Quantifying Dynamic Phenotypic Heterogeneity in Resistant Escherichia coli under Translation‐Inhibiting Antibiotics

Abstract Understanding the phenotypic heterogeneity of antibiotic‐resistant bacteria following treatment and the transitions between different phenotypes is crucial for developing effective infection control strategies. The study expands upon previous work by explicating chloramphenicol‐induced phenotypic heterogeneities in growth rate, gene expression, and morphology of resistant Escherichia coli using time‐lapse microscopy. Correlating the bacterial growth rate and cspC expression, four interchangeable phenotypic subpopulations across varying antibiotic concentrations are identified, surpassing the previously described growth rate bistability. Notably, bacterial cells exhibiting either fast or slow growth rates can concurrently harbor subpopulations characterized by high and low gene expression levels, respectively. To elucidate the mechanisms behind this enhanced heterogeneity, a concise gene expression network model is proposed and the biological significance of the four phenotypes is further explored. Additionally, by employing Hidden Markov Model fitting and integrating the non‐equilibrium landscape and flux theory, the real‐time data encompassing diverse bacterial traits are analyzed. This approach reveals dynamic changes and switching kinetics in different cell fates, facilitating the quantification of observable behaviors and the non‐equilibrium dynamics and thermodynamics at play. The results highlight the multi‐dimensional heterogeneous behaviors of antibiotic‐resistant bacteria under antibiotic stress, providing new insights into the compromised antibiotic efficacy, microbial response, and associated evolution processes.


Quantifying Dynamic Phenotypic Heterogeneity in Resistant E. coli under Translationinhibiting Antibiotics
Haishuang Zhu † , Yixiao Xiong † , Zhenlong Jiang † , Qiong Liu, Jin Wang* Figure S1.A comparison of the fluorescence intensity distribution between the 0.5 mM Cm and the 0.9 mM Cm experimental groups.The data of 2000 bacteria cells induced by (A) 0.5 mM Cm and (B) 0.9 mM, at 10 h, 20 h and 30 h were shown, respectively.Table S2.The switching time (hours) between either two states under 0.6 mM, 0.7 mM, 0.8 mM and 0.9 mM, respectively.Table S3.The mean values, corresponding residence time, transition probabilities, and instantaneous transition rates of the two states after average fluorescence intensity and cell area correlation, obtained from CHMM analysis.The transition probabilities and transition rates were calculated over 12-minute time intervals.

State mean value (Cell area, average intensity (RFU))
Residence time (%)      Movie S1.Some of the E. coli cells in a micro-colony keep dividing in 0.9 mM Cm (bright field).

Movie S2.
Some of the E. coli cells in a micro-colony keep dividing in 0.9 mM Cm (fluorescence field).

Movie S3.
The E. coli cells keep dormant in 0.9 mM Cm (bright field).

Movie S4.
The E. coli cells keep dormant in 0.9 mM Cm (fluorescence field) Movie S5.
Figure S2.A typical cell recovered to grow and started dividing after treatment of 0.9 mM Cm for 12 h.

Figure S3 .
Figure S3.Gaussian Kernel Density Estimation (KDE) contour plots and mesh plots with marginal distributions of the cycle time (*12 min)-average YFP fluorescence intensity (RFU) data under 0.6 mM, 0.7 mM, 0.8 mM, 0.9 mM Cm treatment (from top to bottom).

Figure S4 .
Figure S4.Results from the Elbow method based on K-means clustering of the cycle time (*12 min) -average intensity (RFU) data under 0.6 mM, 0.7 mM, 0.8 mM, 0.9 mM Cm treatment.

Figure S5 .
Figure S5.K-means clustered distribution and CHMM-fitted distribution of the cycle time (* 12 min)-average intensity (RFU) data points under diferent concentrations of Cm treatment.Each state is characterized by a color.The K-means clustering was used solely for confirming the appropriate number of states (via the elbow method) and did not influence the subsequent analysis.

Figure S6 .
Figure S6.(A) Number of switching paths and (B) total switching path distance vs different Cm concentrations.

Figure S7 .
Figure S7.(A) Most probable switching routes (red line) among the four states on the cycle time (×12 min) -average intensity (RFU) landscapes under 0.6 mM Cm inducement.

Figure S7 .
Figure S7.(B) Most probable switching routes (red line) among the four states on the cycle time (×12 min) -average intensity (RFU) landscapes under 0.7 mM Cm inducement.

Figure S7 .
Figure S7.(C) Most probable switching routes (red line) among the four states on the cycle time (×12 min) -average intensity (RFU) landscapes under 0.8 mM Cm inducement.

Figure S7 .
Figure S7.(D) Most probable switching routes (red line) among the four states on the cycle time (×12 min) -average intensity (RFU) landscapes under 0.8 mM Cm inducement.

Figure S8 .
Figure S8.The representative overlay time-lapse images indicated the bacterial cells with dim fluorescence and fast growth rate could be lysed at 0.6 mM Cm and 0.9 mM Cm, respectively.

Table S1
. A comparison of the transition probabilities between the states characterized by either one or two cellular traits at different Cm concentrations.The transition probabilities were calculated over 12-minute time intervals.

Table S4 .
The mean values, corresponding residence time, transition probabilities, and instantaneous transition rates of the two states after cell area and cycle time correlation, obtained from CHMM analysis.The transition probabilities and transition rates were calculated over 12-minute time intervals.
Time-lapse recording of the dynamic behavior of a micro-colony of E. coli cells in 0.8 mM Cm (fluorescence field) Time-lapse recording of the dynamic behavior of another micro-colony of E. coli cells in 0.8 mM Cm (fluorescence field)