ROADMAPS: An Online Database of Response Data, Dosing Regimens, and Toxicities of Approved Oncology Drugs as Single Agents to Guide Preclinical In Vivo Studies

Human tumor xenografts in mice are widely used models for translational studies of experimental anticancer agents (1). Several groups have reported that tumor responses in xenograft models correlate with clinical responses in humans (2–4). Data from patient-derived xenograft (PDX) models have demonstrated consistent results across multiple PDX Development and Trial Centers using varying standard operating procedures (5). Despite the potential predictive value of these experiments for clinical trials in humans, drug-dosing information and vehicle selection from xenograft preclinical studies are not always published. These studies are often the first indicators of reproducible drug activity and/or toxicity in vivo; the results can be used to inform further evaluation of the agents in appropriate solid tumor types (6–8). To our knowledge, these data have never been compiled into a single searchable, publicly available database.

For more than 30 years, the NCI Biological Testing Branch (BTB) has maintained a public repository of human and rodent tumors and cell lines as a research resource for preclinical pharmacologic and pharmacodynamic studies (dtp.cancer.gov/organization/btb/tumor_repositories.htm). In addition, the BTB actively evaluates the in vivo efficacy and the pharmacologic and pharmacodynamic characteristics of potential anticancer compounds (9–11). A critical component of these efforts is the design and conduct of xenograft and allograft studies to define the in vivo efficacy of new drugs and drug combinations (12). To facilitate preclinical study design in the research community, data collected by the BTB have been compiled into a spreadsheet named ROADMAPS (Responses to Oncology Agents and Dosing in Models to Aid Preclinical Studies, dtp.cancer.gov/databases_tools/roadmaps.htm). ROADMAPS includes data on drug dose, dosing schedule, route of administration, responsiveness (i.e., sensitive vs. resistant tumor types), maximum weight loss, drug-related deaths, and the drug formulation vehicle.

The NCI-60 cancer cell panel has been used for years as a screening tool for investigational agents (13). The ROADMAPS spreadsheet includes many cell lines from the NCI-60 panel, but those cell lines constitute less than half of the models included in ROADMAPS at the time of this writing (December, 2021). ROADMAPS, which currently includes data for 70 agents and 140 tumor models, can be filtered to highlight tolerable dosing regimens with activity against a variety of human tumors in different immunocompromised mouse strains and will therefore serve as a valuable resource for investigators in planning preclinical studies.

NCI-60 cell lines used to generate the xenograft data in this report were obtained from the Division of Cancer Treatment and Diagnosis Tumor Repository (Developmental Therapeutics Program, Frederick National Laboratory for Cancer Research, Frederick, MD). Other cells lines or tumor fragments were provided by NCI investigators, purchased from the ATCC, Jackson Laboratories, or other commercial sources. All tissue procurement trials followed protocols approved by Institutional Review Boards and patients provided written informed consent. The identities of all cell lines used in this study were confirmed using Identifiler short tandem repeat (STR) genotyping (Applied Biosystems) since that technology became available; studies conducted prior to the advent of STR profiling could not be characterized in this manner. Each cell line was tested for Mycoplasma when it was accepted into the repository and at each new lot preparation. Routine Mycoplasma testing of cells in culture was not performed. All cell lines were screened for rodent and human viral pathogens prior to inoculation into mice using testing methodologies available at the time (e.g., mouse antibody production testing, PCR).

Agents were selected for testing in consultation with the NCI's Drug Synthesis and Chemistry Branch (DSCB), in response to requests from the NCI's Biological Evaluation Committee within the Developmental Therapeutics Program (DTP), or through other approved methods in the NCI's Division of Cancer Treatment and Diagnosis. Preparation and storage conditions varied by agent. While recommendations from the DSCB informed the handling of each agent, some broad generalizations can be reported. Agents requiring DMSO or ethanol were prepared in stock solutions at 10× the desired concentration and frozen until ready to administer, at which time, stock solutions were thawed and diluted with 9 volumes of the appropriate vehicle. Water-soluble agents were stored as dry powders and resuspended in the appropriate vehicle prior to dosing. Doses were administered on the basis of individual animal body weights rather than group averages. Agents were administered via several routes. For intravenous administration, agents were injected into the lateral tail vein using 27- to 30-gauge needles with mice restrained in commercially available mouse restrainers. For intraperitoneal administration, agents were injected through the abdominal body wall using 23- to 25-gauge needles attached to 1 mL syringes while the mouse was held in the non–syringe-containing hand. For oral administration, agents were administered via 20- to 22-gauge malleable stainless steel feeding needles or flexible oral gavage needles. The standard dosing protocol called for drug solutions to be prepared at concentrations at which 0.1 mL of drug solution was administered per 10 g of body mass (i.e., 0.265 mL of solution would be administered to a 26.5 g mouse). This standardized procedure reduces the risk of dose calculation errors, ensures the dose (mg/kg) is administered accurately, and decreases the time required to perform the injections.

Animal experiments were performed at the Frederick National Laboratory for Cancer Research and the Southern Research Institute (SRI); both are accredited by Association for Assessment and Accreditation of Laboratory Animal Care International and follow the Public Health Service Policy for the Care and Use of Laboratory Animals. Animal care was provided in accordance with the procedures outlined in the Guide for Care and Use of Laboratory Animals (14).

Dosing schedules, route of drug administration, tumor models tested, and vehicle used for agent formulation were determined for each individual study following established methods and study designs developed in the BTB (12). For mouse inoculation, tumor cells were used at the fourth to sixth in vitro passage from cryopreserved cell stocks. Cells (typically 1 × 10⁷ cells/0.1 mL/injection) were subcutaneously inoculated into female mice (nu/nu NCr mice or SCID/NCr mice) and therapeutic studies were initiated upon reaching a target tumor volume of 100 to 400 mm³, depending upon the specific study design. Male mice were used for male-specific tumor models (e.g., prostate cancer models) and for other models when availability of female mice was limited. Many of these studies used serially passaged tumors following previously published methods (12). Briefly, donor tumors were harvested, cut into 2 to 3 mm³ fragments, and implanted subcutaneously using a 9- to 11-gauge tumor implant trocar. Tumors were staged and the mice randomized into groups at the experimentally defined tumor volume ranges (e.g., 125–250 mg, 200–400 mg). More mice were implanted than required for the study so that outlying tumor volumes could be excluded.

Drug-dosing regimens varied and are indicated with each entry in the database. For regimens requiring multiple doses per day, the frequency and number of doses are indicated [e.g., azacitidine was administered twice daily for six doses (BID×6) to mice bearing OVCAR-3 tumors]; other regimens were dosed at multi-day intervals [e.g., methotrexate was dosed against multiple tumor models every 4 days for three doses (Q4D×3)]. Length of drug treatment varied on the basis of the agent and regimen, and mice were followed until tumors reached a calculated mass of 4,000 mg for experiments conducted during or prior to 2000, or 1,500–2,000 mg for experiments conducted from 2001 onward. Control mice were administered drug-free vehicle.

Vehicle control groups typically included 16–20 mice, while drug treatment groups typically included 6–10 mice. If tumors failed to progressively grow in an experimental mouse, that mouse was classified as a “no take.” One or two “no takes” were allowed in an experiment, although these mice were excluded from median tumor mass calculations. If more than two “no takes” occurred in the control group, the experiment was considered to have failed quality control and was not included in ROADMAPS. Drug response (i.e., whether a tumor is responsive or nonresponsive to the regimen tested) was determined by calculating the percent test/control (%T/C) of median tumor weights on each day that tumors were measured during the study. Any %T/C less than 40%, regardless of when it occurred during the study, met the DTP threshold for reporting minimal drug activity (i.e., a 60% reduction in median tumor volume in test mice compared with tumors in control mice treated with drug-free vehicle). Tumor masses in milligrams were calculated as (length × width²)/2, with length and width in millimeters as measured using bidirectional calipers. Both manually read calipers and electronic calipers were used depending on when the studies were conducted. Manual caliper data were collected by hand with subsequent manual entry into an electronic database for endpoint calculations. When StudyLog software (StudyLog Systems) became available, data collection was accomplished with electronic calipers for automated data upload. Mean weight loss, as a percent of the animals’ starting weights, are reported; animals were sacrificed if weight loss exceeded 30% or earlier if there were clinical signs of toxicity in addition to weight loss. Drug-related deaths are also included as a surrogate for toxicity, reported as number of dead animals/total number of animals in a specific dosing cohort.

ROADMAPS includes dose(s) tested, dosing schedule, route of administration, mouse strain, maximum weight loss, drug-related mortality, vehicle, and whether the model was responsive or unresponsive (quantified as %T/C, with positive responses indicating that a drug/dose/route combination resulted in median tumor weights in treated mice that were no more than 40% of the median tumor weight in control mice that received drug-free vehicle at one or more timepoints). Data can be filtered to compare responses with specific agents, dosing regimens, or tumor types. Seventy agents were tested against one or more xenograft models (Table 1). ROADMAPS currently includes data from 140 xenograft models (Table 2). Doxorubicin was tested against more models (76) than any other agent; HCT-116 was tested against more drug/dosing combinations (41) than any other model. A total of 3,161 drug/dosing combinations have been tested at the time of this writing and incorporated into a spreadsheet with 1,212 entries; multiple doses are included in a single entry when other conditions and responses are identical (i.e., methotrexate did not induce responses when dosed Q4D×3 against HOP-92 non–small cell lung cancer (NSCLC) cells at 18, 27, or 45 mg/kg; all three dose levels are included in one entry).

All nine histologies [lung, melanoma, renal, colon, central nervous system (CNS), leukemia, breast, ovarian, and prostate cancers] represented in the NCI-60 panel are also included in ROADMAPS. In addition, ROADMAPS includes lymphoma and bladder cancer models, as well as a collection of “other” models (e.g., head and neck, gastric, leiomyosarcoma; Fig. 1). This initial version of ROADMAPS includes 52 models that are in the NCI-60 panel and an additional 88 models that are not part of the NCI-60. Each model was tested against different agents following differing administration regimens, routes, doses, and vehicles. The number of ROADMAPS entries per model ranges from 13.4 entries per model for renal cancer models to 3.3 entries per model for “other” models; the mean number of entries per model across the entire dataset of xenograft models was 8.4 (1,177 entries for 140 tumor models). An additional 35 entries report results from experiments with five transgenic mouse models (four breast cancer, one prostate cancer) and one canine model (osteosarcoma); these models are included in the downloadable spreadsheet.

Paclitaxel is presented as an example of how ROADMAPS can be filtered. At present, there are 69 unique entries for paclitaxel administered on different schedules, via different routes, in different vehicles, and in different models. To facilitate comparison, these entries were filtered to include combinations involving only intravenous administration of paclitaxel in vehicle containing ethanol and Cremaphor, with 58 entries matching these criteria. Of those 58 entries, tumor responses were observed in 39 entries, while 19 did not yield tumor response (67.2% positive responses). Pharmacokinetic factors associated with paclitaxel monotherapy influence clinical outcomes in patients (15). Paclitaxel was administered daily in 33 entries with 84.8% yielding positive tumor responses (28/33); positive responses were observed in 44.0% (11/25) of entries that utilized other dosing schedules. MDA-MB-231 tumor was responsive when paclitaxel was administered daily for 5 days at doses from 6.7 to 22.5 mg/kg; this model was not responsive to 10 mg/kg paclitaxel administered every 4 days for three doses. These data are consistent with the clinical observation that similar tumors respond differently to an agent based on the administration schedule.

A broad spectrum of tumor histologies are included in the dataset (Fig. 1; Table 2). Colon cancer models are the most abundant group in the current iteration of ROADMAPS, with 173 different entries in the spreadsheet. Included within this group is the single most tested model, HCT-116, with 41 distinct entries (13 responsive entries and 28 unresponsive entries). Nine agents led to tumor responses (dactinomycin, mitomycin C, vincristine, bleomycin, decitabine, sirolimus, clofarabine, topotecan, and irinotecan) via differing routes of administration and dosing schedules, while 24 agents failed to induce tumor response (Table 3). Several agents (mitomycin C, vincristine, bleomycin, and topotecan) led to drug toxicity at higher doses, but all these agents induced tumor response without mortality at lower doses. For example, 2 of 10 mice died after receiving mitomycin C at 4.5 mg/kg, while no deaths occurred in mice receiving mitomycin C at 2 or 3 mg/kg. An additional six entries include data from five cell lines derived from HCT-116 [HCT-116-luc2, HCT-116 (Pommier), HCT-116/Mre11Ch, HCT-116H1, and HCT-116B]; these cell lines are not included in the above description of HCT-116 results. These data indicate ROADMAPS may be used to guide preclinical studies with regard to toxic dose levels in addition to responsiveness.

Use of preclinical xenograft models to evaluate the effects of cancer drugs on human tumor growth in vivo is a well-established component of the drug development pathway (4, 5, 9, 16). As such, the data and methods used to generate data from preclinical studies are not always published. The database of preclinical data outlined in this article addresses an unmet need for such information. Although the results of preclinical drug evaluation studies are not always predictive of human clinical activity and antitumor immune responses cannot be evaluated in immunocompromised mice, evaluating agents against xenograft models can facilitate the identification and optimization of in vivo dosing regimens appropriate for further testing (16). Further studies could assess whether toxicities are comparable in additional mouse strains, as SCID mice have known DNA repair defects (17), making them more sensitive to DNA-damaging agents than athymic nude mice. In addition, the role of metastasis in disease progression must be considered when reviewing data in ROADMAPS, as orthotopic implantation may result in more clinically relevant tumor spread (18) as well as differences in drug exposure at various body sites due to the drug's absorption, distribution, metabolism, and excretion characteristics.

It is anticipated that the data shared in ROADMAPS will help investigators to select tolerable and active dosing regimens for single- and combination-drug studies, as well as to identify suitable, sensitive tumor types. To this end, an optimal 40% T/C threshold was selected for sorting tumor growth into a qualitative yes/no response filter in ROADMAPS; multiple studies over time have used this threshold and statistical analysis has demonstrated sufficient statistical power to evaluate tumor responsiveness (19). One such analysis calculated that groups of 6 mice with a “moderate” coefficient of variation (defined as CV = 0.6) would have 80% power to detect a 60% reduction in mean relative tumor volume (i.e., a T/C of 40%) using a one-sided t test with α = 0.05 and assuming equal numbers of mice in test and control groups (20). Drug-treated groups in ROADMAPS included 6–10 mice, while control groups included 16–20 mice, suggesting that these studies would have at least 80% power to detect 40% T/C ratios given similar CV values.

ROADMAPS includes drugs with a wide variety of mechanisms of action, including cytotoxic agents (e.g., methotrexate, doxorubicin, cyclophosphamide, cisplatin), targeted agents (e.g., imatinib, everolimus, pazopanib, dasatinib), and drugs used in the adjuvant setting (e.g., tamoxifen, abiraterone). Several agents have been tested against multiple models with similar tumor histologies and demonstrated differing responses. As has been reported previously, topotecan has antitumor activity against A375 melanoma xenograft tumors as a single agent, whereas human Colo829 melanoma xenografts are unresponsive to topotecan at the same doses (9). Given the number of agents and models tested in ROADMAPS, no simple nomogram exists for converting mouse dosing regimens to human equivalents. This goal was also hindered by the fact that mouse model experiments are often conducted before human doses are known. Mouse doses presented here may not have had equivalent doses tested in humans due to lack of efficacy or to toxicity. However, an extensive comparison of mouse and human dosing has been published (21).

Relationships derived from these data can be applied to study designs where it may be more efficient to use a sensitive model during early evaluation of a new anticancer agent or agents before committing resources to optimizing the dosing regimen in more resistant models. The four tumor types with the most entries in this dataset (colon, lung, melanoma, and breast; Fig. 1) are all among the five most common cancer types in the United States (breast, prostate, lung, colorectal, and melanoma; https://www.cancer.gov/types/common-cancers). Several models within each tumor type have been tested against many agents (Table 2). Furthermore, the BTB has previously reported details on growth rates and gene expression profiles of a panel including 49 human tumor xenografts (22). Of those 49 cell lines, 42 are included in ROADMAPS and account for 629 of the 1,212 (51.9%) total entries in the database. The growth characteristics of many NCI-60 cell line xenografts were reported by Plowman and colleagues along with their sensitivities to a panel of 12 agents (12). In addition, growth curves for a selection of human tumor xenografts are available at http://dtp.cancer.gov/organization/btb/growth_assay_data.htm. The ROADMAPS database substantially expands on these prior publications in terms of both the number of models and agents tested, and organizes the data into a searchable database. Such a breadth of data on the most common tumor types will streamline the search for appropriate models in which to test novel agents and should facilitate the translation of these agents from preclinical to clinical studies. The data will also help identify doses and tumor types to avoid, whether from lack of activity or association with morbidity.

Complementary pharmacokinetic data to aid determination of desired exposures in nonclinical studies have been published (21). Data presented here demonstrate the relative antitumor efficacies when an agent is administered at differing intervals or via different routes. For example, the Colo829 melanoma model was responsive to intravenous paclitaxel administered daily for 5 days. This model was not responsive to paclitaxel administered weekly at the same doses via the same route. The MDA-MB-231T breast cancer model was responsive to paclitaxel administered intravenously on both daily (QD×5) and weekly (Q7D×3) schedules. However, while weekly dosing was well tolerated (i.e., no mortality or weight loss at any dose tested), daily dosing at the highest dose (15 mg/kg) resulted in drug toxicity for 3 of 8 mice. These examples demonstrate the utility of these data in identifying suitable models and dosing regimens.

ROADMAPS is the first publicly available resource with data compiled over many years available in a filterable format. The NCI's BTB will continue its work testing anticancer agents against various tumor models in mice. Targeted agents tend to be newer and therefore have been tested against fewer models than older agents; multiple ongoing studies are evaluating the efficacy of targeted agents against an array of tumor models. As the BTB expands its repertoire of drug studies in PDX as well as xenograft models, the spreadsheet will be updated periodically, with the most recent version available for download by members of the research community.

No disclosures were reported.

M.G. Hollingshead: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. N. Greenberg: Data curation, validation, investigation, methodology, writing–review and editing. M. Gottholm-Ahalt: Data curation, formal analysis, validation, writing–review and editing. R. Camalier: Supervision, funding acquisition, investigation, methodology, writing–review and editing. B.C. Johnson: Formal analysis, writing–original draft, writing–review and editing. J.M. Collins: Writing–original draft, writing–review and editing. J.H. Doroshow: Funding acquisition, project administration, writing–review and editing.

This project has been funded in whole or in part with federal funds from the NCI, NIH, under contract no. HHSN261200800001E. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.

The authors thank Jacqueline Plowman (NCI), William Waud (Southern Research Institute), and Donald Dykes (SRI) for their efforts. This work spanned many years, and several of those involved have since passed away. The authors recognize the contributions of the late Joseph Mayo (NCI), Betty Abbott (NCI), Daniel Griswold Jr (SRI), John Venditti (NCI), and Russell Laster (SRI) to this project.