The peroxisome proliferator‐activated receptor‐β (PPARβ) has been implicated in tumorigenesis, but its precise role remains unclear. Here, we show that the growth of syngeneic Pparb wild‐type tumors is impaired in Pparb−/− mice, concomitant with a diminished blood flow and an abundance of hyperplastic microvascular structures. Matrigel plugs containing pro‐angiogenic growth factors harbor increased numbers of morphologically immature, proliferating endothelial cells in Pparb−/− mice, and retroviral transduction of Pparb triggers microvessel maturation. We have identified the Cdkn1c gene encoding the cell cycle inhibitor p57Kip2 as a PPARβ target gene and a mediator of the PPARβ‐mediated inhibition of cell proliferation, which provides a possible mechanistic explanation for the observed tumor endothelial hyperplasia and deregulation of tumor angiogenesis in Pparb−/− mice. Our data point to an unexpected essential role for PPARβ in constraining tumor endothelial cell proliferation to allow for the formation of functional tumor microvessels.
Peroxisome proliferator‐activated receptors (PPARs) are nuclear receptors that function as transcription factors and modulate target gene expression in response to endogenous and exogenous ligands (Barish et al, 2006; Desvergne et al, 2006; Feige et al, 2006). The PPAR family consists of three members, PPARα, PPARβ/δ and PPARγ, whose major physiological functions are associated with the regulation of intermediary metabolism. All PPARs are activated by fatty acids and some derivatives, but there are no known subtype‐specific, high‐affinity endogenous ligands, suggesting that PPARs act as intracellular lipid sensors (Forman et al, 1997; Desvergne et al, 2006). However, PPARs are highly relevant drug targets, which has led to the development of several synthetic drug agonists showing increased subtype selectivity and high‐affinity binding (Peraza et al, 2006). PPARs form obligatory heterodimers with the nuclear receptor RXRα on PPAR response elements (PPREs) in their target genes resulting in transcriptional activation. PPARs can also repress genes by directly interacting with specific transcription factors and recruiting corepressor proteins (Lee et al, 2003).
It has recently been shown that PPARβ has a critical role in modulating skeletal muscle lipid catabolism, glucose homeostasis and macrophage activation (Barish et al, 2006; Desvergne et al, 2006). Thus, PPARβ represents a new potential drug target for the treatment of major human diseases such as obesity, metabolic syndrome, inflammation and arteriosclerosis. Intriguingly, PPARβ also promotes cellular differentiation and inhibits proliferation and apoptosis of different cell types (Tan et al, 2001; Schmuth et al, 2004; Kim et al, 2006; Nadra et al, 2006; Varnat et al, 2006). Pparb‐null mice exhibit a defect in wound healing, and increased keratinocyte proliferation occurs in the skin of Pparb‐null mice treated with a tumor promoter (Peters et al, 2000; Michalik et al, 2001). Furthermore, ligand activation of PPARβ selectively stimulates keratinocyte differentiation and inhibits proliferation (Kim et al, 2006), concomitant with a ubiquitin‐mediated downregulation of protein kinase C‐α and MAP kinase signaling (Kim et al, 2005).
Consistent with its function in differentiation and proliferation, PPARβ plays a role in intestinal tumorigenesis. In the human colon cancer cell line HCT116, Pparb disruption has been reported to result in a loss of tumorigenicity (Park et al, 2001). However, in intestinal tumor mouse models, PPARβ affected adenoma growth to a variable extent in different studies (Barak et al, 2002; Gupta et al, 2004; Harman et al, 2004; Reed et al, 2004; Marin et al, 2006). Both, the homozygous deletion of Pparb (Harman et al, 2004; Reed et al, 2004) and the pharmacological activation of PPARβ (Gupta et al, 2004; Marin et al, 2006), have been shown to exert subtle, although statistically significant effects on adenoma growth, but the precise function of PPARβ in intestinal tumor cells remains controversial (Burdick et al, 2006).
None of the mouse studies performed to date addressed the issue as to whether PPARβ might play a role in the cells of the tumor stoma, that is, host cells recruited by the tumor, such as endothelial cells (ECs), fibroblasts and macrophages (Bissell and Radisky, 2001). In the present study, we have specifically addressed this question.
Inhibition of tumor growth in Pparb−/− mice
We first asked whether the genetic status of Pparb in host cells would affect the growth of transplanted syngeneic tumor cells. Pparb−/− mice have been described previously and shown to lack PPARβ protein in all tissues analyzed (Peters et al, 2000; Harman et al, 2004). This was confirmed in the present study for fibroblasts and ECs as typical stromal cell types (Supplementary Figure S1). Lewis lung carcinoma (LLC1) cells were inoculated subcutaneously into Pparb+/+ and Pparb−/− mice. Rapid and progressive tumor growth was observed in all Pparb+/+ mice (n=15), invariably leading to death within 24 days. Tumor growth in Pparb−/− mice (n=11) was initially indistinguishable from Pparb+/+ mice, but halted after approximately 14 days, resulting in 100% survival 32 days after tumor cell implantation, and 90.9% survival after >6 months (Figure 1A; Table I). We also observed a dramatic inhibition of tumor growth with the B16F1 melanoma model (Figure 1B). While the inoculated Pparb+/+ mice invariably succumbed to the B16F1 tumors within 14 days, all Pparb−/− mice showed a clear retardation of tumor growth (Figure 1B; n=8), and exhibited a survival rate of 100% on day 24.
Abnormal microvascular structures in Pparb−/− mice
Blood vessels in LLC1 tumors were visualized by immunohistochemistry using an antibody directed against aquaporin‐1 (AQP‐1), a specific marker for ECs and erythrocytes (Saadoun et al, 2005). Microvessels with a thin lining of ECs and an open lumen with erythrocytes were readily detected in LLC1 tumors from Pparb+/+ mice (Figure 2A). In contrast, microvessels with a normal appearance were rare in LLC1 tumors from Pparb−/− mice, where ECs were found mainly in highly abnormal structures lacking a lumen (Figure 2A). These presumably defective microvessels were frequently located in areas of extensive tumor necrosis, as exemplified in Figure 2A by the cellular debris seen around the abnormal vascular structures. Importantly, immunostaining with antibodies for macrophages, neutrophils or lymphocytes (Supplementary data) revealed only very few cells in tumors of either genotype, clearly suggesting that inflammatory and immune cells did not play an essential role in supporting or restraining LLC1 tumor growth (data not shown).
Surprisingly, a quantitative evaluation revealed a three‐fold higher microvascular density in LLC1 tumors from Pparb−/− mice (Figure 2B, left panel; n=8; P=0.010). However, while 76% of these structures represented morphologically normal microvessels in Pparb+/+ mice, this was reduced to 11.8% in Pparb−/− mice (Figure 2B, right panel; n=8; P=0.010). These observations suggest that an abnormal organization rather than a lack of ECs underlies the abundance of abnormal microvessels in Pparb−/− mice.
Diminished blood flow in LLC1 tumors in Pparb−/− mice
Functional evidence for a tumor vascularization defect in Pparb−/− mice was obtained by dynamic contrast‐enhanced magnetic resonance imaging (DCE‐MRI) (Choyke et al, 2003; Kiessling et al, 2003). Contrast medium (diethylenetriamine pentaacetic acid; Gd‐DTPA) was injected into three PPARb+/+ and three PPARb−/− mice and tumor accumulation was monitored over 9 min. All tumors had a similar volume at the beginning of the analysis (5.3±1.3 mm3; day 0 in Figure 3B). Measurements were repeated after 2 and 6 days with the identical animals. Tumors in Pparb+/+ mice showed continuous progression, reaching sizes of >1000 mm3, whereas tumors in Pparb−/− mice grew much slower followed by a partial regression. Kinetic analysis of Gd‐DTPA accumulation in tumors revealed a rapid and high amplitude wash‐in followed by a clearly discernible wash‐out phase in Pparb+/+ mice at all time points analyzed (Figure 3B), indicative of functional vascularization (Choyke et al, 2003; Kiessling et al, 2003). In contrast, Pparb−/− mice showed a progressive delay of wash‐in and no wash‐out phase, indicating an obstructed blood flow into and through the tumor tissue (Figure 3B). Analysis of dynamic Gd‐DTPA within individual tumors showed a relatively even distribution in tumors in Pparb+/+ mice, whereas large regions of decreased accumulation were seen in Pparb−/− mice, especially in the regressing tumors (green/blue areas in Figure 3A) next to regions showing rapid accumulation (red areas in Figure 3A). These observations are consistent with the histological analyses showing that a large fraction of tumor microvessels is abnormal in Pparb−/− mice.
Endothelial hyperplasia in LLC1 tumors in Pparb−/− mice
To examine the cellular basis underlying the observed vascular phenotype we performed double immunostaining for AQP‐1 and proliferating cell nuclear antigen (PCNA). A considerably higher number of PCNA‐positive cells in tumors was found in Pparb−/− mice as compared with Pparb+/+ mice (38.7 versus 16.6%; Figure 4A; n=8; P=0.010). Consistent with these in vivo observations, we found that the proliferation of primary aortic ECs from Pparb−/− mice was enhanced compared to Pparb+/+ cells cultured on a tumor basement membrane‐derived matrix (matrigel), in the presence of pro‐angiogenic growth factors (Figure 4B).
Microvessels in Pparb−/− mice were typically lined by morphologically less mature ECs frequently protruding into the vessel lumen reminiscent of endothelial hyperplasia. These vessels were surrounded by rings of cells strongly expressing the myofibroblast/pericyte marker smooth muscle α‐actin (SMA) (Bissell and Radisky, 2001) (Figure 5A and B). Hyperplastic tumor endothelial structures were also found in the absence of adjacent perivascular SMA expressing cells, indicating that the occurrence of these SMA‐positive cells is not a prerequisite for vascular dysfunction.
Immature, hyperproliferative ECs in matrigel plugs in Pparb−/− mice
To study the role of PPARβ in an independent in vivo angiogenesis assay, we performed matrigel plug assays (Ley et al, 2004) using prostaglandin E2 (PGE2) and basic fibroblast growth factor (FGF‐2) as angiogenic stimuli. These subcutaneous plugs became rapidly invaded by AQP‐1‐positive cells (Figure 6A and B). In Pparb+/+ mice, the invading cells aligned within 3 days to form network‐like structures consisting of elongated cells connected by dendritic processes (Figure 6A, left panel). These ordered structures were largely absent from matrigel plugs in Pparb−/− mice (Figure 6A, right panel). While morphologically mature cells (flat, elongated cells with processes) represented 39.6±10.0% of cells in microvascular structures in Pparb+/+ mice, the corresponding value was 13.9±2.3% for Pparb−/− mice (Figure 6C, left panel). Matrigel plugs in Pparb+/+ mice also contained more advanced structures lined by elongated ECs (Figure 6B, left panel), whereas morphologically aberrant cells protruding into the lumen were seen in Pparb−/− mice (Figure 6B, right panel). Consistent with the hyperproliferation of tumor ECs shown above, we found a significantly higher proportion of PCNA‐positive cells in matrigel plugs from Pparb−/− mice (37.2±0.7%), compared with Pparb+/+ mice (14.8±0.8%; Figure 6C, right panel).
Consistent with its essential role in tumor vascularization, we also found that Pparb is the predominant Ppar subtype expressed in angiogenic matrigel plugs, in mouse ECs isolated from subcutaneous tumors (Supplementary Figure S2), and in tumor ECs from human lung carcinomas (Supplementary Figure S3), and that it is inducible by angiogenic growth factors in ECs in vitro (Supplementary Figure S4).
Restoration of PPARβ expression triggers microvessel maturation
To determine whether re‐expression of PPARβ in matrigel‐invading cells would restore normal vascularization in Pparb−/− mice, we established a syngeneic fibroblast producer line (3Fb‐p cells) expressing FLAG‐PPARβ and helper retroviruses (Supplementary Figure S1). Matrigel plugs containing these cells were used with the same experimental design described above. Immunohistological analysis of matrigel plugs 3 days post‐implantation showed FLAG‐positive implanted fibroblasts (occurring in small colonies) and FLAG‐positive host ECs, whereas the inclusion of transduced control cells either lacking the FLAG‐PPARβ retrovirus (Lpcx‐p cells) or the helper virus (3Fb‐np non‐producer cells) did not yield any FLAG‐specific host cell staining (Figure 7A). Concomitant with this viral spread and re‐expression of PPARβ in the invading host cells, we observed the appearance of vessels containing erythrocytes lined by morphologically normal ECs, whereas in the presence of either control cell line, microvascular structures remained abnormal and were devoid of red blood cells without exception (Figure 7B and C). This observation provides additional strong evidence for a role of PPARβ in promoting vascularization in a mouse model of tumor angiogenesis.
Altered gene expression in matrigel‐invading cells in Pparb−/− mice
Cells from matrigel plugs were analyzed by qPCR for the expression of EC, fibroblast and macrophage marker genes (Aqp‐1, Cd105, vimentin, Cd14, Cd68, F4/80). No significant differences were detectable between Pparb+/+ and Pparb−/− samples (Supplementary Figure S5, and data not shown), indicating a very similar composition of cell types. To identify angiogenesis‐associated PPARβ target genes, we first analyzed a number of known genes (including all VEGFs, Flk‐1, Angiopoietin‐1/2, Tie‐1/2, VE‐cadherin), but could not detect any significant differences among Pparb+/+ and Pparb−/− samples (data not shown). We therefore compared the gene expression profile of both samples by microarray analysis (ArrayExpress, E‐MEXP‐983). In Pparb+/+ cells, 39 genes were found to be expressed at higher levels, while 27 genes showed lower expression (cut‐off 1.7‐fold; Supplementary Table S1). As expected, several of these genes encode proteins with functions in lipid metabolism that have previously been identified as PPAR target genes, that is, fatty acid binding protein 4 (Fabp4 gene), acyl‐CoA dehydrogenase (long‐chain; Acadl gene), stearoyl‐CoA desaturase 1 (Scd1 gene) and the CD36 fatty acid translocase/thrombospondin receptor (Cd36 gene) (see references in Supplementary Table S1). Differential expression in Pparb+/+ and Pparb−/− cells was verified by qPCR for a subset of genes, with a focus on those coding for regulatory proteins (Supplementary Figure S5). Of these genes, Cdkn1c is of particular interest due to the established direct role of its encoded protein p57KIP2 as a negative regulator of the cell cycle (Lee et al, 1995). We also analyzed the expression of genes coding for other CDK inhibitors (Cdkn1a, Cdkn1b, Cdkn2a, Cdkn2b, Cdkn2c), but were unable to detect significant differences between Pparb+/+ and Pparb−/− mice (Figure 8A, and data not shown). Importantly, we found readily detectable levels of Cdkn1c expression in ECs isolated from four independent samples of primary human lung carcinomas (Supplementary Figure S3), which is consistent with a potential role in tumor vascularization. We therefore focused on Cdkn1c in all subsequent experiments.
Confirmation of Cdkn1c as a PPARβ target gene
To obtain independent evidence that the Cdkn1c gene is regulated by PPARβ, we used fibroblasts established from a mouse strain with a floxed Pparb allele (Barak et al, 2002). Figure 9 shows that the expression of Cdkn1c was reduced ∼2.5‐fold after infection with a Cre expressing retrovirus (Li et al, 1997) relative to control virus infected cells. As a positive control, we included the known PPAR target gene Pdk4 (coding for pyruvate dehydrogenase kinase‐4) (Kitz Kramer et al, 2007), which also showed the expected reduction (∼8‐fold) in the PPARb‐deleted cells.
We next investigated whether the change in Cdkn1c expression in PPARβ‐deficient cells could be ‘rescued’ by retrovirus‐mediated re‐expression of PPARβ. As shown in Figure 8C, Pparb−/− fibroblasts infected with the PPARβ retrovirus expressed the Cdkn1c gene at ∼8‐fold elevated levels. As expected, the bona fide PPAR target genes Pdk4 and Fapb4 also showed a clear upregulation.
We also studied the inducibility of Cdkn1c by the synthetic PPARβ‐selective agonist, GW501516 (1 μM), in human umbilical vein ECs (HuVECs). In this setting, the Pdk4 gene was rapidly induced within 4 h by a factor of ∼4, as expected (Figure 8D). Likewise, Cdkn1c expression increased more than two‐fold at 4 h post‐treatment, decreasing to near basal levels at 24 h, confirming that Cdkn1c is a PPARβ target gene.
Inspection of the mouse Cdkn1c upstream sequence revealed two motifs at positions –1670 and –710 fitting the PPRE consensus sequence (Supplementary Figure S6, top panel). We cloned a 1.7‐kb Cdkn1c promoter fragment harboring these elements, in front of a luciferase reporter gene, and analyzed this construct in mouse 2H11 ECs. We observed a 1.8‐fold induction after cotransfection of PPARβ and RxRα expression plasmids, which was further increased by the PPARβ agonist GW501516 to 2.6‐fold (Supplementary Figure S6). These data clearly confirm that Cdkn1c induction by PPARβ is mediated by a transcriptional mechanism.
PPARβ‐mediated inhibition of cell proliferation involves p57KIP2
We next sought to investigate the growth inhibitory effect of PPARβ on a stroma cell type in further detail using a cell culture model. As shown in Figure 9A, infection of Pparb−/− cells with FLAG‐PPARβ retrovirus resulted in a strong inhibition of cell proliferation compared with control virus infected cells (3F1.13 versus Lpcx cells). This growth inhibitory effect of PPARβ imposes a selection pressure against PPARβ expressing cells. Thus, the continuous passaging of 3F1.13 cells resulted in a clear decrease in PPARb expression (subline 3F1.24; Figure 9B), which correlated with an increase in proliferation (Figure 9A). However, 3F1.24 cells still proliferate more slowly than the Pparb−/− Lpcx cells, suggesting that a relatively low level of PPARβ suffices to attenuate cell proliferation (PPARb expression in 3F1.24 cells is similar to that in wild‐type fibroblasts; data not shown). Importantly, both the level of PPARb and the extent of growth inhibition correlated with the level of Cdkn1c in the cell lines analyzed (Figure 9B).
To obtain direct evidence for a role of Cdkn1c in the inhibition of proliferation by PPARβ, we studied the effect of a Cdkn1c knockdown on the selection against PPARβ expressing cells in culture. Pparb−/− cells infected with a FLAG‐PPARβ retrovirus were repeatedly passaged, and at each passage were treated with either a Cdkn1c‐directed siRNA or an irrelevant control siRNA for a total period of 192 h. As shown in Figure 9C, Cdkn1c siRNA treatment reduced the level of Cdkn1c by 74% compared with the control siRNA at 48 h. Cdkn1c is a cell density‐regulated gene (Samuelsson et al, 1999), which presumably caused the increased levels at 192 h in both Cdkn1c‐ and control siRNA‐treated cells. Nevertheless, Cdkn1c levels were still 52% lower in the presence of Cdkn1c siRNA.
In control siRNA‐treated cells, PPARb expression showed the expected rapid decline (76% after 48 h; 36% after 192 h), reflecting the selection pressure against PPARβ expressing cells. In contrast, Cdkn1c siRNA treatment resulted in significantly increased PPARb expression levels at both time points (92 and 65%, respectively). The clear loss of Pparb expression in the presence of Cdkn1c siRNA strongly suggests that Cdkn1c is a mediator of PPARβ‐dependent inhibition of cell proliferation.
Impaired tumor vascularization and tumor EC hyperplasia in Pparb−/− mice
A hallmark of tumors in Pparb−/− mice is the abundance of histologically highly abnormal microvessels showing a thickened endothelial lining and lacking a lumen, thus appearing dysfunctional (Figure 2). These alterations were associated with a striking increase in both tumor microvessel density and tumor EC proliferation (Figure 4A). Abnormal hyperproliferative endothelial structures were also seen in matrigel plugs (Figure 6), but never in any normal tissue of Pparb−/− mice. Thus, Pparb−/− ECs hyperproliferate under conditions resembling a tumor microenvironment. Concomitant with this hyperproliferation, tumor microvessels in Pparb−/− mice typically consist of immature ECs surrounded by perivascular cells expressing vast amounts of SMA (Figure 5), a condition characteristic of endothelial hyperplasia. That these histological abnormalities are functionally relevant, was demonstrated by kinetic DCE‐MRI analysis of LLC1 tumors (Figure 3), which showed an obstructed tumor blood flow in Pparb−/− mice.
These observations strongly suggest that an abnormal organization caused by a hyperplastic response rather than a lack of ECs underlies the abundance of abnormal microvessels in Pparb−/− mice. However, the presence of a small fraction (11.8%) of morphologically normal tumor microvessels in Pparb−/− mice (Figure 2) obviously suffices to permit the formation of functional blood vessels, as indicated by the DCE‐MRI analysis (Figure 3), which showed small but clearly detectable areas of blood flow even in regressing tumors in Pparb−/− mice. This small fraction of apparently functional tumor microvessels in Pparb−/− mice obviously suffices to permit tumor growth to volumes of <100 mm3 (Figure 1), which is clearly beyond the angiogenic threshold (Folkman, 2002), that is, the maximum size achievable by avascular tumors (approximately 4 mm3). Together, these data indicate that tumor neoangiogenesis is principally functional in Pparb−/− mice, but that a significant fraction of tumor microvessels is abnormal and dysfunctional, leading to an inhibition of tumor growth.
Even though a defect in angiogenesis has not been observed during normal development of Pparb−/− mice (Peters et al, 2000; Michalik et al, 2001; Barak et al, 2002; Nadra et al, 2006), our findings are in agreement with previous findings showing that PPARβ regulates terminal differentiation and has a negative regulatory role in the proliferation of different cell types, including keratinocytes (Tan et al, 2001; Schmuth et al, 2004; Kim et al, 2006; Burdick et al, 2007), trophoblast giant cells (Nadra et al, 2006) and intestinal epithelial cells (Marin et al, 2006; Varnat et al, 2006). This suggests that PPARβ is specifically required by tumor ECs to orchestrate their proliferation and differentiation in an environment, providing an abnormally rich source of growth factors and cytokines. This interpretation is consistent with our observations that the fraction of PCNA‐positive matrigel‐invading cells was 2.5‐fold higher in Pparb−/− mice (Figure 6C), and that primary aortic ECs from Pparb−/− mice showed enhanced proliferation in vitro (Figure 4B). In both situations, ECs were exposed to high levels of soluble pro‐angiogenic growth factors and a tumor basement membrane‐derived matrix (matrigel).
A subtype‐specific role for PPARβ in tumor ECs?
The three known PPAR subtypes have partially overlapping functions. Supplementary Figure S2 shows that in contrast to liver, Pparb is the predominant subtype in matrigel‐invading cells. Likewise, in mouse ECs isolated from subcutaneous tumors (Supplementary Figure S2) and in ECs from primary human lung tumors (NSCLC) Pparb was the predominant subtype (Supplementary Figure S3). These observations point to a PPARβ‐subtype specific function in tumor angiogenesis consistent with the phenotype observed in Pparb−/− mice. This is further supported by the induction of Pparb by angiogenic growth factors in cultured ECs (Supplementary Figure S4).
Altered Cdkn1c gene expression in Pparb−/− stroma cells
To address the molecular mechanisms underlying the observed phenotypic effects, we sought to identify genes that are regulated by PPARβ in vivo under conditions resembling tumor angiogenesis. Microarray and qPCR analysis of RNA from cells in matrigel plugs led to the identification of a set of genes that are differentially expressed in Pparb+/+ and Pparb−/− mice (Supplementary Table S1; Figure 8A; Supplementary Figure S5). In view of the hyperplastic phenotype seen in Pparb−/−, mice we were particularly interested in genes with a direct function in cell cycle regulation. The only gene fitting this criterion was Cdkn1c gene coding for the CDK inhibitor p57KIP2 (Lee et al, 1995). We therefore focused our further investigations on this gene and performed additional experiments to unequivocally confirm Cdkn1c as a PPARβ target gene. This was achieved by different strategies: (i) the Cre‐mediated disruption of the Pparb gene in cultured fibroblasts resulted in the expected reduction of Cdkn1c expression (Figure 8B), (ii) the restoration of PPARβ expression in Pparb−/− fibroblasts led to a clear reactivation of Cdkn1c expression (Figure 8C), (iii) the treatment of primary ECs with the synthetic PPARβ agonist GW501516 rapidly induced the Cdkn1c gene (Figure 8D), and (iv) in silico analysis of the Cdkn1c upstream sequence revealed two potential PPREs, and consistent with this finding, we observed a clear induction of the Cdkn1c promoter by PPARβ/RxRα and GW501516 in transient transfection experiments (Supplementary Figure S6). The two latter observations (iii and iv) are characteristic of a direct PPAR target gene. A detailed analysis of the precise mechanisms underlying this regulation will be the subject of future studies.
Inhibition of cell proliferation by PPARβ
The inhibitory effect of PPARβ on stroma cell proliferation could be recapitulated in a cell culture model. Infection of fibroblasts isolated from Pparb−/− mice with a PPARβ expressing retrovirus resulted in a dose‐dependent inhibition of cell proliferation and, upon continued passaging, in a selection against the PPARβ re‐expressing cells. Two independent lines of evidence indicate that Cdkn1c is a PPARβ target gene that plays a pivotal role in the PPARβ‐mediated attenuation of cell proliferation. First, there is good correlation between the re‐expression of PPARβ, induction of Cdkn1c and the extent of growth inhibition (Figure 9A and B). Second, the selection against PPARβ expressing cells could be greatly diminished by an siRNA‐mediated knockdown of Cdkn1c expression (Figure 9C). These results are consistent with the idea that the hyperproliferation of tumor ECs and matrigel‐invading cells in Pparb−/− mice, results at least in part from a decrease in Cdkn1c expression. Consistent with this model, mice with a targeted disruption of the Cdkn1c gene have been reported to suffer from hyperplasia and an impairment of terminal differentiation in different tissues (Zhang et al, 1997).
Other potential PPARβ target genes
For two other genes downregulated in matrigel‐invading cells from Pparb−/− mice, that is, Cd36 and Thbs2 (Supplementary Figure S5), an inhibitory role in angiogenesis has been established in previous studies (Lee et al, 1995; Armstrong and Bornstein, 2003). Thrombospondins attenuate EC proliferation and migration in vitro and inhibit angiogenesis in vivo, which is strictly dependent on their interaction with the CD36 receptor. CD36 has previously been shown to be regulated by different PPAR subtypes in a variety of cell types (Tontonoz et al, 1998; Westergaard et al, 2001; Sato et al, 2002; Liu et al, 2004), consistent with the present observations. In Pparb−/− cells, both ligand (Thbs2) and receptor (Cd36) genes are downregulated, suggesting that a signaling loop with an essential function in modulating angiogenesis might be impaired in these cells. CD36 and thrombospondin would have a similar effect on EC proliferation as p57KIP2, suggesting that these molecules may act in concert. We have also identified a number of other genes with potential functions in growth control and differentiation that show a decreased expression in matrigel‐invading cells from Pparb−/− mice (see highlighted entries in Supplementary Table S1; Figure 8). Interestingly, further backcrossing of the Pparb−/− strain to C57BL/6N mice results in a lower incidence of tumor resistance (our unpublished observation), suggesting that strain‐specific modifier loci also influence the observed phenotype. At present, it is therefore not possible to distinguish which of these genes play a role in the context of PPARβ‐dependent vascularization, but it is very likely that multiple target genes are involved.
The angiogenic response to growth factors during the final stages of tumor angiogenesis is characterized by an inhibition of EC proliferation and the acquisition of a fully differentiated phenotype (Carmeliet, 2000). Our findings are consistent with a model where PPARβ is required at this stage of tumor angiogenesis, where it functions to attenuate EC proliferation as a prerequisite for microvessel maturation/differentiation. Thus, the loss of PPARβ does not interfere with angiogenesis per se, but rather results in the deregulation of angiogenesis, leading to a hyperplastic phenotype and the formation of non‐functional microvessels. It is likely that the reduced PPARβ‐dependent expression of Cdkn1c is at least in part responsible for the phenotype seen in Pparb−/− mice, as suggested by its role in the PPARβ‐mediated inhibition of cultured cells. Thus, Cdkn1c has the potential to impinge on tumor growth in different ways. When expressed in the tumor cells proper, Cdkn1c inhibits tumorigenesis by a direct inhibitory effect of tumor cell proliferation (Kuang et al, 2007), while its expression in tumor EC cells is required for blood vessel formation and thus tumor growth.
A similar phenotype of enhanced, but non‐productive angiogenesis has very recently been described in mice lacking the Notch ligand Delta‐Like 4 (Dll4), or mice systemically treated with a Dll4‐neutralizing antibody (Noguera‐Troise et al, 2006; Ridgway et al, 2006). The lack of functional Dll4 rendered angiogenic ECs hyperproliferative and markedly increased tumor vascularity, but caused defective microvessel differentiation and blocked syngeneic tumor growth. Even though the histological evidence (hyperplasia and immature microvessels lacking a lumen) and the functional consequences on tumor growth are strikingly similar to our observations with Pparb−/− mice, there are clear differences and probably no functional links. Most importantly, Dll4 is essential for embryonic vascular development and arteriogenesis (Krebs et al, 2004), whereas PPARb is not required for physiological angiogenesis. Furthermore, we were unable to detect any differences in the expression of Dll4 or other key component of Notch signaling between Pparb+/+ and Pparb−/− cells from matrigel plugs (data not shown). This suggests that multiple and presumably independent mechanism are required to prevent the deregulation of tumor EC proliferation and the occurrence of non‐productive angiogenesis. It is possible that due to the highly pathological tumor micro‐environment, tumor‐specific regulatory mechanisms have to be operational to prevent an excessive angiogenic response of the tumor stroma. Our results suggest that Pparβ is such a regulator. Our findings that PPARb is the predominantly expressed Ppar subtype in angiogenic matrigel plugs and human tumor ECs, and is inducible by angiogenic growth factors in vitro, supports this hypothesis.
Previous studies addressing the role of PPARβ in tumorigenesis have yielded partly conflicting results (Park et al, 2001; Barak et al, 2002; Gupta et al, 2004; Harman et al, 2004; Reed et al, 2004; Marin et al, 2006), leaving it unclear whether PPARβ has tumor promoting or suppressing properties. Our findings possibly help to shed light on this issue. PPARβ may have different functions in tumor stroma and in certain tumor cells with opposing effects on tumor growth. Clearly, a detailed understanding of these complexities will be a prerequisite for the development of PPARβ‐directed drugs and their clinical application.
Materials and methods
Lewis lung carcinoma (LLC1) and B16F1 melanoma were obtained from the ATCC and cultured in DMEM plus 10% fetal bovine serum. Aortic ECs were isolated as described (Chen et al, 2004). HuVECs were established and cultured as described (Graulich et al, 1999). Cell lines obtained from Pparb+/+, Pparb−/− and Flox‐Pparb (PPARδck) mice are described in Supplementary data. Retroviral infections, siRNA transfections and luciferase assays are described in Supplementary data.
Pparb−/− and Pparb+/+ mice have previously been described (Peters et al, 2000). All experiments were performed with mice backcrossed with the C57BL/6N strain. PPARδck mice (Barak et al, 2002) harboring a floxed Pparb exon 4 were kindly provided by R Evans (The Salk Institute, La Jolla, CA).
Matrigel plug assay
Matrigel plugs (Ley et al, 2004) were established using the BD Matrigel Matrix (Becton Dickinson), as described by the manufacturer. In brief, 500 μl of liquefied matrigel containing 100 nM PGE2 and 0.6 μg/ml FGF‐2 were injected subcutaneously and plugs were removed 3 days later for analysis. For retroviral ‘rescue’ experiments, 105 stably transduced fibroblasts (see below) were harvested during the exponential growth phase and included in the matrigel.
Dynamic gadolinium Gd‐DTPA enhanced imaging (Choyke et al, 2003; Kiessling et al, 2003) was carried out with a clinical 1.5 T whole‐body MRI‐System (Siemens Sonata, Erlangen, Germany) and a dedicated custom‐made small animal coil. Mice were anesthetized, injected i.v. with 0.5 mmol/kg of Gd‐DTPA, and imaging was started immediately thereafter. We performed 510 measurements per session during a total period of 9 min.
Paraffin‐embedded sections were stained as described in Supplementary data. The following antibodies were used: α‐SMA (peroxidase‐conjugated, Sigma, Munich, Germany), FLAG tag (Becton Dickinson, Heidelberg, Germany), PCNA (polyclonal FL‐261, Santa Cruz Biotechnology, Santa Cruz, California, USA) and AQP‐1 (kind gift from Dr S Nielsen, Aarhus, Denmark) (Gresz et al, 2001). Signals were visualized by biotinylated secondary antibodies and either avidin‐conjugated peroxidase with diaminobenzidine as the substrate, or avidin‐conjugated alkaline phosphatase with Vector® Red (Vector Lab, California, USA). Double immunofluorescence was performed using an FITC‐labeled α‐SMA antibody and the polyclonal rabbit anti‐AQP‐1 antibody with a Cy5‐labeled secondary antibody.
Microarrays were generated using a GMS 417 arrayer (MWG Biotech, Ebersberg, Germany). The chips contained 22 500 clones from the mouse sequence‐verified NIA 15k cDNA library plus the NIA 7.4k cDNA clone set (http://lgsun.grc.nia.nih.gov/cDNA/cDNA.html). Analyses were performed with amplified RNA (Supplementary data). Each experiment was performed as a sandwich hybridization using two arrays, and two independent experiments were performed for each data point. Spot intensities were analyzed from scanned images using Scan Array Express™ (Perkin Elmer, Rodgau, Germany). Data from all four hybridization were averaged before further analysis. All procedures were performed according to the protocols described at http://www.imt.uni-marburg.de/microarray/download.html.
Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).
We thank Dr R Evans (Salk Institute, La Jolla, CA) for PPARδck mice, Dr T Blankenstein (MDC Berlin, Germany) for the retroviral Cre vector, Dr S Nielsen (Aarhus, Denmark) for providing the polyclonal AQP‐1 antibody. We are grateful to Birgit Samans, Dr M Krause and Professor M Eilers for help with microarray analyses, to Lena Holembowski for Cdkn1c promoter experiments, and to Bernhard Wilke, Julia Dick, Margitta Alt, Christine Rösser and Renate Baumann for excellent technical assistance. This work was supported by grants from the Deutsche Forschungsgemeinschaft (SFB‐TR17/A3), the Deutsche Krebshilfe, the Wilhelm‐Sander‐Stiftung and the National Cancer Institute (CA89607, CA97999, CA124533).
- Copyright © 2007 European Molecular Biology Organization