Multilevel Omics-Readouts ofin vitroPerturbation Studies are Determined by Memory Effects from Subculture

AbstractMass spectrometry-based omics technologies are of increasing relevance for drug discovery. The identification of significant molecular events is a prerequisite in order to map treatments effects to biological pathways for mode of action (MoA) deconvolution. The main contributors to statistical significance are the effect size and the variation of a given molecule. While the former is largely determined by the biological system and the drug treatment, the latter is controlled by the experimental workflow. The impact of sample preparation and instrumentation on variation are well described, but the impact of cell culture conditions on the variation of omics readouts is less explored. Here, we provide evidence that memory effects originating from subculture conditions during cell expansion before the perturbation influence the variation of multilevel omics readouts, including eicosadomics, proteomics and phosphoproteomics, as well as phenotypic parameters,e.g. cell growth and motility using a Cellwatcher system. The treatment was carried out in SW480 colon carcinoma cells with the clinically approved anticancer agent arsenic trioxide (ATO), which follows a multi-modal MoA. Experimental variation was assessed using two different subculture conditions. First, the cells were homogeneously expanded in the Log phase before the treatment. Second, the cells were heterogeneously expanded using cells in the Lag, Log and plateau phases of cell growth before the treatment. When comparing the homogeneous and heterogeneous subcultures, the coefficients of variation (CVs) of controls increased from 1% to 3% for eicosadomics, 11% to 23% for proteomics, 24% to 53% for phosphoproteomics, 5% to 13% for growth area occupied and 6% to 14% for cell motility. The ATO treatment did not considerably influence the number of identified molecules by mass spectrometry, nor the fold-change distributions after both subculture conditions. Therefore, the increased number of significant molecular events in drug treatments after homogeneous subculture seems to be mainly caused by the decrease in CVs. This in turn increased the information content in the obtained perturbation network, based on proteomics and phosphoproteomics data, from 58 to 321 causal conjectures after heterogeneous and homogeneous subcultures, respectively. Independently of the subculture conditions, the perturbation networks allowed for delineating the established MoA of ATO, including enhanced kinase signaling and reduced MYC-related gene expression, induction of heat shock proteins and cell cycle disruption. However, only the perturbation network after homogeneous subculture enabled the discovery of unprecedented drug effects, including down-regulation and quadruple dephosphorylation of the transcription factor RB1, among others. Also, the involvement of AP-1, which is central to the MoA of ATO was only detected after homogenous subculture. In summary, memory effects from subculture have broad relevance for MoA testing in drug discovery since they clearly manifest in molecular and phenotypic properties and their control enables a considerably enhanced MoA deconvolution, facilitating discovery efforts.