Deciphering the Signaling Network Landscape of Breast Cancer Improves Drug Sensitivity Prediction

Abstract

Although genetic and epigenetic abnormalities in breast cancer have been extensively studied, it remains difficult to identify those patients who will respond to particular therapies. This is due in part to our lack of understanding of how the variability of cellular signaling affects drug sensitivity. Here, we used mass cytometry to characterize the single-cell signaling landscapes of 62 breast cancer cell lines and five lines from healthy tissue. We quantified 34 markers in each cell line upon stimulation by the growth factor EGF in the presence or absence of five kinase inhibitors. These data – on more than 80 million single cells from 4,000 conditions – were used to fit mechanistic signaling network models that provide unprecedented insights into the biological principles of how cancer cells process information. Our dynamic single-cell-based models more accurately predicted drug sensitivity than static bulk measurements for drugs targeting the PI3K-MTOR signaling pathway. Finally, we identified genomic features associated with drug sensitivity by using signaling phenotypes as proxies, including a missense mutation in DDIT3 predictive of PI3K-inhibition sensitivity. This provides proof of principle that single-cell measurements and modeling could inform matching of patients with appropriate treatments in the future.One-linerSingle-cell proteomics coupled to perturbations improves accuracy of breast tumor drug sensitivity predictions and reveals mechanisms of sensitivity and resistance.HIGHLIGHTSMass cytometry study of signaling responses of 62 breast cancer cell lines and five lines from healthy tissue to EGF stimulation with or without perturbation with five kinase inhibitors.Single-cell signaling features and mechanistic signaling network models predicted drug sensitivity.Mechanistic signaling network models deepen the understanding of drug resistance and sensitivity mechanisms.We identify drug sensitivity-predictive genomic features via proxy signaling phenotypes.

Advances in systems biology modeling: 10 years of crowdsourcing DREAM challenges

“…Single-cell technologies also allow us to revisit challenges with an enhanced resolution. For example, an unpublished dataset that used time course mass cytometry measurements covering 36 phospho-protein markers in over 4,000 experimental conditions totaling 80 million single cells for 67 breast cancer lines (Tognetti et al, 2021) was used to organize a similar challenge as the breast cancer signaling discussed above (Gabor et al, 2021). This time, given the phosphorylation time series data available, we organized a challenge to assess the performance and limits of predictive models of time course response of single cells to stimuli in the presence and absence of kinase inhibitors.…”

Cell-to-cell and type-to-type heterogeneity of signaling networks: Insights from the crowd

“…If the absolute values are important, this can be an important limitation, but not for downstream analysis where the data is rescaled (e.g. (Tognetti et al 2020) ). In addition, the functional effect of at least some main pathways is driven by fold-changes, and thus predicting these, even if not being able to do so accurately for the absolute values, is helpful (Adler and Alon 2018).…”

“…We based the SCSBrC challenge on a single-cell signaling dataset measured with mass cytometry (Tognetti et al 2020). In this dataset, 67 cell lines were stimulated with EGF in combination with one of five kinase inhibitors (PKC, PI3K, mTOR, MEK, EGFR), and in each condition 31 phosphoproteins and 5 cellular markers were measured at 10 different time points over the course of one hour (Figure 1, Supp.…”

Toward Systems Biomarkers of Response to Immune Checkpoint Blockers

“…Dynamic mathematical models can be used to study intra-cellular networks of the different cell types populating the TME ( 96 ). To characterize these networks at the patient-specific level, models of signaling pathways in cancer cells have been trained from perturbation experiments ( 97 , 98 ), gene expression data ( 99 ), or integrating multi-omics data ( 100 ). The resulting parameters corresponding to these personalized models can be relevant biomarkers of clinical outcome ( 99 – 101 ).…”