Mice

Six-week old, sex-matched C57BL/6, Balb/c, or CD45.1⁺ (B6.SJL-Ptprc^aPepc^b/BoyJ and CBy.SJL(B6)-Ptprc^a/J) mice were purchased from Jackson Laboratories. F1 (Balb/cxC57BL/6) mice were bred in-house. RIP1-Tag2 mice were obtained from Mouse Model of Human Cancer Consortium (NCI). Mice with spontaneous invasive pancreatic ductal adenocarcinoma (PDA, Pdx1-Cre LSL-Kras^G12D Pten^L/+) were kindly provided by Dr. DePinho (while at the Dana-Farber Cancer Institute; currently at the MD Anderson Cancer Center) [11].

Cell lines and tumor models

B16, EL4, 4T1, A20, and CT26 cells obtained from ATCC. Pan02 cells were obtained from the NCI DTP repository. The 4T07 cells were obtained from Dr. J. Lieberman (Harvard University, MA); they were expanded, aliquoted, were banked in liquid nitrogen. No additional authentication was performed on these cells. The Her2 cell line was derived from a spontaneous breast tumor in a Balb/c Her-2/neu-transgenic mouse that was found to lose Her-2/neu expression in vitro. Transgenic rat Her-2/neu sequence was inserted into a pMFG retroviral vector to obtain a stably over-expressing cell line. Cell lines were maintained in DMEM (B16 and EL4) or RPMI (4T1, Her2, A20, Pan02, CT26, and 4T07) supplemented with 10% FBS and 1x penicillin/streptomycin.

For subcutaneous (s.c.) tumors, 1–2.5×10⁵ cells were injected into the flanks of naive mice (4T1, Her2, CT26 and 4T07 into Balb/c; B16 and Pan02 into C57BL/6; A20 into Balb/cxCD45.1, EL4 into C57BL/6xCD45.1), and tumor growth was monitored using calipers until tumor size reached 0.5–1 cm diameter. For transgenic models, tumors were monitored and analyzed at different time points based on tumor growth (RIP-Tag2 at 11–12 weeks, and PDA and Her2 transgenic at 11–24 weeks).

Cell preparation and flow cytometry

Spleens and tumors were processed by gentle mechanical disruption with forceps followed by enzymatic digestion using 0.2 mg/mL collagenase P (Roche), 0.8 mg/mL dispase (Invitrogen), and 0.1 mg/ML DNase I (Sigma). Tissues were incubated in digestion medium at 37°C for 30 min; released cells were collected, filtered and the remaining tissue was further processed. Blood was collected into tubes containing 2 mM EDTA, and bone marrow cells were isolated by flushing media through tibias and femurs. Red blood cells were lysed using ACK solution. Cells were resuspended in FACS buffer containing 1% FBS and 2 mM EDTA.

Cells were stained in FACS buffer containing FcR-blocking antibody (2.4G2) with combinations of fluorochrome-conjugated or biotinylated antibodies against CD11b (M1/70), CD11c (N418), CD45 (30.F11), Gr1 (RB6-8C5), MHC class II (M5/114.15.2), CD80 (16-10A1), CD86 (GL-1), CD40 (HM40-3), CD26 (H194-112), CD103 (2E7), cKit (2B8), Flt3 (A2F10) or isotype controls purchased from BioLegend and BD Biosciences. Cells were analyzed using BD FACS Aria II or BD FACSCalibur (BD Biosciences), and FlowJo Software (Tree Star, Inc). For cell sorting from spleens, myeloid cells were enriched by depletion of CD3⁺, CD19⁺ and CD49b⁺ cells using biotinylated antibodies and anti-biotin beads (MACS, Miltenyi). In tumor samples, CD45⁺ cells were enriched by positive selection using a biotinylated antibody and anti-biotin beads (EasySep, StemCell Technologies). After enrichment, cells were stained with fluorochrome-conjugated antibodies and propidium iodide (Sigma) to exclude dead cells. Cells were sorted using a BD FACS Aria II equipped with a 100 μm nozzle running at 20 psi. After an initial sort to verify purity, a second sort was performed to collect myeloid cell populations of 95–100% purity directly into TRIzol reagent (Invitrogen). Sample size for each population was as follows: CD11b^−/intGr1⁻CD11c⁺MHCII⁺ (N=4), CD11b⁺Gr1^lowCD11c⁺MHCII⁺ (N=3), CD11b⁺Gr1^lowCD11c⁻MHCII⁻ (N=3), and CD11b⁺Gr1^hiCD11c⁻MHCII⁻ (N=2) from B16 tumors; CD11b⁺Gr1^lowCD11c⁺MHCII⁺ (N=3), CD11b⁺Gr1^lowCD11c⁻MHCII⁻ (N=3), and CD11b⁺Gr1^hiCD11c⁻MHCII⁻ (N=2) from 4T1 tumors; CD11b⁺Gr1^lowCD11c⁺MHCII⁺ (N=3) and CD11b⁺Gr1^hiCD11c⁻MHCII⁻ (N=3) from Her2 tumors; CD11b⁺Gr1^lowCD11c⁻MHCII⁻ (N=2) and CD11b⁺Gr1^hiCD11c⁻MHCII⁻ (N=3) from spleens of 4T1 tumor bearing mice.

RNA isolation and microarray analysis

Total RNA was prepared from TRIzol using chloroform extraction according to the manufacturer’s protocol, and 100 ng of RNA from each sample was used for amplification, labeling, and hybridization by Expression Analysis, Inc. (Durham, NC). Mouse Gene ST 1.0 chips (Affymetrix) were used for microarray analysis. All data are MIAME-compliant and the raw data generated as part of the Immunological Genome Project (ImmGen) have been deposited in a MIAME-compliant database [NCBI Gene Expression Omnibus (GEO) data repository; record no: GSE15907 and GSE37448]. Various modules included in GenePattern platform (Broad Institute) were used for data analysis. Additional myeloid cell populations from healthy tissues analyzed within the same microchip batch of ImmGen were used as a comparison [F4/80 Macrophages from peritoneal cavity (PC, n=2), Ly6C⁺ DCs from bone marrow (BM, n=2), CD24⁻Sirpa⁺ DC from BM (n=3), CD24⁺Sirpa⁺ DC from BM (n=3), CD4 DC from spleen (n=1), CD11c⁻CD103⁺ DC from small intestine (SI, n=1), neutrophil from BM (C57BL/6, n=2), and neutrophil from BM (Balb.c, n=2)]. Raw data were normalized with ExpressionFileCreator module using the RMA method. The MultiPlot module was used for dataset comparisons including fold-change analysis and statistical filtering. In each analysis, we preprocessed the dataset for populations included in specific analyses. In fold-change analyses, an arbitrary cutoff value of 2 was used, in combination with coefficient of variation (CV)<0.5 for replicates. T-test P values <0.05 were considered statistically significant for each probe. After filtering, we selected all probes with a mean expression above 100 in at least one of the populations analyzed, which indicates actual expression with 95% confidence. For hierarchical clustering, Pearson’s correlation was used with datasets following log2 transformation, row centering, and row normalization in HierarchicalClustering module. Heatmaps were constructed using HeatmapViewer module. Pathway enrichments were analyzed by Ingenuity Pathway Analysis. DCs- and macrophages-associated genes were determined previously [12,13], and a similar analysis was performed to identify neutrophil-associated genes (unpublished data). Principal component analyses were conducted using the PopulationDistances module created by ImmGen (probes with mean expression value >120 selected, dataset log2 transformed, probes with top 15% variability selected).

Lineage determination of tumor-infiltrating myeloid cell subsets based on transcriptional profiling

We next sought to determine the lineage of each tumor-infiltrating myeloid cell population using transcriptomic analysis. To this end, myeloid subsets from 4T1 and Her2 breast carcinomas, and B16 melanoma were sorted to high purity, according to the ImmGen standard operating procedure. Transcriptional profiling data were generated on Affymetrix ST1.0 microarrays adhering to profiling and quality control pipelines of ImmGen [14]. We utilized hierarchical clustering of ~13.5K probes, after excluding those with coefficient of variation values <0.5 and mean expression values <120 to compare populations of tumor-infiltrating myeloid cells with representative myeloid cell subsets (DCs, macrophages, and neutrophils) from lymphoid and non-lymphoid tissues of healthy animals. Dendogram analysis based on gene clustering depicted similarities between neutrophils and tumor-derived Gr1^hi cells, macropahges and Gr1^low cells (regardless of CD11c expression), and conventional DCs and CD11b⁻Gr1⁻CD11c⁺ cells (Fig. 1D). Hierarchical clustering of 139 probes corresponding to signature genes of DCs, macrophages [12,13], and neutrophils (unpublished data) demonstrated that tumor-infiltrating myeloid cell populations are indeed CD11c⁺ and CD11c⁻ TAMs, TANs, and DCs (Fig. 1E and Supplementary Table S1).

Striking heterogeneity in myeloid cell infiltrates across different tumor types

Among the three tumor types analyzed, Her2 tumors were uniquely characterized by the presence of a large CD11c⁺ TAM population whereas 4T1 tumors contained a relatively large TAN population (Fig. 2A). In B16 tumors, myeloid cells comprised only 40% of tumor-infiltrating leukocytes compared to >75% in 4T1 and Her2 tumors (Fig. 2A). The aforementioned differences in myeloid abundance also reflected variation in the abundance of other leukocytes infiltrating tumors including T, B, NK, and NKT cells, and pDCs. For example, B16 tumors were enriched in CD3⁺ T cells compared to 4T1 and Her2 tumors, consistent with a more immunogenic environment in B16 melanoma (Fig. 2B). To further ascertain the degree of myeloid heterogeneity, we extended our analysis to five additional transplanted tumors including 4T07 breast cancer, CT26 colon carcinoma, EL4 T lymphoma, A20 B lymphoma, and Pan02 pancreatic ductal adenocarcinoma (PDA), and three autochthonous mouse models including RIP-Tag2 (insulinoma), Her-2/neu (mammary carcinoma), and Pdx1-Cre LSL-KrasG12D PtenL/L (PDA) mice. The same four major populations were present within each tumor model but at significantly different frequencies across the tumors analyzed (Fig. 2A–C).

Open in a separate window

Figure 2

Abundance of myeloid cell subsets across different tumors

A, Whisker box-plots summarizing the relative frequency of myeloid subsets among hematopoietic cells within 4T1, Her2 and B16 tumors. B, Frequency of CD3⁺ T cells within hematopoietic cells in 4T1, Her2 and B16 tumors. C, Frequency of myeloid cells in 4T07, Pan02, EL4, CT26, PDA, RIP-Tag2 and A20 tumors (n=3–7). * P<0.05, ** P<001, *** P<0.001.

In all tumors, DCs were present at very low frequencies, with the highest levels in B16, EL4 and RIP-Tag2 tumors (4–10%) (Fig. 2). Given that DCs account for only a small percentage of hematopoietic cells within the tumors, and that the major subset expressing CD11c in tumors is TAMs, our study also suggests that CD11c alone is an unreliable marker for DCs in neoplastic lesions. Further analysis comparing DCs in B16 tumors to TAMs indicated that while TAMs expressed a variety of endocytic and regulatory receptors including Pilrb, Sirpa, Lilrb4, and Clec5a (Supplementary Fig. S1B), tumor DCs expressed signature genes such as Dpp4, Flt3, and Kit, as well as CD8/CD103 DC-associated genes [13] including Itgae, Batf3, Tlr3, ifi205, and Xcr1 identifying them as CD11b⁻CD103⁺ tissue-resident DCs. Flow cytometric analysis confirmed that DCs were the myeloid cells expressing high levels of CD26 (Dpp4), Flt3, cKit, and CD103 (Itgae) protein in these tumors (Supplementary Fig. S1C, and data not shown). Based on these distinctions, DCs may be considered as a reference population in future studies to identify unique targets on TAMs and TANs to modulate their tumor-promoting functions. For example, DCs lack the surface receptor Clec5a (MDL-1), whereas expression in TAMs, TANs, and neutrophils is relatively high (>18-fold higher than in DCs), suggesting that it may represent an attractive target for antibody-mediated depletion of CD11b⁺Gr1⁺ cells in tumors [15].

Differences in Gr1^low and Gr1^hi cells across several diverse tumors were not restricted to the tumor lesion itself. For example, a substantial increase in MDSCs was observed in the spleen and blood of 4T1 tumor-bearing mice throughout tumor progression but not in mice bearing other tumor types (Supplementary Fig. S1D and S1E).

Differences in myeloid cell composition arise from specificity of the tumor microenvironment

The myeloid composition of different tumors is likely shaped by a combination of tumor-specified growth factors. Indeed, 4T1, Her2, and B16 cell lines produce unique combinations of cytokines and chemokines, suggesting that each tumor may create a distinctive microenvironment (Supplementary Fig. S2A). These differences may also explain why MDSCs preferentially accumulate in certain tumors. For example, high levels of G-CSF and LIX (CXCL5) expression by 4T1 cells may contribute to systemic accumulation of MDSCs in vivo [16,17]. To assess whether specific tumor microenvironments directly influence the myeloid cell content, we analyzed myeloid cells within the same tumor type growing at different anatomical locations. Comparison of myeloid cell composition in subcutaneous Her2 tumors and the parental spontaneous tumor in mammary glands revealed that both lesions contained high frequencies of CD11c⁺ TAMs (50–72%) and relatively low frequencies of TANs (<5%) among hematopoietic cells (Fig. 3A and Supplementary Fig. S2B). A strong similarity was observed between the myeloid cell composition of subcutaneous 4T1 tumors and spontaneous metastatic lesions in the peritoneal cavity, occurring within 2–3 weeks after tumor challenge. Similar to primary 4T1 tumors, these metastases are enriched with TAMs (~50%, of which >60% are CD11c⁺) and TANs (15–21%) among hematopoietic cells (Fig. 3A and Supplementary Fig. S2C). Furthermore, these results were corroborated in a B16 metastatic model. When B16 cells are injected intravenously, tumor nodules grow in lungs and kidneys within 2–3 weeks and contain a similar myeloid landscape to subcutaneous tumors: ~30% TAMs (~45% expressing CD11c) and 4–9% TANs among hematopoietic cells (Fig. 3A and Supplementary Fig. S2D). Altogether, these data provide evidence that tumor type rather than its anatomic location shapes the myeloid infiltrate.

Open in a separate window

Figure 3

Tumor type dictates the composition of myeloid cell subsets within tumors

A, Myeloid cells within s.c. and mammary gland of transgenic Her2 tumors (n=4–7); within primary s.c. and spontaneous peritoneal metastatic 4T1 tumors (n=3); and within s.c. B16 tumors or lung and kidney metastases (n=3–4). B, Left, Analysis of myeloid subsets within tumors from Balb/c mice with single (Her2 or 4T1, n=5) or double (Her2 and 4T1 on opposite flanks, n=8) tumors. Right, Frequency of myeloid cells in the spleens of tumor-bearing mice. C, Left, Analysis of myeloid cell subsets within tumors from F1 (Balb/c × C57BL/6) mice with single (Her2 or B16, n=3) or double (Her2 and B16 on opposite flanks, n=3) tumors. Right, Frequency of myeloid cells in the spleens of tumor-bearing mice. Numbers indicate percentage of cells for each gate or region. N.S., not significant, * P<0.05, *** P<0.001.

Next, we tested the possible dominant effect of the microenvironment in dictating myeloid cell composition by analyzing myeloid cells in mice with two tumors growing simultaneously. For this purpose, we compared Her2 and 4T1 or Her2 and B16 tumors, as each of these tumors has distinct myeloid infiltrates (Figs. 1 and ​and2).2). Balb/c mice were inoculated with Her2 tumors under one flank and simultaneously with 4T1 tumors under the contralateral flank (Fig. 3B and Supplementary Fig. 2E). Similarly, we established Her2 tumors in the presence of B16 tumors on the opposite flank in F1 (Balb/c × C57BL/6) mice (Figure 3C and Supplementary Fig. 2F). In both models, we did not observe a difference in growth rate of either tumor when compared with tumors grown alone (data not shown). Likewise, the myeloid content of each tumor, either alone or in the presence of a second tumor in the same host was similar (Figs. 3B–C, left and Supplementary Figs. S2E–F), even though there was pronounced systemic accumulation of MDSCs associated with 4T1 tumor growth (Figs. 3B–C, right). Overall, these results suggest that the signals attracting specific myeloid cells into tumors are driven by cancer cells in the primary lesion regardless of anatomic location or presence of distant signals from other microenvironments.

Tumor-infiltrating myeloid cells exhibit a hypoxia-associated signature compared to their splenic counterparts

As shown in Fig. 1, MDSCs accumulate in the spleens of 4T1 tumor-bearing mice at high frequencies. As splenic MDSCs may serve as reservoirs for tumor-infiltrating cells [6], it is important to understand how they relate to their counterparts within tumors. Thus, we compared the gene expression profiles of CD11b⁺Gr1⁺ cells from the spleens and tumors of 4T1 tumor-bearing mice. As shown in Fig. 4A, 263 probes were upregulated in TAMs (both CD11c⁻ and CD11c⁺ subsets) within 4T1 tumors compared to Gr1^low MDSCs from spleens, whereas 152 probes were enriched in splenic cells. In addition, 201 probes were enriched in TANs compared to splenic Gr1^hi MDSCs, while 647 probes were enriched in Gr1^hi MDSCs (Fig. 4B). Next, we compared expression of probes upregulated in 4T1 tumors to that of myeloid cells originated in other tumor models. Hierarchical clustering indicated that TAMs, TANs, and splenic MDSC populations clustered separately, confirming differences between tumor and splenic myeloid cells. Notably, similar profiles were obtained from equivalent populations originated from different tumor types (Fig. 4C), supporting that transcriptomic analysis may illuminate universal features of tumor myeloid cells. Indeed, this analysis identified several genes that highlight the pro-tumorigenic roles of tumor-infiltrating myeloid cells. For example, genes involved in extracellular matrix remodeling (Mmp12, Mmp13, Mmp14, Adam8, Tgm2), immunomodulation (Arg1, Nos2, Cd274, Ptgs2, Spp1) and hypoxia regulation (Vegfa, Hif1a, Slc2a1) were markedly upregulated in TAMs and TANs compared with splenic MDSCs. Interestingly, expression of these genes was similar between CD11b⁻CD103⁺ tumor DCs, and DCs from tumor-free mice. This result suggests that hypoxic responses are not uniform within the same microenvironment and may be tightly regulated among tumor-infiltrating leukocytes. Immunosuppression and tissue remodeling are critical functions for promoting tumor progression, and hypoxia, through hypoxia inducible factor 1α (Hif-1α) [18], can promote transcriptional changes that facilitate the selection of aggressive tumor cells. Expression analysis of Hif-1α-regulated hypoxia-associated genes [19]revealed that these genes are indeed enriched in TAMs and TANs compared to splenic MDSCs and steady-state populations, indicating reprogramming in the tumor environment (Fig. 4D and Supplementary Table S2). Hypoxia and Hif-1α have been implicated in prolonged survival and reduced oxidative burst in neutrophils [20,21], and as such may also contribute to the accumulation of these short-lived cells.

Open in a separate window

Figure 4

Tumor-infiltrating myeloid cells express immunomodulatory genes

A, Comparison of gene expression profiles of CD11c⁺ and CD11c⁻ TAMs to Gr1^low MDSCs from 4T1 tumor-bearing mice based on 2-fold change (fold-change vs fold-change plot). B, Comparison of gene expression profiles of TANs and Gr1^hi MDSCs from 4T1 tumor-bearing mice (fold-change vs t-test P value plot). C, Heatmap generated by hierarchical clustering using probes identified in A and B for all tumor myeloid subsets. D, Heatmap showing the expression of hypoxia-related genes by tumor-associated and steady-state myeloid cells. Selected genes associated with the populations are indicated on the plots.

Neutrophils in the tumor environment are pro-tumorigenic and exhibit an altered functional profile

Our finding that the myeloid cell landscape of each tumor type is distinctive raises the possibility that different tumor microenvironments impart unique functional attributes in the myeloid cells therein. While TAMs in different tumors have been extensively characterized by functional and transcriptional profiling approaches [22–26], our understanding of TANs is limited. Recently, Fridlender et al. compared steady-state neutrophils to splenic MDSCs and TANs in a mesothelioma model by transcriptional profiling and identified striking differences [27]. However, this study focused only on a single tumor and it is not known if such characteristics of TANs are common across different tumors. By comparing TANs from 4T1, B16, and Her2 tumors to Gr1^hi MDSCs from 4T1 tumor-bearing mice and neutrophils from the bone marrow of naive mice (Fig. 4), we identified a conserved expression pattern of immunomodulatory and hypoxia-related genes across these tumors. We also compared genes highly expressed in both Gr1^hi MDSCs and TANs from 4T1 tumors to their expression in steady-state neutrophils (Fig. 5A, left) and found 214 probes selectively enriched in 4T1-associated neutrophils. Interestingly, many of these genes were similarly enriched in TANs from B16 and Her2 tumors compared to steady-state neutrophils (170 out of 214 passed CV<0.5 filter) (Fig. 5A, right). Heatmap analysis also indicated that these probes were highly expressed by all TANs and Gr1^hi MDSCs compared to steady-state neutrophils (Fig. 5B and Supplementary Table S3). Importantly, these genes are involved in pathways critical for tumor growth and immune responses, such as iNOS, IL-10, IL-6 and peroxisome proliferator-activated receptor (PPAR) signaling (Fig. 5C). Collectively, our data support that tumor-associated myeloid cells exhibit a pro-tumorigenic transcriptional program.

Open in a separate window

Figure 5

Expression of pro-tumorigenic genes by TANs

A, Comparison of gene expression profiles of TANs and Gr1^hi MDSCs from 4T1 tumor-bearing mice to naive neutrophils (left, fold-change vs fold-change, 2-fold change). Expression of these probes was further analyzed in fold-change vs fold-change plots comparing TANs from B16 and Her2 tumors to naive neutrophils (right). 170 of the 214 probes passed the CV<0.5 filter. B, Heatmap showing the expression of the 170 probes identified in A across neutrophils. C, Top 20 canonical pathways enriched for TANs and Gr1^hi MDSCs compared to naive neutrophils analyzed by Ingenuity Pathway Analysis.

To further interrogate molecular programs in TANs across different tumors, a side-by-side comparison of TANs from 4T1, Her2 and B16 tumors was performed. We identified 43 probes enriched in 4T1 TANs, 61 in Her2 TANs, and 93 in B16 TANs (Fig. 6A). Genes involved in cell trafficking or differentiation, such as Cxcr1, Foxo3, Slc28a2, were enriched 2–5 fold in 4T1 TANs and may regulate the marked accumulation of TANs in this cancer type [28]. Given that our data point to a direct, tumor cell-specific involvement in the recruitment and expansion of TANs, we analyzed the expression levels of cytokines, cytokine receptors, chemokines, and chemokine receptors by TANs, as these may account for the tumor-myeloid cell cross-talk regulating cell infiltration (Fig. 6B and Supplementary Table S4). Strikingly, we identified different expression profiles for each TAN population, which may reflect the differences observed in the frequencies of these populations in each tumor. In particular, TANs from 4T1 tumors expressed significantly greater levels of Ltb4r1 (BLT1), a leukotriene B4 chemotactic factor for neutrophils [29,30]. Interestingly, we also found enrichment of genes in the anti-inflammatory PPARα signaling pathway (Fig. 5D), which is also a target for leukotriene B4. We further analyzed the expression of PPARα target genes associated with inflammation [31] in neutrophil subsets (Fig. 6C and Supplementary Table S5). While many of these genes are expressed at higher levels in TANs compared to Gr1^hi MDSCs and steady-state neutrophils, neutrophils from 4T1-bearing mice expressed higher levels of Stat1, Stat2, and Stat3, among other genes. Stat3 is a mediator of CXCR2 expression in neutrophils upon G-CSF stimulation [32]. In agreement with this, G-CSF is highly expressed by 4T1 cells (Fig. S2A) and Cxcr2 is expressed at higher levels by 4T1 TANs (Fig. 6B). Overall, these signaling pathways may incite the marked accumulation of TANs and Gr1^hi MDSCs in the 4T1 model. The relationship between neutrophil populations from different tumors was also confirmed by the principal component analysis to highlight functional differences. Neutrophils from naive mice clustered separately and more closely to MDSCs on PC2 (Fig. 6D). On the other hand, while TANs clustered separately from neutrophils and MDSCs, variability on PC1 reflected tumor-to-tumor differences among TANs.

Open in a separate window

Figure 6

Tumor type-specific transcriptional profiles of TANs

A, Fold-change vs fold-change plot comparing 4T1 TANs to TANs in Her2 and B16, based on 2-fold change. Selected genes associated with the populations are indicated on the plots. B, Chemokine, chemokine receptor, cytokine, and cytokine receptor expression profiles of TANs. C, PPARα-associated gene expression profiles of neutrophils. D, Principal component analysis of neutrophil populations.

We thank the Immunological Genome Project team for technical help and discussions; Drs. Dobrin Draganov and Glenn Dranoff for the Her2 cell line; and Dr. Ronald DePinho for PDA mice. This study was supported by research funding from NIH T32 grant to KGE, NIH T32 CA 070083-15 and Postdoctoral Fellowship from the Cancer Research Institute to VC, NIH grants (AG028082, AG033049, AI098108, AI101918) and an Established Investigator Award (0940006N) from American Heart Association to DRG, NIH grants (ROI DK074500, PO1 AI045757, R21 CA182598), American Cancer Society Research Scholar Grant, and Claudia Adams Barr Award for Innovative Cancer Research to SJT.