|Home | About | Journals | Submit | Contact Us | Français|
A diverse set of cellular defects, presumably elicited by multiple genetic alterations, underlies cancer development. Aberrant Hedgehog signaling has recently been implicated in the development and maintenance of breast cancer. However, evidence conclusively showing that activated Hh signaling can induce mammary tumors is lacking. We now show that transgenic expression of the Hh effector protein, GLI1, under the regulation of the MMTV promoter, expressed in the mouse mammary gland, is associated with the appearance of hyperplastic lesions, defective terminal end buds and tumor development. The GLI1 induced tumors are histologically heterogeneous and involve the expansion of a population of epithelial cells expressing the progenitor cell markers keratin 6 and Bmi-1. Moreover, tumor cells express genes involved in proliferation, cell survival and metastasis. GLI1 induced tumors do not fully regress following transgene deinduction indicating that some tumors develop and are maintained autonomously, independent of sustained transgenic GLI1 expression. The data strongly support a role of Hh/GLI signaling in breast cancer development and suggest that inhibition of this signaling pathway represents a new therapeutic opportunity for limiting tumorigenesis and early tumorigenic progression.
Breast cancer is one of the most frequently diagnosed cancers and a leading cause of cancer death. Understanding the genetic and molecular mechanisms underlying this disease is an important challenge and a prerequisite for developing better treatment strategies.
A number of breast cancer subtypes have been defined based on mRNA expression profiling and show differential progression characteristics (1). The basal, aggressive subtype is characterized as hormone-receptor negative, HER2 negative, but positive for basal/myoepithelial cell markers including keratins 5/6 (K5/K6). Most breast tumor patients with germ line mutations in the cancer predisposition gene BRCA1 display this phenotype. This led to the proposal that BRCA1 may regulate mammary stem cell fate (2). Similar tumor types have been described in mouse models where mammary tumorigenesis was driven by a constitutively activated Wnt signaling pathway and presumed to originate in a stem/progenitor cell compartment (3). Two types of stem/progenitor cells have been reported in the normal human mammary gland: a luminal restricted population and a bipotent population giving rise both to luminal and myoepithelial cells (4). The bipotent progenitor-like cells have been found in human breast tumors and breast cancer cell lines (2, 5, 6).
One of the key signaling pathways controlling the embryonic development of certain organs, including the mammary gland, is the Hedgehog (Hh) signaling pathway. This pathway is frequently implicated in the regulation of somatic stem cells as well as cancer development. Studies in humans (5) and mice (7) showed active Hh-signaling in mammary stem/progenitor cells. Constitutive activation of the Hh-pathway by either deletion of one copy of the Patched (Ptch1) gene which encodes a Hh receptor component, or overexpression of a dominant active allele of the Smoothened (Smo) co-receptor in mouse models result in a hyperplastic response but no tumors (8, 9). In a recent study we showed that conditional overexpression of the transcription factor GLI1, a terminal effector in the Hh-pathway, in the murine mammary gland causes a disruption of pregnancy induced gland maturation and induces islands of highly proliferative epithelial cells (10).
Despite a lack of causal evidence, the analysis of a number of Hh gene aberrations has suggested a role for activated Hh-signaling in human breast cancer. Firstly, there is loss of one allele of PTCH1 in a subgroup of breast cancers (11) and a reduction of PTCH1 expression in about 50% of ductal and invasive breast cancers (9, 12). In the latter study, promoter methylation correlated with low PTCH1 expression, thought to be associated with pathway activation. Secondly, GLI1 gene amplification (13) and rare (2/24) GLI1 missense mutations (14) have been shown in breast cancer. The functional properties of the mutated GLI1 proteins are as yet unknown. Also, increased GLI1 mRNA expression in breast cancer epithelial cells (15) and in breast cancer cells with stem cell like properties (5, 16) has been reported. Lastly, overexpression of Gli2 in human mammary stem/progenitor cells leads to an increase in mammosphere formation and dysplastic growths in xenograft assays (5). An RNAi screen aimed at identifying essential genes in human breast cancer cell lines interestingly confirmed GLI2 as a candidate (17).
In spite of all the circumstantial evidence linking Hh-signaling to the development/progression of mammary cancers definitive proof of its capacity to induce mammary tumors when misregulated is lacking. To address this issue we carried out novel investigations using our previously established transgenic model, which conditionally overexpresses GLI1. The new data shows the expansion of proliferative epithelial cells, expressing progenitor cell markers, and the frequent development of tumors with different histological patterns consistent with a progenitor cell origin. The data strongly support a causal role for Hh/GLI signaling in the induction of breast cancer. A translational application of these results would suggest that the inhibition of this pathway could represent a new therapeutic opportunity.
We have previously reported the generation and initial characterization of a bigenic system for the transgenic expression of GLI1 in the mammary gland, utilizing the tetracycline regulatory system (10). In this model, the reverse tetracycline-dependent transcriptional activator (rtTA) is expressed from the mouse mammary tumor virus (MMTV) promoter, and only in the presence of doxycycline can it induce expression of a GLI1 transgene fused to the tetracycline-dependent promoter. Doxycycline (dox, 2 mg/ml) and 5% sucrose were therefore added to the drinking water to induce GLI1 expression. The start of induction was scheduled either during the mating of TREGLI1 to MMTVrtTA mice or at the time of separation, at the age of three weeks. Without doxycycline exposure the bigenic mammary glands did not show detectable GLI1 expression (10). All transgenic mice were generated within an SPF barrier facility according to local and national regulations, and the experimental design was approved by the Stockholm South Animal Ethics Committee. Mammary glands were dissected out and RNA was isolated using RNAzolB (CRP Inc). Random hexamer-primed complementary DNA (cDNA) was generated using the Reverse transcription system (Promega). From this material Gli1 and actin mRNA levels were determined essentially as described (10).
Mammary gland fragments from wt and bitransgenic females were fixed overnight in 4% paraformaldehyde or Feketes fixative, then paraffin embedded, sectioned and hematoxylin/eosin stained. Samples were obtained from virgins (5 and 10 weeks), parous individuals (6.5 days of pregnancy (dpc), age 6-19 weeks) and multiparous (age 14 -52 weeks). Controls from male wt and bitransgenic (age 48-52 weeks) mice were also analyzed. Hematoxylin/eosin staining, immunohistochemistry and immunofluorescence labeling were performed essentially as described (10). Primary antibodies used were: rabbit polyclonal anti-GLI1 (Abcam) at a 1:2000 dilution for immunohistochemistry, rabbit polyclonal anti-GLI1 (Cell signaling) at 1:1000 for immunofluorescence labeling, rabbit polyclonal anti-Snail (Abcam) diluted 1:800, rabbit polyclonal anti-ER-α (Neomarkers) at 1:20, rabbit polyclonal anti-K6 (Babco) at 1:2000, rabbit polyclonal anti-K5 (Babco) at 1:4000, rabbit polyclonal anti-K14 (Babco) at 1:2500, rabbit polyclonal anti-Laminin (Sigma) at 1:100 and rabbit polyclonal anti-Cyclin D1 (Labvision) at 1:100 dilution. The rabbit polyclonal antibody recognizing K17 used at 1:2500 and the rat polyclonal antibody K8 at 1:100 were obtained from Dr. P. Coulombe and Dr. I. Mikaelian, respectively. Detection of BrdU (1:25, Becton Dickinson), P21Crp1/wap (1:25, Santa Cruz), p634A4 (1:200, Santa Cruz), smooth muscle actin (1:50, Novocastra), Histone H3 (1:4000, Abcam) and E-cadherin (1:11000, BD transduction laboratories) was performed using mouse monoclonal antibodies and the Histomouse kit (Zymed). Detection of Bmi-1 (1:100, Upstate) was performed using the Power Vision kit (Immunologic). Negative controls consisted of experiments done without the primary antibody in the presence of equal concentrations of normal rabbit, goat or mouse IgG. The number of GLI1, K6, BrdU and Histone H3 positive mammary epithelial cells were quantified in GLI1 induced hyperplastic regions (1 year). One thousand nuclei from three separate areas in three samples were counted and data presented as a percentage. The routine Giemsa staining method was used to assess the presence of mast cells. To identify glyco-conjugates deparaffinized and rehydrated sections were treated with diastase for the removal of glycogen, and thereafter stained with periodic acid-Schiff (PASD).
Histological analysis of GLI1 expressing female mammary glands revealed that developmental defects appeared during early pregnancy (6.5 dpc) and consisted of defective terminal end buds (TEB) and mammary hyperplasia (Fig.1A: panel b and c). GLI1 expressing TEBs showed a disorganization of cap and body cells, a thickening of body cells, and occasional invasion of epithelial cells, which created a microlumen (Fig. 1A:b). In contrast, wt 6.5 dpc mammary glands showed a normal TEB structure (Fig.1A:a). In 100% of GLI1 expressing parous females (n=27) and in 60% of aged males (n=5), histological analysis revealed large hyperplastic areas comprising polyp-like structures and multiple focal thickenings of the ductal epithelium (Fig.1A:c, d). These protrusions consisted of multiple layers of cells, which were bulging into the luminal space and often spanned the width of the duct. These changes contained glycogenic residues on the basal epithelial membrane when stained with periodic acid-Schiff (PASD) (SI Fig.1A:c). Furthermore, the mammary stroma in the GLI1 expressing mammary glands were occasionally fibrotic and hyalinic (Fig.1A:c, d). Immunohistological analysis of the mammary epithelial cells in the GLI1 expressing hyperplastic regions and tumors (see later sections) demonstrated GLI1 expression in a high percentage of cells (Fig.1B). Consistent with this observation, immunofluorescence revealed a high proportion of GLI1 positive cells in the hyperplastic areas (48.2% ±3.1%) (Fig.3C:a). In addition, immunohistological analysis demonstrated that cells in the GLI1 induced hyperplastic areas mainly expressed the basal cell type cytokeratins 5 (K5), 14 (K14), 17 (K17), the basal marker p63 and the myoepithelial cell marker smooth muscle actin (SMA) with low/negative expression of the luminal cell marker cytokeratin 8 (K8) (Fig.1C:a-c and SI Fig.2). However, the normal profile for SMA staining, evident on the basal mammary epithelial cells of 6.5 dpc wt mammary glands and GLI1 expressing hyperplastic regions (SI Fig.3B:a-h), was dramatically reduced in hyperplastic regions of mammary glands dissected from one year old multiparous GLI1 expressing mice (SI Fig. 3B:i-l), indicating that persistent GLI1 transgene expression leads to disruption of the basement membrane. In contrast, mammary glands from wt virgin or pregnant mice showed a stringent expression pattern of the epithelial markers, with K8 expression on the luminal side and SMA, K5, K14, K17 and p63 on the basal side (SI Fig.2B:a, b and data not shown). No developmental alterations were observed at any time point in control female or male wt or non-induced bigenic mammary glands.
Eighty-eight percent (n=9) of all multiparous bigenic female mice with GLI1 expression targeted to the mammary epithelial compartment developed mammary tumors. Multiple tumors even appeared in the same animal as well as in the same mammary gland (Fig.1A:e and SI Fig.1A:b). The mammary glands were replete with hyperplastic areas, leaving no or very sparse normal mammary epithelium (Fig.1A:e). Mammary tumors started to appear after 18 weeks of doxycycline treatment. Only GLI1 expressing multiparous (2-10 pregnancies) females developed mammary tumors during the experimental time period (one year). However, GLI1 expressing male mice developed hyperplastic lesions in the mammary gland (Fig.1A:d).
A range of histologically distinct tumors appeared in the mammary glands of the GLI1 expressing mice, namely tumors with a solid, ductal or squamous differentiation pattern (Fig.1A:f-h). In all tumors, elevated expression of GLI1 was observed (Fig.1B:b, c and data not shown). The GLI1 induced ductal tumors showed a mammary ductal like growth pattern, whereas GLI1 induced solid tumors showed a more dense cell growth pattern resembling the hyperplastic lesions (Fig.1A:f, g). The tumor appearance, however, was not uniform, and different growth patterns could be observed, including trabecular and papillary projections (SI Fig.1A:c, d). GLI1 induced ductal tumors were occasionally ossified and contained PASD-positive glyco-conjugates and hyperplastic lesions (Fig.1A:g and SI Fig.1A:c). Immunohistological analysis revealed that these tumors contained a substantial amount of K8 positive cells surrounded by a single layer of K5 and SMA positive cells. In contrast, GLI1 induced solid tumors showed a similar cell phenotype to the GLI1 induced hyperplastic lesions seen in old multiparous female mice, characterized by high levels of K5 and low/absent expression of K8 and SMA (Fig.1C:d-f). GLI1 induced tumors with a squamous differentiation pattern contained K5 and K8 positive epithelial cells while no staining was obtained with the anti-SMA antibody (Fig.1C:g-i). In all tumors expression of K14, K17 and p63 showed a basal distribution, resembling the K5 expression pattern (SI Fig.2).
The progression from in situ to invasive and metastatic tumors is characterized by an increased presence of necrotic and mitotic cells, enhanced angiogenesis, an altered cellular differentiation profile and the secretion of proteolytic enzymes (18). We observed that in a proportion of GLI1 induced tumors there were necrotic and mitotically active areas (SI Fig.3A:a, b). Furthermore, the mammary glands of GLI1 expressing mice showed a dramatic increase in the number of mast cells, an event suggested to be important for angiogenesis in the most aggressive forms of human cancers such as malignant melanoma, breast carcinoma and colorectal adenocarcinoma (19). Mast cells were located at the periphery of the GLI1 induced tumors as well as within the hyperplastic regions (SI Fig.3A:c). Staining with an anti-ER-α antibody revealed that GLI1 induced hyperplastic regions and tumors were ER-α negative (Fig.2b-d). This is significant, since aggressive human breast cancers with a high risk for metastasis can be identified with the help of molecular markers, such as estrogen receptor (ER) negativity (20). In contrast, virgin wt mammary glands (10 weeks of age) expressed ER-α in approximately 25% of all ductal epithelial cells (Fig.2a). Furthermore, an epithelial to mesenchymal transition (EMT) is implicated in the progression of primary tumors towards metastasis. One critical step associated with EMT is the repression of the transmembrane protein E-cadherin. It was recently shown that E-cadherin expression was lost in infiltrating human basal cell carcinomas (BCC) as a consequence of induced Snail expression by Gli1 (21). Immunhistological staining with an anti-E-cadherin and an anti-Snail antibody revealed that wt mammary glands expressed E-cadherin on the baso-lateral surface of all alveolar and ductal epithelial cells, while no or very low nuclear Snail expression was detected (Fig.2e, i). In hyperplastic regions and tumors from GLI1 expressing mice E-cadherin expression was, however, reduced or absent, but expression of Snail was nuclear in a large fraction of cells from hyperplastic regions or ductal tumors (Fig.2f-h, j, l), though absent in solid tumors (Fig.2k).
A disrupted basement membrane is necessary for tumor cells to invade neighboring tissues and become metastatic. We showed by immunofluorescence labeling that there was continuous SMA and laminin expression surrounding the basal epithelial cells in the GLI1 induced hyperplastic areas dissected from 6.5 dpc parous female mice (SI Fig.3B:e-h). Similarly, 6.5 dpc control wt mammary glands expressed SMA and laminin around the ductal basal epithelial cells (SI Fig3.B:a-d). However, in mammary glands removed from multiparous mice with induced GLI1 expression (>18 weeks) a disrupted basement membrane and diminished expression of SMA and laminin was observed, possibly indicating an increased risk for tumor progression towards metastasis (SI Fig.3B:i-l).
We previously described an increase in cell proliferation in the mammary gland of GLI1 expressing mice during pregnancy (10). Here, we have compared BrdU incorporation rates and expression patterns of Cyclin D1 and Histone H3 using glands from long-term (1 year) doxycycline treated multiparous females versus wt control mice (Fig.2 and SI Fig.4). A high proportion of BrdU, Cyclin D1 and Histone H3 positive cells were observed in the GLI1 induced hyperplastic regions and tumors (Fig.2 m-t and SI Fig.4A:b-d). We thus correlated transgene expression with proliferative status (Fig.3C and SI Fig.4B-C), since transgenic GLI1 expression was dramatically elevated in only a large fraction of lesional mammary epithelial cells in the multiparous GLI1 expressing mice (48.2%±3.1%) but not in all mammary epithelial cells. Dual immunofluorescence staining for GLI1 and BrdU incorporation revealed that 1.9% ±0.2% of the mammary epithelial cells in the hyperplastic regions were BrdU positive and almost half of these BrdU positive cells were also GLI1 positive (0.9% ±0.1%) (Fig.3C). Furthermore, 5,1% ± 0,6% of the mammary epithelial cells in the hyperplastic regions were Histone H3 positive and essentially all of these mitotically active cells were also positive for GLI1 (SI Fig4B-C). Taken together these results indicate that a significant proportion of the GLI1 expressing cells undergo active proliferation (Fig.3C and SI Fig. 4B-C).
The Hh signaling pathway has been implicated as a central player in the maintenance of stem or progenitor cells in a growing list of adult tissues. We hypothesized that GLI1 expression could alter the number of progenitor cells in the mammary gland. To investigate this possibility we analyzed expression of the known mammary progenitor cell markers, K6, Bmi-1 and p21cip (5, 22, 23). Immunohistochemical analysis of wt mammary glands for K6 expression showed staining of a few mammary cells located between the basal, myoepithelial and luminal cell layers (Fig.3A:a). In mammary glands from GLI1 expressing mice and in particular in hyperplastic regions, K6 was expressed in a markedly higher number of cells (Fig.3A:b). In this context, K6 was heterogeneously expressed with some K6 positive cells located in the mammary ducts, some in the luminal layer and even in the luminal space (Fig.3A:b). The K6 positive cells in the GLI1 induced hyperplastic areas lacked positive staining for other epithelial markers (K5, K8, K14, K17 and SMA) indicating that these cells are unique and may represent stem or progenitor cells (Fig.3A:c and data not shown). In addition, in all tumors from GLI1 expressing mice a high proportion of K6 expressing cells was observed (Fig.3A:d-f). The elevated K6 expression was quantified by immunofluorescence labeling and the analysis revealed that 12.6% (±1.5%) of the mammary epithelial cells in the GLI1 induced hyperplastic areas were K6 positive (Fig.3B). This prompted the question of whether K6 expressing cells were proliferative. Dual immunofluorescence staining for K6 and BrdU revealed that 2.1% (±0.3%) of the mammary epithelial cells in the hyperplastic regions were BrdU positive and that 0.7% (±0.1%) were both K6 and BrdU positive (Fig.3B). Moreover, high levels of the progenitor markers Bmi-1 and p21cip were detected in the GLI1 induced hyperplastic regions (Fig.3A:h-l and data not shown) while wt mammary glands only showed nuclear expression of Bmi-1 and p21cip in a small proportion of epithelial cells at 6.5 dpc (Fig.3A:g and data not shown) supporting a possible expansion of the progenitor cell compartment associated with GLI1 expression.
Studies have shown that inactivation of oncogene expression in conditional transgenic mouse models is not sufficient for the sustained reversal of the tumorigenic process (24, 25). We investigated whether this principle holds for tumors where GLI1 expression is uninduced. GLI1 expression was induced by doxycycline exposure during two pregnancies (15-21 weeks of dox treatment) followed by six additional pregnancies in the absence of induced transgene expression (GLI1(+dox,-dox), 40-45 weeks without dox treatment). Interestingly, tumors reminiscent of the ductal form, observed in long-term doxycycline treated bigenic mice were detected at the time of sacrifice in GLI1(+dox-dox) mammary glands (Fig.4A:a). These GLI1(+dox-dox) tumors were ossified, contained glycogenic fluids, cholesterol deposits and more stroma when compared to tumors with continued transgenic GLI1 expression (Fig.4A:b-c). Immunohistochemical analysis of persistent tumors showed positive staining for K5, K14, K17, p63, K8, K6, SMA and Cyclin D1 (Fig.4A:e-h and data not shown).
Immunohistochemical and RT-PCR analysis revealed that these tumors did not express GLI1 (Fig.4A:d, B) as compared to expression of high GLI1 levels in mammary glands from doxycycline treated bigenic mice at parturition (Fig.4B). We have previously shown that the GLI1 mRNA levels return to basal levels already after the first parturition without doxycycline induction in the GLI1(+dox-dox) mammary glands and bigenic mice not treated with doxycycline does not express detectable levels of GLI1 mRNA in the mammary gland(10). Since the primer pair used for RT-PCR analysis of GLI1 was designed to recognize both human and mouse Gli1 mRNA we conclude that the Hedgehog pathway is not active after doxycycline removal (10).
Activated Hh signaling has long been presumed to be important in the development and progression of breast cancer. In the present study we provide definitive evidence showing that GLI1 can serve as an oncogene in the murine mammary gland. The oncogenic function is associated with a stimulation and expansion of a mammary cell population marked by the expression of K6 and Bmi-1. Such cell expansion may lead to an increase in the number of susceptible cells at risk. A similar expansion of putative progenitor cells followed by tumor development has previously been described in transgenic mice overexpressing wnt signaling components or myc (3, 26). Interestingly, Hh signaling is known to be active in mammary gland stem/progenitor cells both in mice and humans, and is associated with the expression of Bmi-1 (5, 7).
It is, at present, not clear how ectopic GLI1 expression can bring about the expansion of putative progenitor cells. Two possibilities exist, though they are not mutually exclusive. Either GLI1 expression may reverse the endogenous Gli3-mediated target gene repression (27), or GLI1 expression may mimic and amplify the recently described stimulatory effect Ihh has on mammary gland stem cells (7). Of note, dual immunofluorescence staining with GLI1 and cell cycle markers shows that GLI1 expressing mammary cells are proliferative (Fig.3 and SI Fig.4). This contrasts with the different observation in mammary cells from mice expressing a mutant Smo-transgene (9). The reason for this discrepancy is currently unclear but could be due to differing expression levels or signaling output when components from different positions within the Hh-pathway are expressed.
In the GLI1 mouse model of mammary tumorigenesis we observe tumors that display varying tumor differentiation patterns, including ductal, squamous and solid phenotypes. A similar varying differentiation pattern in tumors has previously been described in the wnt/myc driven models of mammary tumorigenesis mentioned above. Similarly, specific human breast epithelial cell populations that become transformed, exhibit different tumor phenotypes reflecting the cell of origin (28). GLI1 induced tumors possess several hallmarks of aggressive tumors. These include necrotic areas, high mitotic activity, a lack of ER-α expression and mast cell infiltration. These features are reminiscent of human basal-subtype breast cancer (2, 6, 29, 30). Stem cells or cells expressing stem cell markers such as Bmi-1 and Nanog/Oct-4 targets also appear to be important in such tumors (2, 29, 30). A model has been proposed, implicating Snail as an upstream activator of FOXC2 expression. The transcription factor FOXC2, in turn, promotes aggressive tumor properties including epithelial-mesenchymal transitions in basal-like breast cancers (31). In light of this, the prominent induction and expression of Snail in the GLI1 transgenic mice is likely to be partially responsible for the aggressive tumor phenotype in this model. Consistent with this notion, Snail promotes tumor recurrence in the HER2/neu mouse model of breast carcinogenesis. High expression of Snail has even been shown to be a negative prognostic factor in human breast cancer (32). We also found that in the GLI1 model some tumors remain/recur in the absence of continued GLI1 transgene expression. Such tumor escape is also known to occur in the c-myc driven mammary tumor model either by reactivation of c-myc expression or in a myc-independent manner (24).
The picture that emerges from the data presented here, in the context of previous results, is that aberrant GLI1 expression leads to the expansion of a cell population potentially associated with progenitor cell properties in the mammary gland. This generates a susceptible target cell population, which can develop into frank tumors. The concomitant GLI1 induced expression of Snail contributes to an aggressive phenotype. Given the presence and role of active Hh-signaling in tumor initiating/cancer stem cells of various tumor types, including breast cancer (5), and the identification of GLI2 as an essential gene in breast cancer cell lines (17), it will be important to determine the functional dependence of both normal mammary stem cells and tumor cells with stem cell-like properties on Hh/GLI signaling. Similarly, further studies of the expression and genetic alterations affecting Hh-pathway components including the association to subtype and progression in human breast cancers are warranted.
Strategies for pharmacological intervention inhibiting Hh signaling at different levels in the pathway are actively being developed (33, 34). We recently identified candidate small molecule inhibitors targeting the terminal GLI dependent step in the Hh-pathway. Targeting GLI is an effective strategy, since it is one of the final stages in the Hh-pathway, irrespective of the mechanism that may aberrantly activate Hh signaling (35). Combinatorial treatment with such inhibitors and established treatments that eliminate the bulk of the tumor cells therefore represents an interesting area for future investigations.
We wish to thank Dr. Coulombe for the rabbit polyclonal antibody recognizing K17 and Dr. Mikaelian for the antibody recognizing K8. The use of animal facilities within the Clinical Research Center, Karolinska University Hospital is gratefully acknowledged. The study was supported by grants from the Swedish Cancer Society, the Swedish Research Council, NIH Program Project Grant AR47898 and NIH/NCI MMHCC Grant U01 CA105491.