NEW YORK – With a combination of multiomic analyses and machine learning, an international team has tracked down endometrial carcinoma (EC) subtypes, along with potential predictive models, treatment targets, and markers for treatment response.
"Analysis of this independent cohort, incorporating pre-existing EC tumor and cell line cohorts, has not only confirmed published findings from our recent exploratory studies, but also provided biological insights relevant to potential therapeutic strategies," co-first author Lizabeth Katsnelson, a graduate student in the Davoli and Fenyö labs at the NYU Grossman School of Medicine, and her colleagues wrote in their paper.
As they reported in Cancer Cell on Thursday, members of the Clinical Proteomic Tumor Analysis Consortium used a combination of exome sequencing, whole-genome sequencing, microRNA sequencing, and array-based methylation profiling — together with targeted and global proteomic analyses, targeted assays, and other approaches — to assess 138 EC tumors and 20 normal endometrium-enriched tissues.
"Our study demonstrates the ability of proteogenomic analysis to increase our understanding of EC tumor biology and to generate new hypotheses," the authors explained. "We have highlighted a few examples of integrative analysis across the omics data modalities that can provide insights with potential clinical applications."
Tapping into this multiomic dataset, the team identified both new and known tumor clusters marked by specific copy number variant (CNV), gene expression, protein, phosphosite, and acetylation site patterns. The work also pointed to frequently mutated genes such as PTEN, PIK3CA, and ARID1A, and to potential treatment response-related biomarkers relevant to immunotherapy and other treatment types.
"Our study validates many previous molecular findings," Katsnelson said in an email. She noted that the team's analyses largely focused on four genomics-based endometrial carcinoma subtypes flagged in a 2013 paper from the Cancer Genome Atlas project: the DNA polymerase epsilon ultramutated, microsatellite instability hypermutated, CNV-high, and CNV-low subtypes.
Even so, the researchers also unearthed previously unappreciated subtypes with potential prognostic implications. In a subset of tumors with lower-than-usual MYC activity, they saw enhanced response to metformin treatment, for example, while tumors with in-frame insertions and deletions in PIK3R1 coincided with an uptick in phosphorylation of AKT1 and tended to show more robust responses to AKT inhibitor treatment. Still other alterations, including phosphorylation site-adjacent hotspot mutations in the CTNNB1 gene, seemed to interfere with the effectiveness of certain drugs.
Consistent with findings from a study the Clinical Proteomic Tumor Analysis Consortium published in Cell in 2020, the latest multiomic analyses revealed a group of EC cases marked by muted antigen presentation machinery (APM), despite the presence of a pronounced mutational burden that should prompt immune responses to antigens corresponding to these mutations.
"This finding may greatly impact the decision to place patients on immunotherapies, as these tumors typically show lower immune infiltration in the tumor microenvironment," Katsnelson said.
By focusing on proteomic patterns in these tumors with the help of machine learning, the team came up with a proteomic assay centered on two peptides that in particular successfully predicted APM status.
Likewise, the investigators' machine learning analyses suggested that it was possible to classify the molecular subtypes and mutation patterns present in EC tumors based on features found in standard histopathology slides.
"Deep learning accurately predicts EC subtypes and mutations from histopathology images, which may be useful for rapid diagnosis," the authors wrote, noting that their work "identified molecular and imaging markers that can be further investigated to guide patient stratification for more precise treatment of EC."