Mouse Clinical Trials (MCTs) are a cornerstone of preclinical oncology research, providing valuable insights into drug efficacy and biomarker discovery. However, the success of these trials hinges on precise study design, appropriate model selection, and rigorous data analysis.
In this post, we’ll explore how to optimize each step of MCTs, from selecting the right indications to refining biomarker discovery. We will discuss how best to leverage key bioinformatics to drive data-based decisions that speed time to clinic and set your trial up for the greatest chance of success.
Selecting Indications and PDX Models
One of the most critical aspects of MCT design is selecting the right Patient-Derived Xenograft (PDX) model. PDX models maintain the genetic and molecular features of original patient tumors, making them highly relevant for studying tumor biology and drug response in a more clinically reflective manner.
The selection of models often requires considering multiple layers of data. For instance, gene expression data can be used to identify models that are more likely to respond to a specific drug. As an example, when studying an EGFR inhibitor, comparing EGFR expression levels across different cancer indications can help narrow down the indications that have high expression of the target (Figure 1).
Figure1. Boxplot of EGFR gene expression levels across multiple cancer indications in HuBase.
In some studies, mutation data may be more relevant. For example, selecting models with the KRAS G12C mutation could be essential when testing the efficacy of a KRAS inhibitor. This approach helps in focusing on models where the drug is more likely to be effective based on the mutation profile.
Beyond gene expression and mutations, pathway activity data can offer additional insights. For instance, when drug efficacy is linked to certain signaling pathways, such as HER2 downstream signaling, models with high activity in these pathways may offer more meaningful results.
Finally, proteomic data can also be a critical factor. In some cases, gene expression does not have high correlation with protein levels. For example, NECTIN4, may exhibit protein levels that are more indicative of response than its gene expression. In such instances, relying on protein expression data can refine model selection and improve study outcomes.
Crown Bioscience offers a database that provides access to a vast collection of PDX models, along with genomic, transcriptomic, and 4D-DIA proteomic data, making it easier to filter models based on these different criteria. This comprehensive data resource supports informed decision-making when selecting the most relevant models for a given study.
Determining the Number of Models and Animals
Once the models are selected, designing the study to ensure statistical power is the next challenge. The number of PDX models and the number of animals per group are important factors that must be optimized to balance study robustness with practical constraints.
Power calculations help determine how many models and animals are needed to ensure that the study can detect statistically significant differences between the test drug and control groups. Linear Mixed Models (LMM) can be employed to generate power curves, showing how different study designs will affect the ability to detect drug efficacy based on the potency of the drug.
In the case depicted below, a power curve shows that for a drug that reduces tumor growth by 30%, using a 1:1 design (one mouse per model per group) might require around 28 PDX models to achieve 80% power. However, using a 3:3 design (three mice per model per group) can reduce the number of PDX models needed to around 10, while still maintaining the same statistical power (Figure 2, right panel). Power curves provide a visual way to explore these trade-offs and guide decisions about study design.
Figure2. Power curves showing the effect of drug potency on the required number of models and mice.
Calculating Sample Size for Specific Models
When focusing on a specific PDX model, accurately calculating the number of animals required in each group is essential for achieving reliable results. This calculation depends on the tumor growth data from previous studies and other relevant historical data.
Using historical data, researchers can estimate the variability in tumor growth across control and treatment groups, and then apply power calculations to determine the sample size needed. These calculations take into account factors such as desired significance level, effect size, and power, ensuring that the study design is robust enough to detect statistically significant differences in drug efficacy.
Balancing Group Allocation in Mouse Clinical Trials
Randomization is a crucial step in experimental design to minimize bias. However, simple randomization schemes can lead to unbalanced groups, particularly when multiple covariates like tumor volume or body weight need to be considered. This imbalance can reduce the statistical power of the study and lead to unreliable results.
In mouse clinical trials, it is often necessary to allocate models into different treatment groups while maintaining balanced baseline covariates. Ensuring that groups are comparable at the start of the experiment is essential for attributing any observed differences in outcomes to the treatment itself rather than pre-existing differences.
To achieve this, more advanced randomization methods that account for multiple covariates can be used. For instance, randomization algorithms can help balance groups based on tumor volumes or other key factors, ensuring that each group starts on equal footing.
Improving Biomarker Discovery with Advanced Statistical Models
Biomarker discovery is a key objective in many preclinical trials. Identifying biomarkers that can predict treatment response helps in patient stratification and can inform clinical trial design. However, simple statistical methods may sometimes lead to false-positive findings, especially when the relationship between gene expression and drug efficacy is complex.
More sophisticated methods, such as Linear Mixed Models (LMM), are better suited for analyzing the complex interactions between gene expression, drug treatment, and tumor response. In a study of cetuximab in gastric cancer PDX models, simple statistical analysis ranked EGFR relatively low as a biomarker. However, when using LMM, which explicitly models the effect of gene expression on tumor growth, EGFR emerged as the most significant biomarker (Figure 3).
This approach reduces the risk of false positives and increases the likelihood of discovering truly predictive biomarkers, leading to more reliable results in preclinical studies.
Figure3. EGFR ranking by various efficacy measurements, showcasing the precision of LMM in identifying predictive biomarkers.
Conclusion
Selecting the right models, determining appropriate sample sizes, and balancing groups are just a few of the critical steps in designing effective Mouse Clinical Trials. By leveraging a combination of gene expression, mutation, pathway activity, and proteomic data, researchers can make more informed decisions about which PDX models to use. Integrating advanced statistical methods such as LMM further ensures that biomarker discovery is accurate and reliable.
Bioinformatics tools, like the ones offered by Crown Bioscience, help streamline this process by providing access to comprehensive data and advanced analysis methods, enabling researchers to optimize every stage of their MCTs.
PDX Mouse Clinical Trials
Translational Models, Predictive Biomarkers, New Indications, Deeper Drug Insights
