Osteosarcoma PDX-Derived Cell Line Models for Preclinical Drug Evaluation Demonstrate Metastasis Inhibition by Dinaciclib through a Genome-Targeted Approach

Osteosarcoma is a rare malignant bone tumor most commonly diagnosed in children, adolescents, and young adults. The majority of patients have micrometastases at diagnosis, requiring the use of intensive adjuvant chemotherapy (1–3). Despite aggressive treatment, progression often presents as distant macrometastatic disease in the lungs or bone, and patients who develop metastases have a 5-year survival rate of 20% to 30% (3–7). Over the past several decades, there has been limited progress in identifying new therapies for patients with osteosarcoma and survival has not substantively improved, particularly for patients with metastatic disease (7–9). This lack of progress may be partly due to a paucity of models that accurately recapitulate osteosarcoma metastatic progression. Osteosarcoma genomes are characterized by multiple copy-number alterations, chromothripsis, and aneuploidy, resulting in significant heterogeneity in oncogenic drivers. Well-characterized models are crucial not only for elucidating mechanisms of metastasis but also for identification of new therapeutic strategies (10, 11).

While in vitro assays are often used to predict or inform on the metastatic capacity of neoplastic cells (e.g., invasion, migration, anoikis, etc.), in vivo modeling remains the gold standard to fully evaluate the metastatic cascade. Experimental metastasis models where cancer cells are disseminated after intravenous injection recapitulate the latter half of the metastatic cascade, including survival within the bloodstream, extravasation, tissue colonization, and tumor establishment. However, to model the full metastatic cascade, models that metastasize spontaneously from a relevant orthotopic site are required.

Only a limited number of established osteosarcoma cell lines have demonstrated the ability to consistently form metastases when injected intravenously (refs. 12–20; Supplementary Table S1). To develop additional models, prior work has sought to increase metastatic capacity by exposure to carcinogens, introduction of oncogenes, or serial passaging of established osteosarcoma cell lines through mice (14, 15, 21–27). These strategies have successfully enhanced the metastatic capacity of several osteosarcoma cell lines in both intravenous and orthotopic models (12, 14). However, these approaches may introduce alterations that are not reflected in endogenous human metastasis. Furthermore, murine orthotopic osteosarcoma models are most commonly generated by implanting tumor cells/tissue into the epiphyseal bone or onto the periosteal surface of the tibia (12, 16, 33–41, 19, 20, 27–32). However, injecting tumor cells/tissue directly into the bone can lead to immediate exit through the vasculature, essentially skipping the initial stages of the metastatic cascade and seeding the lungs directly (42–45). The periosteal implantation strategy to model spontaneous metastasis requires tumor cells to invade through the cortex without risk of inadvertently introducing cells into the vasculature at implantation. Models incorporating amputation of the affected limb, provide additional time for the development of metastases in the lungs or other organs (12, 16, 29, 32, 41). Because surgical removal of the primary tumor is a component of the standard of care for patients with osteosarcoma, inclusion of primary tumor removal in a mouse model more closely recapitulates management of the human disease.

The use of patient-derived xenografts (PDX) to model cancer enhances the probability of retaining key features of the human disease (11, 42, 46–49). However, only a limited number of osteosarcoma PDXs have been evaluated for metastatic potential in vivo (31, 41, 42, 50, 51). Cell lines generated from PDXs may also be useful to establish more consistent and more tractable models for the exploration of molecular pathogenesis and for testing novel therapeutic drugs for pediatric osteosarcoma compared with established commercially available cell lines, many of which have been in culture for decades (52).

To enable the development of new models to study osteosarcoma metastasis, we generated PDXs from clinically annotated patient tumors. Cell lines were then generated from the PDXs and rigorously characterized in vivo using both an intravenous and an orthotopic model (Fig. 1). Whole-genome sequencing (WGS) and RNA sequencing (RNA-seq) were used to evaluate consistency between the PDXs and the patient tumor from which they were derived. Here, we describe the in vivo features of these novel PDX-derived cell lines including tumorigenicity and metastatic capacity in both tail vein intravenous and paratibial orthotopic amputation models. Bioluminescence imaging (BLI) was used to quantify whole body metastatic burden, while IHC was used to aid quantification of metastatic tumor burden in the lung and liver. To demonstrate the utility of PDX-derived cell lines for preclinical testing, we performed a proof-of-concept study which showed that a cyclin-dependent kinase (CDK) inhibitor decreased metastatic tumor growth of a MYC-amplified cell line after amputation of an orthotopic xenograft. We expect that these PDX-derived cell lines will be broadly useful to researchers investigating mechanisms of osteosarcoma metastasis as well as the preclinical development of therapies to treat metastatic disease.

All samples were reviewed by a pathologist (F.K. Hazard or S.J. Cho) and the diagnosis was confirmed as osteosarcoma. Samples were received fresh and were grossly evaluated by a pathologist (F.K. Hazard or S.J. Cho) for viable tumor tissue. A representative sample was reserved for PDX implantation and the remainder of the sample was flash-frozen for DNA/RNA extraction.

For PDX generation, 1 mm³ fragments of fresh or frozen-thawed patient tumor were dipped in Matrigel (Corning) and implanted under the renal capsule of NSG mice. Mice were monitored for tumor growth by abdominal palpation for up to 1 year postimplantation. Successfully engrafted tumors were allowed to reach 1 to 2 cm³ before the mouse was euthanized and tumor harvested. Tumors were digested in a collagenase buffer and filtered through a 70-μm filter. For PDX passaging, cells were implanted subcutaneously in the flank of NSG mice (5×10⁵ cells) in 30 μL of MEM alpha and 20 μL Matrigel. PDX-derived cell lines were generated as described previously (11).

Flash-frozen patient tumors, PDXs, and PDX-derived cell lines were placed into 1.5-mL tubes on ice and disrupted with a plastic tissue homogenizer. Demineralization was not required. Tumors were digested to a single cell suspension in collagenase digest buffer and sorted for human cells using anti-human HLA-ABC (BioLegend, 555555). DNA and RNA were extracted using an AllPrep Kit (Qiagen, 80204) with QIAshredder (Qiagen, 79654). RNA was quantified with a Nanodrop 2000c (Thermo Fisher) and RNA integrity was analyzed on an Agilent 5200 Fragment Analyzer. RNA-seq was performed as described previously (11).

Fastq files and counts were derived using STAR (2.5.3a; RRID:SCR_004463) with the hg38 index (53). Normalized values were calculated using the edgeR package (RRID:SCR_012802) as log CPM with prior count set to 2 (54). ComBat (RRID:SCR_010974) was then used as a batch correction for the sample source (patient, PDX, or cell line; ref. 55). The top 80th percentile most variant genes (based on coefficient of variance) were used for hierarchical clustering using the Manhattan method for distance followed by the Ward.D2 agglomeration method.

Log₂ [transcript per million (TPM) + 1] values were derived from hg38 human index using the UCSC Toil RNA-seq workflow (56) and copy-number assignments (described under Whole-genome sequencing and allele-specific copy-number analysis). Groups were classified as High MYC and Low MYC based on their MYC expression: values exceeding 50 percentile were designated as High MYC, while those below 50 percentile were considered Low MYC. Differentially expressed genes were identified utilizing the limma-voom linear model, incorporating empirical Bayes moderation (57). Genes meeting the criteria of a P value < 0.05, abs [log fold change (FC)] > 1, and a FDR < 0.05 were considered as significantly expressed.

Gene set enrichment analyses were performed using the fgsea R package (58), based on pathways downloaded from MSigDB (RRID:SCR_016863). Genes were ranked according to their logFC between High MYC and Low MYC, and a running-sum statistic was used to calculate enrichment scores for each gene set/pathway. These scores were then normalized to account for the size of each set, resulting in normalized enrichment scores (NES). P values and FDR were derived from a permutation test to assess the statistical significance of the NES.

TARGET (from Genomic Data Commons Data Portal RRID:SCR_014514) and Cancer Cell Line Encyclopedia (CCLE; RRID:SCR_013836) TPM data were downloaded from the UCSC Treehouse Public Data repository (https://treehousegenomics.soe.ucsc.edu/public-data/). For the TARGET dataset, Tumor Compendium v10 Public PolyA TPM expression and clinical datasets released in July 2019 were used and were filtered for osteosarcoma TARGET samples. For CCLE, the Cell Line Compendium v2 TPM expression and clinical datasets released in December 2019 were downloaded and filtered for osteosarcoma cell lines. Patient tumor samples from which cell lines were also derived were excluded from the analysis.

The samples from TARGET, CCLE, and our dataset were processed using the UCSC Toil RNA-seq pipeline (https://github.com/BD2KGenomics/toil-rnaseq) and the RSEM TPM values were used in the correlation analysis. For CCLE and our cell line data, the Toil pipeline was used in-house to generate RSEM gene-level Hugo counts for differential gene expression analysis (56).

The correlation analysis was performed as described by Yu and colleagues (59). Briefly, the 5,000 most variable protein-coding genes by interquartile range (IQR) were selected across the osteosarcoma patient samples. These genes were used to calculate the Spearman's correlation coefficient for each patient sample (TARGET and Sweet-Cordero lab) compared with each cell line (CCLE and PDX-derived cell lines). Patient-to-cell line correlations for CCLE cell lines versus PDX-derived cell lines were compared using the Wilcoxon rank-sum test.

A list of clinically actionable genes was generated by combining the gene lists from several cancer-related gene panels: FoundationOne and FoundationOne Heme (combined, Foundation Medicine, list accessed May 31, 2017), MSK-IMPACT (60), Mi-oncoseq (61), and UCSF 500 cancer gene panel (http://cancer.ucsf.edu/research/molecular-oncology/ucsf500). Genes present in more than one panel were included in an “actionable” gene list. Actionable genes that also had a drug interaction listed in the Drug Gene Interaction Database (DGIdb; accessed June 13, 2018; RRID:SCR_006608; ref. 62) were included in the actionable/druggable gene list, and this gene list was used to identify recurrent somatic copy number. Alignment of FASTQ files, somatic variant calling, and structural variant calling was performed as previously described (11). Copy-number analysis and purity correction was also performed as previously described (11), with the difference of allele-specific copy number being calculated using the B-allele frequency (BAF) based upon germline variant allele frequency (VAF). For samples without germline data available, population-level common SNPs were obtained from dbSNP and used as an alternative to patient-specific germline variants. Concordance of copy number across the genome was calculated by tallying the fraction of the genome with the same integer copy-number value between related samples. Concordance was calculated separately for major and minor alleles. Major copy-number gain is defined as >8 copies and minor is defined as 4 to 8 copies. For copy-number losses, <1.2 copies defines minor and <0.8 copies defines major.

Single-nucleotide variants and small-scale insertions/deletions were called using Mutect2 (GATK v3.8; RRID:SCR_001876; refs. 63, 64) and Strelka (v2.9.10; RRID:SCR_005109; ref. 65). To filter out low-confidence calls, only the variants having at least four supporting reads and called by both Mutect2 and Strelka were retained. In addition, a variant was considered a true positive when the VAF was greater than 5%, the sequencing depth was ≥20, and there were ≥3 sequence reads supporting the variant call. Somatic structural variants were called from WGS data by combining the output of Manta (v1.6.0; RRID:SCR_022997; ref. 66), DELLY (v0.8.1; RRID:SCR_004603; ref. 67), GRIDSS (v2.9.4; ref. 68), and SvABA (v.1.1.3; RRID:SCR_022998; ref. 69). Consensus calls were made by comparing the output of GRIDSS and the other three tools, with a maximum allowed distance of 100 bp as measured pairwise between breakpoints. Two approaches were used to estimate the copy-number profile, ploidy, and purity of a tumor sample from matched normal and tumor WGS data. Approach one implements PURPLE (v3.0; RRID:SCR_022999) which combines BAF from AMBER (v3.3; RRID:SCR_016151), read depth ratios from COBALT (v1.8; RRID:SCR_004152), somatic variant and structural variant calls. Approach two implements alleleCount (v4.2.1) (https://github.com/cancerit/alleleCount) for allele counting and ascatNgs (70, 71) for estimating tumor purity, ploidy, and copy-number profile. Manual inspections were taken when there was a discrepancy between the two approaches. Circos plots were generated using PURPLE (v3.0) to plot final somatic copy-number calls, point mutation calls, and structural variation calls.

Mycoplasma testing and STR analysis (Supplementary Table S2) were performed on cell lines prior to in vitro and in vivo use. Eight PDX-derived cell lines (Table 1) and three established commercially available cell lines were used for the in vivo studies. SJSA (RRID:CVCL_1697) and MG63 (RRID:CVCL_0426) were obtained from the American Type Culture Collection. MG63.3 (RRID:CVCL_WL01) was a gift from Dr. Robbie Majzner (Stanford University). All cells were maintained in a 37°C 5% CO₂ humidified incubator and cultured in standard DMEM media (Corning) supplemented with 10% bovine growth serum (Hyclone SH30541.03) and 1% penicillin–streptomycin–glutamine. Stable luciferase-expressing cell lines were generated using lentiviral transduction (RRID:Addgene_39196) followed by sorting for GFP using a Sony Biotechnology SH800S Cell Sorter (RRID:SCR_018066).

For the cell proliferation assay, cells were seeded in a 96-well plate and allowed to attach overnight. Plates were transferred into an Incucyte S3 Live-Cell Analysis System (Sartorius, RRID:SCR_023147), and whole-well images were collected every 8 hours until wells reached 100% confluence. Incucyte software (Incucyte S3 2019A) was used to create confluence masks for each cell line. Confluence over time was used to identify a period of exponential cell growth and the linear portion of the curve was used to calculate doubling time in R statistical analysis software (72).

For each cell line, a minimum of 4 mice were used for establishment of model endpoint timing. Luciferase-expressing cell lines were used for all PDX-derived cell lines and SJSA. Both parental and luciferase-expressing cell lines were used for a subset of cell lines (OS052, OS152, OS186, and OS526). Only the parental version was used for MG63 and MG63.3. For experimental metastasis studies, 1×10⁶ cells were injected via the lateral tail vein in 100 μL of phosphate buffered saline (PBS); 1×10⁵ cells were used for SJSA. For luciferase-expressing cell lines, BLI was performed every 28 days to monitor metastatic progression using a Perkin Elmer IVIS Spectrum In-Vivo Imaging System (RRID:SCR_020397). On the basis of worsening clinical signs associated with metastases and/or BLI signal intensity, an endpoint for each cell line model was determined.

To model spontaneous metastasis, paratibial orthotopic implantation followed by limb amputation was employed. At implantation, the periosteum of the medial aspect of the mid-tibia was gently scored with a 30-gauge needle prior to injecting 1×10⁶ cells, 1×10⁵ for SJSA, in 10 μL of Matrigel. After initial xenograft detection, tumor growth was monitored with twice-weekly measurements until the limb diameter reached 1 cm, when the affected limb was amputated by coxofemoral disarticulation. Prior to amputation, limb and lung microCTs were performed using the Perkin Elmer Quantum GX2 microCT Imaging System; a pulmonary microCT was also performed prior to euthanasia. If no xenograft could be identified within 6 months of implantation, the cell line was deemed non-tumorigenic. Amputated limbs were demineralized in Immunocal (StatLab) for 24 hours prior to longitudinal bisection for histologic processing. An endpoint for each cell line was determined as it was for the intravenous model. Mice were euthanized 1 to 6 months, cell line dependent, post-injection for the intravenous model or postamputation for the orthotopic model.

For the dinaciclib study, luciferase-expressing OS152 cells were implanted orthotopically, as described above. BLI was performed initially 3 days postamputation and dinaciclib treatment commenced the following day. All mice received treatment for 3 weeks at 20 mg/kg 5 times/week intraperitoneally, and BLI was performed every 7 days. Dinaciclib powder (MedChemExpress) was dissolved in DMSO to 50 mg/mL for stock solution. For working solution, the stock was diluted to 2.5 mg/mL with 20% beta cyclodextrin in PBS immediately before injection. Mice were weighed weekly. BLI signal was tracked over time using Living Image software (Perkin Elmer; RRID:SCR_014247). BLI signal at endpoint was log-transformed and means were compared between the vehicle and control groups with a two-tailed t test, after testing for normality. Differences were considered significant when P < 0.05.

For all in vivo studies, lungs, liver, and tissue from additional metastatic sites were harvested for histologic evaluation and metastatic burden analysis. Prior to collection, 5 mL of PBS was injected into the right ventricle of the heart and the lungs were inflated with 1 mL of formalin via the trachea. Tissues were fixed for 24 to 48 hours in formalin and then transferred to 70% ethanol prior to histologic processing.

IHC was performed on formalin-fixed, paraffin-embedded tissue mounted on glass slides. Standard heat-induced epitope retrieval using sodium citrate buffer (pH 6.0) was performed prior to H₂O₂ quenching of endogenous peroxidase activity. Blocking and antibody dilutions were made in 5% normal horse serum (Vector Labs), and slides were incubated overnight at 4°C in a humidified chamber with anti-human mitochondria antibody (1:1000; Abcam, catalog no. ab92824, RRID:AB_10562769). Slides were incubated in secondary horse anti-mouse antibody (Vector Labs) and avidin-biotin (Vectastain Elite ABC kit) for 30 minutes each at room temperature. Signal was developed with a DAB substrate kit (Abcam, catalog no. ab64238).

Histology slides were scanned by the UCSF Histology & Biomarker Core at 10x using the Zeiss Axio Scan.Z1 Digital Slide Scanner (RRID:SCR_020927). Scanned slides were imported into QuPath Quantitative Pathology & Bioimage Analysis software (RRID:SCR_018257). Tissue was outlined using simple tissue detection prior to manual correction to exclude shadows, debris, and other tissue regions not meant for analysis. Training images were created from multiple regions across a selection of slides and were used to train the pixel classifier to identify DAB positive tissue. The pixel classifier was used to create tumor annotations across all slides with a minimum object size of 250 μm². All annotations were checked manually for accuracy prior to quantification of tumor burden.

Metastatic burden was compared between the treatment and control groups with a two-tailed t test for normally distributed data and the Mann–Whitney test for non-normally distributed data. Differences were considered significant when P < 0.05.

Written informed consent was collected from all patients or parents in the case of patients less than 18 years old. Using established guidelines (Belmont Report), institutional review board approval was granted from each participating institution (73). All procedures involving mice were conducted in accordance with protocols approved by the UCSF Institutional Animal Care and Use Committee.

The sequence data generated by the authors and analyzed in this study are publicly available in database of Genotypes and Phenotypes (dbGAP) at phs002430. Other sequence data analyzed in this study were obtained from TARGET and CCLE at https://treehousegenomics.soe.ucsc.edu/public-data/#tumor_v10_polyA and https://treehousegenomics.soe.ucsc.edu/public-data/previous- compendia.html#cell_line_v2.

Tumor samples obtained from diagnostic biopsies, post-chemotherapy primary resections, or metastases, were collected directly from patients at the time of surgery with institutional review board approval and appropriate consent. Tumor fragments were implanted into the subrenal capsule of NSG mice to generate PDXs. PDXs retain the morphological features of their respective patient tumor based on histology (11), even over multiple passages (Supplementary Fig. S1). From these PDXs, we generated eleven PDX-derived cell lines. Of these cell lines, three were obtained from pretreatment biopsies, three from posttreatment resections, and four from distant metastasis (Table 1).

Where feasible, fresh-frozen tumor samples were analyzed by WGS and RNA-seq. WGS and RNA-seq was also performed for all PDXs and PDX-derived cell lines (Figs. 2 and 3; Supplementary Table S3). Unsupervised hierarchical clustering of RNA-seq data demonstrated remarkable similarity of the patient tumor to the corresponding PDX and PDX-derived cell line (Fig. 3A). In contrast, most established commercially available osteosarcoma cell lines (HOS, 143B, SJSA, U2OS, MG63, and MG63.3) clustered separately from our cohort of patient samples, PDXs, and PDX-derived cell lines (Fig. 3A). Thus, over decades of in vitro passaging, these established commercially available cell lines may have drifted away from patient samples, providing support for the hypothesis that newly derived PDX-derived cell lines retain key features of the gene expression programs of primary tumors. Next, we compared the gene expression profiles of PDX-derived cell lines and established commercially available cell lines to patient primary tumor samples from publicly available data (Fig. 3B and C). Overall, the PDX-derived cell lines had a higher correlation to human osteosarcoma patient samples that was statistically significant (Supplementary Fig. S2A). Across all cell lines tested, the three with the highest mean correlation to human osteosarcoma patient samples were PDX-derived cell lines (Fig. 3B).

Osteosarcoma is genomically complex, with widespread aneuploidy and structural rearrangements (7, 74). We previously described comparison of a limited panel of osteosarcoma PDXs to their corresponding tumor of origin (11). Here we sought to characterize the changes that occurred at the genomic level when comparing this new collection of PDX-derived cell lines to their matched patient and PDX samples. We compared the copy-number profile of patient tumors and their corresponding PDXs and PDX-derived cell lines. Overall, we found a high degree of concordance of copy-number profiles as identified by WGS (Supplementary Fig. S2B and S2C). In two of the cell lines tested in vivo (OS457 and OS186), whole-genome duplication (WGD) led to a wider degree of difference compared with the tumor of origin as quantified by the percentage of genome segments overlapping with regards to copy number (Supplementary Fig. S2B). In one case (OS457), the WGD appears to have occurred in the establishment of the PDX (Supplementary Fig. S2B). In the second case (OS186), the WGD occurred sometime after the establishment of the PDX-derived cell line (Supplementary Fig. S2B). Similarly, Circos plots demonstrated high similarity in the structural rearrangements between patient, PDX, and PDX-derived cell line (Supplementary Fig. S3).

Next, we evaluated the metastatic capacity of the PDXs in vivo. Differences in metastatic potential was assessed using several different methods. Lungs from a subset of mice harboring subcutaneous xenografts, during PDX generation and passaging, were evaluated for the presence of spontaneous metastasis. We also characterized the ability of a subset of these PDXs to form metastases after intravenous injection. Lungs were collected at the time of PDX passaging for mice with subcutaneous PDX implants and 28 days post-injection for mice injected with PDX cells intravenously. Evidence of metastasis was evaluated histologically revealing the presence of both micro- and macro-metastases (Supplementary Fig. S4A–S4C). Of the PDXs evaluated for spontaneous metastasis from subcutaneous tumors, eight of eleven demonstrated the ability to form lung metastasis (Supplementary Fig. S4B). When injected intravenously, three of five PDXs induced lung metastasis (Supplementary Fig. S4C). We observed some discordance in the ability of certain PDXs to form metastasis during subcutaneous passaging versus in the intravenous model. For example, OS128 induced metastasis in the intravenous model but not during subcutaneous passaging, while OS152 and OS186 metastasized during subcutaneous passaging but not after intravenous administration (Supplementary Fig. S4B and S4C).

As we noted a high degree of correlation at the level of transcription between the PDX and the PDX-derived cell lines (Fig. 3A), we reasoned that the latter would be a more tractable model system for metastatic studies. To assess the metastatic capacity of osteosarcoma PDX-derived cell lines in vivo, PDX-derived cell lines were transduced with a lentiviral vector expressing luciferase to facilitate dynamic monitoring of metastatic burden in NSG mice. We first determined that proliferation rates between luciferase-expressing PDX-derived cell lines and the parental version were similar (Supplementary Fig. S5A). Next, cells were introduced via the lateral tail vein and mice were monitored with BLI to estimate whole body metastatic burden (Fig. 4A). Most (6/8) osteosarcoma PDX-derived cell lines tested generated metastases with increasing BLI signal over time (Fig. 4B and C; Supplementary Fig. S6A–S6H). Study duration for each cell line ranged from one to a maximum of 6 months and exhibited a trend of being inversely correlated with BLI signal at endpoint (Table 2 and Fig. 4B and C). A subset of the parental (non–luciferase-expressing) PDX-derived cell lines were also evaluated using the intravenous model (Supplementary Fig. S5B–S5E and Supplementary Table S4).

A wide range of tropism with regards to metastatic sites was observed (Table 2). All six metastatic PDX-derived cell lines induced pulmonary metastases: 5/6 metastasized consistently to lung in every mouse and 1/6 did so frequently (67%; Table 2 and Fig. 4D–F). Lung metastases tended to be more frequent and larger in size toward the pleural surface (Fig. 4F). Four cell lines metastasized to bone (spine, sternum, rib, skull, and long bone) in 20% to 100% of mice per PDX-derived cell line. Lung and bone are the two most common sites of metastasis in patients with osteosarcoma. Liver metastasis (which is uncommon in patients) was also observed for three PDX-derived cell lines in the intravenous model (Supplementary Fig. S7A–S7C), although never in isolation. The non-luciferase versions that were tested metastasized to the same sites as BLI enabled lines with minor differences in location and burden (Supplementary Fig. S5B–S5E and Supplementary Table S4). The three most aggressively metastatic cell lines (OS152, OS742, and OS525) formed metastases in all mice tested, resulting in the highest overall burden, greatest number of sites, and shortest latency compared with the remaining PDX-derived cell lines (Table 2; Fig. 4; Supplementary Figs. S6–S7). With most PDX-derived cell lines demonstrating the ability to metastasize in an intravenous model with high incidence but low burden, we next tested their capacity to spontaneously metastasize in an orthotopic amputation model.

We sought to develop an orthotopic osteosarcoma model to fully evaluate the entire metastatic cascade as a clinically relevant model for preclinical testing. Paratibial implantation of luciferase-expressing PDX-derived cells into NSG mice was performed to assess tumorigenicity and capacity for spontaneous metastasis (Fig. 5A). All but one PDX-derived cell line (OS457) formed xenografts within 6 months of implantation (Fig. 5B; Supplementary Fig. S8A–S8H). This cell line was the most divergent from both the patient tumor and PDX from which it was derived, likely due to a WGD event that occurred while the cell line was established from the PDX (Supplementary Fig. S2B). Time to amputation (when the xenograft reached 1 cm in diameter), ranged from less than 1 month to 8 months for individual mice whereas average time to amputation for each cell line ranged from 1 to 5 months (Table 3; Fig. 5B; Supplementary Fig. S8A–S8G). Each tumorigenic PDX-derived cell line demonstrated a consistent growth rate for orthotopic xenografts except OS384 which varied widely (Fig. 5B; Supplementary Fig. S8G). For another PDX-derived cell line (OS457), BLI signal was detected after 6 months at the orthotopic implant site for all mice (Supplementary Fig. S8I), but no xenografts could be palpated nor were they captured histologically. PDX-derived cell line xenografts exhibited heterogeneous histologic phenotypes (Supplementary Fig. S8J), similar to what is seen in human patients. Most PDX-derived cell line xenografts exhibited regions with similar histologic phenotypes to the patient tumors from which they were derived (Supplementary Fig. S8J).

Cell line xenograft growth was similar regardless of the presence of luciferase, with a tendency for the luciferase-expressing version to reach amputation more slowly after implantation (Supplementary Fig. S9A–S9D). microCT prior to amputation detected tibial osteolysis and medullary cavity or trabecular bone invasion for all tumorigenic PDX-derived cell line xenografts in the orthotopic model, which was confirmed histologically (Fig. 5C), highlighting the aggressiveness of these cell lines.

Postamputation, mice in orthotopic studies were monitored for metastatic burden for up to 6 months using BLI and health monitoring (Table 3; Figs. 5A, 6A and B; Supplementary Fig. S10A–S10G). Study duration after amputation ranged from one to a maximum of 6 months (Table 3; Supplementary Fig. S10A–S10G). Spontaneous metastases developed in mice for five of the seven xenograft forming PDX-derived cell lines with BLI signal increasing over time (Table 3; Supplementary Fig. S10A–S10G). BLI signal at endpoint showed a trend of being inversely correlated with time to endpoint (Table 3; Fig. 6A and B) demonstrating that higher metastatic burden causes increased morbidity in vivo.

As with the intravenous model described above, there was a high degree of tissue-specific tropism in metastasis development (Table 3; Fig. 6A). Two of the five metastatic PDX-derived cell lines (OS742 and OS152) metastasized consistently to the lung in all mice (Table 3; Fig. 6C and D). Of the remaining three xenograft-forming cell lines that metastasized, two (OS525 and OS384) metastasized to lung with high frequency (50% and 88%, respectively) and one (OS052) never metastasized to the lung (Table 3; Fig. 6C and D). PDX-derived cell lines OS152, OS525, and OS742 also metastasized to bone (spine, skull, and long bone) as well as liver in up to 100% of mice (Table 3; Supplementary Figs. S10H–S10I and S11). Four of the PDX-derived cell lines also metastasized to other soft tissue sites (abdomen, kidney, axillary region, and mediastinum) in 13% to 86% of mice (Table 3). Lung burden by area for all mice remained below 4%, whereas liver burden reached as high as nearly 50% (Fig. 6C; Supplementary Fig. S11A). The three most aggressive cell lines (OS152, OS742, and OS525) metastasized in all mice with the highest overall burden, to the most sites and with the shortest latency (Table 3; Fig. 6; Supplementary Figs. S10 and S11) compared with the remaining PDX-derived cell lines. OS052 was one of the fastest cell lines to grow a 1-cm xenograft at the orthotopic site but had low capacity to form spontaneous metastasis (Table 3; Figs. 5B and 6A and B). Using the orthotopic model, metastatic capacity and site tropism between the luciferase and parental versions were similar (Supplementary Fig. S9E–S9H; Supplementary Table S4).

Next, we compared the in vivo metastatic phenotypes of the panel of osteosarcoma PDX-derived cell lines to a small subset of commercially available established osteosarcoma cell lines by evaluating them using the same intravenous and orthotopic models described above.

In the intravenous model, SJSA and MG63.3 caused metastasis-associated morbidity faster than the most aggressive PDX-derived cell lines; the BLI signal for SJSA at endpoint was comparable to the most aggressive PDX-derived cell lines but in half of the time (Supplementary Table S5; Supplementary Fig. S12A and S12B). The metastatic lung burden for SJSA and MG63.3 was much higher than the PDX-derived cell lines, whereas for MG63 it was lower than many of the PDX-cell derived cell lines (Supplementary Fig. S12C and S12D). Unlike with the PDX-derived cell lines, lung metastases formed by established commercially available osteosarcoma cell lines were more evenly distributed throughout the pulmonary parenchyma (Supplementary Fig. S12E). However, MG63.3 tended to track along large airways and the MG63/MG63.3 tumors appear to contain a reduced number of human mitochondria compared with SJSA or PDX-derived cell line metastases (Supplementary Fig. S12E; Fig. 4F and 6E). None of the established commercially available cell lines metastasized to bone and only SJSA metastasized to the liver (Supplementary Fig. S12F–S12H). Because human osteosarcoma frequently metastasizes to bone, a useful feature of the PDX-derived cell line panel described here is the observed bone tropism.

In the orthotopic model, SJSA and MG63.3 xenografts grew quickly with all mice requiring amputation by one month; for MG63, xenografts were sporadically palpable in some mice, but limb diameter was never much larger than ∼6 mm in diameter, even after >250 days postimplantation (Supplementary Table S6 and Supplementary Fig. S13A and S13B). Like the PDX-derived cell line xenografts, SJSA and MG63.3 xenografts invaded through the tibial cortex (Supplementary Fig. S13C and S13D). For SJSA and MG63.3, study duration from amputation to endpoint (Supplementary Table S6) and metastatic lung burden (Supplementary Fig. S13E–S13I) was similar to the most aggressive PDX-derived cell lines; neither of these cell lines metastasized to bone in the orthotopic model. After comparing the in vivo metastatic phenotypes of PDX-derived cell lines to established osteosarcoma cell lines in intravenous and orthotopic models, we next set out to demonstrate their potential utility in genomically informed targeted drug studies.

A primary rationale for the development of an osteosarcoma orthotopic amputation model using PDX-derived cell lines that metastasize spontaneously is to use this model to test new therapeutic approaches that may be effective for treatment of metastatic disease. To this end, we tested the response to targeted therapy in one of the PDX-derived cell lines, OS152, which carries a MYC amplification (Table 1; Fig. 2; Supplementary Table S3). MYC expression is higher and there is MYC target gene set enrichment in the MYC amplified cell lines (Supplementary Fig. S14). We have previously shown that MYC amplified subcutaneous xenografts can be targeted with the CDK inhibitor dinaciclib (11). Dinaciclib inhibits CDK2/5/9 which results in downregulation of MYC expression (75). To determine the utility of this model as a preclinical tool to assess the efficacy of dinaciclib therapy, mice were implanted with OS152 luciferase-expressing cells orthotopically as described above. Intraperitoneal dinaciclib administration (20 mg/kg 5 times/week) commenced 4 days postamputation (Fig. 7A). At endpoint, the BLI signal was significantly lower in dinaciclib treated mice (P = 0.0153; t test; Fig. 7C and D) and the difference between the treatment and control groups increased with time (Fig. 7B). In the treatment group, there was a significant reduction in the average metastatic tumor area (lung: P = 0.0315 and liver: P = 0.0040, Mann–Whitney test; Fig. 7E) and number of metastases per mm² (lung: P = 0.0024, t test and liver: P = 0.0400, Mann–Whitney test; Fig. 7F) in the lung and liver. These results suggest that the orthotopic metastasis model developed here can be used to evaluate novel targeted therapeutic approaches to treat osteosarcoma metastasis.

Osteosarcoma remains a major clinical challenge largely due to its propensity to metastasize. While MYC has been suggested to be associated with aggressive disease (76), there are currently no known mechanisms of metastatic spread in osteosarcoma. The paucity of well-characterized metastasis models for this disease limits the ability to identify mechanisms of metastasis as well as the development of therapies specifically directed at preventing or treating metastatic spread. To address this need, we describe the development and characterization of a large panel of PDXs and PDX-derived cell lines which can be used to study osteosarcoma metastasis. We observed that the PDXs and PDX-derived cell lines largely share the genomic features of their original patient tumors and exhibit genomic complexity that mirrors what is observed in the human disease and is distinct from commercially available cell lines. Similarly, transcriptome clustering indicates that the PDXs and PDX-derived cell lines cluster with their tumor of origin, indicating close fidelity that is maintained over time. We also observed that PDX-derived cell lines cluster with the patient of origin samples rather than with other established commercially available cell lines, suggesting a distinct biology and perhaps a closer similarity to the human disease. Overall, the genomic and transcriptomic studies we performed strongly support the significant added value of developing new PDXs and PDX-derived cell lines for rare cancers rather than relying completely on the often-limited number of highly passaged immortal cell lines currently widely available.

We characterized the in vivo metastatic characteristics of eight PDX-derived cell lines and demonstrated that they have a wide phenotypic range with regard to primary orthotopic growth and metastatic progression. The PDX-derived cell lines are heterogenous with respect to their tumorigenicity as well as their metastatic propensity and tropism. We anticipate that this will be a particularly useful feature of this resource as it will facilitate the selection of cell lines for the investigation of specific aspects of the osteosarcoma metastatic cascade. Three PDX-derived cell lines (OS742, OS152, and OS525) were found to have a consistently high metastatic capacity whereas two cell lines were non-metastatic (OS457 and OS186), and another three had an intermediate metastatic phenotype (OS052, OS526, and OS384). Of the cell lines tested in vivo, OS457 and OS186 were the PDX-derived cell lines most divergent from their patient tumor, due to WGD (Supplementary Fig. S2B). It is possible that this drift has deleterious effects on the in vivo growth of these cell lines. PDX-derived cell lines metastasized most frequently to the lungs, which is also the most common metastatic site in patients. Multiple PDX-derived cell lines also metastasize to bone (OS052, OS152, OS525, and OS742), another important metastatic site in patients with osteosarcoma and an essentially non-existent feature in other previously described models, with the exception of the naturally occurring canine model (28, 45). The two most aggressive cell lines (OS152 and OS742) form invasive orthotopic xenografts that reach the amputation endpoint in <6 weeks and metastasize to lung, liver, bone, and other sites less than 1 month postamputation; these latency characteristics make this an appealing model for mechanistic studies and preclinical therapeutic development. The variation in metastasis predilection sites between osteosarcoma PDX-derived cell lines illustrates the ability of these models to recapitulate the spectrum of the human disease.

In prior work, we demonstrated that the CDK inhibitor dinaciclib leads to decreased subcutaneous xenograft growth for osteosarcoma PDX-derived cell lines that carry MYC amplification (11). Here, we assessed whether dinaciclib could be used to decrease the progression of metastasis in an orthotopic amputation model. This proof-of-concept experiment, using one osteosarcoma PDX-derived cell line (OS152), demonstrated that this model can be used effectively to test the effects of targeting genomic features to inhibit metastasis.

The PDX-derived cell lines described here were derived from both primary and metastatic sites. We did not find a clear association between patient of origin metastatic disease status and the in vivo metastatic phenotype or xenograft growth of the PDX-derived cell lines. The PDX-derived cell line exhibiting the fastest xenograft growth (OS742) and the non-tumorigenic cell line (OS457) were both derived from primary tumors. Similarly, of the two most metastatic cell lines, one is primary in origin (OS742) and the other originates from a metastatic site (OS152). The established commercially available osteosarcoma cell lines evaluated here were all derived from primary tumors and were either highly tumorigenic and produced heavy metastatic burden (MG63.3 and SJSA) or were non-tumorigenic and produced minimal metastatic burden (MG63). For SJSA and MG63.3, especially in the intravenous model, the metastatic burden induced in a short timeframe (∼1 month), caused significant morbidity in the mice and presented challenges with burden quantification.

In patients, conventional osteosarcoma most commonly originates centrally in the metaphyseal region. In our orthotopic model, cells are implanted paratibially to avoid direct seeding into the bloodstream via the bone marrow. However, pre-amputation microCT and postamputation histologic examination of the affected limbs confirmed that all cell lines capable of forming xenografts also invade through the cortex. Confirming central invasion of the osteosarcoma PDX-derived cell line xenografts is probably critical for more accurate modeling of the human disease. In PDX paratibial implantation amputation models, metastasis incidence is reported to be lower in the mouse model than in human patients, ranging from 25% to 50% between PDXs (41, 51). Using PDX-derived cell lines in the orthotopic model, we achieved metastatic rates that were more in line with patient outcome; in three PDX cell lines, the metastatic rate was 100% and a fourth cell line reached 88% in the orthotopic model. In summary, we describe here a panel of new osteosarcoma PDX-derived cell lines that we believe will be of wide use to the osteosarcoma research community.

C.R. Schott reports a grant from Rally Foundation for Childhood Cancer Research during the conduct of the study. C. Curtis reports personal fees from NanoString, Genentech, ResistanceBio, Bristol-Myers Squibb, 3T Biosciences; and personal fees from DeepCell outside the submitted work. E.A. Sweet-Cordero reports grants from NCI and St Baldricks Foundation; as well as grants from Hyundai Hope on Wheels during the conduct of the study. No disclosures were reported by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

C.R. Schott: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. A.L. Koehne: Conceptualization, investigation, methodology. L.C. Sayles: Conceptualization, data curation, methodology, writing–review and editing. E.P. Young: Investigation. C. Luck: Formal analysis, visualization, methodology. K. Yu: Resources. A.G. Lee: Resources, data curation, formal analysis, visualization, methodology. M.R. Breese: Resources, data curation, formal analysis, visualization, methodology. S.G. Leung: Sample acquisition, acquired patient consent. H. Xu: Formal analysis, visualization. A.T. Shah: Data curation. H.-Y. Liu: Sample acquisition, acquired patient consent. A. Spillinger: Sample acquisition, acquired patient consent. I.H. Behroozfard: Sample acquisition, acquired patient consent. K.D. Marini: Investigation. P.T. Dinh: Sample acquisition, acquired patient consent. M.V. Pons Ventura: Sample acquisition, acquired patient consent. E.N. Vanderboon: Formal analysis. F.K. Hazard: Confirmed the diagnosis of osteosarcoma in patient samples. S.-J. Cho: Confirmed the diagnosis of osteosarcoma in patient samples. R.S. Avedian: Sample acquisition. D.G. Mohler: Sample acquisition. M. Zimel: Sample acquisition. R. Wustrack: Sample acquisition. C. Curtis: Resources. M. Sirota: Resources. E.A. Sweet-Cordero: Conceptualization, resources, supervision, funding acquisition, validation, writing--original draft, project administration.

We thank the members of the UCSF Helen Diller Family Comprehensive Cancer Center (HDFCCC) Preclinical Therapeutics Core and Animal Care at UCSF for their useful advice throughout the development of this project. Sequencing was performed at the UCSF Center for Advanced Technology (CAT), supported by UCSF PBBR, RRP IMIA, and NIH 1S10OD028511–01 grants. This work was also supported by NIH S10 instrument grant S10OD025022. Cell sorting was performed at the UCSF Laboratory for Cell Analysis (LCA), supported by the NCI of the NIH under Award Number P30CA082103. C.R. Schott was supported by a Postdoctoral Research Fellow Grant from the Rally Foundation for Childhood Cancer Research. E.A. Sweet-Cordero was supported by the NIH (1R01CA243555) and by the Battle Osteosarcoma grant from the St. Baldrick's Foundation. Dr. Hani Goodarzi gifted the luciferase plasmid. Dr. Robbie Majzner of Stanford University gifted cell line MG63.3. Sara Pyle generated the graphical abstract (Fig. 1). We thank especially all the patients and their families who generously consented to tissue donation to enable this work.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.