sorter FACSaria (Becton Dickinson) to high purity (>97%). We studied formalin-fixed, paraffin-embedded breast cancer tissues from 104 Her-2/neu⁺ patients in Poland (cohort 1; refs. 26, 27), and from 84 breast cancer patients from the University of Michigan School of Medicine (cohort 2) and 90 breast cancer patients from Poland (cohort 3) for this study (Supplementary Table S1). Cohort 1 was initially recruited for clinical trials with Herceptin treatment (26, 27). Cohorts 2 and 3 were randomly recruited regardless of Her-2 expression status. After pathologic review, a tissue microarray (TMA) was constructed from the most representative area of paraffin-embedded breast cancer tissues. The TMAs were used for specific immunohistochemistry staining. Five hundred and ninety three patients with breast cancer were evaluated in The Cancer Genome Atlas (TCGA) breast datasets in Oncomine.org

Immunohistochemistry staining was performed as previously described (28–30). Tissues were stained with polyclonal rabbit anti-human-CD33 (1/10 dilution, DAKO; ref. 13), or mouse anti-human ALDH1 (Becton Dickinson; ref. 24). We used horseradish peroxidase–based detection system to detect positive cells (EnVision, DAKO). The specimens were digitalized with an automated platform (Aperio Technologies) and ScanScope XT and Spectrum Plus using TMA software version 9.1 scanning system. Cores were manually scored in high resolution of ×40. A mean score of duplicate cores from each individual tissue was calculated. Any discrepancies were resolved by subsequent consultation with diagnostic pathologist. The cores were quantified and analyzed for the expression of ALDH1 and CD33 with an Aperio imaging system (Genetix). Cytoplasmic expression of ALDH1 was evaluated, whereas nuclear staining alone was considered nonspecific and was not included in the analysis. Intensity of staining was scored as 0 (no expression), 1+ (less than 1% of positive cells), 2+ (1%–5% of positive cells), 3+ (5%–20% of positive cells), and 4+ (more than 20% positive cells). The intratumoral CD33⁺ cells were quantified and expressed as the numbers of CD33⁺ cells per 0.6 mm² of tumor section. The tissues were divided into high and low CD33⁺ cell infiltration based on the median value. The optimal cutoff points were calculated using the ROC curve. Human vimentin and E-cadherin were detected in paraffin-embedded MCF-7 cells with the PathScan EMT Duplex IF Kit (Cell Signaling Technology).

MCF-7 and MDA-MB-231 cell lines (ATCC) were characterized and authenticated by the vendor using short tandem repeat DNA profiling. The cell lines were passaged in our laboratory for fewer than 6 months. All cell lines were maintained in DMEM supplemented with 10% FBS plus 100 U/mL penicillin and 100 μg/mL streptomycin (P/S).

Human xenograft tumor model was established in female NOD-scid IL2Rγnull (NSG) mice with modifications (13, 24, 25). Briefly, MCF-7 breast cancer cells plus MDSCs were mixed with anti-IL6 (200 μg, mouse IgG2b; R&D Systems) and iNOS inhibitor (L-NMMA, 50 μmol; EMD Millipore), and were subsequently inoculated subcutaneously into NSG mice supplied with estradiol-17β pellet. Tumor development was monitored and tumor size was measured.

CD45⁺CD33⁺CD14⁺CD15^−/dim and CD45⁺CD33⁺CD14⁻CD15^bright MDSC subsets were sorted from breast tumor tissues. MDSCs were cultured with T cells (4 × 10⁴) at different ratio in the presence of 2.5 μg/mL anti-human CD3 and 1.25 μg/mL anti-human CD28 for 3 days. T-cell proliferation was determined by Ki67 expression and CFSE dilution. T-cell cytokine expression was evaluated by intracellular staining. The analysis was gated on CD3⁺ T cells.

MCF7 cells were transfected with GFP-expressing lentiviral vectors encoding shSTAT3 (Supplementary Table S2) or nonfunctional scrambled control. After transfection, the transfected cells were selected and cultured for further experiments.

Single cell suspensions were made from different organs and tumor tissues. Cells were labeled with fluorescence-conjugated antibodies to CD3, CD4, CD5, CD11b, CD14, CD15, CD16, CD19, CD33, CD45, HLA-DR, ALDH, and STAT3 (BD Pharmingen). The cells were analyzed by LSR II (BD Biosciences). The primary cells were sorted from fresh tissues by high speed sorter ARIA (BD Biosciences).

Cell lysates were prepared with SDS lysis buffer and clarified by centrifugation, and protein concentration were determined by the BCA Protein Assay Kit (Thermo Scientific). Equivalent amounts of total cellular proteins were separated by SDS-PAGE, and transferred onto polyvinylidene difluoride membranes. Proteins are detected with the antibodies against STAT3 and p-STAT3 (Y705; Cell Signaling Technology), NOTCH1 (Abcam), NICD (Cleaved NOTCH1 Val1744; Cell Signaling Technology), and GAPDH (6C5; Santa Cruz Biotechnology). Epithelial-mesenchymal transition (EMT)-associated proteins were detected with the Epithelial–Mesenchymal Transition (EMT) proteins Antibody Sampler Kit (Cell Signaling Technology).

Real-time PCR was performed as we described previously (24, 31). All the primers were included in the Supplementary Information (Supplementary Table S2). Cytokine production was measured by ELISA as described by the manufacturer's protocol (R&D Systems; DY206). Nitric oxide (NO) was detected in cultured supernatant according to the manufacturer's protocol (Total NO and Nitrate/Nitrite Parameter Assay Kit; R&D Systems, KGE001) and normalized to the medium control.

MDSCs were sorted from human breast tumor tissues by high-speed sorter (FACSAria; BD Biosciences) as described previously (13). Sphere assay was performed as described previously (24, 32). Briefly, tumor cells or sorted tumor cell subsets were plated in ultra-low attachment plates (Corning) in serum-free EBM-2 or X-VIVO medium (Lonza) supplemented with 5 μg/mL insulin (Sigma), 20 ng/mL human recombinant EGF (Invitrogen), at a density of 1,000 to 10,000 viable cells per well. Spheres (>50 μm) were counted after 1 to 6 weeks.

Human breast cancer cells (10⁶/mL) were cocultured with MDSCs in Transwell system with 3 μmol/L Notch inhibitor, γ-secretase inhibitor I (Z-LLNLe-CHO; Calbiochem) or/and 500 nmol/L STAT3 inhibitor (Cucurbitacin I, Calbiochem). Tumor cells were subjected to genetic and functional analyses.

The Wilcoxon signed-rank test was used to determine pairwise differences and the Mann–Whitney U test was used to determine differences between groups. P < 0.05 was considered as significant. All statistical analysis was done on Statistica software (StatSoft Inc.). Overall patient survival was measured from the date of diagnosis to tumor-related death. Data were censored at the last follow-up for patients who were alive at the time of analysis. Spearman correlation coefficients were computed to assess relationships between MDSCs and ALDH. Survival curves were constructed using the method of Kaplan–Meier and survival differences were assessed using the log-rank test. The Cox proportional hazards model was used to assess the effect of MDSC infiltration after adjusting important prognostic factors, including cohort, histotype, tumor type, stage, grade, and treatment. Statistical significance was defined as a P value of <0.05. All analyses were performed using SAS 9.3 software.

We isolated myeloid cells from human breast cancer for the phenotypic, molecular, functional, and clinical studies. Polychromatic flow-cytometry analysis demonstrated that there existed substantial CD45⁺ immune cell subsets, including CD19⁺CD45⁺ cells, CD5⁺CD45⁺ cells and CD5⁻CD19⁻CD33⁺CD45⁺ cells in fresh breast cancer tissues (Fig. 1A). CD5⁻CD19⁻CD33⁺CD45⁺ cells expressed high levels of CD11b and low to medium levels of CD14, CD15, CD16, and HLA-DR (Fig. 1B and C). There were 22% of lin⁻CD33⁺CD11b⁺CD45⁺ cells in total CD45⁺ immune cells in fresh breast cancer tissues (Fig. 1D). The lin⁻CD33⁺CD11b⁺CD45⁺ cells were sorted by high speed sorter to high purity (>97%) and were subjected to a suppression assay. These cells suppressed T-cell proliferation as shown by reduced Ki67 expression in T cells (Fig. 1E), decreased CFSE-labeled T-cell divisions (Fig. 1F) and effector T cells (Fig. 1G). The percentages of CD14⁺CD15^−/dim cells were higher than that of CD14⁻CD15^high cells (Fig. 1C) and the two subsets were capable of inhibiting T-cell proliferation (Supplementary Fig. S1B). On the basis of the phenotype and immune suppressive capacity, lin⁻CD33⁺CD11b⁺CD45⁺ cells are referred as MDSCs.

Given the high levels of CD33 expression on MDSCs, we attempted to quantify MDSCs with CD33 in the paraffin-fixed breast cancer tissues. When we stained single cells from fresh breast cancer tissue cells with anti-CD33 (Supplementary Fig. S1A), we found that CD33^high cells were basically confined toCD3⁻CD19⁻CD45⁺ cells (Supplementary Fig. S1A). Thus, CD33 may be an operational marker to phenotypically define MDSCs in paraffin-fixed breast cancer tissues. We quantified CD33⁺ cells with immunohistochemical staining (IHC) in three patient cohorts (Supplementary Table S1). The European patient cohort 1 (cohort 1) included 104 treatment-naïve Her-2/neu⁺ primary breast cancer patients. The Michigan patient cohort (cohort 2) and the Poland patient cohort 3 (cohort 3) included 84 and 90 breast cancer patients, respectively, regardless of their Her-2/neu status (Supplementary Table 1S). We observed that the levels of CD33⁺ cells were variable from patient to patient (Fig. 2A). However, the levels of CD33⁺ cells were comparable among three cohorts (Supplementary Fig. S2). For survival analyses, similar to our tumor-associated regulatory T-cell analysis (29), we summed the three cohorts and divided the patients into high and low groups based on the median levels of CD33⁺ cells.

Kaplan–Meier analyses indicated that high levels of CD33⁺ cells correlated with reduced overall survival (OS) compared with low levels of CD33⁺ cells in the total patient population in univariate analysis (Fig. 2B; Supplementary Table S1). As patients in cohort 1 were exclusively Her2⁺, to avoid potential bias due to patient distribution, we independently analyzed the cohort 1 and the combined cohorts 2 and 3. We found that high levels of CD33⁺ cells were associated with reduced OS compared with low levels of CD33⁺ cells in the cohort 1 (Fig. 2C) and the combined cohorts 2 and 3 (Fig. 2D). In a multivariate analysis, including covariates of cohorts, histology, tumor type, stage, treatment, and grade, high MDSC infiltration was again associated with shorter survival in all the three cohorts (Supplementary Table S3), the cohort 1 (Supplementary Table S4) and the combined cohorts 2 and 3 (Supplementary Table S5). Thus, MDSCs are functionally and clinically important in patients with breast cancer.

Next, we investigated the mechanisms by which MDSCs are associated with poor patient outcome. CSCs contribute to tumor progression and therapeutic resistance (7, 8, 33). We reasoned that MDSCs might affect CSC biologic behavior. We showed that human breast cancer–associated MDSCs promoted MCF-7 breast cancer sphere formation (Fig. 3A). ALDH-1⁺ cells are enriched with CSCs in breast and ovarian cancer cells (24, 34). We observed that MDSCs enhanced human breast ALDH⁺ cells (Fig. 3B), stimulated multiple core stem cell gene expression (Fig. 3C), but have no effect on cancer cell proliferation (Supplementary Fig. S3A). To test whether MDSCs were directly associated with CSCs in patients with breast cancer, we quantified ALDH⁺ CSCs in human breast cancer tissues. Breast cancer cells expressed a variety of ALDH levels (Supplementary Fig. S3B). The median levels of CD33⁺ cells (MDSCs) positively correlated with that of ALDH⁺ CSCs (Fig. 3D). Similar results were observed in the cohort 1 (Supplementary Fig. S3C) and cohort 2 (Supplementary Fig. S3B), respectively. Finally, we evaluated the relevance of the interaction between MDSCs and tumor cells in the human xenograft model. Human cancer MDSCs were co-injected with MCF7 breast cancer cells into in female NOD-scid IL2Rγ null (NSG) mice. We found that MDSCs accelerated tumor progression (Fig. 3E). Furthermore, MDSCs increased the incidence of breast cancer tumor formation in NSG model (Fig. 3F). Cancer stemness is often associated with epithelial–mesenchymal transition (EMT). We found that MDSCs increased EMT-related gene expression as shown by Western blotting (Supplementary Fig. S3E) and immunofluorescence staining (Supplementary Fig. S3F). The data support a role of MDSCs in human breast cancer in vivo. Thus, MDSCs are biologically and clinically linked to breast cancer stemness.

STAT3 (18, 20–23) and NOTCH (20, 35, 36) activation is observed in a variety of cancers and may control cancer progression and metastasis. We found that MDSCs strongly induced STAT3 phosphorylation in MCF-7 and MDA-MB-231 breast cancer cells cocultured with MDSCs (Fig. 4A). Interestingly, NOTCH was also activated in MCF-7 breast cancer cells by MDSCs as shown by high expression of NOTCH2, NOTCH3, CHERP, HEY1, and HEY2 transcripts (Fig. 4B) and of intracellular domain of NOTCH (NICD) expression (Fig. 4C). MDSCs promoted breast cancer stemness and were associated with the levels of ALDH⁺ breast CSCs in breast cancer tissues (Fig. 3). We hypothesized that STAT3 and NOTCH activation was involved in MDSC-stimulated breast cancer stemness. To test this hypothesis, we blocked NOTCH and STAT3 signaling in human MDSC and breast cancer cell coculture. The Notch inhibitor or the STAT3 inhibitor partially, and their combination completely, reduced ALDH⁺ breast CSCs induced by MDSCs (Fig. 4D). The data indicate that MDSCs activate STAT3 and NOTCH, and induce human breast CSCs.

As MDSCs induced CSCs partially via STAT3 activation, we examined how MDSCs activated STAT3. IL6 has been reported to activate STAT3 and was associated with tumorigenesis (18, 20–23). Tumor-associated CD14⁺CD15^−/dim MDSCs expressed high levels of IL6 (Fig. 5A; Supplementary Fig. S4A). Breast cancer cells expressed IL6 receptor (Supplementary Fig. S4B) and IL6-activated STAT3 (Supplementary Fig. S4D). STAT3 inhibition partially suppressed breast CSCs induced by MDSCs (Fig. 4C). It suggests that MDSCs may promote CSCs via IL6-mediated STAT3 activation.

As MDSCs induced CSCs partially through NOTCH activation, we further examined how MDSCs activated NOTCH. MDSCs may release reactive nitrogen intermediates to suppress immune responses (37). We detected high levels of NO in the culture supernatants of CD14⁺CD15^−/dim and CD14⁻CD15⁺ MDSCs (Fig. 5B; Supplementary Fig. S4C). GSNO-stimulated NOTCH activation (Fig. 5C) and NOTCH gene expression (Fig. 5D).

Next, we evaluated the relative impact of MDSC-derived IL6 and NO on CSC induction. In the coculture of breast cancer cells and MDSCs, anti-IL6 mAb or iNOS inhibitor partially, and their combination completely, reduced sphere formation stimulated by MDSCs (Fig. 5E). Thus, our results suggest that MDSC-derived IL6 and NO may collaboratively activate STAT3 and NOTCH, and induce breast CSCs.

Finally, we investigated how MDSC-derived IL6 and NO cooperatively activate STAT3 and NOTCH, and support cancer stemness. We first evaluated whether STAT3 and NOTCH signaling reciprocally affected their counterpart's expression in the coculture of MDSCs and breast cancer cells. Genetic knockdown of STAT3 with specific shSTAT3 abrogated the NOTCH activation mediated by MDSCs, as shown by reduced NICD expression (Fig. 6A), and whereas the inhibition of NOTCH with NOTCH inhibitor reduced STAT3 activation mediated by MDSCs (Fig. 6B). STAT3 inhibitor was included as a positive control (Fig. 6B).

We further examined the roles of NO and IL6 in MDSC-mediated STAT3 activation and the kinetics of STAT3 phosphorylation. MDSCs mediated potent tumor STAT3 activation (Fig. 4A; Fig. 6A and B). This activation was partially reduced by iNOS inhibitors and anti-IL6 mAb, and completely blocked by both iNOS inhibitor and anti-IL6 mAb (Fig. 6C). The data raise the possibility that MDSCs induce potent STAT3 activation via IL6 and NO signaling collaboration in human breast cancers.

We further investigated the kinetics and persistence of STAT3 activation via IL6 and NO interaction. We observed that STAT3 activation induced by IL6 was faint within 1 hour (Fig. 6D). In contrast, addition of GSNO sustained and prolonged STAT3 activation (Fig. 6D). Interestingly, GSNO potently activated NOTCH (Fig. 5C and D) and weakly stimulated STAT3 activation (Fig. 6D). Thus, the data suggest that MDSC-derived NO activates NOTCH to facilitate and sustain persistent cancer STAT3 phosphorylation stimulated by MDSC-derived IL6 and IL6 may not be solely responsible for long-lasting STAT3 activation in tumor cells.

We tested the relevance of the interaction between MDSCs and tumor cells in the human xenograft model. Human cancer MDSCs were coinjected with MCF7 breast cancer cells into in female NOD-scid IL2Rγ null (NSG) mice. We found that MDSCs accelerated tumor progression. The effect was blocked by the treatment with anti-IL6 and iNOS inhibitor (Fig. 6E). In line with this, we found that MDSCs increased ALDH1 expression in tumor cells and blockade of iNOS and IL6 abolished this effect (Fig. 6F and G). Furthermore, we analyzed the correlations among ALDH1A1, IL6, and CD33 transcripts in TCGA breast cancer data (Oncomine.org.). We observed strong correlations among ALDH1A1, IL6, and CD33 transcripts (Supplementary Fig. S5). These data support a role of MDSCs and MDSC-derived IL6 and NO in human breast cancer progression in vivo.

Altogether, we have demonstrated that MDSC-derived IL6 initiates STAT3 phosphorylation, MDSC-derived NO activates NOTCH, and NOTCH subsequently and collaboratively acts with IL6 to promote prolonged STAT3 activation. Thus, MDSCs may play a role in stimulating and maintaining CSC pool through the interaction between IL6/STAT3 and NO/NOTCH (Fig. 6H).

In this study, we have generated important novel insights into MDSC and CSC immunobiology and pathology in the context of human breast cancer. (i) MDSCs provide extrinsic signals for CSC renewal and promote tumor metastatic and tumorigenic potential. (ii) MDSCs affect CSC biology through IL6/STAT3 and NO/NOTCH signaling pathways. (iii) NO/NOTCH signaling enforces and sustains persistent and potent IL6/STAT3 activation, and affects cancer stemness. (iv) The interaction between MDSCs and CSCs is biologically and clinically relevant in patients with breast cancer.

Immune-suppressive effects of MDSCs are relatively well studied in tumor-bearing mouse models (38). Myeloid cells, including MDSCs and macrophages, have been linked with cancer stemness (13, 39, 40). However, the non-immunologic effects of MDSCs are poorly understood in human breast cancer. It has been reported that peripheral blood MDSCs correlate with clinical cancer stage, metastatic tumor burden, and doxorubicin–cyclophosphamide chemotherapy (41). In line with this, we have found high numbers of MDSCs in breast cancer tissues. To our surprise, MDSCs directly promote and maintain the CSC pool through two integrated signaling pathways: IL6/STAT3 and NO/NOTCH signaling pathways.

The link between IL6 and STAT3 has been reported in several types of cancer (18, 20–23). Interestingly, IL6 alone induces transient STAT3 phosphorylation, whereas MDSCs induce long-lasting STAT3 activation. MDSC-derived NO activates NOTCH and contributes to sustained STAT3 phosphorylation through IL6 and NO collaborative action. In support of this, it has been demonstrated that NO stimulates NOTCH signaling and delivers a survival signal to glioma cells (42) and drosophila blood cells (43). Thus, although many factors can regulate NOTCH and STAT3 signaling pathways in cancer, our work support the notion that MDSCs integrate the signaling networks between NO/NOTCH and IL6/STAT3 in breast cancer. We propose that MDSCs contribute to persistent and potent STAT3 activation in breast cancer, which promotes and maintains the CSC pool. Given the role of CSCs in cancer metastasis, our work also supports the notion that STAT3 signaling is crucial for myeloid cell colonization at future metastatic sites (19).

After deciphering the molecular and cellular importance of the cross-talk between MDSCs and tumor cells in CSCs, we have further addressed the biological and clinical relevance of this cross-talk in patients with breast cancer. MDSCs correlate with CSCs content in the human breast cancer microenvironment, and are adversely associated with patient survival. It has been reported that response to Herceptin (44) and chemotherapy (45) is in part regulated by immune components in tumor-bearing mouse models. Given the relevance of CSCs in tumor relapse and therapy resistance (7, 8, 33), our data point toward a possibility that immune-suppressive element, MDSCs, directly target the cancer stemness signaling pathway and may potentially affect cancer therapy. Altogether, our results suggest that anticancer therapy should simultaneously target host MDSCs and cancer (stem) cells to improve therapeutic efficiency and abrogate therapy resistance. We have shown that CD33 is an operational marker for human tumor–associated MDSCs (13). Targeting CD33 is considered a strategy to treat patients with acute promyelocytic leukemia (46). Therefore, targeting CD33 signaling may be an optional regimen to treat breast cancer patients.

No potential conflicts of interest were disclosed.

Conception and design: T. Tanikawa, W. Zou, I. Kryczek

Development of methodology: T. Tanikawa, S. Wan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Peng, T. Tanikawa, W. Li, S. Wei, Y. Wang, Y. Liu, E. Staroslawska, F. Szubstarski, J. Rolinski, E. Grywalska, A. Stanisławek, W. Polkowski, A. Kurylcio, C. Kleer, A.E. Chang, M. Sabel, I. Kryczek

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Peng, T. Tanikawa, L. Zhao, M. Sabel, I. Kryczek

Writing, review, and/or revision of the manuscript: D. Peng, M. Wicha, M. Sabel, W. Zou, I. Kryczek

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Vatan, W. Szeliga, E. Staroslawska, F. Szubstarski, J. Rolinski, E. Grywalska, A. Stanisławek, W. Polkowski

Study supervision: M. Wicha, I. Kryczek

The authors thank Daniel Hayes for fruitful discussion and intellectual support, and Deborah Postiff, Michelle Vinco, Jackline Barikdar, and Ronald Craig in the Pathology Department for their technical assistance.

This work was supported in part by research grants from the NIH/NCI R01 grants (W. Zou; CA123088, CA099985, CA193136, and CA152470) and the NIH through the University of Michigan's Cancer Center Support Grant (CA46592).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.