Fig. 3: Artificial translation bottlenecks strongly affect antibiotic efficacy.

a Schematic of synthetic regulation introduced to control the expression of a translation factor x, which creates an artificial bottleneck in translation at a well-defined stage; lacI codes for the Lac repressor, which represses the P_LlacO-1-promoter (Methods²⁷,). b Constructs were made for six translation factors mediating 50S stability (der), initiation (infB), recycling (frr), translocation (fusA), tRNA delivery (tufAB) and GDP release (tsf), respectively. Higher expression alleviates the artificial bottleneck. Thicker lines or arrows indicate higher rates. c Translation factor induction curves (upper row) and response surfaces over the inducer-antibiotic grid for different antibiotics (KSG and FUS, middle and bottom row, respectively) in combination with different bottlenecks (50S stability, initiation, and translocation). Full induction of the translation factor rescues wild type growth; increasing bottleneck severity leads to a smooth decrease in growth rate to zero. Induction curves were measured in n = 8 technical replicates, and the median value of non-zero growth rates was calculated. Comparison of the response surfaces with independent expectation (dashed purple line) identify alleviation (orange line) or aggravation (dark blue line). d Columns show bottleneck dependency vectors in color-code; dependency vectors quantify the response of a given antibiotic to the translation bottlenecks (Supplementary Information). e Clustering of the bottleneck dependency vectors upon dimensionality-reduction by principal component analysis (PCA; Supplementary Information). Circles show dependency vectors projected onto the first two principal components (PC1, and PC2); colors indicate cluster identity. The extended cluster areas shown are convex hulls of bootstrapped projections (denoted by dots). Projections of the three additional antibiotics LAM, NIT, and TMP are denoted by a purple triangle, blue square, and green pentagon, respectively. We estimated the p-value by clustering n = 10⁴ reshuffled datasets with added noise and counting the fraction of instances that matched the shown clustering result. See Supplementary Equations (19–20) and Supplementary Fig. 3e; we did not use a standard statistical test.