A systems-level analysis of gene knockdowns in colorectal cancer

Introduction:: Colorectal cancer is one of the most fatal diseases in the United States, with over 150,000 diagnoses and 50,000 deaths per year. Cancer-associated fibroblasts (CAFs), a subtype of stromal cells, have been shown to impact and aid the growth of colorectal cancer (CRC) cells. CRCs exhibit altered central carbon metabolism, a crucial hallmark of cancer growth and metastasis that is influenced by CAFs. However, the exact mechanisms of how CRC cells and CAFs influence each other’s metabolism remains unknown. Therefore, examining the crosstalk that occurs between these two cells is crucial to better understand the cell reprogramming that occurs in the tumor microenvironment. In addition, the KRAS oncogene, whose mutation has been found to highly elevate CRC fatalities, is also examined. In this work, our group utilizes a systems biology approach by incorporating metabolomics and tumor growth data from cell culture experiments in addition to metabolomics profiles into a constraint-based model for central carbon metabolism. This computational technique allows us to both predict the cells’ metabolic fluxes and to simulate perturbations that could better inform cancer drug therapies.

Materials and Methods:: We use unsteady-state, parsimonious flux balance analysis (upFBA), a computational algorithm which calculates the fluxes for each reaction in a metabolic network according to the choice of a biologically relevant objective function (in this case, biomass). We first use upFBA to compute the fluxes for glycolysis, glutaminolysis, pentose phosphate pathway, and the TCA cycle for both wildtype- and mutant-KRAS cells. Additionally, we performed single and pair-wise targeted enzymatic knockdowns of metabolic reactions, which provides knowledge of vulnerable reactions that are the best candidates for inhibiting certain aspects of CRC cell metabolic activity and growth. Since these simulations produce large matrices, it is difficult to both analyze and select perturbations of interest. Therefore, we utilized a representation learning method that incorporates a Siamese neural network that reduces data inputs into low-dimensional space that allow for more efficient analysis.

Results, Conclusions, and Discussions:: Through upFBA, we predict distinct flux profiles for mutant-KRAS CRC cells grown in CAF-conditioned media compared to wildtype- or mutant-KRAS CRC cells cultured in their normal growth media. Specifically, we see a difference in both the total flux through a reaction and the mechanism by which the CAFs alter various pathways. Through single-enzyme targeted knockdowns of each reaction, we also predict complexities in the outputs of flux fold changes compared to the baseline values (no knockdown) across each reaction. Implementation of this large data set into our representation learning method produces distinct clusters in which the mutant-KRAS cell line in CAF media is the most divergent. The clusters also help pinpoint outlier data points (simulations) within each condition that may be of particular importance to zoom into and examine more closely. This research allows for efficient and thorough insights into the complexities of metabolic pathways that support cancer cell growth and tumor-stromal metabolic crosstalk that are difficult to understand through experimental data alone. Our work also brings forth and validates novel strategies to inhibit cancer cell metabolism that can be further explored and tested in relevant pre-clinical models of CRC.

Acknowledgements (Optional): :

References (Optional): :