Bayesian analysis of longitudinal RB-TnSeq resolves the fitness seascape in fluctuating environments

Temporally structured environments are ubiquitous in nature, but time-dependent fitness effects are difficult to measure and thus understudied. To resolve temporal fitness structure at genome scale, we developed a Bayesian multilevel framework for longitudinal randomly barcoded transposon sequencing (RB-TnSeq) that stabilizes noisy mutant abundance trajectories into time-resolved selection-rate estimates with interpretable uncertainty. Applying this approach to a feast-famine starvation regime in <i>Escherichia coli</i>, we uncovered distinct fitness trajectories underpinned by shared molecular strategies across growth-curve phase, consistent with shifting constraints and antagonistic pleiotropy. Fitness during initial growth strongly constrained cumulative success, such that later advantages under stress could not rescue mutants that were initially deleterious. We then compressed these dynamics with a one-dimensional Fisher's geometric "seascape" model that orders mutants along a latent axis aligning with generalist-specialist and growth-survival trade-offs, providing a compact quantitative description of genome-wide constraints in a fluctuating environment. Finally, our longitudinal estimates and inferred seascape are predictive of both the identity and timing of mutational targets in an extended evolution experiment under similar repeated feast-famine conditions, linking short-term competitive fitness effects, with their related trade-offs and constraints, to long-term adaptive outcomes.