Abstract
Organ-specific metastasis is governed, in part, by interactions between chemokine receptors on cancer cells and matching chemokines in target organs. For example, malignant breast cancer cells express the chemokine receptor CXCR4 and commonly metastasize to organs that are an abundant source of the CXCR4-specific ligand stromal cell-derived factor-1
(ref. 1). It is still uncertain how an evolving tumour cell is reprogrammed to express CXCR4, thus implementing the tendency to metastasize to specific organs. Here we show that the von Hippel–Lindau tumour suppressor protein pVHL negatively regulates CXCR4 expression owing to its capacity to target hypoxia-inducible factor (HIF) for degradation under normoxic conditions. This process is suppressed under hypoxic conditions, resulting in HIF-dependent CXCR4 activation. An analysis of clear cell renal carcinoma that manifests mutation of the VHL gene in most cases revealed an association of strong CXCR4 expression with poor tumour-specific survival. These results suggest a mechanism for CXCR4 activation during tumour cell evolution and imply that VHL inactivation acquired by incipient tumour cells early in tumorigenesis confers not only a selective survival advantage but also the tendency to home to selected organs.
Inactivation of the VHL tumour suppressor gene is linked to the development of several different tumour types in humans, including hereditary and sporadic clear cell carcinoma of the kidney2. The best characterized function of pVHL is as a recognition subunit of an E3 ubiquitin protein ligase complex that targets the -subunits of the DNA-binding transcription factor HIF (ref. 3) for ubiquitin-mediated degradation in the presence of oxygen4. Tumour-derived pVHL mutants are defective in this regard and manifest constitutive activation of HIF target genes5, 6. Indeed, HIF activation is an early event in the evolution of neoplastic kidney lesions in VHL disease7. To identify novel pVHL-regulated genes, we compared the gene expression profile of pVHL-deficient renal cell carcinoma (RCC) cells previously stably transfected with an empty vector (designated A498(neo)) with those engineered to stably produce haemagglutinin (HA)-tagged wild-type pVHL30 (designated A498(HA–pVHL30)) using Affymetrix gene chip technology. This approach identified 101 genes that were significantly upregulated in pVHL-expressing A498 cells relative to vector control cells, and 64 that were significantly downregulated (Supplementary Fig. S1).
We were particularly intrigued by the fact that, among the genes most strongly suppressed by the reintroduction of functional pVHL30 in A498 cells is the gene encoding the G-protein-coupled chemokine receptor CXCR4 (Supplementary Fig. S1), the receptor for the chemokine stromal cell-derived factor-1 (SDF-1; also referred to as CXCL12)8, 9. Chemokine receptor–ligand interactions have been implicated in the homing of various subsets of haematopoietic cells to specific anatomical sites10, 11 and to determine the metastatic destination of malignant breast cancer cells1. The latter finding provided molecular support for the 'chemoattraction theory' of organ-specific metastasis12, 13. A key unanswered question concerns the molecular mechanism through which cells acquire CXCR4 during the evolution of tumour cells. The fact that the gene on top of the 'list' of pVHL-suppressed genes was CXCR4 provided us with an unexpected avenue down which to pursue this question.
To validate the oligonucleotide microarray results, we performed Northern blot analysis. pVHL-negative A498(neo) cells express high levels of CXCR4 messenger RNA (Fig. 1a, lane 1), whereas their pVHL-expressing counterparts, A498(HA–pVHL30), do not (lane 2). The disappearance of CXCR4 mRNA from pVHL-expressing cells was correlated with a downregulation of HIF2 protein expression. A498(neo) cells also exhibited strong surface expression of the CXCR4 receptor as demonstrated by immunofluorescence microscopy (Fig. 1b, upper panel). Thus, pVHL30 negatively regulates, directly or indirectly, CXCR4 mRNA and protein production in RCC cells.
Figure 1: pVHL and hypoxia regulate CXCR4 expression.
a, Panels from the top: CXCR4 mRNA expression in A498(neo) and A498(HA–pVHL30) cells, and immunoblots of HIF2, pVHL and CDK2. b, A498(neo) and A498(HA–pVHL30) cells stained with anti-CXCR4 monoclonal antibody (green) and propidium iodide (red). c, Retrovirally infected A498 cell pools stably expressing pVHL30, pVHL19 or pVHL30(N78S) were analysed for the expression of CXCR4 mRNA and for HIF2, pVHL and CDK2 proteins. d, e, CXCR4 and GLUT3 mRNA expression in HEK-293 cells (d) and RPTECs (e) treated with hypoxia or cobalt chloride (100 
M) as shown. WB, western blot.
The above experiments were performed with an A498 clone expressing pVHL30. To test whether pVHL19 behaves similarly and to assess whether certain naturally occurring tumour-derived mutants of pVHL are defective in CXCR4 regulation, we infected A498 cells with empty control retrovirus or with retroviruses encoding untagged pVHL30, pVHL19 or the HIF degradation-defective tumour mutant pVHL30(N78S)14. Expression of pVHL30 or pVHL19 suppressed CXCR4 mRNA production (Fig. 1c, lanes 2 and 3), whereas expression of pVHL30(N78S) did not (Fig. 1c, lane 4). pVHL30 mutants that exerted residual degradation activity towards HIF2 such as pVHL30(Y98H) or pVHL30(R167W)14 had intermediate suppressive effects on CXCR4 mRNA abundance (data not shown). These results suggest that pVHL is a negative regulator of CXCR4 gene expression in RCC cells and indicate that this function is affected in tumour-derived mutants of pVHL that fail to regulate the HIF system. Moreover, they imply a link between loss of function of pVHL, the ensuing expression of HIF and the overproduction of CXCR4 mRNA.
In keeping with the above, CXCR4 mRNA was strongly induced in pVHL-positive human embryonic kidney (HEK-293) cells and primary human proximal renal tubular epithelia cells (RPTECs) when they were subjected to low oxygen concentrations (1%) (Fig. 1d, e, respectively). In addition, the hypoxia-mimetic substance cobalt chloride induced CXCR4 mRNA in RPTECs (Fig. 1e). The kinetics of CXCR4 mRNA accumulation in response to hypoxia closely followed that of GLUT3, an established HIF target15 (Fig. 1e). Taken together, these results suggest that CXCR4 is a hypoxia-inducible gene.
To substantiate this conclusion we cloned the human CXCR4 promoter including the first intronic region. Sequence analysis of this region revealed four potential hypoxia-response elements (HREs) located within 2.6 kilobases upstream of the transcriptional start site and one at position +1 kb within the intron (Fig. 2a). A luciferase reporter containing this CXCR4 promoter fragment, when introduced into VHL-positive HEK-293 cells, was activated about twofold by hypoxia (data not shown). Co-transfection of wild-type HIF1 enhanced hypoxia-inducible reporter activity (about 10-fold; Fig. 2a) and a prolyl-hydroxylation-defective mutant of HIF1, HIF1(P564A), which escapes pVHL control, activated the CXCR4 promoter also under normoxic conditions (about 7.5-fold; Fig. 2a). A progressive deletion analysis of the CXCR4 promoter revealed that the HRE located at -1.3 kb upstream of the transcriptional start is critical for hypoxia/HIF1-inducible reporter activity (Fig. 2a), and mutation of this sequence in the context of the full-length CXCR4 promoter rendered the promoter insensitive to hypoxia and HIF1 (Fig. 2a). Finally, an oligonucleotide comprising this sequence (referred to hereafter as CXCR4-HRE (CHRE)) bound baculovirus-produced HIF1–ARNT (Aryl hydrocarbon receptor nuclear translocator) heterodimers in electrophoretic mobility-shift assays (Fig. 2b, lane 3). Oligonucleotide binding of HIF1–ARNT complexes was specific, as shown by competition and supershift experiments (Fig. 2b, lanes 4–12 and 13–15, respectively). Similarly, whole cell extracts of A498 cells gave rise to a prominent gel-shift complex (Fig. 2c, lane 2), which competed specifically with wild-type but not with a mutant CHRE oligonucleotide (Fig. 2c, lanes 6–8 and 9–11, respectively). In addition, anti-HIF2 antibodies abolished (Fig. 2c, lane 12) or supershifted (lanes 13 and 14) this complex, suggesting that it contains HIF2. Anti-HIF1 antibodies had no effect (Fig. 2c, lane 15). These results strongly suggest that CXCR4 is a novel target gene of the DNA-binding transcriptional activator HIF.
Figure 2: CXCR4 is a target gene of HIF.
a, Luciferase activity after co-transfection of indicated CXCR4 reporter plasmids and HIF1 alleles in HEK-293 cells. Numbers indicate fold activation relative to 
-galactosidase (-gal) standard and control. Black squares indicate potential HREs. Lower panel: schematic representation of the human CXCR4 locus. b, Gel-shift assay using a probe spanning the CHRE and cell extracts from baculovirus-infected Sf9 cells coexpressing human HIF1 and ARNT. Competitor oligonucleotides (amounts in nanograms) and antibodies used are indicated. c, Extracts of A498 cells were assayed for CHRE DNA binding. Competition and supershift experiments are indicated as described for b. Arrows in b and c indicate specific DNA binding complexes consisting of HIF1/ARNT (b) and HIF2/ARNT (c). WT, wild type.
Next we asked whether the presence of CXCR4 on the surface of A498 cells would confer chemoattraction. As shown in Fig. 3a (left panel), stimulation of A498 cells with increasing concentrations of SDF-1 resulted in directed migration of these cells in a trans-well assay. This response was abrogated by the restoration of wild-type pVHL (Fig. 3a, right panel). Because directional cell migration and tissue invasion require changes in the dynamics of the actin cytoskeleton, we next tested whether CXCR4 signalling in RCC cells results in the activation of downstream effectors such as LIM kinase 1 (LIMK-1)16. The latter is important in controlling actin dynamics through inactivation of the actin depolymerization factor cofilin17. Stimulation of A498 cells with SDF-1 led to a rapid activation of LIMK-1 (Fig. 3b, left panel) and extracellular signal-related kinases (ERKs), which regulate cell proliferation and the cell's motility machinery (Fig. 3c, upper panel). Restoration of pVHL function abrogated these responses (Fig. 3b, c). The loss of pVHL tumour suppressor function in RCC cells is therefore intimately linked to the acquisition of a novel property, namely the enhanced responsiveness of such cells to chemotactic signals.
Figure 3: pVHL suppresses SDF-1
-mediated chemotaxis of RCC cells.
![Figure 3 : pVHL suppresses SDF-1|[alpha]|-mediated chemotaxis of RCC cells. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com](/nature/journal/v425/n6955/images/nature01874-f3.0.jpg)
a, Left: chemotactic response of A498 cells to different concentrations of SDF-1. Results are expressed as the ratio between cells migrating towards the chemokine gradient and cells migrating in the negative control. Right: chemotaxis of A498 cells expressing the indicated pVHL species after exposure to 10 nM SDF-1. b, In vitro LIM kinase-1 assay after treatment of A498, A498(neo) and A498(HA–pVHL30) cells with SDF-1. c, Cells were treated for indicated durations with 10 nM SDF-1 and processed for immunoblotting with anti-phospho-ERK1/ERK2 specific antibodies (upper panels) or anti-ERK1/2 total protein antibodies (lower panels).
Next we asked whether CXCR4 is overexpressed in human renal tumours and, if so, whether its upregulation is correlated with the inactivation of VHL. As a surrogate of the latter, we determined the expression of the HIF target genes encoding carbonic anhydrase (CA9) and glucose transporter 1 (GLUT1) because upregulation of these genes is correlated with VHL status in clear cell RCC. Expression levels of CXCR4, CA9 and GLUT1 were assessed in 29 clear cell RCC and 5 papillary RCC using real-time quantitative polymerase chain reaction (PCR) and compared with those in normal renal tissue (n = 7). As illustrated in Fig. 4a, mRNA levels of CXCR4, CA9 and GLUT1 were significantly higher in clear cell than in papillary RCC (P < 0.05; Supplementary Table S3) or normal renal tissue. Because papillary RCCs are not characterized by VHL inactivation, these findings argue that CXCR4 expression follows VHL inactivation and HIF target gene activation in clear cell RCC.
Figure 4: CXCR4 expression in human renal cell carcinoma.
a, Quantitative reverse transcriptase PCR analysis of normal kidney tissues and renal cell carcinoma. Fold expression levels are normalized to 18S rRNA and compared with the average values in normal kidneys. Red bars, CXCR4; blue bars, GLUT1; green bars, CA9. b–g, Immunohistochemical evaluation of CXCR4 expression. b, Invasive breast cancer. c, Normal renal tissue; arrows indicate a glomerulum (left) and a tubulus (right). Lymphocellular infiltrates stain intensively. Examples of weak (d) and strong (e–g) CXCR4 expression in clear cell RCCs. Original magnifications: b, c, 
400; d, e, 50; f, g, 400. h, Kaplan–Meier analysis of tumour-specific survival in patients with clear cell RCC according to CXCR4 expression. P = 0.001.
To evaluate the expression of CXCR4 in a wide range of RCC, we screened a renal cancer tissue microarray (TMA) containing 532 different elements for CXCR4 protein expression by immunohistochemistry. The antibody used in this study detects high levels of CXCR4 expression in breast cancer tissue (Fig. 4b), consistent with earlier results. Glomeruli and tubules in normal kidney tissue stained weakly, whereas intensive staining of lymphocellular infiltrates was observed (Fig. 4c). With this antibody we observed a pattern of CXCR4 expression on the TMA ranging from absent/weak to strong (Fig. 4d–g, respectively). To determine whether CXCR4 expression is correlated with clinical aggressiveness in clear cell RCC, we focused our analysis on 195 of 407 clear cell carcinoma samples on the TMA for which tumour-specific survival data were available. Within this cohort, strong CXCR4 expression was detected in 93 samples (47.6%). In the remainder of the TMA elements, staining was either weak or absent. Statistical analysis of these data revealed no significant correlation of high levels of CXCR4 expression with tumour stage and/or differentiation grade (Supplementary Table S4). However, we found a striking positive correlation between strong CXCR4 expression and poor tumour-specific survival (P = 0.001) in a univariate analysis (Fig. 4h). Importantly, multivariate Cox proportional hazards analysis suggests that this correlation was independent of tumour stage and differentiation grade (Table 1). No correlation with tumour-specific survival was observed for CA9: it displayed strong staining in 169 (87%) of the 195 samples (data not shown). CA9 expression was also not correlated with tumour stage and grade (data not shown). These results imply that a high level of CXCR4 expression is a predictor of poor tumour-specific survival, which in turn suggests that monitoring CXCR4 expression in patients with RCC may provide additional prognostic information.
The data presented in this report suggest a potential mechanism of how and when evolving clear cell RCC cells may acquire CXCR4 and thus the tendency to metastasize to specific secondary sites. It is known that the survival of patients with clear cell RCC is limited by the development of metastasis. The association of strong CXCR4 expression with poor tumour-specific survival implies that CXCR4 expression levels influence the metastatic behaviour of clear cell RCC. The acquisition of CXCR4 expression is likely to occur during the initial phases of tumour progression, set off by the functional inactivation of pVHL. Whether CXCR4 has additional roles, such as the promotion of cell proliferation, that could confer a selective advantage for the tumour cell at the primary site remains to be investigated. Finally, activation of HIF is also the result of intratumoral hypoxia and non-VHL oncogenic pathways18, 19, 20 (for example, in glioblastoma the expression of CXCR4 is localized to regions of necrosis and angiogenesis21). Hence, the mechanism underlying tumour cell-specific production of CXCR4 proposed here may apply to solid tumours in general.
One of the most remarkable features of metastatic tumours is the degree to which they differ genotypically and phenotypically from their primary tumour. A key question that follows is whether these changes have already taken place in the primary tumour, allowing it to spread to a specific secondary site, or whether primary tumour cells that are carried to secondary organs undergo these changes after they have been exposed to the new environment. The results presented here suggest that the expression of a receptor important for homing to distant organs is directly regulated by the hypoxia pathway and that its upregulation is already seen in early tumour stages. The propensity to metastasize, at least for certain cancers, may therefore be determined by the identities of the mutant alleles acquired relatively early during multistep tumorigenesis22.
Methods
Cell culture, retroviral infections and reporter-gene assays
Cell culture conditions and procedures for retroviral infections as well as plasmid constructions are available from the authors on request. A 2.7-kb fragment of the human CXCR4 promoter and the intronic region was amplified by PCR from A498 genomic DNA and subcloned into pGL2b and pGL2-promoter vector (Promega), respectively. HEK-293 cells were subjected to 12 h of hypoxia in a three-gas incubator (Nabco) 24 h after transfection. Luciferase activity was corrected for -galactosidase activity and at least three independent experiments were performed.
Gene microarray analysis
Analysis was performed with HG_U95A GeneChips (Affymetrix). Scanning was done in an Affymetrix GeneChip scanner. Analysis was performed using Affymetrix Microarray Suite v5 and GeneSpring 4.2.1 (Silicon Genetics). Changes in gene expression were assessed by looking for concordant changes between replicates by using a signed Wilcoxon rank test. The 'change' threshold P was less than 0.003 for increase and more than 0.997 for decrease. After concordance analysis these values become less than 9 10-6 and more than 0.999991 respectively. Any gene whose detection P was more than 0.05 in all experimental conditions was discarded from the analysis. Cluster analysis was performed with dCHIP software (www.biostat.harvard.edu/complab/dchip).
Immunoblotting and northern analysis
Immunoblot analysis for HIF2 ( EPAS1-ab199; Novus Biologicals), pVHL23, CDK2 ( M2; Santa Cruz), ERK (Cell Signaling Technology catalogue no. 9102) and phospho-ERK (Cell Signaling Technology catalogue no. 9101) were performed with standard techniques. Total RNA was prepared with TRIzol reagent (Life Technologies). Blots were hybridized in ExpressHyb (Clontech) to 32P-labelled probes.
DNA binding assays
Preparation of cell lysates and binding reactions were performed as described24, 25. Oligonucleotide sequences for vascular endothelial growth factor and CXCR4 are available from the authors on request. Antibodies used were: H1alpha67 (Novus Biologicals), affinity-purified anti-human HIF1 chicken polyclonal antibody (gift from M. Gassmann), anti-HIF2 ( EPAS1-ab199, HIF2alpha-ab8365, Novus Biologicals; C-16, Santa Cruz) and anti-E2F-1 ( KH-95, Santa Cruz).
Chemotaxis assays
Cell migration was assayed with trans-well inserts with 8-m pore membranes. Human SDF-1 was obtained from R&D Systems. Membranes were precoated with fibronectin (0.5 g ml-1). After starvation for 24 h, cells were detached with PBS containing 2.5 mM EDTA, centrifuged and resuspended in DMEM, 0.1% BSA, 12 mM HEPES pH 7.4 at 2 105 cells per 100 l. After 4 h, cells on the lower surface were fixed and stained with crystal violet. Five different fields of the membrane were counted to assess the number of cells that had migrated.
In vitro kinase reactions
Polyclonal rabbit serum to human LIMK-1 (anti-LIMK-1) was raised against a fusion protein of full-length LIMK-1 and maltose-binding protein (J.L., unpublished work) and affinity-purified23. A498 cells were serum-starved for 24 h, treated with 1 nM SDF-1 for the indicated times, subjected to immunoprecipitation as described26 and processed for kinase reactions in the presence of glutathione S–transferase–cofilin16.
Quantitative PCR
Total RNA was extracted from frozen tissue, treated with DNase I using the RNeasy system (Qiagen) and reverse transcribed with random hexamer primers (Amersham) and AMV reverse transcriptase (Roche). Complementary DNA was analysed by the fluorogenic 5'-nuclease PCR assay. Primers and probes (Applied Biosystems): CXCR4 forward 5'-TGGGTGGTTGTGTTCCAGTTT-3', reverse 5'-ATGCAATAGCAGGACAGGATGA-3', probe 5'-FAM-CATGGTTGGCCTTATCCTGCCTGGTA-TAMRA-3' (where TAMRA represents carboxytetramethylrhodamine). GLUT1, CA9 and 18S rRNA were detected with Assays-on-Demand primer and probe sets Hs00197884_m1, Hs00154208_m1 and Hs99999901_s1, respectively (Applied Biosystems). Gene-specific PCR products were measured continuously by an ABI PRISM 7000 Sequence Detection System (Applied Biosystems) during 40 cycles. 18S rRNA was used for normalization. Expression values for tumour samples were compared with the expression in normal renal tissue. Student's t-test was applied to analyse the expression between different grades, tumour stages and tumour types using StatView 5.0 PPC (SAS Inst. Inc., Cary, North Carolina). P < 0.05 was considered as significant.
Immunofluorescence and tissue microarray analysis
Indirect immunofluorescence staining of CXCR4 was performed as described27 with MAB173 (R&D Systems). Confocal laser scanning microscopy was performed on a FV500/BX61 microscope (Olympus). Immunohistochemistry was performed on a tissue microarray containing 532 renal tumours and normal tissue28 obtained from the renal nephrectomy series at the University of Basel29. Approval to analyse the tissues was obtained from the ethical commission in Basel. Upon antigen retrieval, standard avidin–biotin complex immunohistochemistry ( ABC-Elite; Vectra Laboratories) was used to evaluate CXCR4 ( MAB 172; R&D Systems; dilution 1:2000) and CA9 (M75 (ref. 30), dilution 1:160) expression. Colour was developed with 3-amino-9-ethylcarbazole and sections were counterstained with haematoxylin. CXCR4 and CA9 positivity was assessed semiquantitatively by staining intensity as negative, weak or strongly positive by two independent investigators without prior knowledge of the clinical follow-up data. Cases were accepted only as strongly positive if reviewers independently defined them thus. Contingency table analysis was used to analyse the relationship of CXCR4 and CA9 expression to differentiation grade and tumour stage. Clear cell RCCs were analysed for patient survival with the Kaplan–Meier method. Statistical differences between the groups were determined with the log-rank test. A Cox proportional hazards analysis was used to test for independent prognostic information. Tumour-specific survival data were obtained by reviewing the hospital records, by direct communication with the attending physicians, and from the Cancer Registry of Basel. Patients were evaluated from the time of biopsy diagnosis to the last known follow-up. Tumour-specific clinical follow-up data were available from 195 patients.
