Low Levels of Tumor Necrosis Factor α Increase Tumor Growth by Inducing an Endothelial Phenotype of Monocytes Recruited to the Tumor Site

Tumor-associated myeloid cells modulate several aspects of the malignant phenotype: invasion, tumor growth, angiogenesis, and metastasis (1, 2). It is widely accepted that the tumor microenvironment results in the generation of unique myeloid phenotype(s), although the nature of the phenotypes is just beginning to be elucidated (2, 3).

Several groups, including our own, have shown that circulating human or mouse monocytes can be induced in culture to coexpress endothelial markers [e.g., VE-cadherin (VE-cad; CD144), flk-1 (VEGFR2/KDR), von Willebrand Factor, tie2, endothelial lectins] and exhibit an endothelial phenotype (network formation, synthesis of eNOS, and Weibel Palade bodies) and, when implanted into mice, can improve neovascularization in ischemic injury (4–7). Myeloid to endothelial plasticity in vivo was shown recently through adoptive transfer of immature myeloid progenitors in mice and subsequent confirmation that myeloid cells generate recipient vessels in the liver (8). In addition, recent reports have also identified a myeloid/endothelial biphenotypic leukocyte population within mouse and human tumors.

A tumor-promoting CD11c⁺ myeloid population that coexpresses endothelial markers, P1H12 and VE-cad, was identified in murine cancers (9). A small portion of these cells contributed to tumor microvessels (i.e., participate in vasculogenesis), whereas the majority of the cells retained leukocyte morphology (9). The myeloid/endothelial biphenotypic population has also been characterized in human tumors (10). Another study identified a myeloid (CD11b⁺)/endothelial (tie2⁺) population in mice that promoted tumor vascularity and growth but without significant contribution to mature vessels (11). We have shown that when CD14⁺ human monocytes were admixed with tumor cells and implanted s.c., a portion of the exogenously introduced CD14⁺ cells up-regulated the endothelial marker, VE-cad, suggesting that the tumor microenvironment can induce an endothelial phenotype (6).

Taken together, these data suggest that this “biphenotypic” subset, which will be designated in this report as “vascular leukocytes,” plays an important role in tumor progression and may, in small numbers, directly incorporate into vessels (6, 9–15). It is unknown which signals in the tumor microenvironment generate this subset(s) of vascular leukocytes. We sought to show unique tumor-promoting function of this myeloid/endothelial biphenotypic population and also to identify the tumor signal(s) that generate vascular leukocytes. In this report, we have established a novel role for tumor necrosis factor (TNF)α in myeloid to endothelial differentiation in culture and, moreover, have linked TNFα to tumor colonization by vascular leukocytes in vivo.

Direct evidence for the involvement of TNFα in cancer comes from observations that TNFα knockout mice on 4 different genetic backgrounds were 10-fold more resistant to chemical carcinogenesis of the skin (16–18). Mice deficient in TNFα receptor (TNFR)1 and R2 also showed resistance to skin cancers (18). TNFR1^−/− mice showed reduced liver tumorigenesis and liver metastasis (19). Knockdown of TNFα in ovarian cancer cell lines led to diminished growth and vascular density (20). Hence, although high doses of TNFα to treat cancer have antitumor activity, there is growing data to suggest that endogenous TNFα acts as a tumor promoter. The mechanism of TNFα-mediated tumor promotion, however, is incompletely understood. Our data have identified a novel role for TNFα in myeloid to endothelial differentiation and, moreover, have linked TNFα to tumor vasculogenesis and tumor colonization by myeloid/endothelial biphenotypic vascular leukocytes in vivo.

Mice and cell lines. NOD/LtSz-scid (NOD/SCID), C57Bl/6 (Bl/6, wild-type), flk-1/LacZ on C57Bl/6 (Bl/6) background, and the FVB/nJ strain were purchased from Jackson Laboratory. Mice homozygous mutant for TNFR1 and R2 on a Bl/6 background were obtained from DBP. Stably transfected 1 × 10⁶ B16F10 mouse melanoma cells [American Type Culture Collection (ATCC)], Lewis Lung Carcinoma (LLC) cells (ATCC) or murine breast cancer cells derived from mouse mammary tumor polyomavirus middle T transgenic lines [Py-mT lines were generated by Rebecca Muraoka-Cook (21) and maintained in DMEM, 10% FCS, 5.0 ng/mL 17-β estradiol, and 1 μg/mL progesterone] were implanted in Bl/6 (s.c.) or FVB/nJ (under fourth mammary fat pad) strains, respectively. Tumor volume was calculated as length × width × height × 0.52. For some studies, human myeloid/endothelial biphenotypic cells (CD14⁺/VE-cad⁺/flk-1⁺) or monocytes (CD14⁺/flk-1⁻) were admixed with B16 tumor cells and implanted s.c. in opposing flanks of NOD/SCID mice. Also, mouse myeloid/endothelial biphenotypic cells (CD11b⁺/flk-1⁺) or monocytes (CD11b⁺/flk-1⁻) were admixed with B16 tumor cells and implanted s.c. in opposing flanks of Bl/6 mice. To generate transplanted mice, Bl/6 wild-type, syngeneic littermates received 800 rads of preconditioning irradiation and were transplanted with bone marrow (BM) from flk-1/lacZ transgenic donors generating flk/lacZ/BM transplants (BMT). After engraftment (2 mo), mice were implanted with control or TNFα-producing B16F10 mouse melanoma tumor lines. Some flk/lacZ/BMT experimental groups received tumor admixed with and additional i.p. injections of 1 mg of anti-TNFα antibody (cV1q; Centocor, Inc.), a rat/mouse chimeric IgG2a,k monoclonal antibody (mAb) specific for mouse TNFα or isotype control (Centocor, Inc.). In some cases, BMT animals were injected with Isolectin GS-IB4 (0.5 mg/mouse; Vector Labs) before euthanasia to visualize functional vasculature. All experiments were done in accordance with Association for Assessment and Accreditation of Laboratory Animal Care guidelines and with Vanderbilt University Institutional Animal Care and Use Committee approval.

Stable transfections of tumor lines were done with pcDNA3 plasmid encoding TNFα cDNA driven by the cytomegalovirus promoter carried out with Lipofectamine 2000 (Invitrogen Life Technologies). After selection for 14 d with 1500 μg/mL G418 (Sigma), individual colonies were picked and maintained in medium containing 500 μg/mL G418. Mouse TNFα secreted into culture medium or in plasma was measured as described by ELISA Quantikine kits (R&D systems). Human umbilical vascular endothelial cells (HUVEC; ATCC) were maintained in EGM-2 with supplements.

Cell culture. Peripheral blood (PB) or BM mononuclear cells (MNC) were isolated as previously described (6). MNCs were used to isolate CD14⁺ (human) or CD11b⁺ (mouse) myeloid cells using positive selection strategy with antibody linked microbeads (Miltenyi Biotech) and a magnetic cell sorter device following manufacturer protocol. To generate myeloid/endothelial biphenotypic cells, enriched myeloid populations (human CD14⁺ or murine CD11b⁺ of >95% purity) were plated on fibronectin-coated plates or coverslips in EGM-2 Bullet kit system [Clonetics; consisting of endothelial basal medium, 5% fetal bovine serum, hEGF, vascular endothelial growth factor (VEGF), hFGF-B, insulin-like growth factor, ascorbic acid, and heparin] at a density of 2 × 10⁵ cells/cm² (22, 23), in the absence or presence of recombinant human TNFα (Cell Sciences Corp.) and analyzed by immunofluorescence using species specific anti-flk1 (human, Reasearch Diagnostics; mouse, clone MF1, Imclone) and anti–VE-cad (human, Cell Sciences; mouse, Bender Medsystems). Myeloid/endothelial biphenotypic cells (vascular leukocyte) numbers were assessed by counting 5 random 40× fields of VE-cad⁺/flk-1⁺ cells or analyzed by fluorescence-activated cell sorting (FACS).

FACS analysis. Analysis of the tumor tissues for coexpression of myeloid/endothelial antigens were analyzed by flow cytometry. Mice harboring tumors were perfused with PBS for 10 min before removal of tumors. Tumor-associated cells were obtained for analysis by mincing the tissue to <1 mm³ and digesting at 37°C for 30 min with an enzyme cocktail (collagenase A, Elastase, and DNase I; Roche) and filtering the resulting suspension through with a 70 μm filter. The resulting single-cell suspensions were separated on a density gradient, Lympholyte M, for 30 min at 1500 × g. The interphase was collected, washed thrice in PBS, RBC lysed, and analyzed. For analysis of expression of flk-1 on human CD14⁺ cells after culture, attached cells were detached with EDTA and washed with PBS/0.5% bovine serum albumin before incubation with antibodies. Nonviable cells, identified by 7-aminoactinomycin D (Molecular Probes) staining, were excluded. Myeloid markers were assessed with the following directly conjugated mAbs: F4/80-PE (BM8; eBioscience) and mouse CD11b-PEcy7 (M1/70; R&D systems). Expression of endothelial markers was assessed with mAbs to mouse tie2-biotin (TEK4; eBioscience), flk-1 (DC101; ImClone), and VE-cad (11D4.1; Fitzgerald). For human cells anti–CD14-FITC (eBioscience), or anti–flk-1-Allophycocyanin (APC; R&D), analysis was performed on a LSMII flow cytometer and subsequently analyzed using FACSDiva v5.02 software (Becton Dickinson). Isotype control antibodies were used to establish quadrants.

Additional Methods are in Supplementary Data.

Myeloid/endothelial (CD14⁺/flk-1⁺/VE-cad⁺) cells increased tumor growth. We isolated human PB CD14⁺ monocytes and characterized this population by FACS to be >99% pure and also lacking expression of the endothelial marker, flk-1 (Fig. 1A). In parallel, myeloid/endothelial biphenotypic cells were obtained after 7 days in culture of highly pure populations of CD14⁺ human PB monocytes under angiogenic conditions as previously published (6, 24). The in vitro generated myeloid/endothelial biphenotypic cells were characterized by indirect immunofluorescence for continued expression of CD14 and as well as up-regulation of endothelial markers flk-1 and VE-cad (Fig. 1A). To test whether myeloid/endothelial biphenotypic cells could serve as a tumor-enhancing subset of myeloid cells, paired flank tumors were generated in immunodeficient mice with equal numbers of B16F10 melanoma cells admixed with either human CD14⁺ monocytes or CD14⁺/flk-1⁺/VE-cad⁺ myeloid/endothelial biphenotypic vascular leukocytes (20:1 ratio of B16F10 cells/myeloid cells). After 14 days, the addition of CD14⁺/flk-1⁺/VE-cad⁺ cells resulted in tumors that were 60% larger in volume (P < 0.05; Fig. 1B) and weight (P < 0.05; data not shown) compared with the addition of CD14⁺ monocytes. Moreover, tumors admixed with myeloid/endothelial cells contained ∼2-fold greater vascular density (P < 0.05; Fig. 1C) and less necrosis than those admixed with monocytes, 11.8% ± 8% versus 17.3% ± 7%, respectively, albeit the quantified relative necrotic area was not statistically significant between them (P = 0.07; Fig. 1B). These data provide direct evidence that myeloid cells coexpressing endothelial surface antigens (vascular leukocytes) represent a proangiogenic and growth promoting myeloid subpopulation. We also assessed the ability of mouse in vitro generated biphenotypic cells to promote tumor angiogenesis by producing paired flank tumors in wild-type Bl/6 mice using equal numbers of B16F10 melanoma cells admixed with mouse CD11b⁺/flk-1⁻ cells before culture versus CD11b⁺ cells expressing flk-1 and VE-cad after 4 days in culture (Supplementary Fig. S1A). The resultant tumors containing admixed myeloid/endothelial biphenotypic cells (20:1 ratio of B16F10 cells/myeloid cells) were significantly more vascular (Supplementary Fig. S1B–C), exhibiting a comparable proangiogenic effect as the human biphenotypic cells.

As little is known regarding the mechanisms by which myeloid/endothelial vascular leukocytes are generated, we performed gene expression profiling of these cells generated after culture of human CD14⁺ monocytes (i.e., CD14⁺/flk-1⁺/VE-cad⁺) and compared their expression profile to uncultured CD14⁺ monocytes to try to identify molecular pathways implicated in myeloid to endothelial differentiation. We identified up-regulation of a variety of key proangiogenic genes (Supplementary Table S2, top), which support a paracrine role for this myeloid/endothelial population in mediating tumor angiogenesis. We also identified up-regulation of multiple genes in myeloid/endothelial biphenotypic cells that were known to be transcriptionally up-regulated by TNFα (Supplementary Table S2, bottom). The selective up-regulation of TNFα-inducible transcripts in this biphenotypic population led us to formulate the hypothesis that TNFα promotes endothelial differentiation of myeloid cells in vitro, and more importantly, that it may constitute, at least in part, the in vivo tumor microenvironment signal(s) to promote myeloid to endothelial plasticity.

TNFα enhanced myeloid to endothelial differentiation. We examined the effect of TNFα on the generation of CD14⁺/flk-1⁺/VE-cad⁺ biphenotypic cells in vitro. CD14⁺ cells were cultured on fibronectin-coated plates in medium to promote adherence and endothelial differentiation in the absence or presence of 40 ng/mL TNFα. There was a significant (>3.5-fold) increase with TNFα versus control medium in total numbers of VE-cad⁺/flk-1⁺ adherent outgrowth from human CD14⁺ cells by day 7 of culture (P < 0.01; Fig. 2A). We performed real-time (flt-1 and flk-1) and semiquantitative (VE-cad and tie-2) reverse transcription-PCR (RT-PCR) for endothelial-specific transcripts on human CD14⁺ mononcytes after 4 days in culture to study if TNFα accelerated the time course endothelial transcript expression. flt-1 (VEGFR1) transcripts, expressed on both myeloid and endothelial cells (25), were not affected by TNFα (P > 0.1; Fig. 2B,, graph). However, flk-1 transcripts, which are endothelial-selective (25), were induced ∼2-fold with 10 ng/mL TNFα (P > 0.05, not statistically significant) and by ∼9-fold (P < 0.05) in cells treated with 40 ng/mL TNFα (Fig. 2B,, graph). Transcripts for VE-cad and tie-2, both endothelial-enriched, were detected in cells exposed to 40 ng/mL TNFα but not in its absence (control; Fig. 2B , gel).

TNFα promoted endothelial-enriched transcripts in CD14⁺ cells via TNFα/TNFR interactions. TNFα binds and signals through two distinct receptors, TNFα-R1 (also known as p55) and TNFα-R2 (p75 receptor; ref. 26). To establish the importance of TNFRs on myeloid to endothelial differentiation, we isolated CD11b⁺/CD45⁺ myeloid cells from BM of mice lacking both TNFRs (TNFα-R1 and TNFα-R2 double knockout, TNFR1/2 DKO), and their syngeneic (Bl/6) Wt controls. Culture of Wt CD11b⁺ cells generated adherent cells expressing the endothelial markers VE-cad and flk-1 (staining in inset) after 2 days in culture (Fig. 2C,, top). By contrast, equal numbers of CD11b⁺ cells isolated from TNFR1/2 DKO mice cultured under similar conditions failed to express either marker at that time point (Fig. 2C , bottom); however, continued culture of the DKO-derived cells resulted in endothelial marker expression by day 5. Taken together, the data suggested that TNFα-mediated signaling on myeloid cells accelerated myeloid to endothelial differentiation of myeloid cells resulting in significantly greater yield of myeloid/endothelial biphenotypic cells from cultured monocytes. However, TNFα receptor signaling on monocytes was not absolutely required for endothelial differentiation.

Tumor-derived TNFα promoted local tumor growth. Although multiple cells within the tumor microenvironment (i.e., leukocytes) can secrete TNFα, TNFα can also be produced by tumor epithelial cells themselves at very low, chronic levels (20, 27). To better define the effect of tumor-derived TNFα on the malignant phenotype and assess its role in modulating the phenotype of tumor-associated myeloid cells, three different tumor cell lines that expressed TNFα—B16 murine melanoma cells, LLC, and mammary cancer line, Py-mT—were generated. From the B16 line, we isolated cell lines expressing either high or low levels of TNFα: 3 high expressing (300 ± 50 pg/mL) and 3 low expressing (4.5 ± 1.3 pg/mL) cell lines compared with control tumor lines (undetectable TNFα levels). LLC lines generated low-level TNFα expression, 14.1 ± 0.5 pg/mL of TNFα (n = 2 clones), compared with undetectable TNFα levels in control LLC lines. The TNFα-expressing Py-mT lines generated also expressed a very modest increase in expression compared with control lines, 12 ± 2.1 pg/mL (n = 3 clones) versus 4.6 pg/mL (n = 2), respectively, of TNFα protein. We assessed in vitro proliferation and apoptosis rate of control and TNFα-overexpressing clones using the BrdUrd and Annexin V flow cytometric assays, respectively. As shown in Fig. 3A, TNFα-expressing B16F10 (both low and high-TNFα expressing lines) and Py-mT cells exhibited similar in vitro growth compared with control cell lines. By contrast, a modest increase in TNFα expression in LLC lines actually resulted in a small but statistically significant decrease in in vitro proliferation rate. All TNFα-expressing cell lines tested displayed similar levels of spontaneous apoptosis as their respective control lines. Together, these data suggest that TNFα overexpression did not have significant autocrine effect to promote growth or survival of tumor cells in culture.

B16F10 low or high TNFα-expressing melanoma cells (mixture of >2 clones) were implanted s.c. into the flank of Bl/6 strain; the same number of control cells expressing vector only were implanted into the contralateral flank. After 10 days, tumor volume was determined. Tumors expressing low level TNFα were >4-fold larger in size than control (P < 0.02; Fig. 3B,, left). Importantly, control/TNFα pairs obtained from the same mouse (data points connected by a line) consistently showed comparatively larger tumors resulting from TNFα overexpression compared with contralateral control tumors. To verify our findings, we tested individual TNFα-overexpressing clones to confirm increased growth over control and obtained similar results (Supplementary Fig. S2A). By contrast, high TNFα-expressing tumors did not exhibit a significant increase in growth over control (Supplementary Fig. S2B). Although proliferation in vitro of LLC cells was slightly lower as a result of TNFα expression, in vivo TNFα mediated significantly increased growth (>3-fold; Fig. 3B), similar to that observed with melanoma cells.

To substantiate the tumor-promoting effects of TNFα in a different mouse strain and tumor type, Py-mT mammary carcinoma cells were evaluated. TNFα-expressing Py-mT lines and vector only control (2–3 clones combined for each experiment) were implanted into opposing fourth mammary fat pads of Wt FVB/nJ strain, and tumor growth was monitored over time to show increased average growth evident by day 17, as soon as macroscopic tumors were evident (P < 0.05; Fig. 3B). Similar to B16 melanoma tumors, mammary tumors expressing TNFα generated significantly larger tumors relative to their paired controls, with a >6-fold increase in tumor volume than their paired control tumors on the contralateral mammary gland at 30 days (P < 0.05).

To test if the tumor-promoting effects of TNFα required signaling between TNFα and its cognate receptors on host stroma versus tumor cell-autonomous mechanisms (loss of cell polarity, proliferation, etc.), TNFα-expressing B16 melanoma or LLC lines and equal numbers of control B16 or LLC cells, respectively, were implanted into opposing flanks of TNFR1/2 DKO mice. Unlike in TNFR-competent hosts, TNFα-expressing B16 or LLC lines did not generate larger tumors compared with their paired contralateral control tumors in receptor-deficient mice (Fig. 3C). In fact, TNFα-expressing LLC cells produced significantly smaller tumors in TNFα receptor–deficient host and may reflect the observation that TNFα caused slightly lower proliferation of LLC cells (Fig. 3A).

Because TNFα is a secreted protein, it was of interest to determine if circulating TNFα levels were increased in tumor-bearing mice. PB was collected at the time of sacrifice by cardiac puncture from TNFα-overexpressing B16 melanoma tumor bearing mice (n = 5) and mice harboring only control tumors (n = 6). TNFα levels were measured in plasma by ELISA; plasma levels of TNFα were 15 ± 24pg/mL versus 22 ± 18pg/mL, respectively (P = 0.8, not statistically significant). The lack of change in systemic TNFα levels is consistent with our observations of locally driven tumor progression.

TNFα promoted microvascular density, proliferation of tumor cells, and reduced necrosis in vivo. Both B16 and LLC control tumors showed an extensive amount of necrosis compared with their respective TNFα-expressing tumors (Fig. 4A). The amount of necrosis was quantified within random sections from unpaired control and TNFα-expressing tumor samples from B16, LLC, and Py-mT tumors (Fig. 4B). Although TNFα-expressing B16 and LLC tumors were much larger than control tumors, they exhibited significantly less necrosis (P < 0.05; Fig. 4A). In contrast to the fast growing B16 melanoma and LLC tumors, Py-mT mammary tumors showed very low levels of necrosis in both control and TNFα tumors (Fig. 4A and B). To examine the mechanism by which TNFα enhanced tumor growth, we determined whether TNFα altered tumor vascular density. Control and TNFα-expressing tumor sections from each of the tumor types were immunostained with anti–platelet/endothelial cell adhesion molecule 1 (PECAM-1; CD31) antibody to evaluate microvessel density, and anti-Ki67 to assess tumor cell proliferation, respectively (Fig. 4C). TNFα-overexpressing B16 and Py-mT tumors contained ∼2- to 3-fold more immunoreactivity with the Ki67 antibody than paired control tumors, consistent with the observed increase in growth (Fig. 4D). Quantification of vascular density showed a >2-fold increase in TNFα-overexpressing tumors generated from all tumor lines: melanoma, lung, and mammary (Fig. 4D).

TNFα-promoted vascular leukocytes. We postulated that TNFα may enhance tumor growth and angiogenesis by modulating the activity of tumor-recruited monocytes via generation of biphenotypic vascular leukocytes. Tumor-associated cells obtained from tumor suspensions were evaluated by flow cytometry using the myeloid markers CD11b or F4/80 (28) in conjunction with endothelial markers tie2, flk-1, and VE-cad (Supplementary Fig. S3; Table 1). Tumors expressing TNFα contained significantly higher CD11b and F4/80 myeloid populations coexpressing either flk-1 or VE-cad (P < 0.05; Table 1) than control tumors. Although there was a statistically significant increase in TNFα-induction of CD11b⁺/Tie-2⁺ cells (TEM; ref. 11), no increase was observed in the percentage of F4/80⁺/Tie2⁺ cells. To determine if an intact host TNFα signaling was necessary for TNFα-mediated effects on tumor vascular leukocytes, we quantified melanoma-associated myeloid/endothelial biphenotypic cells from control and TNFα tumors implanted in TNFR1/2 DKO animals. The overall percentage of myeloid/endothelial populations in control tumors was reduced in TNFR1/2 DKO mice from what was observed in wild-type Bl/6 hosts (Supplementary Fig. S3; Table 1). Importantly, there was no increase in any myeloid/endothelial biphenotypic population associated with tumors overexpressing TNFα compared with control from TNFR1/2 DKO mice (Table 1). These data suggested that tumor-derived TNFα significantly increased myeloid/endothelial biphenotypic vascular leukocytes. This effect was abrogated in the TNFR-deficient host, suggesting that intact TNFα signaling in the host stroma was required for this effect, and that this effect was not mediated by secondary factors secreted from the tumor cells. As it is often assumed that TNFα-mediated tumor promotion is likely to be secondary to TNFα-mediated inflammation, we quantified total tumor-associated leukocytes and leukocyte subsets using anti-CD45, anti-CD3, anti-F4/80, and anti-B220 by morphometry (Supplementary Fig. S4A shows representative photomicrographs for melanoma only; however, morphometric data to quantify tumor leukocyte content from both tumors are presented in B). The leukocyte content was higher in the fast-growing melanoma tumors compared with mammary tumors. However, CD45 content was not significantly different between control and TNFα-overexpressing tumors of either type. Similarly, neither T cell nor macrophage numbers were significantly different in either tumor type between control and TNFα. Anti-B220 immunostaining revealed only rare, infiltrating B cells for both experimental groups (data not shown). These data add to the growing evidence that low, chronic TNFα production within tumors promote growth and vascularity, independent of tumor leukocyte recruitment. Moreover, TNFα greatly increased numbers of myeloid/endothelial vascular leukocytes in mouse melanoma tumors in the absence of a concomitant increase in overall F4/80⁺ macrophages.

To further quantify the specificity of TNFα-mediated generation of BM-derived flk-1⁺ leukocyte population and to assess if this population gave rise to functional vessels (vasculogenesis), we used a BMT model using transgenic mice that constitutively expressed β-galactosidase (lacZ) transcriptionally regulated by the flk-1 promoter (29, 30). The use of this model not only enabled clear distinction and quantification of BM-derived flk-1–expressing cells within the tumor but also enabled us to identify BM-derived (lacZ⁺) blood vessels (i.e., vasculogenesis). flk/lacZ/BMT animals were generated by reconstituting syngeneic Wt recipients with BM from transgenic donors as depicted in schematic in Supplementary Fig. S4C and, after hematopoietic engraftment, were implanted with B16 melanoma control or TNFα-expressing tumors. TNFα-expressing tumors had numerous lacZ-positive cells, most of which morphologically resembled leukocytes (Fig. 5A,, top). The control tumors had significantly fewer lacZ-positive cells (Fig. 5A,, top). The donor-derived, lacZ-positive cells showed a predilection to the tumor periphery in both tumor models (data not shown). Tumors from flk/lacZ/BMT transplanted animals were analyzed by coimmunofluorescence to show that donor BM-derived, lacZ-positive cells colocalized with cells expressing the F4/80 antigen, a mature myeloid/macrophage marker (Fig. 5A,, bottom; merged in inset). The effect of TNFα on the tumor-content of BM-derived flk-1⁺ cells was quantified in melanoma tumors of transplanted mice by measuring the fold change in tumor β-gal enzyme activity. The average ratio of β-gal activity between TNFα and control tumors was graphed and showed a statistically significant >3-fold increase with TNFα. The effect of TNFα to increase tumor-lacZ activity (flk-1⁺ content) was specifically inhibited using an anti-TNFα blocking antibody but not by the control antibody (Fig. 5B).

TNFα overexpression induces tumor vasculogenesis. Although the majority of the lacZ⁺/flk-1⁺ BM-derived cells within the tumor resembled leukocytes, a small number of β-gal⁺ microvascular structures were evident on histologic sections (Fig. 6, top). Recent data suggest that, in addition to hematopoietic progenitors, more differentiated myeloid cells can also give rise to functional vessels (8, 9). We used confocal coimmunofluorescence to colocalize the BM marker, β-gal, with infused isolectin, as a marker of functional vasculature (Fig. 6). Within 7 to 10 days after tumor implantation, evidence of BM-derived cells assembled into capillaries anastomosing with the functional host tumor vasculature was evident. The effect of TNFα on tumor vasculogenesis was assessed by counting numbers of areas of BS1-lectin/β-gal colocalization in 10 random sections from 3 experimental animals. No colocalized regions resembling vascular structures were evident in control B16 melanoma tumors. The numbers of donor-derived isolectin⁺ structures were significantly higher with TNFα, albeit comparatively few to overall numbers of vascular structures; at least five to eight areas were identified in each TNFα-expressing melanoma tumors. Hence, TNFα resulted in a significant increase in tumor-associated vasculogenesis.

Myeloid infiltrates can stimulate an antitumor response but more often stimulate tumor development, invasion, angiogenesis, and metastasis (3). It is widely accepted that tumor/myeloid cell crosstalk educate recruited myeloid cells toward a tumor-promoting phenotype(s) (31, 32). Several recently published studies have identified myeloid/endothelial biphenotypic cells in tumors and have shown that they represent a potent tumor-promoting population(s) (6, 9–15). Although a number of factors have been identified that recruit myeloid cells to tumor sites (i.e., VEGF, PlGF, MCP-1, SDF-1, β-defensins; ref. 33), little is known about local signals that modulate the endothelial differentiation of recruited monocytes toward the biphenotypic, myeloid/endothelial vascular leukocytes. Because both myeloid cells and myeloid/endothelial biphenotypic cells are implicated in proangiogenic and tumor-promoting role, we directly compared these two cell populations in the context of tumor growth. Our findings are the first to show that the biphenotypic myeloid/endothelial cells generated larger, more vascular tumors than monocytes lacking endothelial markers. Moreover, our data have identified a novel role for TNFα in myeloid to endothelial differentiation and have linked TNFα to tumor vasculogenesis and tumor colonization by myeloid/endothelial cells.

Our rationale for detailed study of the role of TNFα in the generation of biphenotypic, vascular leukocytes transpired from our findings that TNFα mediated a significant, dose-dependent up-regulation of endothelial antigens' (flk-1, tie-2, and VE-cad) mRNA expression on cultured human monocytes. Ultimately, TNFα resulted in almost 4-fold increase in total yield of myeloid/endothelial biphenotypic cells from plated CD14⁺ PB monocytes. Using murine monocytes from the BM, we showed that the effect of TNFα-mediated acceleration of endothelial-like cells required the presence of TNFRs on monocytes and was not likely to be mediated by up-regulation of secondary factors.

Increased expression of TNFα within the tumor microenvironment resulted in a local, tumor-associated increase in populations of myeloid/endothelial vascular leukocytes, including CD11b/VE-cad and the tie2-expressing myeloid populations, tie2⁺/CD11b⁺ (also known as TEM; ref. 11). The absence of intact TNFα receptors (TNFR1/2 DKO host) not only reduced overall numbers of tumor-associated vascular leukocytes but also abolished their induction by tumor-derived TNFα. Our findings were further supported by studies using the BM-transplant model. TNFα induced a large increase in BM-derived flk-1⁺ tumor leukocytes that colocalized with the late myeloid marker, F4/80. Moreover, TNFα-blocking antibodies eliminated TNFα-induced increase in BM-derived flk-1⁺ cells in the BMT model. Together, our data indicate that the TNFα-mediated induction and acceleration of myeloid to endothelial differentiation required intact TNFα receptor signaling on myeloid cells. Our data suggest that there may be multiple distinct populations of myeloid/endothelial biphenotypic cells. It is not known if these differences translate into disparate effects on tumor angiogenesis/vasculogenesis and/or progression. Published reports of vascular leukocytes, such as CD11b⁺/VE-cad⁺, suggest that they are present in very low numbers or absent in peripheral circulation and enriched in tumor sites (6, 10). Hence, our data suggest that TNFα acts in a paracrine manner on tumor-recruited monocytes as a local inducer of this cellular phenotype. On the other hand, work by De Palma and colleagues (11, 13) have shown that TEMs (CD11b⁺/tie2⁺) in mice also exist in significant numbers in peripheral circulation as well as within tumors, and recent work suggests that their tumor promoting function may be regulated at the level of recruitment to tumor sites by hypoxia and angiopoietin 2 (34). Hence, the regulation of tumor content of specific populations of vascular leukocytes, such as TEMs, may be multifactorial, both by modulating recruitment as well as by their local induction by such factors as TNFα.

There are several studies that suggest that myeloid lineage cells can also serve as endothelial progenitor cells by contributing to functional vessels (i.e., vasculogenesis; refs. 8, 9, 14). Moreover, myeloid/endothelial biphenotypic cells from both human PB or purified monocytes have been identified in neovasculature after injury and tumor growth (6, 22). Using a BMT model, we determined that TNFα expression mediated a significant increase in BM-derived functional vessels, albeit the overall numbers of BM-derived vessels were relatively few. Our study does not specifically show that TNFα-dependent tumor vasculogenesis occurred via a myeloid intermediate and not from hematopoietic progenitor populations (CD34⁺/lin⁻). However, the addition of TNFα at varying concentrations (10–40 ng/mL) failed to promote generation of endothelial-like cells from human CD34⁺/lin⁻ progenitors in culture but, instead, resulted in significant cell death (data not shown). Taken together, the data support that TNFα-mediated vasculogenesis likely occurred through a myeloid intermediate. The data also suggest that the mechanism of generating endothelial phenotype from myeloid versus hematopoietic progenitor populations is distinct, and the effect of TNFα on this process in vivo may help us discern the “origin” of tissue BM-derived endothelium in specific contexts. A recent study showed that vasculogenesis after ischemic limb injury was significantly reduced in TNFR2-deficient mice, and this was dependent on TNFR2 expression on BM-derived cells and further support a role for TNFα in modulating vasculogenesis (35).

Chronic TNFα expression by tumor cells has been shown in several model systems to play a role in tumor progression, although the mechanisms are not yet understood (36). We used both heterotopic and orthotopic implanted tumor models on distinct mouse strains to show that cancer cell clones exhibiting a very modest increase in TNFα resulted in a significant increase in tumor volume. Importantly, this low level, chronic TNFα production by our modified tumor cells parallels published levels in unmanipulated human malignant cells (20, 27). By contrast, B16 melanoma clones that expressed very high TNFα levels did not mediate enhanced growth. These findings are consistent with the observation that high TNFα levels (e.g., pharmacologic doses) are used as antitumor therapy, whereas endogenous, low-level TNFα in this study as well as other recent reports were growth promoting (36, 37).

Low-level TNFα expression by tumor cells led to increased tumor growth characterized by reduced tumor necrosis in melanoma tumors (despite larger tumor size), locally increased microvascular density, and increased proliferation index in all three cancer models. The mechanism(s) of tumor growth by TNFα are not yet understood. An unanswered question is whether TNFα primarily acts in a cell autonomous manner to promote cell growth, survival, or expression of growth factors. TNFα expressing clones of either melanoma or breast tumor type did not affect the rate of proliferation or apoptosis in vitro. Interestingly, TNFα caused a small but statistically significantly reduction in LLC proliferation. These data strongly suggest that TNFα did not drive increased in vivo growth/proliferation by a direct autocrine mechanism via TNFα receptors on tumors. These results are consistent with a recent study in which TNFα knockdown in ovarian cancer lines diminished in vivo tumor xenograft growth without altering in vitro proliferation rate (20). However, this study identified reduced tumor-derived expression of specific chemokines/factors (e.g, CCL2, CXCL12, VEGF, IL-6, and MIF) resulting from knockdown of TNFα expression, suggesting that a potential mechanism of tumor promotion by TNFα may be via autocrine up-regulation of tumor-promoting factors (20). However, our observation that TNFα-expressing B16 or LLC lines did not mediate TNFα-directed growth acceleration (in fact LLC lines actually showed reduced growth) in TNFR1/2 DKO mice suggests that a significant aspect of the growth-promoting properties of TNFα rely on paracrine interactions with tumor stroma requiring intact host TNFα signaling. An earlier study that showed mice lacking TNFRs were resistant to chemically induced skin cancers supports this finding (18).

We observed significant increase in vascular density in TNFα-expressing tumors in both tumor models. Although our data suggest that this is due, at least in part, to TNFα-induced increase in myeloid/endothelial vascular leukocytes, it is important to note that TNFα is also reported to directly modulate endothelial cells (38). In rodent choroidal endothelial cells, TNFα increased the expression of angiopoietin (Ang)2, Ang1, and VEGF and inhibition of Ang2 secretion in this model blunted TNFα-induced neovascularization in vivo (38). Hence, TNFα-mediated increase in tumor vascularity may be multifactorial.

It is frequently inferred that TNFα mediates cancer promotion through its proinflammatory activity, resulting in increased numbers of leukocytes (39). However, the degree of leukocyte recruitment did not show a statistically significant difference between TNFα expressing and control tumors. Leukocyte subpopulations detected by anti-F4/80 (macrophage), anti-CD3 (T cell), and anti-CD45/B220 (B-cell) also were similar between control and TNFα-expressing tumors. Although unexpected, the absence of inflammation in response to locally elevated TNFα has been reported. In a corneal limbus model, a potent “angiogenic” response in the absence of infiltrating leukocytes was observed when hydron implants containing 3.5 ng of TNFα were implanted in the cornea of a rat (38, 40, 41). Using PVA sponges as a model to study granulation tissue-associated angiogenesis, it was shown that genetic ablation of TNFR1 resulted in markedly decreased vascular density without significant changes in leukocyte infiltrate (42). Finally, transgenic mice that stably expressed a membrane tethered TNFα on endothelial cells did not exhibit increased inflammation until later time points in 7- to 8-month-old animals (43). The B16 melanoma and LLC tumors were excised between 10 to 14 days, and the Py-MT mammary tumors were excised 28 to 30 days after implantation. Hence, it is possible that if allowed to persist over longer time periods, even small increases in TNFα may enhance overall local inflammation. Based on our findings, however, increased inflammation is an unlikely explanation for TNFα-mediated tumor promotion in this study.

Our study shows that TNFα significantly facilitates the endothelial differentiation of myeloid cells in vitro and in vivo. Because not all tumors display a convincing association between prognosis and myeloid cell accumulation, it may be valuable to examine specific vascular leukocyte subsets of tumor-associated myeloid cells to better define the role of myeloid cells in tumor progression.

No potential conflicts of interest were disclosed.

Grant support: This work was supported by the National Institutes of Health grant HL088424 and fro17-7488m the Department of Veterans Affairs (both to P.P. Young) and DK56008 (D.B. Polk).

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.