Shh Pathway Activity Is Down-Regulated in Cultured Medulloblastoma Cells: Implications for Preclinical Studies

Medulloblastoma is the most common malignant pediatric brain tumor, accounting for ∼20% of cases. Current treatment, which involves surgical resection, chemotherapy, and radiation of the craniospinal axis, is often associated with major long-term side effects, particularly in young children (1–3). Therefore, there is a great need for new nontoxic therapies. However, the identification of improved therapies for medulloblastoma has been hampered by the lack of adequate model systems for testing novel drugs. Recently, genetically defined mouse models have shown promise as cancer model systems in which tumors arise spontaneously and recapitulate the histologic, spatial, and temporal context of the corresponding human disease (4, 5). Such models are particularly useful for proof-of-concept studies of molecular targeted therapies. Using the Ptc1^+/−p53^−/− mouse model of medulloblastoma (6), we successfully showed that a small-molecule inhibitor (HhAntag-691, hereafter referred to as HhAntag) of Smoothened (Smo; refs. 7, 8) could suppress the Sonic Hedgehog (Shh) pathway in vivo leading to the elimination of spontaneous mouse medulloblastoma (9). However, genetic models recapitulate only a specific disease subset and they do not encompass the full spectrum of the corresponding human cancer. In addition, such models are often not suited to high-throughput studies of potential new treatments. This is particularly the case for brain tumor models in which the analysis can be complex and very labor-intensive. Therefore, it is important to complement studies on genetic models with approaches involving cancer cells propagated in vitro and in vivo.

The majority of preclinical cancer studies carried out to date employ human tumor cell lines or xenografts established in the flank of immunocompromised mice (10, 11). Although these models are often assumed to be representative of the original disease, the tumor cells have been removed from their native environment and selected for their ability to grow under artificial conditions. In addition, in some cancers, such as medulloblastoma, only a small percentage of tumors can be established as cell lines (12–14) or transplants, and it is unclear whether such experimental models replicate the phenotype of the original tumor cells.

In the present study, we compared the molecular properties of medulloblastoma cell lines and allografts with those of the original tumors. We found that Shh pathway activity was maintained only in allografts derived directly from the original tumors but not in medulloblastoma cultures or allografts established using cultured tumor cells. In addition, the Shh pathway activity was maintained in serially transplanted allografts established directly from medulloblastomas. We found that whereas HhAntag was very effective at eliminating tumors in which Smo was active, it was completely ineffective against allografts in which the Shh pathway was down-regulated.

Cell culture. Mouse medulloblastoma cell cultures (SJMM cells) were established as described previously (15). Cells were maintained in DMEM (Cambrex Bio Science Walkersville, Walkersville, MD), supplemented with 10% FCS, 1 mmol/L Gluta Max (Invitrogen, Carlsbad, CA), 50 units/mL penicillin G and 50 μg/mL streptomycin, under a humidified atmosphere of 90% air and 10% CO₂.

Allograft propagation. All mouse studies were carried out according to protocols approved by the institutional Animal Care and Use Committee at St. Jude Children's Research Hospital. Six-week-old female athymic nude mice (CD1 nu/nu; Charles River Breeding Laboratory, Wilmington, MA) were injected s.c. in the flank with single cell suspensions of 5 × 10⁶ SJMM cells. For transplantation of fresh tumors, either a small piece of solid fragment or suspended tumor cells derived from Ptc1^+/−p53^+/−, Ptc1^+/−p53^−/− (6), and Lig4^−/−p53^−/− (16) mice were implanted. For tumor cell suspension, small fragments of medulloblastoma were incubated with 500 μg/mL collagenase type IV, (Invitrogen) and 500 μg/mL hyaluronidase (Atlanta Biologicals, Lawrenceville, GA) for 30 minutes, then strained using a Cell-Strainer (40 mm, BD Bioscience, San Jose, CA) and resuspended in Matrigel (BD Bioscience); 2 to 3 × 10⁶ cells were injected. Tumor sizes were measured using digital calipers (TRACEABLE, 0-200 mm; Control Company, Friendswood, TX) and tumor volumes were calculated according to the following formula, (π/6) × d³, where d is the mean diameter (17). Allografts with volumes in the 400 to 600 mm³ range were harvested and snap-frozen.

HhAntag treatment. Cohorts of mice bearing allografts of ∼200 mm³ were treated with vehicle alone or with 100 mg kg⁻¹ of HhAntag twice a day by oral gavage. Four hours after the last dose, mice were euthanized and allografts were fixed in 10% formalin or snap-frozen. HhAntag was obtained from Curis (Boston, MA) and Genentech (South San Francisco, CA).

Northern and Southern analyses. Total RNA and genomic DNA were isolated from snap-frozen tissues or cultured cells using the TriPure reagent (Roche Applied Science, Indianapolis, IN) and Genomic tip (Qiagen, Valencia, CA), respectively. RNA was analyzed by Northern blotting as described previously (18), using antisense RNA probes synthesized by the Strip-EZ RNA kit (Ambion, Austin, TX; refs. 9, 18). For Southern analysis, genomic DNA (5 μg) was digested with XbaI, and fragments were separated by agarose gel electrophoresis and transferred onto a NytranSPC membrane (Schleicher & Schuell BioScience, Keene, NH). Transferred DNA was hybridized with a radiolabeled Gli1 probe (corresponding to Met⁶⁹⁹-Pro⁸⁷⁰ of mouse Gli1), synthesized using the Rediprime II labeling system (Amersham, Piscataway, NJ). The intensity of signal was quantitated using the FluorChem 8900 system (Alpha Innotech, San Leandro, CA).

Immunoblotting. Immunoblotting was done as described previously (19), and protein concentrations were determined using the bicinchoninic acid protein assay kit (Pierce, Rockford, IL).

Histologic analysis and immunohistochemistry. Allografts fixed in 10% formalin were embedded, sectioned and mounted on Superfrost+ slides (Fisher Scientific, Pittsburgh, PA), as described previously (9). Paraffin sections were stained with H&E, and proliferation was quantitated using the Ki67 antibody (Novocastra Laboratory, Newcastle, United Kingdom), as described previously (9).

Fluorescence in situ hybridization. Five-micron-thick sections of allografts and medulloblastomas were mounted on poly-l-lysine-coated slides for dual-color fluorescence in situ hybridization (FISH) experiments. Bacterial artificial chromosome–derived probes specific for Gli1 gene (RP23-416B23, Research Genetics) and mouse chromosome 10 (RP23-43F3) were labeled with rhodamine and fluorescein, respectively. The labeling and FISH experiments were carried out as described previously (20).

Spectral karyotype analyses. Cells derived from SJMM-12 allografts were dissociated, cultured, and analyzed by spectral karyotype (SKY), using the Applied Spectral Imaging System (Carlsbad, CA; ref. 21). Briefly, after 4 hours of incubation with colcemid, cells were harvested, hybridized with probes, and imaged using the SkyPaint kit. Images were acquired with a fluorescence microscope (Nikon, Japan) equipped with an interferometer (Spectra Cube, Applied Spectral Imaging) and a custom designed filter cube (Chroma Technology Corporation, Rockingham, VT). SKY analysis was completed using the SKY View version 2.1 software. A total of 15 metaphase cells were analyzed for each sample.

Microarray gene expression analyses. Total RNA was extracted from four medulloblastomas (tumor-1, tumor-2, tumor-12, and tumor-21), four medulloblastoma cell cultures [SJMM-1, which was derived from tumor-1 (n = 2) and SJMM-12, which was derived from tumor-12 (n = 2)], seven culture-derived allografts (three from SJMM-1 cells, two from SJMM-2 cells, which were derived from tumor-2, and two from SJMM-12 cells), and three directly transplanted allografts (one from tumor-2 and two from tumor-21). The integrity of all RNA samples was verified using the Agilent 2100 Bioanalyzer. RNA samples were processed for hybridization without amplification and hybridized with the mouse genome 430 2.0 array (Affymetrix, Santa Clara, CA), which includes 45,101 probe sets representing ∼34,000 annotated genes. Labeling, hybridization, and scanning were done essentially as described previously, using the Affymetrix Microarray Suite version 5.0 software (MAS v5; ref. 22). MAS v5 data were normalized by global scaling. To stabilize variance, a small offset of 20 was added to each signal prior to log transformation. A t test was done on each probe set and the probe sets were ranked by the P value. Heat maps were generated using the z score of the log signal in Spotfire 8.1. Log ratios of signal are log₂ ratios of geometric means of the stabilized signals, which were generated using STATA/SE 8.2 for Linux. To identify the changes in gene expression among medulloblastomas, tumor cell cultures, and allografts, the number of probe sets scored as overexpressed (>2-fold increase, log ratio > 1) and underexpressed (>2-fold decrease, log ratio < 1) were counted.

The loss of one copy of Patched-1 (Ptc1) derepresses the Shh pathway and leads to medulloblastoma formation in Ptc1^+/− (19, 23, 24) and Ptc1^+/−p53^−/− mice (6). These tumors express high levels of Gli1, indicating that the Shh pathway is activated (6, 19, 24). Previously, we showed that the treatment of mice with HhAntag inhibited Gli1 expression in tumor cells, blocked tumor cell proliferation, increased apoptosis, and eradicated tumor mass, indicating that Shh pathway activity is required for the maintenance of medulloblastoma in vivo (9). However, this is not the case in vitro, as cultured tumor cells, derived from Ptc1^+/− and Ptc1^+/−p53^−/− mice, proliferated readily although they expressed only very low levels of Gli1 mRNA (lanes 2, 6, and 9, Fig. 1A; Supplemental Fig. S1) and as reported previously (9). Furthermore, the expression levels of two additional downstream target genes in the pathway that are up-regulated in tumors, Ptc2 and Sfrp1 (25), were also decreased in all mouse medulloblastoma cell lines (SJMM cells; lanes 2, 6, and 9, Fig. 1A; Supplemental Fig. S1). Thus, it seems that the Shh pathway is suppressed in medulloblastoma cultures and that Smo activity is not required for the proliferation of tumor cells in vitro. We were also unable to restore the activity of the Shh pathway by culturing the medulloblastoma cells in low serum conditions. This was the case even when Shh was added to the culture medium. Thus, it is unlikely that the rapid silencing of the pathway is simply due to the selected growth conditions. A previous report indicated that the Shh pathway remained active in cultured medulloblastoma cells grown in 0.5% FCS, however, in that study, lacZ activity from the mutated Ptc1 allele was used to monitor the pathway status (26). We found that this was not reliable and instead we measured the level of endogenous Gli1 expression directly to assay pathway activity. Silencing of the Shh pathway in medulloblastoma cells in vitro indicates that the tumor environment plays a significant role in regulating the Shh pathway. We propose that additional factors are present in vivo that are required to maintain Smo activity, therefore, we attempted to restore pathway activity by transplanting SJMM cells into the flank of nude mice.

All of the SJMM cell lines tested could be propagated as allografts, although the growth rates varied among the samples (Table 1). In allografts derived from SJMM-2 (lanes 7 and 8, Fig. 1A; and lane 3, Fig. 1B), SJMM-11 (lane 4, Fig. 1B) and SJMM-14 cells (lane 6, Fig. 1B), the level of Gli1 expression was very low, indicating that the Shh pathway remained inactive. In contrast, SJMM-12 allografts expressed extremely high levels of Gli1 (lanes 10 and 11, Fig. 1A; lane 5, Fig. 1B), much greater than in Ptc1^+/− medulloblastoma (lane 1, Fig. 1A; lane 1, Fig. 1B), suggesting that the Shh pathway might be overactivated in SJMM-12 allografts. However, the level of Sfrp1 mRNA in SJMM-12 allografts was much lower than in medulloblastomas and it was not obviously increased compared with cultured SJMM-12 cells (lanes 1 and 9-11, Fig. 1A). Therefore, it was unlikely that the Shh pathway was activated in SJMM-12 allografts, even though Gli1 levels were extremely high. Interestingly, one of five SJMM-1 allografts exhibited Gli1 mRNA expression levels almost identical to that of the tumor (lane 3, Fig. 1A; Table 1). However, the Shh pathway seemed to be inactive in this allograft because Ptc2 and Sfrp1 were not increased (lanes 3-5, Fig. 1A). Thus, the high levels of Gli1 expression observed in some allografts derived from cultured tumor cells may occur independently of the Shh pathway.

To understand the mechanism responsible for the high level of Gli1 expression in SJMM-12 allografts, we investigated the copy number of Gli1. Southern analysis of DNA from cultured SJMM-12 cells (lane 1, Fig. 2A), SJMM-12 allografts (lanes 2 and 3, Fig. 2A), and the matched original tumor (tumor-12; lane 4, Fig. 2A) indicated that a dramatic amplification of the Gli1 locus (∼30-fold) had occurred following transplantation. FISH analyses revealed a homogeneous staining region (hsr) containing Gli1 in the SJMM-12 allograft that was not detected in a SJMM-2 allograft or in medulloblastoma tumor tissue (Fig. 2B). SKY analysis also showed a large hsr on chromosome 10 in all of metaphase cells cultured from the SJMM-12 allograft (right, Fig. 2C). These results clearly indicate that increased Gli1 mRNA in SJMM-12 allografts is a consequence of gene amplification as opposed to the activation of the Shh pathway. The hsr on chromosome 10 preexisted in cultured SJMM-12 cells as it was observed in one of 15 metaphases from the starting material (data not shown), although the majority of cells had no hsr (left, Fig. 2C). It seems that the amplicon containing Gli1 conferred a selective advantage to transplanted SJMM-12 cells as all of the allografts exhibited amplification. Collectively, we did not observe reactivation of the Shh pathway in any of the transplanted tumor cell lines.

In general, xenografts derived directly from patient biopsies retain more morphologic and molecular marker properties of the source tumors than those derived from cell lines (10, 27). Thus, we established allografts by directly transplanting medulloblastoma tumor tissue isolated from Ptc1^+/−p53^−/− mice (hereafter termed direct allografts) to compare their histologic and molecular features with those of the original tumors and with allografts derived from cell cultures. The direct allografts were morphologically similar to the original tumors arising within the mouse cerebellum, which showed histologic features similar to those of human large cell anaplastic medulloblastoma (Fig. 3A). On the other hand, the cells in the SJMM-12 allograft were slightly larger than those seen in medulloblastomas or in direct allografts and they displayed a more spindle-shaped morphology, although they also had some anaplastic features (Fig. 3A).

We compared the response of allografts derived from cell lines and direct allografts to HhAntag. Treatment of mice bearing SJMM-12 allografts with HhAntag had no effect on tumor growth (Fig. 3B) and did not reduce the expression levels of Gli1 or Ptc2 (lanes 7-10, Fig. 3C). Similarly, SJMM-2 allografts also did not respond to HhAntag treatment (Fig. 3B), indicating that the growth of culture-derived allografts is independent of the Shh pathway. Conversely, the direct allografts maintained the molecular signature of Shh pathway activation, characteristic of the original medulloblastomas, they responded to treatment with HhAntag (Fig. 3B) and were eliminated after 2 weeks of treatment. In addition, the mRNA levels of Gli1, Ptc2, and Sfrp1 in tumor cells were clearly reduced after treatment with eight doses of 100 mg kg⁻¹ HhAntag (lanes 3-6, Fig. 3C). Cell proliferation was also decreased after HhAntag treatment of the direct allografts but not after treatment of allografts derived from cell culture (Fig. 3D). These results are consistent with our previous report, indicating that HhAntag inhibits the Shh pathway, suppresses the expression of target genes, and eliminates medulloblastoma in Ptc1^+/−p53^−/− mice (9).

Using the direct allograft system, we also tested the effect of HhAntag on medulloblastomas derived from Ptc1^+/−p53^+/− (6) and Lig4^−/−p53^−/− mice (16). In both cases, HhAntag was very effective at inhibiting allograft growth (Fig. 3B). Furthermore, the ability of these allografts to respond to HhAntag was maintained for up to 3 passages in vivo (data not shown). Thus, the direct allograft system may provide a useful preclinical model to study the activity of Shh pathway antagonists in medulloblastoma.

To investigate the degree to which tumor allografts replicate the original medulloblastomas, we compared gene expression profiles by cDNA microarray analysis. The expression profiles of the Shh pathway genes were slightly different among all of the SJMM cell lines (Supplemental Fig. S1) and allografts (Fig. 1), indicating that each tumor-derived sample has a unique signature, perhaps reflecting the presence of additional gene mutations or variable epigenetic alterations. For this reason, each tumor-derived sample was compared with a matched sample of the original tumor.

The gene expression profiles of direct allografts were very similar to those of the original medulloblastoma with only 18% (tumor-2 versus direct allograft-2) or 14% (tumor-21 versus direct allograft-21) of the probe sets indicating differential expression in each one-chip to one-chip comparison (Supplemental Table S1). Interestingly, the ratio of differentially expressed genes remained relatively similar (13-19%), when we compared the gene expression profile of one medulloblastoma with direct allografts derived from other medulloblastomas (highlighted cells, Supplemental Table S1). Conversely, the gene expression profiles of cultured medulloblastoma cells were appreciably different from those of the original medulloblastomas with ∼40% of the probe sets indicating differential expression (tumor-1 versus SJMM-1 cells and tumor-12 versus SJMM-12 cells, Supplemental Table S1). A number of genes (30-44%) were differentially expressed when cultured tumor cells were propagated in vivo (Supplemental Table S1). The original gene expression profiles were not restored in culture-derived allografts, as the profiles of culture-derived allografts differed from the original medulloblastoma, with 37% (tumor-1 versus SJMM-1 allograft), 31% (tumor-2 versus SJMM-2 allograft), and 35% (tumor-12 versus SJMM-12 allograft) of probe sets indicating differential expression (Supplemental Table S1). These comparisons indicate that direct allografts recapitulated the gene expression profiles of the original medulloblastoma tumors and that both cultured tumor cells and allografts generated from cultured tumor cells were quite distinct from the original tumor. These data strongly support the use of direct medulloblastoma allografts to test the efficacy of Shh pathway inhibitors as well as other agents that target additional signaling pathways that are active in the original tumors.

To highlight the effects of growth environment on the gene expression signatures, we compared profiles among subgroups, including tumors (n = 4), cultured tumor cells (n = 4), culture-derived allografts (n = 7), and direct allografts (n = 3). These comparisons confirmed the observation that the gene expression profiles of direct allografts were very similar to those of medulloblastoma with only 1.8% (∼800) of the probe sets indicating differential expression, in contrast to cultured medulloblastoma cells and culture-derived allografts (Fig. 4A). Interestingly, ∼4,000 of the probe sets were differentially underexpressed in cultured medulloblastoma cells compared with tumors, which is much higher than the number of overexpressed genes (Fig. 4A). This indicates that many genes that are highly expressed in tumors are down-regulated or silenced during cell culture, which is consistent with the data from individual one-chip to one-chip comparisons (Supplemental Table S1).

To obtain insights into the mechanisms responsible for the suppression of the Shh pathway, we compared the top 500 highly significant genes that distinguished medulloblastoma tumors from medulloblastoma cell cultures. Of the 500 genes analyzed, 161 were overexpressed (log ratio > 2) and 172 were underexpressed (log ratio < 2) in medulloblastomas compared with the cultures. An expression heat-map was generated using these 333 probe sets to illustrate the profiles (Fig. 4B). We found that the profiles of the individual culture-derived allografts, as opposed to the individual cell cultures, were quite variable (Fig. 4B), indicating that the allografts had diverged significantly, perhaps as a consequence of selective pressure for allograft growth. An expanded view of the selected region of Fig. 4B, illustrating the genes overexpressed in tumors and direct allografts but down-regulated in both culture and allografts derived from cultured cells, is shown in Fig. 4C. The expression of these genes correlates with Shh pathway activity and they may represent genes required to maintain the activity of Smo in medulloblastoma or target genes of the Shh pathway. Some of the genes in this group, including Atoh1, Nmyc1 (Fig. 4C), and Sfrp1 (Fig. 4D), were previously identified as up-regulated genes in medulloblastoma (25). In support of this idea is the observation that treatment of spontaneous medulloblastomas with HhAntag results in down-regulation of this set of genes (9). However, they may not simply represent targets of Gli1 alone as they were not up-regulated in SJMM-12 allografts, which express very high levels of Gli1 (Figs. 1A and 4D), or in SJMM cells infected with a retrovirus expressing Gli1 (data not shown). In contrast, Ptc2 expression correlated with Gli1 expression both in vivo and in vitro (9).

Genetically defined mouse models of human cancer provide important tools for proof-of-concept studies and for evaluating new therapies (4, 5, 9). However, transplantation cancer models offer several advantages, including the possibility of high-throughput drug testing. Previously, we used Ptc1^+/−p53^−/− mice to show the usefulness of a small-molecule inhibitor (HhAntag) of the Shh pathway for the treatment of medulloblastoma (9). Here, we extend these studies to include allografts, derived either from cultured tumor cells or directly from several mouse medulloblastoma models. We found that cultured tumor cells rapidly lose their dependence on Shh pathway activity. The pathway activity is not restored when such cells are transplanted in vivo because the transplanted tumors are now resistant to treatment with HhAntag. However, direct allografts maintain the histologic features and gene expression profiles of the original medulloblastomas and they respond well to treatment with HhAntag. The ability of allografts to respond to HhAntag was maintained after serial passage, indicating that this model will be useful for preclinical studies of Shh pathway inhibitors and perhaps other anticancer compounds.

The finding that Smo activity could not be restored by transplantation of cultured tumor cells into immunocompromised mice suggests that the tumor cells might have undergone a permanent epigenetic change, irrespective of the culture conditions. This effect occurs very rapidly after one passage in culture, hence, it is not likely to represent genetic selection in vitro. Genetic selection can occur, particularly after transplantation because a rare amplicon containing Gli1 was selected following transplantation of SJMM12 cells. It is possible that the tumor microenvironment, including stromal cells, maintains the ability of Smo to be inactivated by HhAntag, and this influence is lost when cells are placed in a monolayer culture. In the case of direct allografts, the tumor microenvironment may be retained by donor stromal cells for a long enough period to allow host stromal cells to take over this function. Alternatively, extracellular matrix and soluble factors that are present in vivo, but not in cell culture, may be critical for maintaining Smo function. Thus, it is possible that alternative culture conditions (including coculture with stromal cells, modulation of oxygen levels, neurosphere conditions, or culture in the presence of Shh) may allow medulloblastoma cells to maintain the Shh pathway activity in vitro.

The gene expression profiles of signaling pathway genes in direct allografts were very similar to those of the original tumors (Supplemental Table S2). In contrast, the gene expression profiles of tumor cell cultures and their allografts were strikingly different from those of the original medulloblastomas. As expected, cultured tumor cells failed to express many of the Shh pathway target genes (Figs. 1A and 4C and D; Supplemental Fig. S1). Furthermore, many other genes important for tumorigenesis, including mitogen-activated protein kinase and transforming growth factor-β pathway genes were differentially expressed in cultured medulloblastoma cells (Supplemental Table S2), and the expression profiles for almost all of those genes were not restored in culture-derived allografts (data not shown). However, interestingly, the profiles of integrin and matrix metalloproteinase–related genes, the expression levels of which were changed in culture (Supplemental Table S2), were restored in culture-derived allografts (data not shown), suggesting that substrate adhesion in monolayer culture might specifically influence these signaling pathways. Taken together, the expression data suggests that many signaling pathways that could be important for tumorigenesis are altered during the growth of tumor cells in vitro and in vivo. This raises concerns about many of the human tumor xenografts that are currently being used to test anticancer drugs as it is not yet clear if any of these have maintained an active Shh pathway.

As described above, medulloblastoma cultures and culture-derived allograft systems do not replicate the phenotypes of in situ tumors and are inappropriate for evaluating molecularly targeted agents. In the preclinical studies of medulloblastoma (28–30) and other cancers in which the Shh pathway is important for tumor growth (31–34), several tumor cell cultures and xenografts have been employed and characterized based only on their histology and the expression of a few marker genes. However, based on the data presented here, such limited evaluation might not be sufficient as we have shown that the Shh pathway remained suppressed in culture-derived allografts, despite the expression of selected marker genes.

Although HhAntag eliminated allografts in which the Shh pathway was active, it was completely ineffective against allografts that grew independently of the pathway. This suggests that HhAntag may work specifically on a subset of human medulloblastoma tumors in which components of the Shh pathway (including Shh, Ptc1, Smo, and Sufu) have been altered by mutation. Approximately 25% of sporadic human medulloblastomas are associated with mutations in different components of the Shh pathway (3, 35–39). Our comparative analyses using different mouse systems strongly suggest that direct transplant models of human tumors may be most successful as preclinical models and that the molecular signature of the established xenografts must be compared with those of original tumors. We believe that the combined use of genetically altered mice and appropriate transplantation models of human tumors is important for preclinical drug development.

Grant support: Accelerate Brain Cancer Cure, the Pediatric Brain Tumor Foundation, grants PO1-CA096832 and P30-CA21765 from the NIH, and the American Lebanese and Syrian Association Charities.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Christopher Calabrese for help with the allograft experiments, James Dalton, Marc Valentine, and Virginia Valentine for FISH and SKY analyses, members of the Curran Lab for helpful discussions and comments on the manuscript, and Curis and Genentech for providing HhAntag.