Bacterial persistence is a phenomenon in which a small fraction of isogenic bacterial cells survives a lethal dose of antibiotics. Although the refractoriness of persistent cell populations has classically been attributed to the presence of growth-inactive cells generated prior to drug exposure, evidence is accumulating that actively growing cell fractions can also produce persister cells, depending on bacterial species, antibiotics, and culture conditions. However, the direct elucidation of survival modes, drug response diversity, and growth history dependence by single-cell observation is limited due to the extremely low frequencies of persisters. Here, we visualize the responses of more than 106 individual cells of wildtype Escherichia coli to lethal doses of antibiotics, sampling cells from different growth phases and culture media into a microfluidic device. We show that when cells sampled from exponentially growing cell populations in batch cultures were treated with ampicillin or ciprofloxacin, all persisters for which we identified their single-cell histories were growing prior to antibiotic treatment. We detected pre-existing growth-arrested non-growing persisters to ampicillin for an E. coli strain expressing a defective general stress response regulator, but these constituted minor fractions of persister cells. Growing persisters exhibit heterogeneous survival dynamics in response to drug exposure, including continuous growth and fission with L-form-like morphologies, responsive growth arrest, or post-exposure filamentation. Incubating cells under stationary phase conditions increases both the frequency and the probability of survival of non-growing persisters to ampicillin. However, no non-growing persisters were identified for ciprofloxacin treatment, even in post-stationary phase cell populations. These results demonstrate diverse persistence dynamics at the single-cell level depending on antibiotic types and the pre-exposure cultivation history.

Recent advances in single-cell time-lapse microscopy have revealed non-genetic heterogeneity and temporal fluctuations of cellular phenotypes. While different phenotypic traits such as abundance of growth-related proteins in single cells may have differential effects on the reproductive success of cells, rigorous experimental quantification of this process has remained elusive due to the complexity of single cell physiology within the context of a proliferating population. We introduce and apply a practical empirical method to quantify the fitness landscapes of arbitrary phenotypic traits, using genealogical data in the form of population lineage trees which can include phenotypic data of various kinds. Our inference methodology for fitness landscapes determines how reproductivity is correlated to cellular phenotypes, and provides a natural generalization of bulk growth rate measures for single-cell histories. Using this technique, we quantify the strength of selection acting on different cellular phenotypic traits within populations, which allows us to determine whether a change in population growth is caused by individual cells' response, selection within a population, or by a mixture of these two processes. By applying these methods to single-cell time-lapse data of growing bacterial populations that express a resistance-conferring protein under antibiotic stress, we show how the distributions, fitness landscapes, and selection strength of single-cell phenotypes are affected by the drug. Our work provides a unified and practical framework for quantitative measurements of fitness landscapes and selection strength for any statistical quantities definable on lineages, and thus elucidates the adaptive significance of phenotypic states in time series data. The method is applicable in diverse fields, from single cell biology to stem cell differentiation and viral evolution.

Bacteria prudently regulate their metabolic phenotypes by sensing the availability of specific nutrients, expressing the required genes for their metabolism, and repressing them after specific metabolites are depleted. It is unclear, however, how genetic networks maintain and transmit phenotypic states between generations under rapidly fluctuating environments. By subjecting bacteria to fluctuating carbon sources (glucose and lactose) using microfluidics, we discover two types of non-genetic memory in Escherichia coli and analyze their benefits. First, phenotypic memory conferred by transmission of stable intracellular lac proteins dramatically reduces lag phases under cyclical fluctuations with intermediate timescales (1–10 generations). Second, response memory, a hysteretic behavior in which gene expression persists after removal of its external inducer, enhances adaptation when environments fluctuate over short timescales (<1 generation). Using a mathematical model we analyze the benefits of memory across environmental fluctuation timescales. We show that memory mechanisms provide an important class of survival strategies in biology that improve long-term fitness under fluctuating environments. These results can be used to understand how organisms adapt to fluctuating levels of nutrients, antibiotics, and other environmental stresses.

We present a formulation of branching and aging processes that allows distributions along lineages to be studied within populations, and provides a new interpretation of classical results in the theory of aging. We establish a variational principle for the stable age distribution along lineages. Using this optimal lineage principle, we show that the response of a population’s growth rate to age-specific changes in mortality and fecundity – a key quantity which was first calculated by Hamilton – is given directly by the age distribution along lineages. We apply our method also to the Bellman-Harris process, in which both mother and progeny are rejuvenated at each reproduction event, and show that this process can be mapped to the classic aging process such that age statistics in the population and along lineages are identical. Our approach provides both a theoretical framework for understanding the statistics of aging in a population, and a new method of analytical calculations for populations with age structure. We discuss generalizations for populations with multiple phenotypes, and more complex aging processes. We also provide a first experimental test of our theory applied to bacterial populations growing in a microfluidics device.