Failure to prolyl hydroxylate hypoxia‐inducible factor α phenocopies VHL inactivation in vivo

William Y Kim, Michal Safran, Marshall RM Buckley, Benjamin L Ebert, Jonathan Glickman, Marcus Bosenberg, Meredith Regan, William G Kaelin

Author Affiliations

  1. William Y Kim1,
  2. Michal Safran1,
  3. Marshall RM Buckley1,
  4. Benjamin L Ebert1,2,
  5. Jonathan Glickman3,
  6. Marcus Bosenberg4,
  7. Meredith Regan5 and
  8. William G Kaelin Jr*,1,6
  1. 1 Department of Medical Oncology, Dana‐Farber Cancer Institute and Brigham and Women's Hospital, Harvard Medical School, Boston, MA, USA
  2. 2 Broad Institute of MIT and Harvard, Cambridge, MA, USA
  3. 3 Department of Pathology, Brigham and Women's Hospital, Harvard Medical School, Boston, MA, USA
  4. 4 Department of Pathology, University of Vermont, Burlington, VT, USA
  5. 5 Department of Biostatistics and Computational Biology, Dana‐Farber Cancer Institute, Boston, MA and Department of Medicine, Harvard Medical School, Boston, MA, USA
  6. 6 Howard Hughes Medical Institute, Chevy Chase, MD, USA
  1. *Corresponding author. Department of Medical Oncology, Dana‐Farber Cancer Institute and Brigham and Women's Hospital, Harvard Medical School, 44 Binney Street, Mayer 457, Boston, MA 02115, USA. Tel.: +1 617 632 3975; Fax: +1 617 632 4760; E-mail: William_kaelin{at}
View Full Text


Many functions have been assigned to the von Hippel‐Lindau tumor suppressor gene product (pVHL), including targeting the alpha subunits of the heterodimeric transcription factor HIF (hypoxia‐inducible factor) for destruction. The binding of pVHL to HIFα requires that HIFα be hydroxylated on one of two prolyl residues. We introduced HIF1α and HIF2α variants that cannot be hydroxylated on these sites into the ubiquitously expressed ROSA26 locus along with a Lox‐stop‐Lox cassette that renders their expression Cre‐dependent. Expression of the HIF2α variant in the skin and liver induced changes that were highly similar to those seen when pVHL is lost in these organs. Dual expression of the HIF1α and HIF2α variants in liver, however, more closely phenocopied the changes seen after pVHL inactivation than did the HIF2α variant alone. Moreover, gene expression profiling confirmed that the genes regulated by HIF1α and HIF2α in the liver are overlapping but non‐identical. Therefore, the pathological changes caused by pVHL inactivation in skin and liver are due largely to dysregulation of HIF target genes.


von Hippel‐Lindau (VHL) disease is a hereditary cancer syndrome characterized by the development of various tumors including retinal and central nervous system (CNS) hemangioblastomas, pheochromocytomas, and clear cell renal carcinomas (Kim and Kaelin, 2004). VHL patients carry one wild‐type VHL allele and one mutated VHL allele in their germline. Pathology develops when the wild‐type allele is somatically inactivated in a susceptible cell. Mutation or hypermethylation of the VHL locus is also very common in sporadic renal cell carcinomas and sporadic CNS hemangioblastomas (Kim and Kaelin, 2004).

The VHL gene product, pVHL, is the substrate recognition module for an E3 ubiquitin ligase complex that contains elongin B, elongin C, Cul2, and Rbx1 (Kaelin, 2002). Cells lacking pVHL constitutively over‐produce hypoxia‐inducible mRNAs (Gnarra et al, 1996; Iliopoulos et al, 1996; Siemeister et al, 1996; Stratmann et al, 1997). Many of these mRNAs are controlled by a heterodimeric transcription factor called hypoxia‐inducible factor (HIF), which consists of a labile alpha subunit and a stable beta subunit. Humans possess three alpha subunit genes (HIF1α, HIF2α, or HIF3α) and three HIFβ genes (also called the aryl hydrocarbon receptor nuclear translocator (ARNT)) (Semenza, 2003). pVHL binds directly to HIFα subunits and orchestrates their polyubiquitination, leading to their destruction by the proteasome (Kaelin, 2002). The binding of pVHL to HIF is governed by an oxygen‐dependent, post‐translational modification (prolyl hydroxylation) of HIFα subunits catalyzed by EGLN family members (also called PHDs or HPHs) (Kaelin, 2005; Schofield and Ratcliffe, 2005). Thus, HIFα subunits are normally degraded in the presence of oxygen but are stabilized under hypoxic conditions or in the setting of pVHL loss, leading to transactivation of HIF target genes.

Over 60 HIF target genes have been identified, many of which play critical roles in cellular and systemic responses to hypoxia (Semenza, 2003). Target genes such as VEGF and PDGF‐B are implicated in angiogenesis and likely contribute to the hypervascularity observed in VHL‐associated tumors through their actions on endothelial cells and pericytes, respectively. Another HIF target, TGFα, is a powerful renal epithelial cell mitogen and probably contributes to the development of both hemangioblastoma and renal cell carcinoma (Kim and Kaelin, 2004).

HIF1α and HIF2α are not wholly redundant. HIF1α−/− mice die in midgestation and exhibit severe blood vessel defects (Iyer et al, 1998). HIF2α−/− mice also exhibit embryonic or early postnatal lethality but as a consequence of catecholamine deficiency or lung maturation defects (Tian et al, 1998; Peng et al, 2000; Compernolle et al, 2002). In a recent study, embryonic stem (ES) cells in which an HIF2α cDNA was homozygously inserted into the HIF1α locus were more angiogenic and tumorigenic than their wild‐type counterparts (Covello et al, 2005). Finally, HIF1α and HIF2α produce overlapping yet distinct gene expression profiles. For example, HIF1α appears to be more important than HIF2α with respect to the hypoxic induction of glycolytic genes and the proapoptotic gene BNip3 (Hu et al, 2003; Raval et al, 2005; Wang et al, 2005).

HIF2α appears to play a special role in VHL−/− renal carcinomas. First, VHL−/− renal carcinomas express both HIF1α and HIF2α or exclusively HIF2α (Maxwell et al, 1999). Second, an apparent switch from HIF1α to HIF2α expression occurs in preneoplastic lesions arising in human VHL+/− kidneys in association with increasing dysplasia and cellular atypia (Mandriota et al, 2002). Third, inhibition of HIF2α by shRNA is sufficient to suppress VHL−/− tumor growth in vivo (Kondo et al, 2003; Zimmer et al, 2004). Finally, the ability of pVHL to suppress tumor formation by VHL−/− cells can be overridden by an HIF2α variant that cannot be prolyl hydroxylated, but not by the corresponding HIF1α variant (Maranchie et al, 2002; Kondo et al, 2002, 2003; Raval et al, 2005).

Genotype–phenotype correlations in VHL disease suggest, however, that pVHL has functions in addition to regulating HIF. For example, pVHL mutants that retain the ability to bind and polyubiquitinate HIF have been linked to familial pheochromocytoma (so‐called type 2C VHL disease) (Cockman et al, 2000; Hoffman et al, 2001). Moreover, pVHL mutants linked to type 2A VHL disease (high risk of hemangioblastoma and pheochromocytoma and low risk of renal cell carcinoma) are, like 2B mutants (high risk of all three tumors), defective with respect to polyubiquitinating HIF, suggesting that a second pVHL function modulates renal cancer risk in the setting of VHL disease. Finally, patients with Chuvash polycythemia, who are homozygous for a hypomorphic VHL allele, develop polycythemia (presumably due to a quantitative HIF defect leading to increased erythropoietin (EPO) production) but are not tumor prone (Ang et al, 2002; Pastore et al, 2003; Gordeuk et al, 2004).

Biochemical studies also suggest that pVHL has multiple functions. pVHL has been reported to polyubiquitinate atypical protein kinase C, two deubiquitinating enzymes (VHL‐interacting deubiquitinating enzyme 1 and 2), and two subunits of RNA polymerase II (hsRBP7 and the hyperphosphorylated from of Rbp1) (Czyzyk‐Krzeska and Meller, 2004). Furthermore, pVHL has been reported to bind directly to the plant homeodomain protein, Jade‐1, and to the KRAB‐A domain‐containing protein, VHLaK (Czyzyk‐Krzeska and Meller, 2004). Gene expression studies also support that pVHL has HIF‐independent functions (Wykoff et al, 2000, 2004; Zatyka et al, 2002; Jiang et al, 2003).

The existence of HIF‐independent pVHL functions might explain the observation that forced activation of HIF target genes in animal models has not caused tumors. Production of nondegradable forms of HIF or HIF‐like proteins in mouse skin and rabbit musculature induces normal appearing blood vessels (Vincent et al, 2000; Elson et al, 2001). This might suggest that dysregulation of HIF is not sufficient to promote tumor formation. A caveat, however, is that these studies were conducted using HIF1α (or variants thereof) leaving open the possibility that increased HIF2α or HIF3α activity, alone or together with HIF1α, is required for tumor formation. It is also possible tumor promotion by HIF is context dependent with respect to tissue and species. This would explain the limited spectrum of tumors observed in VHL patients and the discordance between mice and man with respect to the tissues affected by pVHL loss.

VHL−/− mice die during embryogenesis due to placental insufficiency while VHL+/− mice develop cavernous liver hemangiomas (Gnarra et al, 1997; Haase et al, 2001; Ma et al, 2003). Similar liver abnormalities are observed with shorter latency and higher penetrance when pVHL is conditionally inactivated in hepatocytes of VHL flox/flox mice. In addition, such mice are runted, die prematurely, and develop steatosis (Haase et al, 2001; Ma et al, 2003). A recent study showed that these hepatic lesions could be prevented by simultaneous deletion of ARNT (but not by elimination HIF1α) implying that dysregulation of at least one ARNT partner other than HIF1α is necessary for the development of these lesions (Rankin et al, 2005). In this report, we show that dysregulation of HIF2α is sufficient to partially phenocopy pVHL loss in the skin and liver and that simultaneous dysregulation of HIF1α and HIF2α in the liver produces lesions that are virtually indistinguishable from those observed after pVHL inactivation in that organ.


Generation of mouse strains that conditionally produce HIF1α or HIF2α variants

We set out to make mice in which hemagglutinin (HA)‐tagged HIF1α or HIF2α variants that escape recognition by pVHL due to proline to alanine substitutions (hereafter called HIF1dPA and HIF2dPA for simplicity) could be conditionally expressed in tissues of interest. Earlier studies showed that these variants are relatively stable compared to their wild‐type counterparts and retain the ability to transcriptionally activate HIF target genes (Masson et al, 2001; Hu et al, 2003; Kondo et al, 2003). In this way, we hoped to ask whether activation of HIF target genes is sufficient to induce pathological changes observed when pVHL function is compromised.

Toward this end we created targeting vectors for inserting the HIF1dPA or HIF2dPA cDNA into the mouse ROSA26 locus by homologous recombination (Figure 1A). Both targeting vectors also introduced a Lox‐stop‐Lox cassette 5′ of the cDNAs so that their transcription would be dependent upon Cre‐mediated excision of the stop element (Jackson et al, 2001; Mao et al, 2001; Srinivas et al, 2001). The ROSA26 locus is dispensible for normal development and heterologous genes placed under its control are ubiquitously expressed (Zambrowicz et al, 1997; Soriano, 1999; Mao et al, 2001; Srinivas et al, 2001). ES cells that had undergone homologous recombination with the targeting constructs for ROSA26‐HA‐HIF1dPA and ROSA26‐HA‐HIF2dPA, as determined by Southern blot analysis (Figure 1B) and PCR (data not shown), were microinjected into C57/B6 blastocysts to generate chimeric mice. Germline transmission was confirmed by detection of the targeted ROSA26 locus in the offspring of chimeras (data not shown).

Figure 1.

Insertion of HA‐HIF1dPA and HA‐HIF2dPA cassettes into ROSA26 locus by homologous recombination. (A) Schematic of the pRosa26‐HA‐HIF1dPA and pRosa26‐HA‐HIF2dPA targeting vectors and the Rosa26 locus before and after Cre‐mediated excision of the transcriptional stop element. LoxP sites are represented by the arrowheads and the 5′ external probe is indicated by the hatched rectangle. Shown also are the primers (arrows) used for the generation of (C). (B) Representative Southern blots of individual ES clones after electroporation with indicated targeting vectors and drug selection. Genomic DNA was digested with EcoRV and hybridized to 5′ external probe. The wild‐type locus produces an 11.2 kb fragment (arrow). Successful recombination is indicated by the presence of a 3.8 kb fragment (arrowhead), which arises because of an internal EcoRV site in the stop element. (C) MEFs were infected with an adenovirus expressing Cre recombinase. DNA was extracted from the MEFs 48 h later and PCR was performed using primers (see (A)) that differentiate the targeted allele (300 bp product; arrowhead) (primers 1 and 2) from the recombined alleles (400 bp product for HIF1dPA; arrow) (primers 1 and 3) and 700 bp product for HIF2dPA; dashed arrow (primers 1 and 4)). (D) MEFs with the indicated genotypes were infected with adenoviral Cre. The proteasome inhibitor MG132 (+) or DMSO (−) was added 48 h later. Immunoblot analysis was performed 6 h later with the indicated antibodies.

In pilot experiments, mouse embryo fibroblasts (MEFs) were isolated from embryos of HIF1dPA/+ and HIF2dPA/+, or wild‐type littermate controls and infected with adenovirus encoding Cre recombinase. Exposure to Cre led to successful recombination of the targeted allele (Figure 1C). As expected, HA‐HIF1α and HA‐HIF2α were detectable in lysates derived from HIF1dPA/+ and HIF2dPA/+ MEFs infected with adenoviral‐Cre, but not in wild‐type littermate control MEFs infected with adenoviral‐Cre. HA‐HIF1α and HA‐HIF2α were not detectable in uninfected MEFs (Figure 1D). The levels of HA‐HIF1α were lower than that of HA‐HIF2α, possibly due to the existence of pVHL‐independent ubiquitin ligases that can target HIF1α (Ravi et al, 2000; Isaacs et al, 2002). In support of this idea, HA‐HIF1α, but not HA‐HIF2α, was sensitive to the addition of the proteasome inhibitor MG132 (Figure 1D). Likewise, the abundance of endogenous HIF1α, but not HIF2α, increased after treatment of VHL−/− human renal carcinoma cells with MG132, indicating that the behavior of HA‐HIF1α was not peculiar to the transgene (Figure 2A).

Figure 2.

Expression levels of HA‐HIF1α and HA‐HIF2α. (A) Immunoblot analysis of HeLa and RCC4 cells treated with the proteasome inhibitor MG132 (+) or DMSO (−) for 6 h. (B) Immunoblot analysis of RCC4 cells or MEFs with the indicated genotypes. Where indicated, cells were infected with an adenovirus encoding Cre recombinase 72 h before lysis.

The levels of HA‐HIF1α and HA‐HIF2α achieved after Cre‐mediated activation in HIFdPA/+ MEFs were comparable to the levels of their endogenous murine counterparts in VHL−/− MEFs, as determined by immunoblot analysis with antibodies against epitopes that are conserved between human and mouse HIF1α or human and mouse HIF2α, respectively (Figure 2B). Similar results were obtained with two independent antibodies for each HIFα protein (data not shown). The levels of HA‐HIF1α and HA‐HIF2α were somewhat lower than observed in VHL−/− human renal carcinoma cells (Figure 2B). Although the significance of this finding is not clear, it suggests that the results described below are not due to gross overexpression of the HIFα transgenes. The levels of HA‐HIF1α and HA‐HIF2α achieved following Cre exposure were associated with induction of known HIF target genes, such as VEGF, ADRP, Egl nine homolog 3 (EglN3) and Glut1 (Supplementary Figure 1A and B).

HIF2α causes skin changes that are similar to those observed after pVHL loss

To test whether the conditional alleles were functional in vivo, we next crossed the HIF1dPA and HIF2dPA mice to mice (K14‐Cre) expressing Cre recombinase under the control of the keratin14 promoter. In parallel, we crossed mice carrying a conditional VHL allele (VHLflox/flox, hereafter referred to as VHLf/f for simplicity) (Haase et al, 2001) to K14‐Cre mice. The K14‐Cre transgene efficiently deletes LoxP containing alleles in the basal keratinocytes of the epidermis (Indra et al, 2000). Earlier transgenic experiments indicated that overproduction of specific isoforms of the HIF target gene VEGF in keratinocytes induces the formation of leaky blood vessels and skin ulceration (Larcher et al, 1998; Thurston et al, 1999), while overproduction of a stabilized HIF1α deletion mutant leads to the robust induction of normal appearing blood vessels (Elson et al, 2001).

HA‐HIF1α and HA‐HIF2α were, as expected, detected specifically in extracts prepared from skin of K14‐Cre;HIF1dPA and K14‐Cre;HIF2dPA mice, respectively (Figure 3A). However, HIF1dPA levels were much lower than HIF2dPA, which is similar to the results obtained with MEFs (Figure 1D). By postnatal day 5 the skin of K14‐Cre;HIF2dPA and K14‐Cre;VHLf/f mice was noticeably red and could easily be distinguished from the skin of littermate controls (data not shown). By weaning age, the mice exhibited skin erythema, partial alopecia, and runting (Figure 3B and C). These phenotypes were only observed in animals that carried both the K14‐Cre transgene and either the HIF2dPA allele or homozygous VHL floxed alleles. We did not observe a gross skin phenotype in the K14‐Cre;HIF1dPA mice (data not shown). This might reflect the lower levels of HIF1α achieved in this system relative to HIF2α (Figure 3A).

Figure 3.

Gross phenotypes of K14‐Cre;HIF2dPA mice and K14‐Cre;VHLf/f mice. (A) Immunoblot analysis of skin biopsy homogenates prepared from mice with the indicated genotypes. (B) Photographs of representative K14‐Cre;HIF2dPA and K14‐Cre;VHLf/f mice alongside age and sex‐matched K14‐Cre control. (C) Growth curves of mean body weight of K14‐Cre;HIF2dPA, K14‐Cre;VHLf/f mice, and K14‐Cre mice.

HIF2α, like pVHL loss, induces the formation of nonleaky blood vessels and epidermal hyperproliferation

H&E staining of furred and unfurred skin from both K14‐Cre;HIF2dPA and K14‐Cre;VHLf/f mice showed epidermal hyperplasia and an increased number of dilated dermal blood vessels (Figure 4A). CD31 staining confirmed an increase in vascular density of the skin from K14‐Cre;HIF2dPA and K14‐Cre;VHLf/f mice (particularly notable at the dermal–epidermal junction) and the appearance of ectatic vessels deeper within the dermis (P<0.01 and <0.05, respectively) (Figure 4A and B). Both K14‐Cre;HIF2dPA and K14‐Cre;VHLf/f mice displayed an increase in Ki67 staining of keratinocytes relative to K14‐Cre mice, indicative of increased keratinocyte proliferation (P<0.05 and <0.01 respectively) (Figure 4C).

Figure 4.

Histological and functional changes in the skin and vasculature of K14‐Cre;HIF2dPA and K14‐Cre;VHLf/f mice. (A) Representative histological sections of skin from 6‐ to 8‐week‐old mice with the indicated genotypes after staining with hematoxylin and eosin (H&E) ( × 100) or with antibodies against CD31 ( × 100) or Ki67 ( × 400). CD31 and Ki67 sections were counterstained with hematoxylin. (B) Quantitation of microvessel density. Error bars=1 s.d. (*P<0.05, **P<0.01). (C) Quantitation of Ki67+ cells. Error bars=1 s.d. (*P<0.05, **P<0.01). (D) Evans blue vascular permeability assay comparing mean dye extravasation at baseline (white) and after topical application of mustard oil (black). Error bars=1 s.d. (*P<0.05).

To better understand the effects of HIF2α stabilization and pVHL loss on the function of the dermal vasculature, we performed Evans blue vascular permeability assays on K14‐Cre;HIF2dPA and K14‐Cre;VHLf/f mice. Baseline vascular leakage of both K14‐Cre;HIF2dPA and K14‐Cre;VHLf/f mice was similar to that of K14‐Cre mice following Evans blue dye injection (Figure 4D). After application of mustard oil (an inflammatory stimulus) both K14‐Cre;HIF2dPA and K14‐Cre;VHLf/f mice showed a decreased level of blood vessel leakage compared to K14‐Cre controls (P<0.05) (Figure 4D), consistent with an earlier study in which a nondegradable version of HIF1α was expressed under the control of the K14 promoter (Elson et al, 2001). Therefore, the skin vessels that form as a consequence of HIF2α activation or pVHL loss in keratinocytes are not conspicuously leaky.

Gross liver changes induced by HIF activation in hepatocytes

Having established that the HIF1dPA and HIF2dPA alleles could be successfully recombined with Cre in vivo, and that HIF2α dysregulation could phenocopy pVHL loss in the skin, we next turned our attention to the liver because earlier work showed that VHL inactivation in this organ, either stochastically (in VHL+/− mice) or conditionally (in VHLf/f mice) in mice, leads to hepatomegaly, steatosis, and blood vessel proliferation, including hemangioma‐like lesions (Haase et al, 2001; Ma et al, 2003).

To this end, we crossed the transgenic mice expressing Cre recombinase under the control of a hepatocyte‐specific promoter (Alb‐Cre) with the HIF1dPA mice, HIF2dPA mice, and VHLf/f mice. The Alb‐Cre transgene efficiently deletes loxP containing alleles in hepatocytes of mouse livers (Postic et al, 1999). The hepatic HA‐HIF1dPA levels in Alb‐Cre;HIF1dPA mice were much lower than the hepatic HA‐HIF2dPA levels in Alb‐Cre;HIF2dPA mice (Figure 5A), consistent with the differences we observed in MEFs (Figure 1D) and skin (Figure 3A). As expected, Alb‐Cre;VHLf/f mice were runted relative to littermate controls (Figure 5B) and died at approximately 6–8 weeks of age (data not shown). Interestingly, Alb‐Cre;HIF2dPA and Alb‐Cre;HIF1dPA/HIF2dPA mice also died at 6–8 weeks of age but maintained a normal body weight (Figure 5B and data not shown). In contrast, Alb‐Cre;HIF1dPA mice were indistinguishable from littermate controls and did not appear to have a shortened lifespan.

Figure 5.

Body weight and liver weight changes induced by pVHL loss or HIF2 activation. (A) Immunoblot analysis of liver homogenates prepared from mice with the indicated genotypes. (B) Growth curves of mean body weight of mice with the indicated genotypes. (C) Mean liver weights (as a percentage of body weight) of 6‐ to 8‐week‐old mice with the indicated genotypes. Error bars=1 s.d. (**P<0.01).

Alb‐Cre;VHLf/f, Alb‐Cre;HIF2dPA, and Alb‐Cre;HIF1dPA/HIF2dPA mice all had marked hepatomegaly with liver weights (as a percentage of body weight) approximately twice that of controls (each P<0.01) when killed at 6 weeks of age (Figure 5C). Alb‐Cre;VHLf/f and Alb‐Cre;HIF1dPA/HIF2dPA livers were indistinguishable from one another on gross inspection and clearly abnormal. They were friable and appeared stippled with irregular, yellow spots (suggestive of steatosis) on a reddish background (Figure 6). Alb‐Cre;HIF2dPA livers were also easy to visually discriminate from controls by their homogeneous deep maroon color. Alb‐Cre;HIF1dPA livers were similar in weight and appearance to control livers (Figures 5C and 6).

Figure 6.

Gross and histological hepatic changes induced by pVHL loss or HIF activation. Representative photographs of livers from 6‐ to 8‐week‐old mice with the indicated genotypes (left column) along with histological sections after staining with H&E at × 100 (middle column) and × 400 (right column). Arrowheads indicate macrovesicular steatosis, arrows indicate microvesicular steatosis.

Simultaneous activation of HIF1α and HIF2α in hepatocytes is sufficient to induce histological changes observed after pVHL inactivation in the liver

The livers of Alb‐Cre;VHLf/f and Alb‐Cre;HIF1dPA/HIF2dPA mice were virtually indistinguishable when stained with H&E and examined histologically. Both showed collections of irregular, dilated, blood filled sinusoids and cytoplasmic vacuolization in hepatocytes with eccentric nuclei (suggestive of steatois) (Figure 6). H&E stains of Alb‐Cre;HIF2dPA livers demonstrated similar vascular lesions but with minimal evidence of vacuolization (Figure 6). Alb‐Cre;HIF1dPA livers had a fine vacuolization of hepatocytes but were otherwise similar to controls. Staining with CD34 and quantitation of microvascular density confirmed the apparent increase in vascularity in Alb‐Cre;HIF2dPA, Alb‐Cre;HIF1dPA/HIF2dPA, and Alb‐Cre;VHLf/f livers relative to controls (P<0.01, <0.01, and <0.5, respectively) (Figure 7A and B).

Figure 7.

Hepatic changes in vascularity, fat accumulation, and proliferation induced by pVHL loss or HIF activation. (A) Histological sections of livers obtained from 6‐ to 8‐week‐old mice with the indicated genotypes after staining for antibodies against CD34 ( × 100) or Ki67 ( × 200) or with Oil Red O ( × 200). (B) Quantitation of microvessel density. Error bars=1 s.d. (*P<0.05; **P<0.01). (C) Quantitation of Ki67+ cells. Error bars=1 s.d. (**P<0.01; ***P<0.001).

Consistent with their macroscopic and H&E appearance, Alb‐Cre;VHLf/f and Alb‐Cre;HIF1dPA/HIF2dPA livers had a clear increase in lipid accumulation within hepatocytes in a mixed micro and macrovesicular steatotic pattern, as determined by Oil Red O staining, whereas Alb‐Cre;HIF1dPA livers showed moderate lipid accumulation in a microvesicular pattern (Figure 7A). Alb‐Cre;HIF2dPA livers had minimal lipid accumulation over controls.

Ki67 staining of Alb‐Cre;HIF2dPA, Alb‐Cre;HIF1dPA/HIF2dPA, and Alb‐Cre;VHLf/f and livers revealed a clear increase in Ki67 staining of hepatocytes indicative of increased proliferation relative to littermate controls (P<0.01, <0.01, and <0.001, respectively) (Figure 7A and C). A number of different cell types exhibited Ki67 positivity including hepatocytes and uncharacterized cells located within the hepatic sinusoids. The proliferation rate of Alb‐Cre;HIF1dPA hepatocytes was normal.

Hematological effects of HIF activation or pVHL loss in the liver

The liver is capable of producing EPO and Alb‐Cre;VHLf/f mice have been reported to develop polycythemia secondary to elevated levels of this hormone (Haase et al, 2001). In our experiments, Alb‐Cre;HIF2dPA, Alb‐Cre;HIF1dPA/HIF2dPA, and Alb‐Cre;VHLf/f mice all displayed noticeable erythema of the paws and unfurred skin by 4–6 weeks of age (data not shown). We postulated this was secondary to polycythemia. Indeed, hematocrit, reticulocyte, and EPO levels were all elevated in Alb‐Cre;HIF2dPA, Alb‐Cre;HIF1dPA/HIF2dPA, and Alb‐Cre;VHLf/f mice relative to controls and Alb‐Cre;HIF1dPA mice (Figure 8A–C). The increased EPO levels observed after pVHL inactivation relative to HIF2α activation (with or without HIF1α activation) was of borderline statistical significance because of the large variance within the VHLf/f group. Polycythemia might have contributed to the earlier deaths of these animals.

Figure 8.

Hematological changes induced by pVHL loss or HIF activation. Mean hematocrits (A), reticulocyte counts (B), and serum EPO levels (C) from 6‐ to 8‐week‐old mice with the indicated genotypes. Error bars=1 s.d. (**P<0.01; ***P<0.001).

Gene expression changes after HIF activation or pVHL loss in the liver

The livers of mice in which HIF1α and HIF2α were simultaneously activated appeared strikingly similar to those in which pVHL had been lost. We next employed oligonucleotide microarrays to compare gene expression in the livers producing HIF1dPA, HIF2dPA, both HIF1dPA and HIF2dPA, or lacking pVHL. We first identified the top 200 genes that are upregulated in the Alb‐Cre;VHLf/f mice compared to Alb‐Cre control mice, all of which were significant at P<0.01 by permutation testing. Using gene set enrichment analysis (GSEA) (Subramanian et al, 2005), we asked whether activation of HIF1α, HIF2α, or both could induce the signature of pVHL inactivation. All genes on the microarray were ranked by signal‐to‐noise ratio (Golub et al, 1999) as compared to the Alb‐Cre control. The enrichment score was derived from the position of the genes in the VHL signature in these ranked lists. Activation of HIF1α in hepatocytes led to an enrichment score of 0.71 (P=0.056), indicating that the similarity between the genes regulated by HIF1α and VHL in this system approaches but does not reach statistical significance (Figure 9A). In contrast, activation of HIF2α or simultaneous activation of both HIF1α and HIF2α resulted in highly statistically significant enrichment scores of 0.93 and 0.91, respectively (both P<0.001) (Figure 9A). Hence, the vast majority of genes upregulated upon pVHL loss are also induced by HIF2α (with or without HIF1α activation).

Figure 9.

Gene‐expression changes induced by pVHL loss or HIF activation. (A) Enrichment score as determined by GSEA comparing genes upregulated by pVHL loss in the liver to the expression profiles of livers from mice of the indicated genotype. Using the signal to noise ratio score, all genes on the array were ranked by their differential expression in samples from mice expressing constitutively active HIF1α (left), HIF2α (center), or both HIF1α and HIF2α (right) compared to samples from mice expressing the Alb‐Cre control. GSEA was performed using a VHL signature comprised of the top 200 genes upregulated by VHL inactivation. The graphs in red show the enrichment scores. The vertical blue bars represent the position of each of the 200 genes in the VHL gene set in the ranked lists of the comparison groups. The graph in gray at the bottom shows the signal‐to‐noise score for each gene in the ranked list. (B) Top 100 genes differentially regulated by HIF1α or HIF2α. Each column represents a different sample and each row represents a different gene. The highest expression for each gene across samples is portrayed as red and the lowest expression in blue. The list of differentially regulated genes is included in the Supplementary data.

The presence of microvesicular steatosis in the livers of mice expressing the activated form of HIF1α strongly suggested that some genes were specifically regulated by HIF1α and not HIF2α. This existence of such genes was confirmed in the microarray analysis (Figure 9B). For example, certain glycolytic genes such as enolase 1, pyruvate kinase L, and aldolase C were selectively activated by HIF1α, in keeping with earlier work in other models (Hu et al, 2003; Raval et al, 2005; Wang et al, 2005) while other known HIF target genes such as PAI‐1, annexin V, and EglN3 (Aprelikova et al, 2004; Schofield and Ratcliffe, 2004) were specifically upregulated by HIF2α (Figure 9B and Supplementary data). Therefore, in this model, HIF1α and HIF2α regulate an overlapping but distinct set of genes.


We found that production of an HIF2α variant (HIF2dPA) that escapes recognition by pVHL in keratinocytes of genetically engineered mice is sufficient to phenocopy the runting, epidermal hyperplasia, and hypervascularity seen when pVHL is lost in this cellular compartment. Interestingly, production of this HIF2α variant within hepatocytes is sufficient to recapitulate the vascular lesions seen when pVHL is inactivated in liver parenchyma, while production of the corresponding HIF1α variant (HIF1dPA) causes microvesicular steatosis. Simultaneous expression of both HIFα variants in hepatocytes causes pathological changes, including increased vascularity and macrovesicular steatosis, that are indistinguishable from those seen when pVHL is lost in these cells. Additionally, activation of HIF2α (with or without HIF1α) induces the gene expression signature of pVHL loss. Collectively these results suggest that the pathological changes seen after pVHL inactivation in skin and liver are primarily due to dysregulation of HIF target genes. Moreover, both the histological changes and the microarray analyses suggest that HIF1α and HIF2α are not wholly redundant with respect to these hepatic changes but instead cooperate with one another.

The experiments described were designed to compare the consequences of pVHL loss to HIFα activation and some of the differences observed between HIF1α and HIF2α should be interpreted cautiously. HIF1dPA and HIF2dPA were both under the control of a foreign promoter (ROSA26) and both lacked their native 5′ and 3′UTRs. It is also possible that the functional (e.g., DNA‐binding and transactivation) assays performed so far are not sensitive enough to detect biologically relevant changes, beyond changes in protein stability, induced by the proline to alanine substitutions. The ability of HIF1dPA to induce certain genes and phenotypes, however, despite its lower levels relative to HIF2dPA, argues strongly that such genes and phenotypes are indeed preferentially regulated by HIF1α. On the other hand, failure of HIF1dPA to induce certain genes and phenotypes could be due to the technical considerations outlined above.

For example, we did not observe a skin phenotype in mice expressing HIF1dPA in keratinocytes under control of the ROSA26 promoter. This is in contrast to Arbeit and coworkers who observed the induction of normal appearing blood vessels in transgenic mice expressing this same mutant (Jeff Arbeit, personal communication), or an HIF1α variant lacking the oxygen‐dependent degradation domain (Elson et al, 2001), under the control of the K14 promoter. This discrepancy might be due to technical differences, such as differences in the strengths of the ROSA26 and K14 promoters. We did, however, observe changes similar to those reported by Elson and co‐workers when HIF2dPA was expressed in keratinocytes, probably due to the higher levels of HIF2dPA we achieved in our mice relative to HIF1dPA. This would argue that either HIF1α or HIF2α is capable of inducing skin blood vessel formation. Expression of HIF2dPA or inactivation of pVHL in the mouse epidermis also causes systemic weight loss. A similar phenotype after pVHL inactivation in the skin has been observed by Randy Johnson (R Johnson, personal communication), who documented increased heat loss and basal metabolic rate as possible causes of the runting.

Earlier work showed that VHL inactivation in the liver, either stochastically (in VHL+/− mice) or conditionally (in VHLf/f mice) in mice leads to hepatomegaly, steatosis, and blood vessel proliferation, including hemangioma‐like lesions (Haase et al, 2001; Ma et al, 2003). A recent study showed that these lesions can be prevented by simultaneous inactivation of ARNT, but not by simultaneous inactivation of HIF1α (Rankin et al, 2005). This suggested that dysregulation of one or more ARNT partners is necessary, possibly in collaboration with other pVHL targets, for the development of these lesions.

Remarkably, we found that stabilization of HIF2α is sufficient to recapitulate the vascular liver lesions and hepatomegaly seen after pVHL loss and that dual stabilization of HIF1α and HIF2α is sufficient to induce to the other hepatic phenotypes linked to pVHL inactivation including macrovesicular steatosis. The apparent cooperation between HIF1α and HIF2α in the liver is consistent with earlier studies that concluded these two proteins are not wholly redundant. For example, it was shown earlier in other systems that the genes regulated by these two proteins, while overlapping, are non‐identical (Hu et al, 2003; Raval et al, 2005; Wang et al, 2005), a result in keeping with our own microarray data. HIF2α, but not HIF1α, appears to act as an oncogene in pVHL‐defective renal carcinomas, which also argues that the functions of HIF1α and HIF2α are not identical (Kondo et al, 2002, 2003; Maranchie et al, 2002; Zimmer et al, 2004; Raval et al, 2005).

Therefore, none of the many HIF‐independent functions that have been ascribed to pVHL appear to play major roles in the skin and liver phenotypes observed so far after pVHL inactivation in keratinocytes and hepatocytes, respectively, with the caveat that the runting observed after hepatic pVHL inactivation was not recapitulated in the HIFdPA mice. This could reflect strain differences (although all phenotypes described were compared to littermate controls) or could reflect an HIF1/2‐independent pVHL function. It also remains possible that HIF‐independent pVHL functions are important in cell types and conditions that were not examined here. Nonetheless, our findings further validate HIF as a therapeutic target in pVHL‐defective neoplasms, especially with respect to the angiogenesis that is characteristic of such lesions. In this regard, the mice described here might be useful for testing agents that inhibit HIF‐dependent angiogenesis.

Neither pVHL inactivation nor HIFα stabilization has, to date, caused the emergence of solid tumors in mice. In particular, pVHL inactivation (or HIFα activation) in the mouse has not caused the development of clear cell renal carcinomas and pheochromocytomas observed in human VHL disease and the vascular lesions seen in the liver only loosely resemble hemangioblastomas. This might be because the correct cell has not yet been targeted in the mouse or because mice, in contrast to humans, are not permissive for the development of these tumors following pVHL inactivation (or HIFα activation). There are many examples of human cancer genes that, when mutated in the mouse, cause a different spectrum of tumors than observed in humans. This presumably reflects a variety of biological differences between mice and men that impact upon tissue tropism. Nonetheless, the HIFdPA mice should be useful for asking whether HIF is sufficient to induce tumors in certain contexts and whether HIF, as suspected, can act as a progression factor in established tumors.

Materials and methods


pRosa26‐PA and pBig‐T vectors were gifts of Phillipe Soriano and Frank Constantini (Srinivas et al, 2001). The pBig‐T plasmid contains two LoxP sites that flank a neomycin resistance gene, which is followed by a transcriptional Stop element. A multiple cloning site is present after the second LoxP site. The pRosa26‐PA plasmid contains ROSA26 genomic sequences into which a multiple cloning site containing AscI and PacI restriction sites has been inserted and a diphtheria toxin cassette 3′ of the ROSA26 DNA for positive selection against nonhomologous recombinants.

pcDNA3.0‐HA‐HIF1α P402A;P564A and pcDNA3.0‐HA‐HIF2α P405A;P531A (Kondo et al, 2003) were cut with HindIII, blunt ended with T4 DNA polymerase (New England Biolabs), and recircularized to convert the HindIII site into an NheI restriction site. The HIFα cDNAs were then excised with NheI and NotI and ligated into pBigT that had been cut with these two enzymes to make pBigT‐HA‐HIF1α P402A;P564A and pBigT‐HA‐HIF2α P405A;P531A. The PacI–AscI fragments from these two plasmids were ligated into pRosa26‐PA cut with the same enzymes to make the pRosa26‐PA‐HA‐HIF1α P402A;P564A and pRosa26‐PA‐HA‐HIF2α P405A;P531A targeting constructs.

Homologous recombination

The pRosa26‐HA‐HIF1α P402A;P564A and pRosa26‐PA‐HIF2α P405A;P531A targeting constructs were linearized with KpnI and electroporated into TC1 ES cells derived from the 129 SVEV strain. G418 resistant HA‐HIF1α P402A;P564A and HA‐HIF2α P405A;P531A ES cell clones found to have undergone successful homologous recombination as detected by Southern blot analysis with a 5′ external ROSA26 probe were microinjected into C57/BL6 blastocysts and implanted into pseudopregnant mothers to generate LSL‐HIF1dPA and LSL‐HIF2dPA chimeric mice that were backcrossed to Balb/C mice for three generations.

Mouse colony generation and maintenance

The albumin‐Cre (Alb‐Cre) C57/B6 mouse strain was purchased from Jackson Labs. The keratin14‐Cre (K14Cre) FVB mouse strain was a generous gift from Pierre Chambon (IGBMC, University of Stasbourg) and the VHLf/f Balb/C mouse a generous gift from Volker Haase (University of Pennsylvania). Genomic DNA was isolated from mouse tails using a Puregene DNA Isolation Kit (Gentra Systems). Mice were genotyped by PCR (primers and conditions available upon request). All procedures involving mice were performed in accordance with the NIH guidelines for the use and care of live animals and were approved by the Dana‐Farber Cancer Center IACUC.

Mouse embryo fibroblasts

MEFs were prepared using standard techniques and cultured in DMEM (Gibco) with 10% fetal bovine serum (Hyclone). Early passage MEFs was infected with adenoviral Cre (University of Iowa) at a MOI of 10 in the presence of 4 μg/ml of polybrene. MEFs were treated with MG132 (CalBiochem) for 6 h at a concentration of 30 μM.

Western blot analysis

Whole tissue or cell pellets were homogenized in 20 mM Tris (pH 7.5), 250 mM NaCl, 1 mM EDTA, 1 mM EGTA, and 1% Triton X‐100 in the presence of a protease inhibitor cocktail (Roche). Whole‐cell lysate (50 μg) or 100 μg of tissue homgenate, as determined by the Bradford method, was resolved by SDS–PAGE and transferred onto PVDF membranes. After blocking with Tris‐buffered saline (TBS) with 4% non‐fat milk the membranes were probed with anti‐HA (HA‐11) 1:1000 (Covance), anti‐tubulin (DM1A) 1:2000 (Sigma), anti‐HIF2α (BC100‐480) 1:1000 (Novus), anti‐HIF2α (NB100‐122) 1:1000 (Novus), anti‐HIF1α 1:5000 (Bethyl), and anti‐HIF1α (NB100‐449) 1:1000 (Novus), antibodies diluted in TBS with 4% bovine serum albumin. Bound antibody was detected with horse radish peroxidase‐conjugated secondary antibody (Pierce) and SuperSignal West Pico chemiluminescent substrate (Pierce).

Histological studies

Tumor samples were fixed in 10% formalin overnight at room temperature, paraffin embedded, and stained with hematoxylin and eosin by standard methods. For immunohistochemisty, 5 μm sections were deparaffinized in xylene and rehydrated sequentially in ethanol. Antigen retrieval was performed by boiling in 10 mM sodium citrate (pH 6.0) for 30 min. Sections were stained with antibodies to CD31 (Pharmingen, 1:50), CD34 (Abcam, 1:500), and Ki67 (Vector Labs, 1:50). Oil red O (Sigma) staining was carried out on 10 μm frozen sections. The number of Ki67 positive cells was expressed as a percentage of positive staining cells over the total number of keratinocytes or hepatocytes per high powered field. Microvessel density was expressed as the number of microvessels per high powered field.

Evans blue vascular permeability assay

Mice (8–12 weeks old) were anesthetized (ketamine/xylazine) and administered 30 mg/kg Evans blue dye (Sigma) intravenously. Heavy mineral oil or mustard oil (Sigma; diluted to 5% in heavy mineral oil) was applied 10 and 20 min later. At 10 min after the last oil application, the mice were killed and the ears were dissected and weighed. Evans blue dye was extracted in 250 μl of 100% formamide overnight at 72°C. Extravasated dye was quantitated using a spectrophotometer (absorbance 610 nm) and corrected for background absorbance.

Hematologic studies

Whole blood was obtained by cardiac puncture of anesthetized mice just prior to euthanasia. Serum was prepared from whole blood using serum separator tubes and frozen at −80°C until use. Hematocrit levels and reticulocyte counts were measured using an Advia 120 hematology system on EDTA anticoagulated whole blood. Serum EPO was measured using an ELISA kit for mouse EPO (R&D Systems).

Statistical analysis

Statistical analyses were carried out with the variables in either their natural units or after log transformation. The data were analyzed using ANOVA followed by pairwise t‐tests for comparisons of interest.

Gene expression analysis

Total RNA was isolated from mouse tissue using Trizol Reagent (Gibco) and then purified by the RNAeasy kit (Qiagen). Biotin labeled cRNA prepared from 15 μg of total RNA was fragmented and hybridized to mouse oligonucleotide arrays (Affymetrix, 430A 2.0 array). Preprocessing and normalization of the data were performed using Robust Multichip Averaging (Bolstad et al, 2003). Gene expression values less than a minimum threshold of 20 or a maximum threshold of 16 000 were set to 20 and 16 000, respectively. Genes with minimal variation across the data set were discarded (maximum/minimum <3 or maximum – minimum <100). Class neighbors were selected using GenePattern software ( using the signal to noise metric (Golub et al, 1999). GSEA was performed as described (Subramanian et al, 2005).

Quantitative real‐time PCR

Total RNA was isolated from MEFs using a QIAGEN RNeasy RNA isolation kit. Conversion into cDNA was performed using oligo‐dT(12–18) (Invitrogen) and ImProm‐II reverse transcriptase (Promega). PCR reactions using sequence specific primers and probes obtained from Applied Biosystems were performed in the ABI PRISM 7700 Sequence Detection System (Applied Biosystems). The PCR reactions were performed in duplicate for each cDNA, averaged, and normalized to endogenous ribosomal 18S reference transcripts.

Supplementary data

Supplementary data are available at The EMBO Journal Online (

Supplementary Information

Supplementary Figure 1 [emboj7601300-sup-0001.pdf]

Supplementary data [emboj7601300-sup-0002.pdf]


We thank Volker Haase and Pierre Chambon for mouse strains, Nabeel Bardeesy, Ron DePinho, Norman Sharpless, and James Horner for helpful advice, and members of the Kaelin Laboratory for useful discussions. This work was supported by grants from NIH to WYK and WGK and by a gift from the Murray Foundation. WGK is an HHMI Investigator.


View Abstract