Trends in Oncology

Imagine a world where doctors can predict how a tumor will respond to treatment before a patient starts therapy. For decades, cancer researchers have relied on The Cancer Genome Atlas (TCGA), a massive dataset of 20,000 samples across 33 cancer types, to decode the molecular secrets of tumors. But TCGA has a blind spot: it mostly studies untreated, primary tumors, leaving critical questions about advanced cancers and treatment responses unanswered.

Enter the Champions Oncology TumorGraft^® platform—a game-changer in cancer research. With 1,500 patient-derived tumor models from over 50 cancer types, TumorGraft^® captures the real-world complexity of advanced, metastatic, and heavily treated tumors. By combining molecular data with detailed pretreatment histories and treatment response insights, it’s opening new doors for predicting drug effectiveness, uncovering resistance mechanisms, and designing smarter therapies. In this post, we’ll explore how TumorGraft^® complements TCGA, using a fascinating case study on mutational signatures to show its power.

TCGA: The Gold Standard with Limits

TCGA is a cornerstone of cancer research. It's publicly accessible, high-quality data, spanning DNA, RNA, and proteins, has fueled countless discoveries about how tumors develop. Researchers use it to characterize tumors, improve diagnoses, and identify molecular drivers of cancer.

But TCGA isn’t perfect. Most of its samples come from primary, untreated tumors, with only a small fraction (about 42 patients) having received neoadjuvant treatment. This makes TCGA ideal for studying cancer’s early stages but less useful for advanced, metastatic, or post-treatment tumors. It also underrepresents certain populations and cancer stages, lacks pretreatment histories, and doesn’t allow access to physical samples for follow-up experiments. If you’re studying treatment resistance or real-world patient outcomes, TCGA’s data can only take you so far.

TumorGraft^®: A Window into Advanced Cancers

The TumorGraft^® platform, developed by Champions Oncology, flips the script. Its 1,500 patient-derived xenografts (PDXs)—tumors grown in mice to mimic human cancer—represent over 50 cancer types, focusing on advanced-stage, metastatic, and pre-treated tumors. These models reflect the diversity and complexity of patients seen in clinics, where cancers often evolve under the pressure of multiple therapies.

Unlike TCGA, TumorGraft^® includes detailed pretreatment information, capturing the therapies patients received before their tumors were sampled. This data, which can be further mined, offers a window into how prior treatments shape tumor biology, enabling researchers to study real-world clinical scenarios. Combined with treatment response data from in vivo experiments, where Champions Oncology has tested drugs representing multiple standards of care and measured Tumor Growth Inhibition (TGI)—how much a drug slows tumor growth—TumorGraft^® unlocks use cases TCGA can’t touch:

•    Predicting Treatment Responses: See how a tumor’s molecular profile and pretreatment history predict its reaction to specific drugs.
•    Discovering Biomarkers: Identify markers that signal whether a treatment will work, informed by prior therapies.
•    Understanding Resistance: Study why some tumors resist therapy and find ways to overcome it, leveraging pretreatment data.
•    Improving Combination Therapies: Test drug combinations to find the most effective mixes, considering treatment histories.
•    Finding New Drug Targets: Link molecular features and pretreatment patterns to treatment outcomes to uncover novel therapies.

Mutational Signatures: A Shared Language

To prove TumorGraft’s^® reliability, researchers compared it to TCGA using mutational signatures—DNA damage patterns that act like fingerprints, revealing what caused a tumor, like UV light or faulty DNA repair. These signatures, cataloged in the COSMIC database, help identify vulnerabilities in tumors and guide treatment strategies. The analysis began by curating COSMIC signatures and calculating their exposure and frequency in both TCGA and TumorGraft^® datasets.

Figure 1: Diagram showing calculated exposure/frequency of tumors in TumorGraft^® (left) and TCGA (right).

The results were striking. Both datasets flagged similar patterns. For example, SBS7a—a signature tied to UV light exposure (COSMIC SBS7a)—was strongly linked to melanoma in both TumorGraft’s^® PDX models and TCGA’s samples. Analysis of 1,155 PDXs revealed SBS7a as a hallmark of melanoma, mirroring TCGA’s findings. This alignment shows that TumorGraft’s^® data is as biologically accurate as TCGA’s, despite its focus on advanced, treated tumors.

Figure 2: Chart of multi-omic data analysis from 1,155 TumorGraft^® PDXs, highlighting trends in mutational signatures across tumor types.

Figure 3: Bar chart showing SBS7a mutational signature prevalence in melanoma PDXs from TumorGraft^®, with high association compared to other cancer types.

Researchers analyzed signatures using different motifs—single base substitutions (SBS96), double base substitutions (DBS78), and insertions/deletions (IND83). All signatures found in TumorGraft^® appeared in TCGA, with TCGA’s larger sample size revealing a few extra patterns due to its scale.

Figure 4: Comparison of mutational signatures across SBS96, DBS78, and IND83 motifs in TumorGraft^® and TCGA datasets.

A UMAP plot, a visual tool that maps data similarity, confirmed that tumors cluster by type (e.g., melanoma, lung), not by whether they came from TCGA or TumorGraft^®. This suggests that biology, not the data source, drives the differences—a green light for using TumorGraft^® alongside TCGA.

Figure 5: UMAP scatterplot showing tumors clustered by type, not data source, confirming TumorGraft’s^® biological consistency with TCGA.

Figure 6: Dot plot showing associations between mutational signatures and metadata, with dot size indicating event frequency and color representing p-value.

Notably, tumors with signatures SBS6, SBS15, and SBS10b showed resistance to alkylating agents and platinum-based chemotherapies, exhibiting lower TGI. According to COSMIC, SBS6 and SBS15 are linked to defective DNA mismatch repair, common in microsatellite-unstable tumors, while SBS10b is tied to mutations in DNA polymerase epsilon, often seen in hypermutator tumors. By mining pretreatment data, researchers can explore how prior therapies influence these resistance patterns, offering clues to personalize treatments and avoid ineffective drugs.

Figure 7: Graph showing correlation between SBS6, SBS15, and SBS10b signatures and resistance to alkylating agents/platinum in TumorGraft^® PDXs, with lower tumor growth inhibition (TGI).

This is just the beginning. TumorGraft’s^® pretreatment data, combined with patient treatment histories and in vivo responses to thousands of standard-of-care drugs, is a goldmine for studying how prior therapies shape tumor evolution and treatment outcomes.

Why TumorGraft^® Matters

The similarities between TCGA and TumorGraft’s^® mutational signatures prove that TumorGraft^® is a reliable partner to the gold standard. But its focus on advanced, treated tumors, detailed pretreatment information, and treatment response data takes cancer research to new heights. The ability to mine pretreatment histories—unavailable in TCGA—enables researchers to uncover how past therapies influence tumor biology, paving the way for personalized medicine and better patient outcomes. Whether you’re developing new drugs, tackling treatment resistance, or designing combination therapies, TumorGraft^® provides insights that TCGA can’t.

Looking ahead, researchers can dig deeper with TumorGraft^® by exploring copy number signatures, RNA sequencing, or proteomics. Mining pretreatment data alongside these analyses could reveal even more about how tumors evolve under therapeutic pressure, driving breakthroughs in cancer care.

Ready to Transform Cancer Research?

The TumorGraft^® platform is more than a dataset—it’s a bridge to personalized medicine. By combining the molecular depth of TCGA with TumorGraft’s^® real-world treatment and pretreatment insights, researchers can unlock answers that bring us closer to curing cancer.
Want to explore TumorGraft^® for your next study? Learn how our platform can power your research today.

Note: Data sourced from Champions Oncology and validated against TCGA’s data using the mutational motifs in COSMIC.

Gene copy number in a tumor cell is a significant indicator of the implication of a given gene in several oncogenic processes such as uncontrolled proliferation, elusion of programmed cell death, and resistance to treatments.

At Champions Oncology, gene copy number analysis is performed by using the EXCAVATOR2^[1] tool on whole exome sequencing (WES) data generated to characterize our patient-derived xenograft (PDX) models. This tool allows for classifying each segmented region into five qualitative genomic states (two-copy deletion, one-copy deletion, normal, one-copy duplication, and multiple-copy amplification) and quantifying the number of chromosomal copies.

All our model characterization data can be explored in Lumin, a unique solution integrating Champions’ tumor model multi-omic data and public datasets in one accessible platform for model selection and data interpretation.

In Lumin, gene copy number analysis results are presented in the format shown in the example below:

Here we answer the most common questions about CNV data reporting to help you navigate WES data in our Lumin platform as well as interpret your own study data.

Q: What does Log2R mean and why sometimes is it marked as NA?

A: The Log2R (Log2 ratio) value represents the number of copies relative to the normal reference sample (NA12878). The EXCAVATOR2 algorithm used to calculate CNV uses a median normalization approach, with the log-transformed ratio (Log2R) being calculated from the window mean read count (WMRC) values of the test sample compared to the normal reference. When Log2R value is marked as NA, no significant copy number alteration was detected.

Q: What are the Call values and how are they defined?

A: The Call value is calculated using the FastCall algorithm^[2] and classifies each segmented region as one of five possible states: 2 copy deletion= -2; one copy deletion= -1; normal= 0; one copy duplication= 1; and multiple copy amplification= 2.

Q: Are the copy number values the absolute or the relative copy numbers detected?

A: Copy number values represent the absolute copy number detected, which is derived from the Copy Number Fraction and is rounded to the nearest integer.

Q: Do “Amp” and "Del" always mean that there is a gain or loss in copy number? Or is this only the case if the “Alteration” column says “Gain” or “HomoDel”/ “Hetloss”?

A: The "amp_del" column definitions are derived from the Call values, whereas the "Alteration" column classifies the alteration based on the copy number detected. In some instances, there may be discordance between the copy number and Call (as shown for FAM231C gene in the example table above), as the two values are derived by approximation of continuous values. In this specific example, the conflict between the two annotations is indicative of the tumor gene copy number being between 1 and 2, suggesting the presence of both cells with 1 and cells with 2 copies of that genomic region. For additional investigation, we recommend looking at the continuous values in the raw data.

Q: What columns should I consider when I want to search for a model with an amplification/deletion for a certain gene?

A: We would first recommend using the "CopyNumber" column to identify models with an amplification or deletion of a specific gene. Once you have filtered the models based on this, you can then use the “Alteration” column to verify whether Call, copy number, and the “amp_del” column values are concordant.