Endothelial Side Population Cells Contribute to Tumor Angiogenesis and Antiangiogenic Drug Resistance

Lung cancer is the leading cause of cancer death worldwide. Although there have been recent advances in diagnosis and treatment, the prognosis is still very poor. Angiogenesis, the development of new blood vessels from preexisting vessels supplying oxygen and nutrients to the tumor, is regarded as a hallmark of cancer development (1). Therapeutic strategies aimed at destroying newly formed blood vessels with antiangiogenic drugs have now been adopted clinically for many types of solid tumors, including lung cancer. Many antiangiogenic agents have been developed targeting VEGF itself or VEGFRs and downstream signaling pathways (2). Although therapy with these agents has resulted in retardation of tumor progression in some patients, the results are more modest than expected (3). This may be for several reasons, including the possibility that many cancer patients are intrinsically refractory to, or develop resistance to, antiangiogenic therapy and respond only minimally (4). Moreover, inhibition of VEGF/VEGFR induces pruning of abnormal tumor blood vessels while maintaining less abnormal blood vessels, which in turn continue to provide a blood supply to the tumor (5). Several mechanisms of resistance to antiangiogenic drugs have been proposed, but little attention has been paid to the heterogeneity of resident vascular endothelial cells (EC). New blood vessels in the tumor were originally considered to be formed by angiogenesis. However, in 1997, the concept was proposed that vasculogenesis, that is, the recruitment of bone marrow–derived endothelial precursor cells (EPC) into new blood vessels, also persists into adult life and contributes to the formation of new blood vessels (6). Notably, it has been shown that bone marrow–derived ECs could constitute >50% of all ECs in the tumor vasculature (7). In contrast, recent reports have failed to show a direct contribution of EPCs to newly formed ECs (8–10), suggesting an exclusive role of resident ECs for neovascular formation.

The adult lung has long been viewed as a very quiescent tissue, with pulmonary circulation incapable of supporting the growth of new vessels. However, accumulating evidence suggests that the lung harbors tissue-resident stem cells and has remarkable reparative capacity (11, 12). Moreover, the existence of vascular-resident progenitor cells that are able to differentiate into ECs has been proposed (13). However, their characteristics and contribution to the tumor vasculature have not been clearly defined. In earlier work, we identified stem/progenitor–like ECs in peripheral blood vessels based on their ability to efflux Hoechst 33342 dye (14). Cells that do this are termed side population (SP) cells because of their characteristic appearance in flow cytometry. SP cells appear as a discrete subpopulation at the side of main population (MP) cells. The Hoechst method was first developed as a purification method for hematopoietic stem cells (15) and is now applied to a wide variety of tissue-resident stem cells and cancer cells (16, 17). We showed that these endothelial-SP (EC-SP) cells possess EC colony–forming potential in vitro and contribute to angiogenesis by generating functional mature blood vessels when transplanted into an ischemic limb. In this study, we characterize EC-SP cells in the lung vasculature and examine the origin of these cells using a Cre/loxP–based lineage-tracing system and bone marrow transplantation model. Furthermore, we show a contribution of EC-SP cells to the tumor vasculature and their potential role in resistance to small-molecule antiangiogenic drugs. Our data suggest that this heterogeneity of ECs may be a mechanism contributing to resistance to antiangiogenic therapy.

C57BL/6, DBA2, and C57BL/6-Tg (CAG-EGFP) mice [enhanced GFP (EGFP) mice that express GFP in all tissues] were purchased from Japan SLC. Mice 8 to 12 weeks of age were used for all experiments. VE-Cadherin (BAC) CreERT2 (18) and Flox-CAT-EGFP mice (19, 20) were provided by Drs. Yoshiaki Kubota (Keio University, Tokyo, Japan) and Toshio Suda (Kumamoto University, Kumamoto, Japan), respectively. Vav1-Cre mice were purchased from The Jackson Laboratory. VE-Cadherin (BAC) CreERT2 mice were crossed with Flox-CAT-EGFP mice, and recombination was induced by intraperitoneal injection of tamoxifen (Sigma) at adult ages (older than 2 months). All experimental procedures in this study were approved by the institutional Animal Care and Use Committee of Osaka University (Osaka, Japan).

Cells from lung and tumor were isolated as described previously, with slight modification (21). Briefly, mice were euthanized and organs were excised, minced, and digested with Dispase II (Roche Applied Science), collagenase (Wako), and type II collagenase (Worthington Biochemical Corp.) with continuous shaking at 37°C. The digested tissue was passed through 40-μm filters to yield single-cell suspensions. Red blood cells (RBC) were lysed with ACK buffer (0.15 mol/L NH₄Cl, 10 mmol/L KHCO₃, and 0.1 mmol/L Na₂-EDTA).

Hoechst staining was performed as described previously (14). Briefly, cell-surface antigen staining was performed, and cell suspensions were incubated with Hoechst 33342 (5 μg/mL; Sigma) at 37°C for 90 minutes in DMEM [2% FCS (Sigma), 1 mmol/L HEPES (Sigma)] at a concentration of 10⁶ nucleated cells/mL in the presence or absence of verapamil (50 μmol/L; Sigma). Cell-surface antigen staining was performed as described previously (22). The mAbs used in immunofluorescence staining were anti-CD31, -CD34, -CD44, -CD45, -VE-cadherin, -Flk1, -Sca1, and -c-Kit mAbs (BD Biosciences). Respective isotype controls (BD Biosciences) were used as negative controls. Propidium iodide (PI, 2 μg/mL; Sigma) was added before FACS analysis to exclude dead cells. Flow cytometry of stained cells was performed on a SORP FACSAria (BD Biosciences), and data were analyzed using FlowJo Software (Tree Star Software). A UV laser (emitting at 355 nm) was used to excite the Hoechst dye. To analyze cell-cycle status using Pyronin Y (Sigma), cells were first stained with Hoechst 33342 at 37°C for 45 minutes, and then 0.5 μg/mL of Pyronin Y was added and incubation continued at 37°C for another 45 minutes.

Lewis lung carcinoma (LLC; RIKEN Cell Bank) was maintained in DMEM (Sigma) supplemented with 10% FCS and 1% penicillin/streptomycin (Life Technologies). The KLN205 (mouse squamous cell carcinoma; RIKEN Cell Bank) cell line was cultured in αMEM (Sigma) supplemented with 10% FCS, 1% penicillin/streptomycin, 2 mmol/L l-glutamine (Life technologies), and 1% nonessential amino acids (Life technologies). The OP9 stromal cells (RIKEN Cell Bank) were maintained in DMEM with 20% FCS, 2 mmol/L l-glutamine, and 1% penicillin/streptomycin. Cell lines utilized are mycoplasma free, authenticated by supplier based on morphology, growth curve analysis, and isoenzyme analysis, and were passaged for fewer than 6 months after resuscitation.

The sorted EC-SP or -MP cells were seeded into plates and cocultured on OP9 stromal cells in RPMI (Sigma), supplemented with 10% FCS and 10⁻⁵ mol/L 2-mercaptoethanol (Life Technologies). VEGF (10 ng/mL; PeproTech) was added every three days. For the in vitro VEGFR inhibition assay, vandetanib (LC Laboratories) was added to the EC-SP or -MP coculture dishes at different concentrations. For the drug accumulation study, nonspecific inhibitors of the ATP-binding cassette (ABC) transporter, cyclosporine A (10 nmol/L, Sigma) or verapamil (5 nmol/L) was added to the VEGFR inhibition assay. Cells were fixed for immunostaining after 10 days.

Eight- to 12-week-old C57BL/6 mice underwent bone marrow transplantation from age-matched donor EGFP mice as described previously (23). Briefly, bone marrow cells were obtained by flushing the tibias and femurs. RBCs were depleted using ACK buffer. Transplantation was performed into C57BL/6 mice lethally irradiated with 10.0 Gy by intravenous infusion of approximately 1 × 10⁶ donor RBC-lysed bone marrow cells. At 24 weeks after transplantation, by which time recipient bone marrow was reconstituted, the mice were used for the experiments. The percent reconstitution of the bone marrow was confirmed in all mice at the time of the experiment.

RNA was extracted using RNeasy Mini Kits (Qiagen), and cDNA was generated using reverse transcriptase from the ExScript RT Reagent Kit (TaKaRa). Real-time PCR was performed using a Stratagene Mx3000P (Stratagene). PCR was performed on cDNA using specific primers. The sequences of the gene-specific primers were as follows: 5′-TGG CAA AGT GGA GAT TGT TGC C-3′ and 5′-AAG ATG GTG ATG GGC TTC CCG-3′ for GAPDH; 5′-CCA GCA GTC AGT GTG CTT ACA-3′ and 5′-GCC ACT CCA TGG ATA ATA GCA-3′ for ABCG2; 5′-CCA GCA GTC AGT GTG CTT ACA-3′ and 5′-GCC ACT CCA TGG ATA ATA GCA-3′ for ABCB1a; and 5′-TGA TCA TCA GCA ACA GCA GTC-3′ and 5′-TGA AAC CTG GAT GTA GGC AAC-3′ for ABCB1b. Expression level of the target gene was normalized to the GAPDH level in each sample.

LLC or KLN205 cells (2 × 10⁵) were mixed with 30-μL Matrigel (BD Biosciences) and injected into the lung intercostally. Tumors were studied 2 weeks after implantation when they were 5 to 10 mm in diameter. The lung metastasis model was established by injecting 1 × 10⁵ LLC cells intravenously. For the tumor cell and EC-SP or -MP cell coinoculation assays, 2 × 10⁵ LLC cells and 1 × 10⁴ EC-SP or -MP cells sorted from EGFP mice were mixed together and injected into the right lungs of wild-type mice. After 10 days, those tumors >3 mm in diameter were fixed and prepared for immunostaining. The number of GFP⁺ vascular colonies was counted in three different sections of each tumor. The sum of these counts was taken as the number of colonies in that tumor. For drug treatment, vandetanib and axitinib (Selleck Chemicals) were used as angiogenesis inhibitors as described previously (24). Vandetanib was dissolved in DMSO (Sigma)–PEG-400 (Sigma; 1:1) and injected at a dose of 30 mg/kg i.p. for 3 days. Axitinib was dissolved in a solution of PEG-400–acidified (pH 2–3) water (3:7) and injected at a dose of 25 mg/kg i.p. for 3 days.

For IHC, anti-CD31 antibody (BD Biosciences) was used for staining, and biotin-conjugated polyclonal anti-rat IgG (Dako) was used as the secondary antibody. Biotinylated secondary antibodies were developed using ABC Kits (Vector Laboratories). DAB/NiCl₂ (Sigma) was used for the color reaction. For immunofluorescence studies, anti-CD31 antibody, Cy3-conjugated anti-SMA antibody (Sigma), and anti-GFP antibody (Invitrogen) were used for staining and anti-rat IgG Alexa Fluor-546, -647 and anti-rabbit IgG Alexa Fluor-488 (Invitrogen) as the secondary antibodies. Cell nuclei were visualized with TO-PRO-3 (Invitrogen) or Hoechst dye (Sigma). Samples were imaged using an Olympus IX-70 and Leica TCS/SP5 confocal microscope. Images were processed with the Leica Application Suite (Leica) and Adobe Photoshop CS6 software (Adobe Systems).

All data are presented as the mean ± SEM. Statistical analyses were performed using Statcel 3 (OMS). Data were analyzed by ANOVA, followed by Tukey–Kramer multiple comparison tests. When only two groups were compared, the two-sided Student t test was used. P < 0.01 was considered significant.

We first performed flow cytometric analysis of cells isolated from adult mouse lung to identify EC-SP cells in the lung vasculature using a UV laser emitting at 355 nm. Among cells positive for the EC marker CD31 and negative for the hematopoietic cell marker CD45 (CD31⁺CD45⁻ ECs; Fig. 1A), 0.67 ± 0.24% were found in the Hoechst low-fluorescent SP fraction of the dot plot relative to Hoechst bright-fluorescent MP cells (Fig. 1B), in line with our previous work (14). Low intracellular accumulation of Hoechst is characteristic of SP cells, and this cell population disappears when cells are treated with ABC transporter inhibitors, such as verapamil (Fig. 1C). To evaluate the proliferative capacity of lung EC-SP cells in vitro, we cultured them on OP9 stromal cells, which can support EC growth (25). After 10 days, EC-SP cells generated a higher number of CD31⁺ EC colonies than did EC-MP cells (Fig. 1D–F). These data document the presence of EC-SP cells in the lung vasculature and their potential to generate EC colonies in vitro.

Next, we characterized the phenotype of lung EC-SP cells and found that they express the EC markers VE-cadherin, Flk1, Sca1, and CD34, as do EC-MP cells (Fig. 1G). The CD44 antigen, which is expressed by EC-MP cells but not by EC-SP cells isolated from limb muscle (14), was not expressed by either EC-SP or EC-MP cells of the lung. The c-Kit antigen, a well-known marker for hematopoietic progenitor cells, was widely expressed by EC-MP cells but not EC-SP cells (Fig. 1G). We next analyzed the expression of a set of genes known to be arterial or venous markers. The expression of the arterial markers Ephrin B2, Hey1, and Hey2 was lower in EC-SP than EC-MP cells. In contrast, venous markers EphB4 and COUP-TF2 did not differ in the two groups. Glycam1, which is upregulated in lower limb EC-SP cells (14), was not detectable in either lung EC-SP or -MP cells (Fig. 1H). These results indicate that EC-SP cells in the lung vasculature are phenotypically indistinguishable from MP cells by well-known EC markers. Furthermore, the EC-SP markers found in lower limb EC-SP cells are not applicable to lung EC-SP cells.

We next examined whether lung EC-SP cells are committed to the EC lineage by analyzing the lung vasculature of VE-cadherin CreERT2 (BAC)/Flox-CAT-EGFP mice. As expected, GFP was detected only in the ECs at the innermost layer of peripheral blood vessels by immunostaining (Fig. 2A and B) and not in peripheral blood (Fig. 2C) in these mice. Among CD31⁺CD45⁻ cells, >90% were GFP⁺ (Fig. 2C and D), suggesting that most ECs successfully underwent Cre-lox recombination. EC-SP cells are present in the GFP⁺CD31⁺CD45⁻ EC fraction (Fig. 2E). When cultured on OP9 stromal cells, these GFP⁺ SP cells generated EC colonies (Fig. 2F). Previous work by other investigators has demonstrated that lung SP cells from adult mice are composed of CD45⁺ and CD45⁻ SP cells (26). They also showed that the small percentage of CD45⁻ SP cells is derived from the bone marrow (27). Therefore, we examined whether lung EC-SP cells originate from bone marrow by analyzing wild-type mice that had undergone bone marrow transplantation from EGFP donor mice. Unlike in CD45⁺ lung hematopoietic cells, there were no highly GFP⁺ cells in the CD31⁺CD45⁻EC fraction nor in the EC-SP fraction. Most of the EC-SP cells were negative for GFP, although 3.12 ± 0.28% were weakly positive (GFP^dim; Fig. 2G). To evaluate colony-forming capacity of GFP^dim and GFP⁻ EC-SP cells, we cultured each fraction on OP9 stromal cells. After 10 days, only GFP⁻ EC-SP cells generated CD31⁺ colonies in numbers comparable with wild-type mice (Fig. 2H and I). These results indicate that EC-SP cells are phenotypically identical to terminally differentiated ECs in the endothelial-specific gene recombination model and that colony-forming EC-SP cells do not originate from bone marrow.

To study the potential of EC-SP cells in the lung to contribute to the formation of new blood vessels in the tumor, we first investigated their proliferative capacity in the tumor vasculature. We implanted LLC tumor cells orthotopically into the lung, and after 14 days, tumors were dissected for Hoechst analysis (Fig. 3A). In the tumor, the percentage of EC-SP cells was higher (7.02 ± 2.81%) than the control lung obtained from the other side of the animal (Fig. 3B–E). We confirmed that SP-ECs in the tumor were indeed SP cells by their response to verapamil treatment (Fig. 3C and E). We next analyzed the percentages of tumor EC-SP cells in the lung tumor metastasis model using LLC cells. The metastatic small nodules were gathered, minced, and stained with Hoechst dye. In the CD31⁺CD45⁻ EC fraction, the percentage of SP cells was significantly higher (2.81 ± 0.38%) than normal lung (Fig. 3F). To determine the mitotic state of these EC-SP and -MP cells, we next performed cell-cycle analysis with Hoechst dye and the RNA-binding dye Pyronin Y to identify cells in G₀ and G₁ (28). In the normal lung, 96.6 ± 1.4% of EC-SP cells and 96.7 ± 1.0% of EC-MP cells were in the Pyronin Y–negative G₀ phase (Fig. 3G and Supplementary Fig. S1A). In the tumor microenvironment, the number of Pyronin Y–positive cells, reflecting the number of cells in the active phase of the cell cycle, was significantly increased compared with the normal steady state (Fig. 3H and I and Supplementary Fig. S1B and S1C). Moreover, we examined the expression of several key factors for the cell cycle and for maintenance of quiescence using quantitative RT-PCR. It has been reported that progression of the cell cycle is dependent on members of the Cip/Kip family of cyclin-dependent kinase (CDK) inhibitors (p21, p27, and p57; ref. 29). Accordingly, the expression of all these factors was found to be significantly lower in tumor EC-SP and -MP cells compared with normal lung cells (Supplementary Fig. S1D–S1F). Furthermore, the percentage of Ki67-positive cells, indicating cell proliferation (30), was higher in tumor EC-SP and -MP cells compared with normal controls (Supplementary Fig. S1G). These results indicate that both EC-SP and -MP cells in the tumor are Ki67-positive actively proliferating cells in which p21, p27, and p57 were downregulated. Next, to study whether the origin of tumor ECs is EC-SP cells, we first cultured sorted tumor EC-SP and EC-MP cells on OP9 feeder cells to evaluate their colony-forming capacity. After 10 days, tumor EC-SP cells gave rise to EC colonies comparable with normal lung EC-SP cells (Fig. 3J). On the other hand, although cell-cycle analysis revealed that EC-MP cells were actively cycling, they generated significantly lower numbers of EC colonies, and the colony size was smaller (Fig. 3K and L). Furthermore, we inoculated tumor cells with normal lung EC-SP or -MP cells to confirm their contribution to the tumor vasculature in vivo. Lung EC-SP cells were isolated from EGFP mice and inoculated together with LLC tumor cells into the right lungs of wild-type mice. As expected, many large GFP⁺ colony-like vascular structures were formed in the EC-SP–coinoculated tumor. On the other hand, EC-MP cells only formed a few GFP⁺ small dots (Fig. 4A and B). These results indicate that although both EC-SP and -MP cells are activated in the tumor, the former generate more ECs than the latter and effectively contribute to the new blood vessels being formed in the tumor microenvironment.

Next, we compared the effectiveness of antiangiogenic drugs on tumor EC-SP and -MP cells. After 3-day treatment of tumor-bearing mice with the small-molecule tyrosine kinase inhibitors (TKI) axitinib or vandetanib targeting primarily the VEGFR, the tumor vascularity decreased by 59% and 61%, respectively (Fig. 5A–D). Although tumor vascular density decreased on treatment with these two drugs, FACS analysis revealed that the percentage of EC-SP cells in the tumor ECs increased to 17.6 ± 6.3% and 19.9 ± 6.5%, respectively (Fig. 5E–H). To determine the sensitivity of EC-SP and -MP cells to these antiangiogenic drugs, we next cultured them on OP9 stromal cells together with different doses of vandetanib. Addition of a high concentration (100–500 nmol/L) of vandetanib inhibited colony formation by both EC-SP and -MP cells (Fig. 6A). However, although colony formation by EC-MP cells was significantly inhibited, EC-SP cells maintained their colony formation potential at a low concentration (20 nmol/L) of vandetanib (Fig. 6B). It has been reported that several drug transporters are highly expressed in tissue-resident stem cells and cancer stem-like cells that have an SP phenotype (31, 32). The ABC transporter family of proteins is also well characterized for its ability to efflux a wide range of small molecules and drugs. It was also reported that tumor ECs are resistant to low-dose chemotherapy through expression of ABCB1 (33). Therefore, to clarify the mechanism responsible for this differential sensitivity to antiangiogenic drugs, we investigated the expression of mRNA for the ABC drug transporters. The level of expression of mRNA for several ABC drug transporters was indeed significantly higher in tumor EC-SP cells (Fig. 6D). Next, to confirm that antiangiogenic drugs were actually exported through these ABC transporters, tumor EC-SP and -MP cells were cultured on OP9 stromal cells, together with a low concentration (20 nmol/L) of vandetanib with or without ABC transporter inhibitors. Although addition of verapamil or cyclosporine A, which are known to inhibit the multiple drug transporters, including ABCB1, itself partially blocked the appearance of EC-SP–derived colonies, a combination of these drugs with vandetanib blocked colony formation more markedly (Fig. 6E and F). These results indicate that EC-SP cells are VEGF dependent, as is the case for regular ECs. However, in the presence of a low concentration of an antiangiogenic inhibitor, EC-SP cells are more resistant to the drugs, at least in part via ABC drug transporter activity.

Furthermore, we examined the origin of tumor EC-SP cells before and after antiangiogenic therapy to exclude the possibility that the increasing EC-SP cell population after antiangiogenic therapy derived from hematopoietic lineage cells. We analyzed tumor-bearing Vav1 Cre/Flox-CAT-EGFP (Vav1-GFP) mice and VE-cadherin CreERT2 (BAC)/Flox-CAT-EGFP (VE-cadherin-GFP) mice. In these transgenic mice, hematopoietic lineage cells are GFP⁺ in Vav1-GFP mice, and endothelial lineage cells are GFP⁺ in VE-cadherin-GFP mice. FACS analysis of cells from tumors generated in VE-cadherin-GFP mice revealed that the percentage of SP cells was increased in the GFP⁺ cell fraction to the same extent as in wild-type mice. This indicates that tumor EC-SP cells are true ECs (Fig. 7A). FACS analysis of Vav1-GFP mice revealed that all the ECs, including EC-SP cells, were GFP⁻, showing that they are not of hematopoietic origin (Fig. 7B). Moreover, tumor GFP⁺ ECs isolated from the VE-cadherin-GFP mice treated with antiangiogenic drugs harbored high percentages of EC-SP cells (Fig. 7C). On the other hand, none of the ECs, including EC-SP cells, were GFP⁺ in Vav1-GFP mice treated with antiangiogenesis drugs (Fig. 7D). These results suggest that tumor EC-SP cells originate from resident ECs, not from hematopoietic cells, and selectively survive after VEGFR inhibition via expression of drug transporters.

The presence of SP cells in the lung has been reported by several groups (34–36). Although such studies indicated that some of these SP cells are positive for endothelial markers (34, 37), their identity has not been clearly established. We previously reported the existence of stem/progenitor–like endothelial SP cells by the Hoechst method in the peripheral vasculature (14, 23). Therefore, to determine the characteristics of lung SP cells within the ECs, we gated the EC fraction to include cells expressing the endothelial marker CD31 but to exclude CD45⁺ hematopoietic cells and then applied the Hoechst assay to that fraction. We showed that ECs indeed harbor SP cells. Among the cells positive for CD31 and negative for CD45, about 0.7% were in the SP region; these cells possessed colony-forming potential in the OP9 coculture model. Since the concept of adult vasculogenesis by bone marrow derived EPCs was proposed (6), the importance of their contribution to new blood vessel formation has been intensively studied. The formation of EC colonies is a feature shared by both EPCs and EC-SP cells; however, we found that EC-SP cells are EC lineage-committed cells, which do not originate from the bone marrow. In the lung vasculature, the existence of highly proliferative resident progenitor cells that possess an endothelial colony–forming cell (ECFC)–like phenotype has been reported (38, 39). ECFCs were originally described in human peripheral and umbilical cord blood based on the timing of their emergence in culture (40). Considering that ECFCs are believed to originate from resident vascular endothelium (41), EC-SP cells may overlap with this population. However, as with ECFC, we could not find any specific or restricted molecules that permit discrimination of EC-SP cells from mature endothelium; identification of specific markers for EC-SP cells within resident vascular endothelium will be required.

In the tumor vasculature, we found that the percentage of EC-SP cells, which can generate EC colonies is increased and that these cells are actively proliferating. Furthermore, we found that tumor blood vessels are predominantly formed by EC-SP cells in a transplantation model. There is a report demonstrating a major contribution of EPC to ECs in tumors (42), but in another study, it was reported that bone marrow–derived cells do not integrate, but associate with ECs and promote angiogenesis only indirectly (9). Although there are controversies in the field of EC differentiation from EPCs, the formation of new blood vessels as a result of the mitotic division of ECs of existing blood vessels is beyond doubt (43). Our results suggest that resident vascular EC-SP cells may play an important role in tumor angiogenesis as EC-supplying cells. However, it is possible that our in vitro culture system and cotransplantation model does not accurately reflect the tumor microenvironment; therefore, lineage tracing of EC-SP cells with specific markers would be needed to provide definitive results.

Interestingly, the percentage of tumor EC-SP cells was increased after treatment with small-molecule TKIs of VEGF receptors. We found that total blood vessels decreased to about one third, and the percentage of EC-SP cells increased 2.5-fold after antiangiogenic therapy. This strongly suggests that the number of EC-SP cells did not increase but remained the same after therapy. Although the EC-SP phenotype cannot be maintained in the OP9 coculture assay, EC colonies generated from EC-SP cells were more resistant to vandetanib, at least partly via ABC transporter activity. Recently, development of resistance to antiangiogenic agents has become apparent as a clinical problem that must be overcome (4, 44). Several mechanisms for acquired resistance to antiangiogenic therapy, such as mutation of tumor cells imbuing greater resistance to hypoxia (45, 46), the compensatory increased secretion of multiple angiogenic factors by tumor cells and stromal cells (47), or increased pericyte coverage that protects from vessel regression (48) have all been suggested. Because EC-SP cells selectively remain after antiangiogenic therapy, we propose that this heterogeneity of ECs may be another mechanism for acquired resistance. However, because axitinib is known to inhibit the kinase activity of c-Kit in addition to the VEGF receptors and PDGFR-β, it is possible that EC-MP cells, which express c-Kit, are more sensitive to treatment with this drug.

In summary, we characterized vascular EC-SP cells in the lung vasculature and documented their contribution to tumor angiogenesis. We used the genetic recombination model to demonstrate that these EC-SP cells are indeed ECs and do not originate from hematopoietic cells. Furthermore, we showed that tumor EC-SP cells selectively remained after treatment with small-molecule antiangiogenic agents. Our data suggest that targeting this specialized EC population may offer a new strategy to overcome antiangiogenic drug resistance.

No potential conflicts of interest were disclosed.

Conception and design: H. Naito, N. Takakura

Development of methodology: H. Naito, N. Takakura

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Naito, T. Wakabayashi, H. Kidoya, F. Muramatsu, K. Takara, D. Eino, K. Yamane, T. Iba

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Naito, T. Wakabayashi

Writing, review, and/or revision of the manuscript: H. Naito, N. Takakura

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Kidoya, N. Takakura

Study supervision: H. Kidoya, N. Takakura

The authors thank Drs. Toshio Suda (Kumamoto University, Kumamoto, Japan) and Yoshiaki Kubota (Keio University, Tokyo, Japan) for providing them with Flox-CAT-EGFP mice and VE-cadherin (BAC) CreERT2 mice, respectively. The authors also thank M. Ishida, N. Fujimoto, C. Takeshita, and K. Fukuhara for technical assistance.

This work was supported by the Ministry of Education, Culture, Sports, Science and Technology (MEXT) Grant-in-Aid for Scientific Research on Innovative Areas grant number 22112005, the Japan Science and Technology Agency (JST) Projects for Technological Development, Research Center Network for Realization of Regenerative Medicine, Japan Society for the Promotion of Science (JSPS) Grants-in-Aid for Young Scientists B grant number 25830080, 15K18409, Takeda Science Foundation, Foundation for Promotion of Cancer Research in Japan, and Tokyo Biochemical Research Foundation.

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.