PDX-MI: Minimal Information for Patient-Derived Tumor Xenograft Models Free

Patient-derived tumor xenograft (PDX) models are created by implanting tumor cells or fragments from patients with cancer into a transplant-compliant mouse host (Supplementary Fig. S1; refs. 1, 2). Human tumors that engraft successfully in host mice are subsequently fragmented and passaged multiple times to generate large cohorts of tumor-bearing mice. PDX models accurately reflect the patient's tumor properties, creating a powerful platform to study the molecular mechanisms of tumor growth and drug resistance as well as serving as patient “avatars” for predicting response to anticancer therapeutic compounds (3–5). The host strains for PDX model development are typically severely immunodeficient; however, “humanized” immune system mice engrafted with human immune cells are increasingly being used in xenograft studies to explore in vivo interactions between the immune system and cancer (6, 7).

Although many academic and commercial sources of PDX models have emerged in recent years, the size of the resources and the processes for creating and characterizing PDX models is quite variable. Crucial information about tumors, host strains, transplant, and quality assurance processes are inconsistently presented in both the scientific literature and in database resources, limiting the ability of researchers to find relevant models and associated data. A standardized data exchange format is needed to foster the ability of researchers to identify appropriate PDX models and to share information about them. As developers of NCI-funded informatics resources, we obtained the internal standards developed by four independent PDX model resources [the EurOPDX consortium (5), the IMODI consortium (France), the Patient-Derived Models Repository at NCI-Frederick, and The Jackson Laboratory PDX Resource (8)]. After comparing standards in use across these resources, we generated a draft PDX-minimal information standard (PDX-MI) that was reviewed and modified by the authors of this report. We propose that the standards described here serve as the starting point for community-wide adoption.

The PDX-MI consists of four modules that reflect the process of generating, validating, and using a PDX model: clinical, model creation, model quality assurance, model study, and an additional associated metadata category (Table 1). Within each module, we define “essential” attributes that are required for accurate description and reporting on PDX models and “desirable” attributes that are frequently recorded by PDX producers and should be available.

The clinical module is divided into two submodules: “clinical/patient” and “clinical/tumor.” “Clinical/patient” requires information about the patient from which the engrafted tumor originates, including age, sex, ethnicity, and disease diagnosis. To reduce the possibility of patient identification, PDX-MI recommends grouping ages into 5-year groups, although more granular groupings may be used in cases such as pediatric tumors if approved by a contributor's Institutional Review Board. Reporting on patient consent is considered essential as well. Some attributes of patient treatment history are listed as “desirable” as they can impact the characteristics of resulting PDX models but may be challenging to provide due to patient privacy or data inaccessibility. The “clinical/tumor” submodule reports on information about the originating tumor from which the PDX model is derived and includes tumor classification, anatomic location, and tumor histopathology. The presence or absence of specific diagnostic markers is listed as “essential” for tumor types where testing for such marker(s) is considered the clinical standard of care (e.g., FLT3 genotype in acute myelogenous leukemia). In addition, patient viral infection status has implications for disease biology as well as occupational safety and is included as a desirable field.

The model creation module of PDX-MI captures critical attributes in the creation of a PDX model. Host strain is reported using official strain nomenclature (http://www.informatics.jax.org/mgihome/nomen/index.shtml) as well as strain source and any modifications that “humanize” the host strain through engraftment of human immune-progenitor cells (6). Initial engraftment of the tumor describing processing of the tumor (solid or cell suspension) and the anatomic site of implantation (subcutaneous or orthotopic) is represented. Other model generation characteristics such as engraftment rates and therapeutic response data are considered desirable. A “subline” field indicates when a PDX model is derived from an existing model that has changed characteristics (e.g., loss/gain of a biomarker, change in therapy response).

The model quality assurance module captures information about tissue provenance and fidelity of the passaged tumor with respect to key characteristics of the patient tumor. Validation is required to confirm the PDX tumor is of the appropriate patient and not of murine origin nor consisting primarily of Epstein–Barr virus human B lymphocytic cells as both are frequently observed in PDX model creation (9). Other “desirable” quality assurance methods vary with tumor types and can include histopathology, assessment of human cancer biomarkers by IHC, in situ hybridization, and assessment of gene mutations and rearrangements, DNA methylation, or gene expression profiling. Some producers evaluate how well a PDX model recapitulates the originating tumor's response by measuring PDX tumor growth response to standard-of-care treatment and this is included as a desirable attribute. Additional desirable information includes DNA profiling of serial passages to corroborate lineage fidelity and animal health status from standard health surveillance programs. The current PDX-MI requires evidence of quality assurance but does not require every possible technique be performed as methods vary across resources.

Tumors from PDX often undergo comprehensive genomic characterization and/or treatment in controlled dosing studies to define therapeutic response and resistance. PDX-MI includes “desirable” fields in the reporting of these studies that supplement existing guidelines for reporting on in vivo biomedical research (10). Additional optional metadata are accession IDs from data archives and citation IDs (including digital object identifiers) for publications describing the PDX model(s).

PDX models present unique challenges due to the specific approaches needed for the diversity of cancer subtypes. One challenge is that a subset of PDX models will require reporting on diagnostic biomarkers. For example, in breast cancer, testing for certain pathologic markers (estrogen receptor, progesterone receptor, and human epidermal growth factor 2) is considered the clinical standard of care for prognostic and predictive purposes and should therefore be considered essential for PDX-MI. Another challenge is that tumor grades and disease stages captured in the “clinical module,” which drive patient diagnosis and treatment, may be derived from scoring systems with diagnostic and geographic variation (11). PDX-MI will be flexible and allow users to report the system used clinically rather than enforce a particular one.

PDX resources employ a combination of custom and community-developed vocabularies. This presents challenges in data integration, as it takes expert knowledge to map the divergent systems. For example, cancer diagnoses are represented within different PDX resources by terms from the NCI-Thesaurus (12), SNO-MED CT (13), MeSH (14), and the Disease Ontology (15). Free text descriptions are used for many PDX model attributes and a mix of generic, commercial, and chemical labels are used for drugs. Ontology resources and community model organism databases have been developing tools to semiautomate mapping of standards that produce unified indices to facilitate data query and discovery. Rather than impose a limited set of terms to describe a given minimal information attribute, PDX-MI will allow the reporting of a resource's internal standards. We will ensure the quality of standard mappings by facilitating feedback between the PDX producers and the developers of ontology tools.

The current version of PDX-MI describes the minimal information needed to report on a PDX models to facilitate data integration and resource sharing. The authors of this report hope PDX-MI will serve as a guide for authors and journal editors in promoting rigorous yet attainable publication standards and as a template for managers of public molecular archives in the capturing of critical metadata required for submission of PDX model data. PDX-MI standards will also be implemented in an online resource being jointly developed by EMBL-EBI and the Jackson Laboratory called PDX Finder, www.pdxfinder.org (see Supplementary Video S1). This resource currently in the prototype phase will provide a comprehensive global catalog of PDX models available for researchers and their associated data across distributed repositories when formally launched at the end of 2017. PDX-MI will be used to validate data submissions from producers of PDX models and from data curated from the literature. PDX-MI will also inform scoring algorithms being developed in the NCI Oncology Models Forum to assess how well PDX models recapitulate hallmarks of human cancers.

Future versions of PDX-MI will capture additional details as procedures become more standardized. Input from clinical and translational professional societies will inform evolving requirements for diagnostic markers on a disease-specific basis. Given the recent success of immune checkpoint inhibitors in the treatment of cancer, improving “humanized immune system” PDX models is an area of intense research and PDX-MI will evolve to represent this. Other aspects of PDX models that are rapidly changing include improved surgical techniques and quality assurance methods. As we develop resources to capture and disseminate data related to PDX models, we will continue to improve and version PDX-MI to reflect the state of the art in the field. A web-based form to allow feedback from the community about the standard described here can be accessed at the Mouse Tumor Biology database website (http://tumor.informatics.jax.org). As has been demonstrated across multiple disciplines, a minimal standard adopted by a research community accelerates the rate of scientific discovery while reducing unnecessary duplication.

S. Ferretti is a laboratory head at Novartis. M.T. Lewis is a manager and limited partner at StemMed Holdings. E. Vinolo is a consultant/advisory board member for EurOPDX Consortium and XenTech. D.M. Weinstock reports receiving a commercial research grant from Novartis, AstraZeneca, Abbvie, Aileron, Roche, and Novartis, has ownership interest (including patents) in Travera, and has provided expert testimony for Monsanto. S.J. Weroha reports receiving a commercial research grant from Novartis, Genentech, and Tesaro and has ownership interest (including patents) in Mayo Clinic Ventures. A.J. Butte is a cofounder and scientific advisor at Personalis, Inc. and NuMedii, Inc., reports receiving a commercial research grant from Progenity, Inc., and has ownership interest (including patents) in NuMedii, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T.F. Meehan, T. Goldstein, G. Inghirami, S. Brabetz, Z. Gu, D.A. Begley, D.M. Krupke, C. Caldas, D.A. Clark, J.R. Dry, Y.A. Evrard, N.C. Goodwin, D. Greenawalt, E. Hermans, P.J. Houghton, T.O. Khor, J.M. Olson, J.K. Seong, R.J. Wechsler-Reya, S.J. Weroha, A.J. Butte, H. Parkinson, C.J. Bult

Development of methodology: T.F. Meehan, T. Goldstein, J.A. Wiser, P. Dunn, D.A. Begley, D.M. Krupke, J.H. Doroshow, N.C. Goodwin, D. Greenawalt, M.A. Haendel, E. Hermans, P.J. Houghton, T.O. Khor, M.T. Lewis, J.M. Olson, L. Trusolino, C.J. Bult

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.F. Meehan, N. Conte, D.A. Begley, Y.A. Evrard, E. Hermans, J. Jonkers, J.M. Olson, L. Trusolino, E. Vinolo, F. Amant, C.J. Bult

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Conte, T. Goldstein, G. Inghirami, D.A. Begley, D.M. Krupke, H. Dowst, J.H. Doroshow, J.M. Olson, O.M. Rueda, J.K. Seong

Writing, review, and/or revision of the manuscript: T.F. Meehan, N. Conte, T. Goldstein, G. Inghirami, M.A. Murakami, S. Brabetz, Z. Gu, P. Dunn, D.A. Begley, D.M. Krupke, A. Bertotti, A. Bruna, M.H. Brush, A.T. Byrne, C. Caldas, A.L. Christie, D.A. Clark, J.R. Dry, O. Duchamp, Y.A. Evrard, S. Ferretti, K.K. Frese, N.C. Goodwin, D. Greenawalt, M.A. Haendel, E. Hermans, P.J. Houghton, K. Kemper, M.T. Lewis, K.C.K. Lloyd, E. Medico, S.B. Neuhauser, D.S. Peeper, J.K. Seong, E. Vinolo, R.J. Wechsler-Reya, D.M. Weinstock, A. Welm, S.J. Weroha, F. Amant, S.M. Pfister, M. Kool, A.J. Butte, C.J. Bult

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.F. Meehan, N. Conte, M.A. Murakami, P. Dunn, D.A. Begley, D.M. Krupke, C. Caldas, O. Duchamp, K.C.K. Lloyd, J. Mason, S.B. Neuhauser, S.M. Pfister, C.J. Bult

Study supervision: N. Conte, T.O. Khor, H. Parkinson

Other (management of industry stakeholders): D.A. Clark

This project has been funded in whole or in part with federal funds from the National Cancer Institute, NIH: R01CA089713D (to C.J. Bult), P30CA034196 (to C.J. Bult), U24CA204781 (to H. Parkinson), U24CA195858 (to A.J. Butte), contract no. HHSN261200800001E (to J.H. Doroshow, Y.A. Evrard), R24OD011883 (M.H. Brush, M.A. Haendel), the Leukemia & Lymphoma Society SCOR grant (2015, 2016; to G. Inghirami), and the Science Foundation Ireland grants 13/CDA/2183 and 15/TIDA/2963 and the Irish Cancer Society Collaborative Cancer Research Centre BREAST-PREDICT Grant CCRC13GAL (A.T. Byrne). G. Inghirami, A. Bertotti, A. Bruna, A.T. Byrne, C. Caldas, E. Hermans, J. Jonkers, K. Kemper, E. Medico, D.S. Peeper, O.M. Rueda, L. Trusolino, and F. Amant are members of the EurOPDX Consortium.