CycleFlow quantifies cell-cycle heterogeneityin vivo

Abstract

While the average cell-cycle length in a cell population can be derived from pulse-chase experiments, proliferative heterogeneity has been difficult to quantify. Here we describe CycleFlow, a broadly applicable method that applies Bayesian inference to combined measurements of EdU incorporation and DNA content. CycleFlow accurately quantifies the fraction of proliferating versus quiescent cells and the durations of cell-cycle phases of the proliferating cells in vitro and in vivo.

Modelling the Dynamics of T-Cell Development in the Thymus

“…This model was useful in comparing the cycling behavior of cells in two environments for which the EdU/BrdU labeling were already indicative, but additionally providing an estimate of the difference in cell-cycle duration. Notably, an earlier model, assuming that labeled cells cannot return to S phase during the 4 h of staining, inferred very short cell-cycle durations in the range of 3 to 4 h from the same data (Figure 5B, left equation) [85]. This example highlights the impact of model design on the inferred cycle duration values, and underscores that single linear ODEs generate an exponential residence time of cells at each stage, requiring some care in model design or interpretation.…”

Modeling the Dynamics of T-Cell Development in the Thymus

“…Jolly et al [ 86 ] have proposed an ODE-based model that solves this problem ( Figure 5 C) by separating each cycle phase into many sequential steps, and applied it on a EdU labeling kinetics scheme in both cell cultures and in vivo . This model would also be valid for dual pulse.…”

“…Although ODE models could raise a good fit to experimental data with very scarce time-points, it is not sure whether they would manage to reproduce time-resolved datasets with many time-points. This can be solved either by duplicating each phase into sub-phases as in Jolly et al [ 86 ] and performing parameter optimization, or in the general case by considering age-structured models. Age-structured models offer analytical formula under balanced growth, or can be simulated as PDEs or agent-based models, with higher flexibility to account for experimental biases (efficiency of labeling or duration of the label pulse; possibility of inflow of labeled cells from progenitors if the experiment timeframe is long).…”

“…A recommendation for future experimental-modeling hybrid approaches would be to perform test experiments with many time-points to assess whether the model structure is suitable (possibly in vitro , as in Jolly et al [ 86 ]), and which time-points are most informative (as suggested in Weber et al [ 89 ]) before analyzing in vivo data-points with reduced time resolution. The discrepancy in gating strategies, in particular the boundaries between DN4, pre-selection DPs and post-selection DP, would be beneficial to be provided, because it forbids to compare the results between studies.…”

“…Most studies neglected the iSP8 population, whose high proliferation might artificially contaminate the inferred proliferation rate of SP8 cells, and would be beneficial to separate as an additional population. Finally, it appeared that the dynamics of labeling is not always enough to infer the full cycle duration (as in [ 86 ]), possibly because with few time-points they can only inform on the ratio between each phase, rather than the absolute cycle duration. Such studies additionally used the amounts of cells at steady state in each cell phase to infer the cycle duration.…”