High-throughput genomic and molecular profiling of tumors is emerging as an important clinical approach. Molecular profiling is increasingly being used to guide cancer patient care, especially in advanced and incurable cancers. However, navigating the scientific literature to make evidence-based clinical decisions based on molecular profiling results is overwhelming for many oncology clinicians and researchers. The Personalized Cancer Therapy website (www.personalizedcancertherapy.org) was created to provide an online resource for clinicians and researchers to facilitate navigation of available data. Specifically, this resource can be used to help identify potential therapy options for patients harboring oncogenic genomic alterations. Herein, we describe how content on www.personalizedcancertherapy.org is generated and maintained. We end with case scenarios to illustrate the clinical utility of the website. The goal of this publicly available resource is to provide easily accessible information to a broad oncology audience, as this may help ease the information retrieval burden facing participants in the precision oncology field. Cancer Res; 77(21); e123–6. ©2017 AACR.
With the increasing number of potential therapeutic targets and novel therapeutic agents (1), clinicians and researchers may be overwhelmed by the amount of information available in public repositories and in the published literature. At The University of Texas MD Anderson Cancer Center, we have established a team of scientists, clinicians, and bioinformatics experts whose mission is to “provide the scientific support and rationale for all cancer patients and physicians to use as a resource in guiding potential personalized cancer therapy options” (2–4). Our Precision Oncology Decision Support (PODS) team has developed the Personalized Cancer Therapy website (www.personalizedcancertherapy.org) to help clinicians and researchers navigate the growing amount of available information. We developed the website with the intention of being gene-centric and therefore “patient tumor type-agnostic,” and our goal was to facilitate clinical use of this complex information. Specifically, we envision that patient care team members would use this resource during patient encounters, particularly to identify available genotype-matched therapies.
Individual gene sheets form the basis for the layout of the website. To be fully annotated and published on www.personalizedcancertherapy.org, a gene must fulfill an initial set of actionability criteria: (i) the gene must be a cancer-associated gene (e.g., associated with cancer development or progression, or with cancer cell proliferation or survival); (ii) there must be evidence that targeting the gene or its pathway therapeutically may result in tumor suppression; and (iii) drugs targeting the gene product or its downstream consequence (e.g., downstream signaling) are either FDA-approved or are being clinically investigated. Members of the PODS team select genes for inclusion on the website based upon the strength of available scientific data, with input from MD Anderson clinicians and researchers. Once we select a gene for annotation, a systematic scientific literature review is performed to generate content for the sections of the gene page on the website (Fig. 1). The majority of this content is generated using manual literature reviews of PubMed or internet search engines, but we are also developing and using semiautomated approaches (5).
Pathway and gene image
The MD Anderson Medical Graphics Department assists with the generation of gene and pathway images. For gene images, we retrieve chromosomal location, transcript information from NM_transcript, and protein domain information from Uniprot (6). Information is then aggregated automatically from the Entrez database (7). For creation of condensed signaling pathway images, we use manual review of the scientific literature.
Functional annotation of single nucleotide or gene fusion variants
For functional annotation of variants, we compile unique alterations within the PODS database. This includes somatic or germline single-nucleotide alterations, and small insertions or deletions. Resources used for functional annotation include cBIO (8), COSMIC (9), alterations included on institutional next-generation sequencing platforms, alterations encountered during annotation of individual patient reports, and published findings. We use manual review of the scientific literature for annotation of copy number variations (i.e., amplifications and deletions) and gene fusion variants. Relevant literature is identified and retrieved using a specially designed information retrieval tool called RetriLite (10), which we developed to streamline the literature review process. This customizable information retrieval tool systematically formulates queries that describe the specific information needs. The list of retrieved documents is ranked by relevancy, and search terms are highlighted in the results for ease of interpretation. Additional publicly available databases, such as ClinVar (11), dbSNP (12), The Breast Cancer Information Core (13), and internet search engines may also be manually queried.
We manually review the retrieved literature to write functional annotation of variants. Each variant is assigned a priority based on its reported frequency within the consulted databases and the existence of published functional or therapeutic data. Only alterations with clear clinical or preclinical data are published to the website.
Drug annotation: identifying drug–gene associations
To generate a list of drugs relevant to the gene of interest, we catalog drugs using online databases and scientific literature. Such databases include the National Cancer Institute (NCI) drug dictionary (14), Selleckchem.com (15), medkoo.com (16), the Drug Gene Interaction Database (DGIdb; ref. 17), and ClinicalTrials.gov (18). Drugs are curated and annotated within the PODS database. Although the annotator must identify some information manually (i.e., drug-gene associations and phase of drug development), the database is able to automatically retrieve much of the information (i.e., NCI definition and aliases from NCI Thesaurus; ref. 19).
We then tag each drug with potentially relevant biomarkers. Currently for PODS, a biomarker refers to an alteration at the genomic or protein expression level. We record the reported IC50 of a given drug from the scientific literature in the context of each alteration. We then order each of the direct and indirect drug targets by specificity. A gene is considered to be a direct target of a drug if the drug is capable of inhibiting the activity of the gene-encoded protein with an IC50 < 1 μmol/L within biochemical assays. A gene is considered to be an indirect target of a drug if the protein targeted by the drug is within the pathway of the protein encoded by the gene of interest. Additionally, there must be evidence that the drug has antitumor properties when an alteration within the gene of interest is present. For indirect targets, knowledge of the pathway and biologic relevance impact the decision to list a gene as applicable to that specific drug.
Drugs with FDA-approval are also tagged with each approved disease type, biomarker, and any other drugs used in combination. Only clinically available drugs (i.e., drugs that are FDA-approved or being actively investigated within clinical trials) that directly or indirectly target the gene of interest are automatically published to the “Drugs” tab of their respective gene pages.
We annotate therapeutic implications to describe the level of evidence supporting the effectiveness of a specific drug in a specific tumor type harboring a specific biomarker. RetriLite is used to retrieve scientific literature relevant to potential therapeutic implications from MEDLINE abstracts. Literature is then manually reviewed and stratified for drug, gene alteration, disease, and level of evidence.
Only literature that fulfills our previously described level of evidence criteria (2) is published to the website. The level of evidence scale is as follows: level 1 includes prospective studies with strong evidence of therapeutic efficacy linked to a biomarker evaluation (in the context of a specific disease), usually with FDA approval of the drug in the context of a biomarker; level 2 includes retrospective studies in a specific tumor type or clinical data from another tumor type; level 3 includes case studies with a scientific rationale or high quality preclinical data linking a biomarker to drug efficacy (2). We also report conflicting data, as well as evidence that a biomarker may confer resistance to a particular drug or class of drugs.
Clinical trial annotation
We use an in-house informatics tool to identify clinical trials. Retrieval of trials leverages the direct and indirect gene target information from drugs annotated in the PODS database. Once a list of trials has been generated for the gene of interest, each trial is annotated by aggregating information from ClinicalTrials.gov and MD Anderson's internal Clinical Trial Management System.
When a gene is entered into the system, trials using drugs that have been previously annotated in the database as being relevant to that gene are identified. The clinical trials informatics tool programmatically interacts with ClinicalTrials.gov's web application programming interface (API). Through this, we can retrieve open trials that use one or more targeted therapies of interest to treat cancer patients. The PODS team then manually annotates these trials for drugs, diseases, and biomarkers. Trials are designated as “genotype-selected” (i.e., a trial that requires a certain genomic alteration for enrollment) or “genotype-relevant” (i.e., a trial that uses a drug relevant to a genomic alteration, but does not require that alteration for enrollment; ref. 10).
Gene page development workflow
The workflow process for developing a gene page is outlined in Fig. 1. Before publication, scientists on the PODS team, oncologists, disease experts, and gene domain experts extensively review the content for a gene page. Level of annotator effort can be broken down as follows: approximately 30% on functional variants, 30% on therapeutic implications, 20% clinical trials, 10% on drugs, and 10% on the overviews and frequency/outcomes sections. Time to completion can vary depending on the amount of information to be reviewed and access to expert reviewers.
Our goal is to make major updates to therapeutic implications, frequencies, and outcomes annually using published literature and abstracts presented at major cancer meetings. Our Sheikh Khalifa Bin Zayed Al Nahyan Institute for Personalized Cancer Therapy Medical Director and/or oncologists specializing in the implicated gene or diseases review any significant changes before publication. However, we update www.personalizedcancertherapy.org daily for alterations that are annotated for MD Anderson patient reports. This includes alterations with either deleterious or likely benign functional significance. MD Anderson clinical trials, which can capture new drugs, are curated and annotated every two weeks.
Comparison to other available decision support resources
There are a variety of resources that contain information relevant to precision oncology (20). Given the great need for decision support for genomic medicine, several academic and commercial entities are trying to meet this need. Two other publicly available websites with similar goals to our website are My Cancer Genome (www.mycancergenome.org) and OncoKB (www.oncokb.org). Similar to www.personalizedcancertherapy.org, My Cancer Genome contains information regarding alterations, actionability, and therapeutic implications. However, alteration-specific information is organized by tumor type rather than by alteration. This difference in organization may make it more difficult to extrapolate alteration information to other tumor types. Alternatively, OncoKB is organized by alteration, and also provides information regarding actionability and therapeutic implications. However, clinical trials are not included in this resource, and users must navigate to a different webpage to search for relevant clinical trials.
We present the following two case scenarios to highlight how our resource might be used in a clinical setting. We have also included a virtual tour of www.personalizedcancertherapy.org (see Supplementary Video S1).
Mr. B is a patient with cholangiocarcinoma who is being seen for treatment planning. At the last visit, his physician, Dr. A, ordered a clinically available next-generation sequencing panel test to identify actionable alterations. Dr. A is now reviewing the report before her scheduled discussion with Mr. B. The panel testing revealed an ERBB2 gene amplification. Using www.personalizedcancertherapy.org, Dr. A identifies that there are no HER2-targeted therapies that are FDA-approved for cholangiocarcinoma. However, she is able to identify several genotype-selected clinical trials that are currently available to Mr. B in his home state of California.
Mrs. C is a patient with HER2 nonoverexpressing and non-amplified breast cancer who has progression of her disease. She is referred to the Phase I Clinic for consideration of new therapeutic options. She had previously undergone next-generation sequencing panel testing that revealed an ERBB2 single nucleotide variant S310F. The clinical research nurse in charge of enrolling on a trial for a HER2 inhibitor reviews her results in anticipation of her consult visit. Using www.personalizedcancertherapy.org, she determines that S310F is a known activating mutation, and thus is actionable. Consequently, the team is able to offer Mrs. C enrollment on their clinical trial.
One of the important aims of www.personalizedcancertherapy.org is to curate updated information for immediate clinical use. As such, the website is frequently evolving to the needs of our audience. We are currently working toward adding several additional biomarkers, including biomarkers relevant to immunotherapy: microsatellite instability status, PD1/PDL1, and tumor mutational burden. In addition, we plan to continue to expand the filters on our clinical trial search function to improve generalizability to other practice settings. In the future, we hope that other institutions will be able to tailor some of the features of our website to their specific needs.
www.personalizedcancertherapy.org aggregates decision support information that is critical to the practice of precision oncology into a single, well-supported platform. Although it is not the only resource of its kind, we believe that it is a valuable addition to the current publicly available resources. Connecting clinicians and researchers with available trials through the use of highly curated and organized complex data is a daunting task, and is becoming increasingly difficult as our knowledge base grows. The goal of our website is to provide easily accessible and interpretable information to a broad oncology audience to improve the real-world practice of precision oncology.
Disclosure of Potential Conflicts of Interest
B. Litzenburger is a Senior Somatic Scientist in QIAGEN. G.B. Mills reports receiving a commercial research grant from Adelson Medical Research Foundation, AstraZeneca, Critical Outcome Technologies, Komen Research Foundation, Nanostring, Breast Cancer Research Foundation, Karus, Illumina, Takeda/Millenium Pharmaceuticals, Pfizer; has received speakers bureau honoraria from Symphogen, MedImmune, AstraZeneca, ISIS Pharmaceuticals, Lilly, Novartis, ImmunoMet, Allostery, Tarveda, Pfizer; has ownership interest (including patents) in Catena Pharmaceuticals, PTV Ventures, Spindletop Ventures, Myriad Genetics, ImmunoMet; and is a consultant/advisory board member for, Adventist Health, AstraZeneca, Provista Diagnostics, Signalchem Lifesciences, Symphogen, Lilly, Novartis, Tarveda, Tau Therapeutics, Allostery, Catena Pharmaceuticals, Critical Outcome Technologies, ISIS Pharmaceuticals, Immunomet, Takeda/Millenium, MedImmune, Precision Medicine. E.V. Bernstam has received speakers bureau honoraria from Roche. No potential conflicts of interest were disclosed by the other authors.
Conception and design: K.C. Kurnit, J. Zeng, B.C. Litzenburger, G.B. Mills, E. Bernstam, K. Shaw, F. Meric-Bernstam
Development of methodology: K.C. Kurnit, A.M. Bailey, J. Zeng, A.M. Johnson, B.C. Litzenburger, N.S. Sánchez, Y.B. Khotskaya, E. Bernstam, F. Meric-Bernstam
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.C. Litzenburger, N.S. Sánchez, Y.B. Khotskaya, V. Holla, A. Simpson, G.B. Mills, F. Meric-Bernstam
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Md. A. Shufean, B.C. Litzenburger, V. Holla, G.B. Mills, E. Bernstam, F. Meric-Bernstam
Writing, review, and/or revision of the manuscript: K.C. Kurnit, A.M. Bailey, J. Zeng, A.M. Johnson, Md. A. Shufean, L. Brusco, B.C. Litzenburger, N.S. Sánchez, Y.B. Khotskaya, V. Holla, G.B. Mills, J. Mendelsohn, E. Bernstam, F. Meric-Bernstam
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.M. Bailey, A.M. Johnson, J. Mendelsohn, F. Meric-Bernstam
Study supervision: K. Shaw, F. Meric-Bernstam
This work was supported in part by 1U01 CA180964 (to F. Meric-Bernstam and E. Bernstam), NIH Research Training Grant T32 CA101642 (to K.C. Kurnit), The Cancer Prevention and Research Institute of Texas (RP150535 to A.M. Bailey, J. Zeng, A.M. Johnson, B.C. Litzenburger, N.S. Sanchez, Y.B. Khotskaya, V. Holla, J. Mendelsohn, E. Bernstam, K. Shaw, and F. Meric-Bernstam), the Sheikh Khalifa Bin Zayed Al Nahyan Institute for Personalized Cancer Therapy (to A.M. Bailey, J. Zeng, A.M. Johnson, M.A. Shufean, B.C. Litzenburger, N.S. Sanchez, Y.B. Khotskaya, V. Holla, A. Simpson, J. Mendelsohn, K. Shaw, and F. Meric-Bernstam), NCATS Grant UL1 TR000371 (Center for Clinical and Translational Sciences; to E. Bernstam and F. Meric-Bernstam), The Bosarge Family Foundation (to G.B. Mills and F. Meric-Bernstam), and the MD Anderson Cancer Center Support Grant (P30 CA016672).