Correspondence to Nature Genetics

To the Editor

Genomic information about predisposing germline mutations within normal cells as well as acquired somatic lesions within cancer cells will enable the development and delivery of individualized cancer therapies. Ongoing global initiatives have revealed that the spectrum of somatic and germline genetic lesions in pediatric cancer are distinct from those found in adult cancer^1,2,3. However, existing cancer genome data portals (cBioPortal⁴ and COSMIC⁵) have focused primarily on presenting data generated from adult cancer studies. They also lack features for exploring pathogenic germline mutations, gene fusions, and mutation stratification by cancer subtypes, all of which are of great importance in pediatric cancer.

Here we describe ProteinPaint, a web application for simultaneously visualizing genetic lesions (including sequence mutations and gene fusions) and RNA expression of pediatric cancer. The pediatric data set consists of 27,188 validated somatic coding lesions acquired at diagnosis or relapse from 17 subtypes of pediatric cancer, 252 pathogenic or loss-of-function germline lesions detected in >1000 pediatric cancer patients of 21 subtypes⁶, and RNA-Seq of 928 pediatric tumors from 36 subtypes (Supplementary Notes). The data were compiled from five major studies (Supplementary Notes) and will be expanded with the publication of additional pediatric cancer studies.

Genetic lesions of pediatric cancer are shown on a protein panel (Fig. 1) with the option for a parallel view of a curated version of published somatic mutations in COSMIC database (Supplementary Notes). This enables the use of adult data for interpreting the significance of rare genetic lesions in pediatric cancer (Supplementary Figs. 1, 2) and vice versa (Supplementary Fig. 3). To ensure consistency, all variants were reannotated with a modified version of Annovar⁷. As an example, we show how this presentation has enabled the detection of aberrant splicing caused by recurrent “silent” mutations in TP53. This finding also provided insight into the pathogenicity of matching germline variants found in patients with cancer predisposition syndromes (Supplementary Fig. 1, Supplementary Notes). Additionally, presentation of mutant allele fractions in DNA and RNA facilitates evaluation of tumor heterogeneity related to cancer relapse (Supplementary Fig. 3) as well as detection of allelic imbalance in DNA or RNA caused by a second genetic or epigenetic hit in tumor (Supplementary Fig. 1, Supplementary Notes). Loss-of-heterozygosity (LOH), which was computed by the CONSERTING⁸ algorithm in the pediatric cancer genomes we analyzed, is shown to further facilitate the identification of double-hit mutations (Supplementary Fig. 4).

Open in a separate window

Figure 1

Comprehensive Visualization of Sequence Mutations, Gene Fusions, and RNA Expression Using ProteinPaint. (a) TP53 mutation profile in the Pediatric data set (top) and COSMIC database (bottom). The number of samples affected by each mutation is indicated by the text within each disc, as well as disc size. The arc outside each disc indicates the proportion of samples that are germline (solid) or relapsed tumor (hollow). The full legend is shown in Supplementary Fig. 1. The manually-curated “NLS” domain reveals a hotspot nonsense mutation R306* that disrupts a known nucleotide localization signal⁹. (b) JAK2 gene fusion (left) and expression (right). Left: JAK2 fusions are shown along with sequence mutations affecting the pseudokinase and kinase domains of JAK2. A half-filled disc represents a gene fusion, with the filled section representing the N- or C- terminal of the protein involved in the fusion. The arrow points to the PAX5-JAK2 fusion detected in 7 tumors of Ph-like B-cell acute lymphoblastic leukemia¹⁰. The fusion protein involves the C terminal of JAK2. Right: JAK2 expression level in Pediatric samples. The horizontal axis represents range of FPKM values. Gray circles represent samples in descending order of JAK2 FPKM value. Samples highlighted by the filled red circles are those with PAX5-JAK2 fusion selected by the user. The ratio of PAX5-JAK2 fusion transcript to the overall expression of its two partner genes, PAX5 and JAK2 are labeled in the red text. Boxplots represent FPKM value distribution in pediatric cancer cohorts, labeled by the disease names and cohort sizes.

The expression panel presents the rank and quantity of gene expression of each sample with superimposed boxplots summarizing the expression range of the entire cohort, or user-selected subtypes. Selecting a genetic lesion such as the PAX5-JAK2 fusion on the protein panel automatically highlights the mutated samples on the expression panel; in the case of PAX5-JAK2 fusion, this reveals the aberrantly high expression of JAK2 caused by gene fusion (Fig. 1b). Conversely, examination of aberrant expression in a tumor may lead to novel insight into its causal genetic lesion. We show an example of how outlier expression of FLT3 in a leukemia with kinase activation signature led to discovery of a high-level FLT3 amplification resulting from replication of an episome formed by a complex rearrangement involving three chromosomes (Supplementary Fig. 5, Supplementary Notes).

ProteinPaint is designed to deliver a premium visualization experience with interactive and animated features. Novel “disc-on-stem” skewer graphs were implemented to depict the diverse prevalence, complex allelic alteration, and temporal origin of mutations and gene fusions at a glance (Fig. 1, Supplementary Figs. 6–7). Customized views include display of mutation and expression by cancer subtypes or tumor tissues, dynamic zoom, and integration of user-provided data with new features implemented based on user feedback. Data in mutation annotation format (MAF) generated by studies such as The Cancer Genome Atlas (TCGA) or individual research labs can be uploaded to ProteinPaint to enable data visualization and cross-study comparison for the broad genetic research community (Supplementary Fig. 8, Supplementary Tutorial). Manually-curated protein domains have been incorporated for genes frequently mutated in pediatric cancer to facilitate the interpretation of mutation pathogenicity (Fig. 1, Supplementary Fig. 6). ProteinPaint complements existing cancer genome portals by providing a comprehensive and intuitive view of pediatric cancer genomic data with advanced visualization features, as well as integration of expression and adult cancer data (Supplementary Figs. 6–7, Supplementary Notes). Taken as a whole, these novel features make ProteinPaint a powerful tool for employing genomic data to enhance pediatric cancer research, collaboration, and clinical care.

Supplementary Material

Acknowledgments

Footnotes

References