Multimodal Metabolic Analysis of Tumors and Therapeutic Responses

Dr. Reznik and Dr. Hakimi will integrate two distinct data modalities, metabolomics, and transcriptomics, in order to study metabolic adaptations across cancer. Specifically, they will develop integrative analytical approaches to study the metabolic features defining distinct immune-hot/immune-cold microenvironments, and develop transcriptome-based signatures for the in silico inference of metabolite levels.

Final Progress Report, September 2021

Below we summarize the key findings associated with each Specific Aim. We have achieved all stated goals of the project, and the work funded by this grant is associated with 3 key publications from the lab (Freeman et al, Genome Biology 2022, Benedetti et al, In Revision at Nature Metabolism, and Tang et al, In Revision at Cell Metabolism).

Specific Aims:
  1. Specific Aim 1: Define multimodal metabolic portraits of tumors and their microenvironment. We have completed this analysis, which essentially resulted in a manuscript describing the multimodal metabolic landscape of cancers and the architecture of gene-metabolite coregulation. The remarkable conclusion derived from this analysis is that metabolomic and transcriptomic changes across tumors are highly discordant, i.e. metabolites and genes in the same pathway pervasively show completely distinct patterns of up- or down-regulation in tumors relative to normal. Related to this analysis, we have also identified a small number of lineage-agnostic gene-metabolite interactions (“GMIs”), corresponding to gene-metabolite pairs which show recurrent covariation across numerous tissue and cancer lineages. We have determined the mechanistic basis of a subset of GMIs.
  2. Specific Aim 2: RNA-based signatures of metabolite abundance for in silico metabolomics. We have developed a method, MIRTH, which uses a missing-data-tolerant form of non-negative matrix factorization to impute unmeasured metabolite levels by leveraging a reference dataset where those metabolites are measured alongside another molecular data modality of interest. We have applied MIRTH to the problem of imputing metabolite levels from RNA sequencing data in several distinct datasets, including our pan-cancer multimodal metabolomics database and the CCLE database. MIRTH shows excellent performance, and (for example) in the CCLE can impute ~95% of metabolites from RNA data alone when samples are split into 50% train/50% test.
  3. Design and initial implementation of MMM data portal (Aim 3). This has been completed, and an initial data portal is available at https://rezniklab.shinyapps.io/cAMP-shiny-app/. This data portal is part of a manuscript in review now at Nature Metabolism.