Vascular leukocytes contribute to tumor vascularization

Tumors require blood supply for expansive growth. With increasing distance from vessels, hypoxic tumor cells produce angiogenic factors that induce the formation of neovessels.¹  These are different from vessels of normal tissue at the morphologic and molecular levels.^2,3  Until recently, angiogenesis, or sprouting of endothelial cells (ECs) from existing vessels, was the only accepted mechanism of tumor vascularization. Recent studies have suggested that vasculogenesis, or recruitment of endothelial progenitors that differentiate into endothelial cells, might contribute to the formation of tumor neovessels.⁴  However, the exact nature of endothelial progenitors remains controversial. Although initially it was thought that they arise from a CD34⁺ hematopoietic stem cell population, only a small percentage of these cells appears to be incorporated into the neovasculature.⁵  Emerging studies have reported the capability of a CD34^- monocyte/macrophage–containing mononuclear cell population to differentiate into endothelial-like cells in vitro.^6-9  However, definitive in vivo evidence for the existence of leukocytes contributing to postnatal vascularization is lacking. Investigating the existence of leukocytes with angiogenic potential infiltrating ovarian carcinomas, we found definitive evidence that leukocytes can function as endothelial precursors in postnatal life.

Specimen preparation, cell sorting, and flow cytometry are detailed as supplementary information (available on the Blood website; see the Supplemental Document link at the top of the online article).

To evaluate the angiogenic potential of vascular leukocytes (VLCs), we labeled sorted CD45⁺VE-cadherin⁺ cells from 3 different specimens with cell tracker fluorochrome carboxyfluorescein diacetate (CFSE). We suspended more than 200 000 VLCs or control cells in 150 μL endothelial culture media (Cambrex, Rutherford, NJ) containing 400 μg/mL human VEGF₁₆₅ (Peprotech, Rocky Hill, NJ) plus 40 μL rabbit antiasialo GM1 (Wako Chemicals, Richmond, VA), mixed them with 150 μL cold Matrigel (BD Biosciences, Bedford, MA), and injected them subcutaneously into BALB/c SCID mice (Jackson Laboratory, Bar Harbor, ME). After 2 weeks, we injected biotinylated Lycopersicon esculentum (tomato) lectin (Vector Laboratories, Burlingame, CA) through the left ventricle (10 μg/g body weight). Tomato lectin selectively binds to fucose residues on the endothelial cell surface.¹⁰  Animals were killed by perfusion fixation,¹¹  and Matrigel plugs were resected. Biotinylated L. esculentum lectin was detected by Cy3-streptavidin (Arcturus, Mountain View, CA). Counterstaining was performed using Vectashield with DAPI (Vector Laborato).

We analyzed tumor-infiltrating host cells from 10 consecutive stage III ovarian carcinomas for their expression of leukocyte marker CD45 or endothelial marker VE-cadherin⁴  by flow cytometry. Strikingly, 3 distinct cell fractions were clearly discernible in all specimens: CD45⁺VE-cadherin^- (leukocytes), CD45^-VE-cadherin⁺ (endothelial cells), and, surprisingly, CD45⁺VE-cadherin⁺ (Figure 1A). CD45⁺VE-cadherin^- cells comprised mainly tumor-associated lymphocytes, based on forward scatter (FSC) and side scatter (SSC) profiles. The CD45^-VE-cadherin⁺ fraction corresponded to bona fide endothelial cells (ECs). Notably, the population coexpressing CD45 and VE-cadherin exhibited a markedly high frequency and represented 52% (31%-84%) of CD45⁺ cells and 74.3% (34%-96%) of VE-cadherin⁺ cells. This was not caused by nonspecific binding of antibodies; because the cells were thoroughly blocked with 10% AB serum or 20% fetal bovine serum, isotype controls produced no significant signal. Furthermore, CD45⁺VE-cadherin⁺ cells were rarely detectable (less than 1%) in ascites or peripheral blood from patients with cancer (Figure 1A). Finally, CD45⁺VE-cadherin⁺ cells were not found in most normal mouse tissues tested, with the exception of spleen, where they represented approximately 2% of cells (Supplemental Document 1). Thus, the coexpression of leukocyte-specific marker CD45 and endothelium-specific marker VE-cadherin¹²  identifies a novel population of cells with leukocyte/endothelial-like phenotype, which we named vascular leukocytes, or VLCs.

To further characterize VLC markers, we sorted the CD45^-VE-cadherin⁺ and CD45⁺VE-cadherin⁺ fractions from solid specimens. More than 90% of ECs or VLCs expressed typical endothelial markers P1H12, CD34, and CD31⁺ (Figure 1B). As expected, the CD45^-VE-cadherin⁺ subset (ECs) did not express the leukocyte-specific markers CD11c, CD14, or CD8α. ECs were also CD123^-, and more than 90% were AC133^-, reflecting a mature EC phenotype. In contrast, more than 90% CD45⁺VE-cadherin⁺ cells (VLCs) were MHC class II (MHC-2⁺) and CD86⁺ (not shown), whereas more than 86% expressed the monocytic and dendritic cell (DC) marker CD11c. Sixty percent of VLCs also expressed CD8α (range, 51%-91%), 50% expressed CCR6 (range, 48%-55%), 40% expressed CD14 (range, 15%-80%), and 20% expressed CD123 (range, 6%-34%). Importantly, 45% of VLCs were AC133⁺ (range, 15%-80%), suggesting an immature phenotype in at least some VLCs because committed myeloid cells lose AC133.⁵  VLCs differed in size (FSC) and granularity (SSC). They were large medium light– to high light–scattering cells, reflecting heterogeneity within this population, though most of them exhibited FSC and SSC characteristics of monocytes and DCs (Figure 1C).

Real-time reverse transcription–polymerase chain reaction (RT-PCR) of sorted ECs and VLCs, human umbilical vein endothelial cells (HUVECs), and normal human spleen revealed similar levels of CD31 mRNA in VLCs and HUVECs (Figure 1C), whereas tumor-derived ECs exhibited significantly higher expression. Confirming their different lineage, we observed differential expression of CD45 mRNA in VLCs and spleen (very high) compared with ECs and HUVECs (absent or very low). TEM1 and TEM7, previously proposed as tumor endothelium–specific makers,¹³  were both found at comparable levels in tumor ECs and VLCs, but not in HUVECs. Thy-1, a lymphocyte marker also overexpressed on neoplastic endothelium,¹³  was found at higher levels in VLCs than in spleen. Interestingly, sorted tumor ECs expressed CD11c mRNA, though at much lower levels than VLCs.

Next, we observed that VLCs cultured on fibronectin-coated plates for 2 weeks developed cytoplasmic projections and intercellular junctions (Figure 2B). To verify their angiogenic potential, freshly sorted CD45⁺VE-cadherin⁺ cells were CFSE-labeled, suspended in Matrigel containing endothelial culture media plus antiasialo GM1 antibody (to inhibit natural-killer [NK]–mediated rejection), and xenotransplanted into the flanks of SCID mice. After 2 weeks, all subcutaneous Matrigel plugs exhibited tomato lectin–perfusionable capillaries that were assembled by green fluorescent cells that had anastomosed with the host's vascular tree (Figure 2). Cells were CFSE^bright, suggesting that no significant proliferation occurred. As expected, sorted tumor ECs and HUVECs also formed functional capillaries in Matrigel (HUVECs formed large plugs and were CFSE^dim, indicating proliferation), whereas CD45⁺VE-cadherin^- cells and did not partake in the formation of blood vessels (Figure 2A and Supplementary Figure 1). Control plugs injected with activated CD8⁺ T cells¹⁴  were reabsorbed after 2 weeks, consistent with the absence of vessel formation. Finally, select tumors exhibited capillary structures harboring CD45⁺ cells that coexpressed VE-cadherin (Figure 2C). Taken together, our data indicate that VLCs display morphologic and functional properties of endothelial-like cells and have the capacity to functionally contribute to vasculogenesis in vivo.

The phenotypic overlapping between hematopoietic and endothelial cells has complicated the characterization of EPCs.⁷  We show for the first time a population of CD45⁺VE-cadherin⁺P1H12⁺CD34⁺CD31⁺TEM1⁺TEM7⁺ cells expressing typical leukocyte markers in vivo, which are able to create functional blood vessels. It is intriguing to speculate that tumor VLCs are analogous to CD34⁺VE-cadherin⁺ endothelial progenitors previously identified by Burger et al,¹⁵  though CD45 was not included during that analysis. Unexpectedly, VLCs in our samples were more prevalent than canonical ECs, suggesting a critical and ignored role for hematopoietic cells in tumor vascularization. However, we cannot exclude an excessive proportion of VLCs because of our dissociation procedure. Mechanical-only dissociation of tumor specimens might have favored the retrieval of VLCs from tumor islets over bona fide ECs from tumor stroma or, alternatively, might have preserved epitope exposure compared with enzymatic separation. The degree of contribution of VLCs to human tumor vascularization warrants further investigation because they appear to be a promising novel therapeutic target.