PCAT: an integrated portal for genomic and preclinical testing data of pediatric cancer patient-derived xenograft models

Abstract

Although cancer is the leading cause of disease-related mortality in children, the relative rarity of pediatric cancers poses a significant challenge for developing novel therapeutics to further improve prognosis. Patient-derived xenograft (PDX) models, which are usually developed from high-risk tumors, are a useful platform to study molecular driver events, identify biomarkers and prioritize therapeutic agents. Here, we develop PDX for Childhood Cancer Therapeutics (PCAT), a new integrated portal for pediatric cancer PDX models. Distinct from previously reported PDX portals, PCAT is focused on pediatric cancer models and provides intuitive interfaces for querying and data mining. The current release comprises 324 models and their associated clinical and genomic data, including gene expression, mutation and copy number alteration. Importantly, PCAT curates preclinical testing results for 68 models and 79 therapeutic agents manually collected from individual agent testing studies published since 2008. To facilitate comparisons of patterns between patient tumors and PDX models, PCAT curates clinical and molecular data of patient tumors from the TARGET project. In addition, PCAT provides access to gene fusions identified in nearly 1000 TARGET samples. PCAT was built using R-shiny and MySQL. The portal can be accessed at http://pcat.zhenglab.info or http://www.pedtranscriptome.org.

Figure 1.

A summary of PDX models in the current PCAT release. (A) A pie chart illustrating histology of the 324 models. (B) An overview of molecular and preclinical testing data of the 324 models. Each column represents one model. Tumors are separated into liquid cancer (dark green) and solid tumors (maroon). Blue bar denotes data are available for the model.

Figure 2.

Gene fusions identified in TARGET samples. (A) Benchmarking of fusion identification against 234 driver events annotated in ALL and AML clinical data. Each column represents one fusion. Solid color indicates the fusion was found by the corresponding tool noted on the left. (B) Distribution of fusion events across cancer types and sample types (NBL, neuroblastoma; RT, rhabdoid tumor; ALL, acute lymphoblastic leukemia; AML, acute myeloid leukemia; WT, Wilms’ tumor; CCSK, clear cell sarcoma kidney). Each dot represents a cancer sample. Fusion counts of post-treatment samples are significantly higher than primary and recurrent samples in ALM (both P-values <0.001, Wilcoxon rank sum test).

Figure 3.

The display page of PDX and fusion search results. (A) PDX information is displayed in a tabular page divided into clinical, mutation, fusion and preclinical testing data if available. (B) Fusion page displays identification information (evidence, junctions, etc.), a circos plot illustrating all fusions identified in this case and a scatter plot showing expression of the two partner genes. BCR-ABL1 is used in this example (red in circos plot).

Figure 4.

Examples demonstrating the utility of the analysis modules. (A) Expression of PDGFRA, a mesenchymal stem cell marker, across PDX models. (B) Correlation of CDKN2A expression and deletion in ALL PDXs. −2: homozygous deletion; −1: shallow deletion; 0: copy number neutral. (C) Correlation of TERT expression and patient overall survival in TARGET neuroblastoma. TERT expression is divided into high and low based on mean expression. (D) ssGESA results for top 20 MYCN targets in TARGET neuroblastoma. The color bar on top of the heat map indicates MYCN status per clinical data annotation. Samples are ordered ascendingly per ssGSEA score, as indicated by the top bar.

Figure 5.

Molecular correlates of drug responses. (A) Model response to 19D12, an IGF1R antibody inhibitor, is correlated with high expression of the target gene IGF1R. X-axis reflects response levels as follows: PD1/2, progressive disease; CR, complete response; MCR, maintained complete response. Each dot represents a PDX model. A total of 31 models are included here, including 8 ALLs, 6 neuroblastomas, 5 osteosarcomas, 5 Ewing’s sarcomas and 7 others. P-value indicates the difference between responsive groups (CR, MCR) and resistant groups (PD1, PD2). (B) Limiting the analysis to osteosarcoma models only (n = 5) sustains the association, suggesting it is not a tissue effect; i.e. osteosarcoma has high IGF1R expression and the drug works better on osteosarcoma. The outlier OS-2 was excluded from the t-test in this panel.