NEW YORK – A bioinformatics tool developed by researchers at MD Anderson Cancer Center may help researchers find interactions between tumor-associated genes and immune checkpoints that are predictive of immunotherapy response.
Immune checkpoints that are associated with tumor-related processes and immune phenotypes are known to be drivers of immune evasion and may be better therapeutic targets than checkpoints without those dual associations. As an example, expression of the gene SERPINB9 is upregulated in tumor cells by the presence of interferon gamma in the tumor microenvironment, and this can make patients resistant to CTLA4 checkpoint inhibition. Thus, targeting SERPINB9 and CTLA4 at the same time has the potential to kill cancer cells while overcoming resistance. However, relatively few such tumor-immune interactions are documented.
In a publication last month in Communications Biology, researchers led by Anil Korkut, an assistant professor of bioinformatics and computational biology at MD Anderson, described the development of a tool to map those interactions, ImogiMap, and use of it to identify tumor-immune interactions in uterine corpus endometrial cancer and basal-like breast cancers that co-associate with interferon gamma.
Currently available tools such as TIMER, CIBERSORT, and Kassandra have been used to study immune signatures and explore interactions between the immune system and gene expression but haven't been rigorously statistically validated, according to the study authors.
"This is one of many papers we've published in a series of attempts to understand how cancer cells are interacting with their surroundings and with drugs, mainly precision therapy agents," said Korkut, adding that the starting point for the study was to identify immune checkpoint mechanisms and explore functional tumor interactions, which in turn could lead to the identification of new drug targets for immunotherapies.
In developing ImogiMap, Korkut and his collaborators measured how the expression of immune checkpoint and tumor-intrinsic pathways impact certain immune phenotypes such as overall immune enrichment in the system, interferon gamma response, or other key immune events that could be predictive of immunotherapy response. He and his colleagues computed tumor-immune interactions for nearly 10,000 patients with 32 types of cancer and a range of oncogenic pathways and immune checkpoint molecules.
To validate the tool, Korkut and colleagues explored interactions between genes involved in T-cell dysfunction and interferon gamma gene expression in uterine corpus endometrial carcinoma and breast cancer. They found a set of significant interactions associated with interferon gamma gene expression and confirmed SERPINB9's previously documented role in T-cell dysfunction.
Researchers can access and use ImogiMap through a web app. "People can type the genes that they are most interested in and see if there is any association between immune checkpoints and immunologic responses in a particular cancer type," said Korkut. The results come with statistical analysis, network models, and graphical representations of interactions and pathways. "These are ready-to-go, completed results for nearly 10,000 patients," he noted.
Korkut said at this point ImogiMap is still a discovery tool that gives immunotherapy response predictions for further testing in the laboratory and in clinical trials. The tool right now could be useful, he believes, for researchers who want to better understand how tumor cells evade the immune system. "With this framework, we can identify these interactions, and we can use these network models to build more predictive models of drug response," Korkut said.
The goal, ultimately, is to use the tool to match patients to immunotherapy drugs, and Korkut's team is doing more work in this regard. "The next step, which we are pursuing in the lab, is to put this into an AI framework where we can make more systematic predictions on which particular interactions will predict responses to immunotherapy on a case-by-case basis for patients or groups of patients," said Korkut.
Korkut also plans to incorporate spatial omics analysis into studies of tumor-immune interactions using ImogiMap. While the ImogiMap algorithm currently provides insights into tumor-immune interactions for groups of patients, future iterations will allow detection of interactions at the single-cell level with positional information within tumors. "These improvements will enable us to understand the heterogeneity of immune-tumor interactions at the finest possible detail in each tumor," he said.
Korkut is also working with other translational researchers at MD Anderson to apply his computational methods in the ARTEMIS trial. In that study, cosponsored by MD Anderson and the National Cancer Institute, investigators are using ImogiMap to analyze the genomic signature of triple-negative breast tumors in newly diagnosed patients and home in on the treatment most likely to benefit them. Following several cycles of standard chemotherapy, patients may choose to continue with chemotherapy or participate in a drug trial based on their breast cancer subtype and the molecular profile of their tumor. Researchers will track patients' responses to the treatments they receive and secondary endpoints like survival.
"We can use this clinical framework to find groups of patients that may have improved response to neoadjuvant therapy with new agents," Korkut said. "That's an ongoing effort with these tools as well as other tools from our groups."
Korkut's group has also developed the Recurrent Features Leveraged for Combination Therapy, or REFLECT, platform, which identifies drug targets by using machine learning and cancer informatics algorithms to analyze tumor features such as mutations, copy number variations, gene expression, and protein aberrations. In 2022, the team published results from a retrospective validation study in Cancer Discovery showing that drug combinations identified by the platform decreased median tumor volume by 34.5 percent in preclinical patient-derived xenografts models, whereas tumor volume increased by a median of 5.1 percent in models treated without the help of the platform.