|Home | About | Journals | Submit | Contact Us | Français|
The highly aggressive muscle cancer alveolar rhabdomyosarcoma (ARMS) is one of the most common soft tissue sarcoma of childhood, yet the outcome for unresectable and metastatic disease is dismal and unchanged for nearly 3 decades. To better understand the pathogenesis of this disease and to facilitate novel preclinical approaches, we previously developed a conditional mouse model of ARMS by faithfully recapitulating the genetic mutations observed in the human disease, i. e. activation of Pax3:Fkhr fusion gene with either p53 or Cdkn2a inactivation. In this report we show that this model recapitulates the immunohistochemical profile and the rapid progression of the human disease. We demonstrate that Pax3:Fkhr expression increases during late preneoplasia, but that tumor cells undergoing metastasis are under apparent selection for Pax3:Fkhr expression. At a whole genome level, a cross-species gene set enrichment analysis and metagene projection study showed that our mouse model is most similar to human ARMS when compared to other pediatric cancers. We have defined an expression profile conserved between mouse and human ARMS as well as a Pax3:Fkhr signature, including the target gene, SKP2. We further identified 7 “druggable” kinases over-expressed across species. The data affirms the accuracy of this genetically engineered mouse model.
Rhabdomyosarcoma is the most common soft tissue tumor in childhood (1). Pediatric rhabdomyosarcoma can be divided into two major subtypes, embryonal rhabdomyosarcoma (ERMS) and alveolar rhabdomyosarcoma (ARMS) (1). ERMS comprises 50–60% of all rhabdomyosarcoma cases and typically manifests a favorable outcome, while 20–30% of rhabdomyosarcoma are the more aggressive alveolar subtype that is associated with frequent metastasis at the time of initial diagnosis (2). The development of more effective therapies in ARMS, however, has been hampered by a lack of knowledge about basic molecular mechanisms of tumor development. Cytogenetic and molecular studies show that 70–85% of ARMS have balanced chromosomal translocations of t(2;13) or t(1;13), which lead to the formation of chimeric transcription factors consisting of the N-terminal regions of Pax3 or Pax7 fused to the C-terminal region of Fkhr (3). Pax3:Fkhr-positive ARMS is more aggressive than Pax7:Fkhr-positive or fusion-negative ARMS, and thus Pax3:Fkhr-positive ARMS represents the most clinically intractable subset of ARMS (4).
We previously generated a conditional knock-in allele of Pax3:Fkhr in Pax3 locus and established a mouse model of ARMS by simultaneously activating Pax3:Fkhr expression and inactivating p53 or Cdkn2a in Myf6-expressing maturing myofibers (5–7).In the current study, we demonstrate that this model authentically recapitulates the natural history, histological features and genetic features of the human disease, and we demonstrate this model’s utility in understanding aspects of disease progression and therapeutic target identification.
The conditional models of ARMS have been previously described (5). At necropsy, animals were sacrificed by CO2 asphyxiation in accordance with an approved IACUC protocol. Characteristics of mouse tumor and skeletal muscle samples used for microarray and quantitative RT-PCR are described in Supplementary Table S1 and S2.
Quantitative reverse transcription-PCR (qRT-PCR) analyses were performed by a Taqman assay for mouse Pax3:Fkhr expression or by SYBR Green assay (PE Applied Biosystems) for other genes of interest. Primer and probe sequences are shown in Supplementary Table S3 and S4.
Gene expression analysis was performed using Affymetrix Mouse 430A arrays (Affymetrix, Santa Clara, CA). Original CEL files of the mouse ARMS are uploaded in Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/). For human tumors, published data sets of rhabdomyosarcomas (8, 9), juvenile and old skeletal muscles (10), Duchene muscular dystrophy (11) and a series of mesenchymal tumors (12, 13) and pediatric malignancies (14) were used (Supplementary table S5). For mouse tumors, published datasets of osteosarcoma (15) and medulloblastoma (16) were utilized. Methods of microarray analysis including GSEA and metagene analysis are described in Supplementary Methods.
CAT constructs containing SKP2 promoter were described previously (17). The 220bp genomic fragment 49kb 3’ to Skp2 gene was inserted into pGL4.24 vector (Promega). Reporter plasmids were co-transfected with Pax3:Fkhr and p53 into NIH3T3 cells or p53-deficient mouse embryonic fibroblasts (MEFs).
Western blotting was performed as previously described (18). Antibodies against p27Kip1 (C-19), Skp2 (H-435), and Fkhr (C-20) were from Santa Cruz. Pax3 antibody (ab-2) was from Geneka. α-tubulin antibody was from Oncogene.
The mean latency of ARMS development was 110 days with 100% penetrance of ARMS when bi-allelic activation of conditional Pax3:Fkhr allele was combined with homozygous deletion of conditional p53 allele (Figure1A). However, when the mice had homozygous Pax3:Fkhr and heterozygous p53 mutant alleles, or heterozygous Pax3:Fkhr and homozygous p53 mutant allele combinations, tumor incidence was significantly lower than for double homozygous alleles (p<0.001), indicating a mutation dosage effect. As previously described, activation of Pax3:Fkhr was necessary but not sufficient for ARMS development (5). When Pax3:Fkhr allele was combined with conditional Cdkn2a mutation, mice still required bi-allelic activation of both mutations to develop ARMS at 100% penetrance (Figure 1B). There was no significant difference in ARMS development between Pax3:Fkhr-p53 mice and Pax3:Fkhr-Cdkn2a mice (Figure 1C). All ARMS cases were diagnosed by a qualified pathologist based upon histology as well as MyoD and Myogenin immunohistochemistry (Figure1D). To determine the relative contribution of p53 mutation to the development of ARMS, Myf6ICNm/WTPax3P3Fm/P3Fmp53F2-10/F2-10 tumors were compared to Myf6ICNm/WTPax3WT/WTp53F2-10/F2-10 tumors (Supplementary Figure S1). Myf6ICNm/WTPax3WT/WTp53F2-10/F2-10 tumors developed pleomorphic rhabdomyosarcomas, at a much lower frequency than the ARMS seen in Myf6ICNm/WTPax3P3Fm/P3Fmp53F2-10/F2-10 mice (p<0.001, log-rank test). These findings suggest that the tumors from Myf6ICNm/WTPax3P3Fm/P3Fmp53F2-10/F2-10 mice are not caused solely by p53 mutation, but cooperating effects of Pax3:Fkhr and p53 mutation.
The sites of tumors and stages at necropsy are summarized in Supplementary Tables 7A and 7B. Both Pax3:Fkhr, p53 and Pax3:Fkhr, Cdkn2a models developed advanced ARMS tumors although the frequency of distant hematogenous metastasis in Pax3:Fkhr, p53 model was significantly higher than Pax3:Fkhr, Cdkn2a model (chi-square test, p<0.0001). These conditional mouse models showed a predisposition to rapid disease progression including rapid local tumor growth and invasion, regional lymph node involvement, and distant hematogenous metastasis (Figure 2A–D, Supplementary Table S7B). For the latter, microCT demonstrated both macro-metastases as well as alveolar macrophages associated with micro-metastases (Figure 2C, D).
Although expression of Pax3:Fkhr fusion gene is driven by Pax3 promoter in both human ARMS and in our conditional mouse models, promoter activity of Pax3 is predicted to be low in mature myofibers (19). To monitor expression level of Pax3:Fkhr during disease progression, we performed quantitative RT-PCR of Pax3:Fkhr in adult skeletal muscles from wildtype and Myf6ICNm/WTPax3P3Fm/P3Fmp53F2-10/F2-10 mice as well as primary and metastatic ARMS tumors from Myf6ICNm/WTPax3P3Fm/P3Fmp53F2-10/F2-10 mice (Figure 3A).Samples are detailed in Supplementary table S2. As expected, expression of Pax3:Fkhr in Myf6ICNm/WTPax3P3Fm/P3Fmp53F2-10/F2-10 preneoplastic skeletal muscle was low, while Pax3:Fkhr expression was more than 100 fold higher in ARMS tumors. Metastatic tumors expressed Pax3:Fkhr at incrementally higher levels than the primary tumors. Pax3:Fkhr in the mouse tumor tissues were also detected at protein level using anti-Fkhr antibody (Figure 3B). We also performed immunofluorescent analysis of corresponding tissue samples using anti-GFP antibody as an in situ correlate of Pax3:Fkhr expression (Figure 3C). In our mice, Pax3:Fkhr is followed by an internal ribosomal entry site and the eYFP gene, therefore eYFP expression corresponds to transcriptional activation of Pax3:Fkhr in these tissues (5). While eYFP was undetectable in wildtype and preneoplastic adolescent skeletal muscle, primary and metastatic ARMS tumors strongly expressed eYFP. Expression pattern of eYFP in primary tumors was heterogeneous compared to the uniform expression pattern in metastatic tumors; quantitatively, the number of eYFP-expressing cells in metastatic ARMS tumors was higher than primary tumors (90% vs 33%, p<0.001). These results suggest that the level of the transcriptional activation of Pax3:Fkhr is linked to tumor development and progression, and that cooperative cellular events are required in the transformation from preneoplasia to tumor in order to activate Pax3:Fkhr transcription. Later, higher quantitative Pax3:Fkhr levels by RT-PCR for metastatic tumor lesions appears to be attributable to the more uniform expression of Pax3:Fkhr in tumor cells, but not necessarily higher expression in any individual cell.
Gene set enrichment analysis (GSEA) is a computational method for assessing that has been successfully used to assess whether pathways are conserved between zebrafish and human rhabdomyosarcoma (20, 21). For our GSEA, we tested whether the gene sets upregulated in mouse ARMS are enriched in human ARMS when compared to other mesenchymal malignancies. The differentially expressed genes was selected by comparing mouse ARMS to 4-week-old wildtype skeletal muscle for which p-value <0.01. Using published database of human sarcomas (13), we performed GSEA with up-regulated gene sets of mouse ARMS (list is shown in Supplementary Table S8). The gene set up-regulated in mouse ARMS was enriched most significantly in human ARMS among all human mesenchymal malignancies (normalized enrichment score=2.0720, FDR qval<0.001; Supplementary Table S9). ERMS scored lower (normalized enrichment score=1.5773, FDR qval=0.0038). Additional GSEA results using a human rhabdomyosarcoma dataset (9) is given in Supplementary Figure S2.
Tamayo et al recently developed a metagene projection methodology to enable a direct cross-species and cross-platform comparison (22). This method can be used to assess the degree to which mouse ARMS displays a transcriptional profile comparable to other human tumors (15). For the purpose of further investigating whether mouse ARMS shares genetic features of human ARMS, metagene projection analysis was undertaken. To define a metagene for human ARMS compared to other human tumors, we utilized previously published datasets of human mesenchymal tumors (13) and pediatric tumors (14). Although a metagene was defined for each human malignancy, the projected clustering could not entirely separate human ARMS and ERMS (Supplementary Figure S3A). In addition to our 6 cases of mouse ARMS samples, published mouse osteosarcoma samples (15) and mouse medulloblastoma samples (16) were utilized as testing samples. The projected clustering of mouse tumors demonstrated mouse ARMS cluster with human rhabdomyosarcomas (both ARMS and ERMS), and that mouse osteosarcomas and medulloblastomas also clustered with their human counterparts. Another metagene analysis was performed using a well-characterized dataset of human rhabdomyosarcoma (9). Again, however, a defined metagene failed to separate human ARMS and human ERMS completely, instead; 7 out of 22 human ARMS clustered with human ERMS (Supplementary Figure S3B). Hierarchial clustering after metagene projection demonstrated that mouse ARMS cluster with those 7 cases of human ARMS, which confirmed that mouse ARMS recapitulates, at least, a subset of human ARMS cases.
The previous literatures (8, 9, 23) have identified a subset of genes which are specifically overexpressed in human ARMS compared to ERMS. Lae et al (9) compared those gene sets and identified 11 genes that are shared in all of those 3 publications. To further validate that the mouse tumors share the genetic features of human ARMS, expression of those 11 genes as well as Mycn, another representative alveolar specific gene, were examined by quantitative RT-PCR (Figure 4). Among those 12 genes, 9 genes (Ass1, Cnr1, Dcx, Ela1, Foxf1a, Pipox, Tcfap2b, Wscd1, and Mycn) were significantly overexpressed compared to skeletal muscle. Thus, collectively mouse ARMS tumors share a common core expression profile with human ARMS tumors.
To identify a conserved molecular profile of ARMS across species, genes differentially expressed in mouse tumors compared to wildtype skeletal muscle were projected into human rhabdomyosarcoma vs. skeletal muscle. For the human data, published datasets of human young, old, and pathologic skeletal muscle (10, 11) and human rhabdomyosarcoma (8) were used. 1624 genes were differentially expressed in mouse ARMS vs. skeletal muscle (673 genes upregulated and 951 genes downregulated in mouse ARMS; Supplementary Table S8). Among those 1624 genes, 1046 genes (392/673 upregulated genes and 654/951 downregulated genes) were also differentially expressed in human ARMS (p<0.01 by t-test in tumors of both species compared to skeletal muscle; Supplementary Figure S4A). This list may be a mixture of tumor-related and non-tumor related genes, especially knowing that in vivo studies have shown that Pax3:Fkhr not only can cause tumors but also can lead to abnormally developed, disordered (dystrophic) muscle (7, 24). Therefore, we sought to enrich for tumor-specific genes by excluding genes differentially expressed by diseased muscle, thus highlighting 368 genes in the cross-species molecular profile of ARMS (158 upregulated and 210 downregulated; Supplementary Table S8).
Using a different approach, we went on to identify a Pax3:Fkhr molecular signature conserved across species by combining the 1624 mouse genes differentially expressed between mouse ARMS with the set of human Pax3:Fkhr-positive ARMS vs. fusion-negative ARMS (p<0.01, Supplementary Figure S3B, a gene list in Supplementary Table S10). Fifty-six intersecting genes were identified, which may be Pax3:Fkhr direct or indirect transcriptional targets.
Among those genes was SKP2, whose expression has been reported to be upregulated by Pax3:Fkhr (25). The overexpression of SKP2 gene in both mouse and human fusion-positive rhabdomyosarcoma was confirmed by quantitative RT-PCR (Supplementary Figure S3C). To determine whether SKP2 transcription is regulated by Pax3:Fkhr, NIH3T3 cells were infected with a retrovirus carrying Pax3:Fkhr then treated with cycloheximide for up to 8 hours (Figure 5A). Treatment with cycloheximide did not affect SKP2 levels during this time course, suggesting that SKP2 transcription may be directly regulated by Pax3:Fkhr. To further study whether SKP2 is a direct transcriptional target of Pax3:Fkhr, a reporter assay was performed using the SKP2 promoter (Figure 5B). Serially deleted genomic fragments from the SKP2 promoter region (spanning the 3723bp fragment 5’ upstream of SKP2) were tested for the response to Pax3:Fkhr overexpression in NIH3T3 cells. However, the SKP2 promoter fragments did not show a transcriptional response to Pax3:Fkhr, although the SKP2 promoter did respond to E2F1, a known direct transcription activator of SKP2 gene (Figure 5B) (17).
Whereas the proximal 3.7kb SKP2 promoter had no activity in response to Pax3:Fkhr, we speculated that Pax3:Fkhr may be upregulating SKP2 through a cis-element. In keeping with this hypothesis, Barber et al. reported from a chromatin immunoprecipitation screen that Pax3:Fkhr can bind to a 220bp-genomic fragment which is 49kb downstream (3’) to the SKP2 gene transcription initiation site (Supplementary Figure S5) (26). The distance, albeit long, is not unprecedented for genes involved in myogenic programming (27).This potential cis-element is conserved across species (Supplementary Figure S5A). A reporter assay using this 220bp genomic fragment demonstrated increased luciferase activity when NIH3T3 cells or p53-deficient MEFs were cotransfected with Pax3:Fkhr, and like the PDGFRA reporter control (18), p53 may antagonize Pax3:Fkhr-mediated transcriptional activation of the SKP2 cis-element depending upon the cellular context (antagonism was seen in NIH3T3 cells, but not in p53-deficient MEFs, Figure 5C). Thus, this cis-element may be at least one site by which Pax3:Fkhr regulates SKP2. A definite link between this Pax3:Fkhr responsive element and transcription of the SKP2 gene will likely require future generation of new transgenic animals.
To determine the relevance of SKP2 upregulation by Pax3:Fkhr, we performed functional studies in human ARMS cells. SKP2 has been reported to be involved in cell cycle-dependent control of p27kip1 ubiqitination and thus cell cycle entry/tumor cell growth. To determine whether SKP2 repression can affect the cell growth, the human ARMS cell line Rh30 was stably transfected with SKP2-specific short hairpin RNA (shRNA) as described previously (28) (Figure 5D). Increased protein level of p27kip1, as well as reduced expression of SKP2, was confirmed in SKP2-shRNA cells by Western blotting. Rh30 cells infected with SKP2-shRNA showed substantially reduced cell growth compared to control-shRNA cells. This effect was also confirmed in mouse ARMS cells, derived from a Myf6ICNm/WTPax3P3Fm/P3Fmp53F2-10/F2-10 tumor (Supplementary Figure S5B). Collectively, these data indicate that SKP2 is a potential transcriptional target of Pax3:Fkhr via a 3’ cis-element and that SKP2 plays a major role in cell proliferation of ARMS. More broadly, these results suggest that the mouse model of ARMS can serve to identify a Pax3:Fkhr molecular signature and Pax3:Fkhr target genes conserved across species.
This mouse model was previously used to validate a receptor tyrosine kinase, PDGFRA, as a direct transcriptional target of Pax3:Fkhr as well as therapeutic target (18). To identify other potential druggable targets in ARMS, we selected a subset of protein kinase genes that were up-regulated in both mouse and human ARMS tumors (Figure 6A). Among 19 protein kinases up-regulated in mouse tumors, up-regulation of 16 kinases was conserved in human ARMS. From this set, kinase inhibitors are available against 7 genes including VRK1, AURKB, PLK2, PLK4, CDK4, CHEK1, and TK1 (29–31). Overexpression of these kinases was confirmed by quantitative RT-PCR in a larger set of mouse tumors (Figure 6B). These results validate the future use of this mouse model as a preclinical tool for the study of therapeutic kinase inhibitor strategies in ARMS.
In this paper, we present a cross-species validation of a genetically-engineered mouse model of ARMS. The implicit advantage of using conditional genetic models for preclinical therapeutic testing are that tumors arise in an authentic microenvironment, i.e. skeletal muscle, and that the immune system is intact. The latter may be especially important for the promising cadre of monoclonal antibodies, for which antibody-dependent cellular cytotoxicity may require immunocompetence (32)
Our study shows that this ARMS model is advantageous for preclinical therapeutics for several reasons. We show that the Pax3:Fkhr, p53 model has 100% penetrance by 150 days (young adulthood in a mouse) with a spectrum of disease sites that are comparable to human rhabdomyosarcoma. Histology and immunohistochemical markers also mimic the human disease, as reported here and previously (5, 7). Furthermore, the progression of disease in terms of primary tumor growth and extent of disease are as rapid as or more rapid than the human disease, making the model useful for understanding the underlying disease mechanisms that allow unresectable or metastatic rhabdomyosarcoma to elude therapy.
We demonstrate at a cellular level that cooperative factors other than the Pax3:Fkhr fusion or p53 inactivation are likely to be responsible for Pax3:Fkhr transcriptional regulation in preneoplastic muscle. However, once the primary tumor has formed, tumor cells that metastasize appear to be under selection for Pax3:Fkhr expression. Whereas targeting transcription factors such as Pax3:Fkhr is therapeutically challenging, one can hope that cooperative factors that facilitate high Pax3:Fkhr transcription might include cell surface receptors or proteins sensitive to small molecule inhibitors. The identification of these cooperative factors which modulate Pax3:Fkhr expression is the subject of ongoing studies.
To validate our model on a whole genome basis, we performed a cross-species gene expression analysis. Gene set enrichment analysis confirmed that our model is most related to human ARMS amongst a variety of human sarcomas. We also performed metagene projection. This powerful method of cross-species, cross-platform analysis (22) has been used recently to compare mouse and human pediatric cancer models amongst a variety of cancer subtypes. However, this method warrants some caution because results are dependent upon a training set with a large, homogeneous collection of each tumor subtype. For rhabdomyosarcomas, which are relatively rare, sample size has been problematic in other studies (15). Nevertheless, we were able to demonstrate that in comparison to other pediatric cancers, our mouse model is most similar to human rhabdomyosarcomas and specifically human ARMS. We found, however, that despite using the best available microarray dataset for rhabdomyosarcoma subtypes, metagene analysis could no better separate human Pax:Fkhr-positive ARMS from ERMS than the original report for this dataset (9). This result may be due to a technical limitation of this approach and small sample size or may suggest that ARMS and ERMS (as defined by histology) may be a continuous spectrum of disease. This later possibility taken in a positive light suggests that rhabdomyosarcomas might still be further subclassified on molecular criteria beyond, or in addition to, Pax:Fkhr fusion status and histology.
We went on to identify 368 tumor-specific genes in common between mouse and human tumors that could neither be explained as being related to a normal muscle or degenerative muscle phenotype. Next, we employed this genetic model to identify potential downstream targets of Pax3:Fkhr. Identifying Pax3:Fkhr targets has been the subject of numerous antecedent studies using many different valid approaches (ie., transfection of rhabdomyosarcoma or non-rhabdomyosarcoma cells with Pax3:Fkhr, comparison between primary tumors, or combinations thereof) (23, 26, 32–37). Our approach is meant only to be complementary. In the end, primary human tumor samples (Pax3:Fkhr-positive vs. fusion-negative ARMS) are the definitive study set for such determinations, but getting large numbers of high quality rapidly-processed samples of these rare tumors has been a challenge for the field. Nevertheless, our cross-species approach identified 56 candidate target genes of Pax3:Fkhr, including SKP2 (Figure 5A-right). SKP2 has been suggested to be a target gene of Pax3:Fkhr but not Pax3 in fibroblasts (25). We have extended this result by validating SKP2 as a Pax3:Fkhr target in vivo. SKP2 is a component of the SCF (SKP1–CUL1–F-box) protein complex that mediates the ubiquitination and proteasomal degradation of cell cycle regulatory genes including p27 (25, 38), thereby accelerating cell cycle progression. Ironically, SKP2 also interacts with and promotes the ubiquitin-mediated degradation of Fkhr (FoxO1A)(39). This SKP2-mediated degradation of Fkhr requires phosphorylation of Fkhr at Ser-256 (39), which is in fact retained by Pax3:Fkhr (40). Interestingly, Fkhr Ser-256 phosphorylation also reduces binding of Fkhr to DNA, and causes nuclear exclusion of Fkhr when Thr-24 and Ser-319 are also phosphorylated (41). The extent to which the phosphorylation of this serine residue in Pax3:Fkhr can be enforced to take advantage of SKP2 over-expression, SCF-mediated degradation and Pax3:Fkhr nuclear exclusion is the topic of ongoing investigation.
In order to identify new therapeutic targets, we examined the expression of potentially "druggable" kinases. The range of available kinase inhibitors is growing rapidly, and therefore we examined the cross-species rhabdomyosarcoma expression of kinases known to have an inhibitor available preclinically or clinically. We identified seven kinases, including an aurora kinase and 2 polo-like kinases.
For all of the strengths of this five allele genetically engineered model (more alleles if you include reporter genes for non-invasive imaging), significant infrastructure investments are required to maintain this disease model system. Because tumors can arise from deep sites, specialized small animal imaging technology is necessary (42) since traditional measurement with calipers at the skin surface nearly always under-estimates the extent of disease. Luciferase has been suprisingly non-informative in our model system because tumors have a tendency to be centrally hypovascular and hypoxic (42), thereby unable to have access to the oxygen required by luciferase (unpublished result). The financial investment in maintaining mouse stock lines, husbandry and genotyping is also non-trivial; therefore, alternative models such as very successful rhabdomyosarcoma xenograft systems (43) and a recently-reported ectopic allograft model (44) are warranted options to our transgenic model. In some instances, certain targets identified from human tumors are not expressed in the cell lines used for xenografts (18). In these cases, the genetically-engineered model may be not only essential but also extremely productive. Our laboratory recently identified PDGFRA as a potential therapeutic target from the study of the ARMS preclinical model we report here (18). To follow this example and to make our model more practical for widespread use, we will be soon participating in the NCI Pediatric Preclinical Testing Program (45, 46), with the intent of examining efficacy of novel targeted therapies. We will also be providing preclinical testing for outside investigators on a high volume, low cost basis. In this cooperative framework, the outlook for new therapies in ARMS may be significantly improved.
This work was funded by Bradley J. Breidinger Memorial Research Award from the Sarcoma Foundation of America to C.K., by NIH grant CA074907 to C.W., by NIH grant CA64202 to F.B., by an Alex’s Lemonade Stand Foundation Grant to K.N., and by the Scott Carter Foundation to C.K. and K.N. C.K. is a member of the Clinical Trial Research Center (P30CA54174). Rh30 was graciously provided by Dr. Peter Houghton. We thank Drs. Louis Kunkel and Peter B. Kang for datasets and kind review of this manuscript.
Disclosure of Potential Conflicts of Interest
C.K. is co-founder of Numira Biosciences, which has licensed micro-CT-based Virtual Histology from UTHSCSA.