Overexpression of Platelet-Derived Growth Factor-BB Increases Tumor Pericyte Content via Stromal-Derived Factor-1α/CXCR4 Axis

The platelet-derived growth factor (PDGF) family of growth factors consists of five different disulfide-linked dimers: PDGF-AA, PDGF-AB, PDGF-BB, PDGF-CC, and PDGF-DD, which exert their biological effects through their receptors, PDGFRα and PDGFRβ (1). The essential role of the PDGF-BB/PDGFRβ signaling in vascular maturation has been well documented by genetic approaches (2, 3). During the vascular development, endothelial cells (EC) secrete PDGF-BB, which not only enhances the motility of pericytes but also forms a chemotaxis-like gradient to facilitate the pericyte recruitment (4, 5). Recently, many studies have been performed to correlate PDGF-BB signaling with vascular remodeling within tumors and show that transgenic PDGF-BB expression in tumors can increase pericyte density (6–8). Because tumor-derived PDGF-BB cannot contribute to the formation of the chemotaxis-like gradient around the endothelium (6), the underlying mechanism that tumor-derived PDGF-BB promotes pericyte recruitment remains unclear.

Here, we show that SDF-1α/CXCR4 axis compensates with PDGF-BB–induced pericyte recruitment during cancerous vascular remodeling. PDGF-BB can increase SDF-1α secretion of ECs due to the up-regulation of the transcription. This induction is mainly dependent on the activation of HIF-1α through the phosphatidylinositol 3-kinase (PI3K)/Akt/mTOR pathway. These data suggest a novel role of SDF-1α in angiogenesis and reveal an intriguing mechanism of PDGF-BB in pericyte recruitment during cancerous vascular formation process.

Cell lines and culture. HMEC is an HDMEC cell line (Sciencell) that was transfected with SV40 large T antigen, cultured with DMEM with 10% fetal bovine serum according to previous report (9). Tumor-associated ECs (TEC) were isolated from mouse tumor specimens by magnetic activated cell sorting (MACS), following the report of Camussi and his colleagues (10). All other cells were obtained from American Type Culture Collection.

Cell recruitment assay. HMECs (10⁴ cells/mL) and HASMCs (5 × 10³ cells/mL), stained by CMPTX and CMFDA (Invitrogen), respectively, were cocultured in Matrigel-coated culture slides for 16 h. Quantitation was done by measuring the merged cells from five fields of three independent experiments.

Real-time quantitative reverse transcription-PCR. RNA was isolated by TRIzol according to the manufacturer's instructions. Real-time reverse transcription-PCR (RT-PCR) was performed using the TaqMan RT-PCR Master Mix Reagents kit. Primers were synthesized by Invitrogen (sequences available on request).

Animal study. The Institutional Animal Care and Use Committee of Tsinghua University approved all experiments. Female nude mice (4–6 wk old; Vital River) were anesthetized with ketamine and xylazine. Inoculums of 10⁶ A549 or MCF-7 tumor cells in 0.1 mL of PBS was mixed with Matrigel at 4°C and then injected into the s.c. space on the flanks of nude mice. A549 tumor-bearing mice were also treated with AMD3100 (s.c. injection of 1.25 mg/kg), saline, anti–PDGF-BB antibody (i.p. injection of 25 mg/kg), or control IgG every 2 d, respectively. For another assay, in vivo JetPEI transfecting reagent was used to inject 10 μg of PDGF-B overexpression plasmid or control DNA into MCF-7 tumors or 10 μg of PDGF-B short hairpin RNA (shRNA) or control DNA into A549 tumors every 2 d. Both MCF-7 and A549 tumor-bearing mice were then treated with AMD3100 or saline every 2 d. Tumor tissues were harvested and placed in 10% formalin for fixation as previous reports (11, 12). Immunohistochemistry and immunofluorescence assays were preformed, as described by the antibody's instruction. For quantitation, at least five independent fields per section were independently analyzed by two investigators.

Statistical methods. Statistical significance was assessed by the Student's t test. P values were obtained by comparing with indicated groups. A P value of <0.01 was considered statistically significant.

SDF-1α mediates PDGF-BB–induced pericyte recruitment in vivo. To investigate the role of tumor-derived PDGF-BB in pericyte recruitment, we first established an A549 tumor xenograft model because A549 showed a high level of PDGF-BB secretion, which is consistent with the mRNA and protein levels of PDGF-BB precursor detected in vivo and in vitro, respectively (Supplementary Fig. S1). The effect of PDGF-BB on the A549 tumor vasculature was first analyzed by staining two endothelial markers (CD31 and CD105) and one pericyte marker (NG2). As shown in Supplementary Table S1, a high level of pericyte coverage (65.1 ± 13.1%) was observed within the A549 tumors, which was attenuated by anti–PDGF-BB antibody treatment (18.5 ± 6.6%). Correspondingly, blockade of PDGF-BB also led to the ECs attrition due to the loss of the mechanical support provided by pericytes. All these results show that PDGF-BB within A549 indeed regulates the pericyte recruitment.

Then a PDGF-BB low-expressing A549 tumor (A549-shPDGF-BB) was generated by the in vivo transfection with PDGF-BB shRNA. Both the low- and high-expressing A549-bearing mice were treated with AMD3100 (a CXCR4 antagonist that can block SDF-1α signaling) or saline every 2 days, respectively. As shown in Table 1 and Supplementary Fig. S2A, the levels of both vascular density and pericyte coverage decreased within A549-shPDGF-BB compared with that in A549-S.C. (A549 transfected with scramble shRNA). Intriguingly, AMD3100 treatment dramatically decreased the level of pericyte coverage in PDGF-BB A549-S.C. while having a slight effect in A549-shPDGF-BB. To further confirm this observation, MCF-7 cells, tumor cells with low PDGF-BB expression level, were s.c. injected and then transfected with a plasmid encoding pdgfb gene (MCF-7-PDGF-BB) or a control vector (MCF-7-Mock). Consistently, after treatment with AMD3100 or saline, we observed that overexpression of PDGF-BB in MCF-7 led to a remarkable increase of both the pericyte coverage and the vascular density (Supplementary Fig. S2B; Table 1). Whereas, AMD3100 treatment reversed this elevated pericyte content and increased the EC attrition in the MCF-7-PDGF-BB tumors. Based on the above findings that AMD3100, which specifically disrupts SDF-1α signaling, can block the pericyte coverage, especially in PDGF-BB high-expressing tumors, we indicate that SDF-1α/CXCR4 axis has some functional effects in the PDGF-BB–mediated vascular remodeling.

SDF-1α mediates PDGF-BB–induced pericyte-like cell recruitment in vitro. Because the role of SDF-1α in the recruitment of proangiogenic progenitor cells has been well studied (13–16), we then hypothesized that the SDF-1α/CXCR4 axis also regulates the recruitment of pericytes. Firstly, the expression pattern of CXCR4 was investigated in the primary pericytes. As shown in Supplementary Fig. S3A, CXCR4 colocalized with NG2 in the A549 tumor tissues. Moreover, after the isolation by MACS with anti-NG2, the surface expression of CXCR4 can also be detected in these NG2⁺ cells (Supplementary Fig. S3B), supporting the potential role of SDF-1α on pericytes recruitment.

To determine the in vitro effect of SDF-1α on pericytes, we used two pericyte cell models (HASMCs and 10T1/2; refs. 17, 18). Western blot and RT-PCR revealed that CXCR4 was indeed expressed in HASMCs (Supplementary Fig. S3C) and 10T1/2 (data not shown). In addition, the surface-bound SDF-1α in HASMCs could also be detected by flow cytometry, whereas AMD3100 attenuated this binding dramatically (Supplementary Fig. S3D). We then detected whether SDF-1α could induce the HASMC motility. In both transwell and wound healing assays, significantly more migrated HASMCs were observed in SDF-1α–treated group than those in the control group, whereas AMD3100 (10 nmol/L) markedly inhibited this SDF-1α–induced HASMC migration (Supplementary Fig. S4A and B). On the other hand, neither the activation of PDGFRβ nor the viability of HASMCs was affected by AMD3100 treatment (Supplementary Fig. S4C and D). These results were further confirmed by the same approaches using 10T 1/2 cells (data not shown).

To elucidate the functional significance of SDF-1α/CXCR4 axis in pericyte recruitment, a modified transwell assay was introduced to mimic the in vivo tumor microenvironment. HMECs were cultured in the lower wells to confluence. HASMCs transfected with PDGFRβ small interfering RNA (siRNA; HASMC^PDGFRβ-KD) or scrambled siRNA (HASMC^Control) were seeded in the upper chamber. As shown in Fig. 1A and B, HMECs conditioned with PDGF-BB recruited HASMC^Control cells more significantly than did HMECs in the normal condition. Intriguingly, HASMC^PDGFRβ-KD exhibited similar migration activity compared with HASMC^Control, whereas either AMD3100 or anti–SDF-1α antibody blocked the maintenance of HASMC^PDGFRβ-KD migration compared with the control group. A three-dimensional culture assay was developed to confirm this observation. As shown in Fig. 1C and D, HMECs and HASMCs (HASMC^PDGFRβ-KD or HASMC^Control) were labeled with CMTPX or CMFDA, respectively, and then mixed together with Matrigel. Consistently, PDGF-BB increased the amount of attached HASMCs (both HASMC^PDGFRβ-KD and HASMC^Control), which was significantly attenuated by AMD3100 and anti–SDF-1α antibody. Taken together, all these results reveal that although the PDGF signaling is disrupted by knockdown of PDGFRβ in HAMSCs, SDF-1α can serve as a compensating factor to facilitate the pericyte recruitment.

PDGF-BB induces SDF-1α expression in ECs. Based on our previous in vivo and in vitro studies, we showed that the SDF-1α/CXCR4 axis coincides with the PDGF-BB–induced pericyte recruitment during cancerous vascular remodeling. We then hypothesized that PDGF-BB could induce the SDF-1α secretion in ECs. HMECs were used as the in vitro EC model (9). As PDGFRβ is highly expressed on TECs (19–23), we first investigated the expression of PDGFRβ in HMECs. As shown in Supplementary Fig. S5A–C, PDGFRβ was indeed expressed in HMECs. Confocal Z sections also revealed a prominent PDGFRβ localization on the surface of HMECs (Supplementary Fig. S5D; Supplementary Video S1).

HMECs were treated either with different doses (0, 5, 10, or 20 ng/mL PDGF-BB for 8 hours) or for different time periods (20 ng/mL PDGF-BB for 0, 4, 8, or 12 hours). The conditioned medium was evaluated by SDF-1α ELISA quantitation kit (R&D Systems). The result showed that SDF-1α secretion was both dose-dependent and time-dependent (Fig. 2A), consistent with SDF-1α protein levels of the whole-cell lysate (Fig. 2B). Quantitative real-time RT-PCR analysis revealed similar induction trends of sdf-1α transcripts (Fig. 2C). Moreover, after treatment with PDGF-BB, the luciferase activity of SDF-1α promoter was elevated 3-fold (Fig. 2D), suggesting that the PDGF-BB–induced SDF-1α secretion occurs at the level of transcription. On the other hand, the phosphorylation of PDGFRβ was also induced in HMECs after PDGF-BB treatment. Both the SDF-1α mRNA level and PDGFRβ phosphorylation were attenuated by either anti-PDGFRβ antibody, PDGFRβ tyrosine kinase inhibitor III, or PDGFRβ siRNA (Supplementary Fig. S5E), indicating that the activation of PDGFRβ is essential for the induction of SDF-1α.

To further confirm this PDGF-BB–induced SDF-1α expression in TECs, we isolated the primary ECs from tumor specimens by MACS with anti-CD105 antibody following the protocol reported by Camussi and his colleagues (10). As shown in Fig. 3A and B, the CD105⁺ TECs did express not only PDGFRβ but also other angiogenic EC markers, including Tie-2 and vascular endothelial growth factor receptor 2 (VEGFR2). In addition, these TECs were negative for CD45 and NG2, suggesting that these cells were not contaminated by either pericytes or hematopoietic stem cells (Supplementary Fig. S5F). To avoid the contamination of fibroblasts and myoepithelial cells, we further detected the level of α-SMA. As expected, <0.5% of these isolated TECs was α-SMA positive (data not shown). To fully characterize the isolated TECs, several TEC markers were analyzed by RT-PCR using mouse-specific primers (22). CD31, VEGFR1, VEGFR2, Tie-1, Tie-2, TEM-1, TEM-5, and TEM-8 were all positive (Fig. 3C). Consistent with the previous results (Fig. 2A–C), after treatment with PDGF-BB, both the mRNA and the protein level of SDF-1α were dramatically elevated (Fig. 3C and D), indicating that PDGF-BB indeed induces the SDF-1α expression in ECs.

PDGF-BB induces SDF-1α expression through HIF-1α activation. HIF-1α is the central mediator of the cellular response to hypoxia, regulating over 60 genes to modulate cell metabolism in ischemic conditions, including sdf-1α and vegf (24–26). Thus, we hypothesized that the PDGF-BB–induced SDF-1α expression is regulated by HIF-1α. As shown in Fig. 4A and Supplementary Fig. S6A and B, both time-dependent and dose-dependent induction of HIF-1α mRNA were observed after PDGF-BB treatment, with a peak at 4 hours, whereas the protein level was dramatically elevated after 8 hours. Then, the effect of PDGF-BB on the translocation of HIF-1α was examined. As shown in Fig. 4B, a large amount of HIF-1α was detected in the nucleus fraction after PDGF-BB treatment. Consistent result was also observed by immunofluorescence assay (Supplementary Fig. S6C). To confirm the role of HIF-1α on the PDGF-BB–induced SDF-1α expression, two specific HIF-1α siRNAs (si-HIF-1 and si-HIF-2) were synthesized. After knocking down HIF-1α, the protein level of SDF-1α was totally abolished in the PDGF-BB–treated HMECs (Fig. 4C). Moreover, the activities of hypoxia-responsive element (as a positive control) and SDF-1α promoter were markedly blocked by these HIF-1α–targeted siRNAs, suggesting that HIF-1α activation is essential for the sdf-1α gene transcription (Fig. 4D).

PDGF-BB mediates SDF-1α expression through the PI3K/Akt pathway. Because activated PDGFRβ initiates several signaling pathways, including PI3K/Akt and ERK/MEK (27), we then tested which downstream could regulate SDF-1α expression. As shown in Supplementary Fig. S7A, U0126 (MEK/ERK inhibitor) showed little effect on SDF-1α expression whereas triciribine (Akt inhibitor) could dramatically block this induction. Thus, we focused on the regulatory role of the PI3K/Akt pathway in this induction. Wang and colleagues showed that vegf, a target gene of HIF-1α, can be induced by PDGF-BB through the PI3K/Akt pathway in ECs (28). mTOR, a downstream target of Akt kinase involved in protein synthesis, regulates the accumulation of HIF-1α (29). We, thus, hypothesized that the PI3K/Akt/mTOR pathway may result in the activation of HIF-1α, which in turn regulates the SDF-1α expression. To determine the functional significance of this pathway, HMECs were treated with rapamycin, triciribine, LY294002, or BAY 11-7082, which inhibit the kinase activity of mTOR, Akt, PI3K, or nuclear factor-κB (negative control), respectively. The HIF-1α expression induced by PDGF-BB was significantly decreased by rapamycin, triciribine, or LY294002, respectively (Supplementary Fig. S7B). In contrast, BAY 11-7082 did not alter HIF-1α expression in HMECs. In addition, immunofluorescence assay was performed to confirm the regulatory role of the PI3K/Akt/mTOR pathway in the HIF-1α translocation (Supplementary Fig. S7C). All these data show that the PI3K/Akt/mTOR pathway is indeed involved in the PDGF-BB–induced SDF-1α expression, and this mechanism is mainly dependent on HIF-1α accumulation and activation.

Angiogenesis is a multistep process involving EC sprouting and pericyte recruitment (30–32). Recently, antiangiogenic targeting has been expanded to include pericytes that provide both survival signals and structural support to ECs (33). Variable degrees of pericyte coverage in tumors have been reported (7, 8). However, the exact molecular mechanism mediating cancerous pericyte coverage remains unclear. Here, we show that the SDF-1α/CXCR4 axis essentially coincides with the PDGF-BB–induced pericyte recruitment during cancerous vascular remodeling (Fig. 5). Blockade of the SDF-1α/CXCR4 axis not only abolishes the PDGF-BB–induced pericyte recruitment in vitro but also significantly decreases pericyte coverage of PDGF-BB–overexpressing tumors. Moreover, we show that PDGF-BB can increase the secretion of SDF-1α in ECs due to the up-regulation of the mRNA synthesis of SDF-1α. This up-regulation mechanism is mainly dependent on the activation of HIF-1α through the PI3K/Akt/mTOR signaling pathway.

The role of PDGF-BB/PDGFRβ axis in vascular remodeling. The effect of PDGF-BB on cancerous pericyte recruitment has been reported by many independent studies (6–8). Betsholtz and his colleagues reported that transgenic PDGF-BB expression in tumor cells can increase the pericyte density in both wild-type and pdgfb^ret/ret mice (6). Ellis and his colleagues also provided evidence that overexpression of PDGF-BB leads to the increase of the tumor pericyte content, showing that the recruitment of pericytes within tumors is in a PDGF-BB dose-dependent manner (8). The more PDGF-BB is produced, the more pericytes are recruited. Of note, this tumor-derived PDGF-BB does not contribute to the chemotaxis gradient lining the endothelium and conversely disrupts the preexist gradient formed by endothelium, suggesting that some compensating mechanism may cooperate with the PDGF-BB/PDGFRβ axis. Because CXCR4 expression was observed in the pericyte progenitor cells (16), mesenchymal stem cells (a potential origin of pericytes in tumor tissues; ref. 34), and primary pericytes (35), we hypothesize that SDF-1α serves as a compensating factor for this PDGF-BB–induced pericytes recruitment.

Based on the current study, we observed that blockade of SDF-1α/CXCR4 do not show marked effects on the pericyte recruitment in the PDGF-BB knocking-down A549 tumors. One proposed mechanism is that the induction of SDF-1α signaling is regulated by PDGF-BB. As shown in Supplementary Fig. S1C, the expression level of SDF-1α in the PDGF-BB knocking-down A549 tumor was dramatically diminished compared with the PDGF-BB–overexpressing A549 tumor, supporting that the amount of PDGF-BB in tumor tissues indeed contributes to the induction of SDF-1α expression. Regarding the potential role of endothelial-derived PDGF-BB mainly produced from the leading endothelial tip cells (5), the overall effect may not be as potent as PDGF-BB that was continuously secreted from tumor cells, because the PDGF-BB–induced SDF-1α expression is in a dose-dependent manner (Fig. 2). However, unlike tumor-derived PDGF-BB, the role of endothelial-derived PDGF-BB is more than a mediator for the pericytes recruitment. Betsholtz and his colleagues showed that, other than the chemoattractant effect, endothelial-derived PDGF-BB retention is indispensable for proper integration of pericytes in the vessel walls (6). Whether SDF-1α has a similar effect on the integration of pericytes remains unclear.

On the other hand, we also detected the expression of SDF-1 in both vSMC/pericyte cell lines and primary pericytes isolated from A549 tumors. As shown in Supplementary Fig. 8, both HASMC and 10T1/2 can express SDF-1 upon PDGF-BB stimulation in vitro. Intriguingly, the primary pericytes isolated from A549 tumors did not show as high mRNA level of SDF-1 as TECs did. Based on the findings of Cheresh and his colleagues, the activation of PDGFRβ signaling in vSMC/pericytes is suppressed by the assembly of a PDGFRβ-VEGFR2 receptor under the induction of VEGF (21). Thus, one proposed reason for these different SDF-1 mRNA levels is that, in the A549 tumor microenvironment, which exhibits high VEGF concentration, the response of vSMC/pericytes on PDGF-BB stimulation may also be inhibited by the assembly of the PDGFRβ-VEGFR2 receptor.

The role of SDF-1α/CXCR4 axis in angiogenesis. SDF-1α is a constitutively expressed and inducible chemokine that regulates multiple physiologic processes, including embryonic development and organ homeostasis (36–38). The cognate receptor for SDF-1α, CXCR4, is widely and constitutively expressed by numerous cells, including pericytes and smooth muscle cells (16, 37, 39). SDF-1α/CXCR4 axis has a major role in the recruitment of CXCR4⁺ bone marrow cells to the neoangiogenic niches supporting vascular remodeling and tumor growth; it also traps and correctly positions the proangiogenic myeloid cells around the growing vessels in tissues (13). In the current study, we have linked SDF-1α/CXCR4 axis to the PDGF-BB–induced pericyte recruitment, based on the identification of CXCR4⁺PDGFRβ⁺cKit⁺ pericytes and CXCR4⁺ smooth muscle progenitor cells (35). During vascular development, these progenitor cells can be recruited through the SDF-1α chemotaxis gradient and differentiate into vSMC/pericytes to facilitate the blood vessel formation. The crucial roles of SDF-1α and CXCR4 in embryonic vasculogenesis are shown by the blood vessel abnormalities manifested in the SDF-1α^−/− and CXCR4^−/− mice (40). These studies strongly suggest that EC-derived SDF-1α is highly possible to be involved in the cancerous pericyte recruitment process. It has also been reported that SDF-1α can recruit several immune cells to promote angiogenesis (14, 15, 37, 39). Here, we detected some immune cell markers, including CD4, CD8, CD11b, CD14, and CD45 (Supplementary Fig. S9). Among them, CD11b and CD14 showed corresponding trends with SDF-1α expression, supporting the previous reports that Gr-1⁺CD11b⁺ myeloid cell and CD14⁺ dendritic cells can be recruited to facilitate tumor progression by SDF-1α (41, 42).

AMD3100, composed of 1,4,8,11-tetraazacyclotetradecane moieties, binds CXCR4 with extreme high specificity and, thus, is an ideal tool to evaluate the physiologic and pathologic importance of SDF-1α/CXCR4 interactions (43). However, whether AMD3100 will affect the viability of pericytes and/or other kinase activation is still unclear. To confirm this issue, we investigated the effect of AMD3100 on the PDGFRβ phosphorylation first. As shown in Supplementary Fig. S4C, no inhibition effect was observed after AMD treatment (0–50 nmol/L). We also detected the protein levels of several apoptosis-related proteins, including caspase-8, AIF-1, Bcl-2, and Bcl-Xs/l. No significant variations of these markers were observed (Supplementary Fig. S4D).

The expression of PDGFRβ in angiogenic ECs. There is little consensus in the literature as to whether or under what circumstances ECs express PDGFRβ in vivo (20, 21, 23, 44). St. Croix and his colleagues compared gene expression patterns of ECs between physiologic and pathologic angiogenesis and showed that several markers are only selectively overexpressed in the pathologic angiogenesis of tumors, but not the physiologic angiogenesis (45). Recently, Wurdinger and colleagues illustrated that miR-296 can up-regulate the expression of tumor angiogenesis-related receptors, such as PDGFRβ and VEGFR2 (23). However, whether this regulatory mechanism is dependent on certain microenvironments remains unclear. Thus, more efforts are still needed to unravel the underlining mechanism that regulates the expression of the PDGFRβ in the tumor angiogenic ECs.

In the current study, both HMEC and human umbilical vascular EC (HUVEC) were used as EC models. As shown in Supplementary Fig. S5C, we detected several angiogenic-related receptors in HMECs and HUVECs (with/without A549 conditioned media). The expression levels of VEGFR2, Tie-2, CD105, and PDGFRβ were higher in HMECs and HUVECs treated with tumor-conditioned media than that in HUVECs in the resting conditions (Supplementary Fig. S5C). In addition, we isolated TECs from mouse tumor specimens by CD105 (10, 46). The results of fluorescence-activated cell sorting (FACS) and RT-PCR analysis indicated that these CD105⁺ cells indeed had the expression of several angiogenic EC markers, including PDGFRβ (Fig. 2B and C).

In summary, the present study provides evidence that the SDF-1α/CXCR4 axis is required for the PDGF-BB–mediated pericyte recruitment within tumor tissues. Through PDGFRβ phosphorylation, tumor-derived PDGF-BB induces SDF-1α expression through the HIF-1α activation in ECs. This endothelial-derived SDF-1α forms a chemotaxis gradient and recruits pericytes to facilitate vascular maturation. These discoveries show a novel role for SDF-1α in angiogenesis and establish an intriguing compensating mechanism of PDGF-BB in pericyte recruitment during vascular formation and remodeling.

No potential conflicts of interest were disclosed.

Grant support: General Programs of National Natural Science Foundation of China grants 30670419 and 30771083, Major Program of National Natural Science Foundation of China grant 30490171, National High Technology Research and Development Program of China grant 2007AA02Z155, and State Key Development Program for Basic Research of China grant 2006CB910305.

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.

We thank Yamin Tian (University of Oxford) for providing the pGL-HRE-luc plasmid for the luciferase studies and Yun Zhu and Kenichi Harimoto (Tsinghua University) for their help in performing flow cytometry.