Chemical-genetic profiling reveals limited cross-resistance between antimicrobial peptides with different modes of action

Abstract

Antimicrobial peptides (AMPs) are key effectors of the innate immune system and promising therapeutic agents. Yet, knowledge on how to design AMPs with minimal cross-resistance to human host-defense peptides remains limited. Here, we systematically assess the resistance determinants of Escherichia coli against 15 different AMPs using chemical-genetics and compare to the cross-resistance spectra of laboratory-evolved AMP-resistant strains. Although generalizations about AMP resistance are common in the literature, we find that AMPs with different physicochemical properties and cellular targets vary considerably in their resistance determinants. As a consequence, cross-resistance is prevalent only between AMPs with similar modes of action. Finally, our screen reveals several genes that shape susceptibility to membrane- and intracellular-targeting AMPs in an antagonistic manner. We anticipate that chemical-genetic approaches could inform future efforts to minimize cross-resistance between therapeutic and human host AMPs.

Fig. 1. Chemical-genetic profiling of AMPs.

a Schematic representation of the chemical-genetic pipeline. The chemical-genetic interactions of ~4400 single gene-overexpressions and 15 different AMPs were measured using a pooled fitness assay with a deep sequencing readout (see Methods). b A density scatter plot showing the overall correlation of replicate measurements of the chemical-genetic scores (log₂ fold-change in the relative abundance of each gene in the presence vs absence of each AMP) across all genes and AMPs (r = 0.63 and P = 2.2 × 10^–16, Pearson’s correlation, n = 53,292). c Heatmap showing the chemical-genetic interaction scores. Resistance-enhancing and sensitivity-enhancing chemical-genetic scores are represented by blue and red, respectively. Groups C1–C4 refer to clusters defined in Fig. 2. Source data are provided as Supplementary Data 2.

Fig. 2. Chemical-genetic profiling discriminates membrane-targeting and intracellular-targeting AMPs.

a Heatmap showing the ensemble clustering of the AMPs based on their chemical-genetic profiles (see Methods). For each AMP pair, the color code represents the frequency of being closest neighbors across the ensemble of clusters (n = 75,000 clustering). The four major clusters are labeled as C1, C2, C3, and C4. Membrane-targeting and intracellular-targeting broad modes of action are labeled with pink and orange, respectively, on the rightmost side of the figure. Gray color indicates that the specific broad mode of action has not been described or not tested (see Table 1). References describing these activities are provided in Supplementary Data 7. b Most important physicochemical properties that differentiated AMPs in cluster C1, C2 from AMPs in cluster C3, C4. Significant differences: **P = 0.0026 and 0.0012 for isoelectric point and relative number of prolines, respectively, * P = 0.0391 and P = 0.0154 for hydropathicity and total aggregation hotspot area, respectively, two-sided Mann–Whitney U test, n = 9 and n = 6 for C1, C2 and C3, C4, respectively. c, Physicochemical properties that distinguished the clusters when the 4 main AMP clusters were considered separately (p < 0.05 ANOVA, Tukey post-hoc test, n = 15). Significant differences: ***P = 1.1 × 10^–6, P = 1.3 × 10⁻⁶ and P = 4 × 10⁻⁶ for C1 vs C4, C2 vs C4 and C3 vs C4, respectively in the case of number of disordered amino acids. *P = 0.034 and P = 0.027 for C1 vs C3 and C2 vs C3, respectively. **P = 0.0022 for C3 vs C4. ***P = 5.5 × 10⁻⁵ and P = 4.2 × 10⁻⁵ for C1 vs C4 and C2 vs C4, respectively, in the case of relative number of prolines. Central horizontal lines represent median values. Source data are provided as Supplementary Data 3.

Fig. 3. Functionally diverse latent and intrinsic AMP resistomes.

a Heatmap shows the corrected Jaccard similarity indices calculated for resistance-enhancing genes (blue) and sensitivity-enhancing genes (red) between AMP pairs based on the overexpression screen (see Methods for calculation of corrected Jaccard indices, n = 210, that is, the number of AMP pairs). The darker the color the higher the overlap of gene sets between AMP pairs. Source data are provided as a Source Data file. b The overlaps in the latent resistomes (genes enhancing resistance upon overexpression) between AMP pairs belonging to different chemical-genetic clusters. Significant difference: **P = 0.0045 from two-tailed unpaired t-test, n = 54 and n = 51 for between C1, C2, and C3, C4, and others, respectively. Boxplots show the median (center horizontal line), the first and third quartiles (bottom and top of box, respectively), with whiskers showing either the maximum (minimum) value or 1.5 times the interquartile range of the data. c Heatmap shows the corrected Jaccard similarity indices calculated for resistant (blue) and sensitive (red) chemical-genetic interactions with partially depleted essential genes (see Methods). Source data are provided as Supplementary Data 5. d Schematic figure showing sets of essential genes that simultaneously enhance AMP resistance when overexpressed and sensitivity when depleted. Color code is explained in the figure.

Fig. 4. Collateral sensitivity (CS) interactions are frequent between AMPs with different modes of action.

a Heatmap depicting the overrepresentation of collateral sensitivity-enhancing genes for each AMP pair over random expectation (n = 210 AMP pairs). Random expectation is calculated using the number of resistance-enhancing genes and sensitivity-enhancing genes for each AMP (see Methods). b Collateral sensitivity effects were especially pronounced between AMP pairs with different broad mode of action, that is, between membrane-targeting (C1, C2) and intracellular-targeting (C3, C4), as compared to AMP pairs from the same cluster. Significant difference: ***P = 1.7 × 10⁻⁰⁸ from two-tailed unpaired t-test, n = 108 and 46 for pairs of AMPs between C1, C2, and C3, C4, and those within cluster, respectively. Y-axis shows odds ratio (log2) of enrichment of collateral sensitivity interactions between AMP pairs. Boxplots show the median (center horizontal line), the first and third quartiles (bottom and top of box, respectively), with whiskers showing either the maximum (minimum) value or 1.5 times the interquartile range of the data. Source data are provided as a Source Data file.

Fig. 5. Mutation in mlaD influences AMP susceptibilities through antagonistic mutational effects.

a Relative change in MICs of the mlaD overexpression and deletion strains (ΔmlaD) to a representative set of membrane-targeting and intracellular-targeting AMPs. MICs were compared to corresponding wild-type control strains (see Supplementary Figures 12, 13). Dashed lines represent previously defined cut-offs for resistance ( ≥ 1.2 x MIC of the control) and sensitivity ( ≤ 0.8 x MIC of the control). b Decreased net negative surface charge of the mlaD overexpression and deletion strains. Significant differences: **P = 0.0021 and P = 0.0021 for WT + empty vector vs overexpression and WT vs deletion strain, respectively, from two-sided Mann–Whitney U test, n = 6 biological replicates for each genotype. Charge measurement was done using FITC-labeled poly-L-lysine (FITC-PLL) assay where the fluorescence signal is proportional to the binding of the FITC-PLL molecules. A lower binding of FITC-PLL indicates a less net negative surface charge of the outer bacterial membrane (see Methods). c Increased membrane potentials of the mlaD overexpression and deletion strains. Significant differences: **P = 0.007, P = 0.0079 and P = 0.0079 for WT + empty vector CCCP control vs WT + empty vector, WT + empty vector vs WT + mlaD overexpression and WT vs. ΔmlaD, respectively, two-sided Mann–Whitney U test, n = 5 biological replicates for each genotype. Relative membrane potential was measured by determining relative fluorescence (RFU) using a carbocyanine dye DiOC2(3) assay (see Methods). Red/green ratios were calculated using population mean fluorescence intensities. WT E. coli carrying the empty vector treated with cyanide-m-chlorophenylhydrazone (CCCP, a chemical inhibitor of proton motive force) was used as an experimental control for diminished membrane potential. Central horizontal lines represent mean values of biological replicates. Source data are provided as a Source Data file and in Supplementary Fig. 16.

Fig. 6. Mode of action informs on cross-resistance spectra of AMP-evolved lines.

a Cross-resistance interactions (i.e., defined as 2-fold increase in MIC) are significantly overrepresented between AMP pairs either from the group of exclusively membrane-targeting AMPs (TPII, CP1, PGLA, LL37, PEX) or from the group of exclusively intracellular-targeting AMPs (PR39 and BAC5) as compared to pairs of AMPs between the two mode of action groups (MOA). Significant difference: ***P = 1.333 × 10⁻¹⁰ from two-sided Fisher’s exact test, n = 119 and 93 for within and between mode of action groups, respectively. b Collateral-sensitivity interactions (i.e., defined as ≥ 20% decrease in MIC) are overrepresented between the groups of membrane-targeting and intracellular-targeting AMPs (i.e., between MOA). Significant difference: ***P = 4.83 × 10⁻⁵ from two-sided Fisher’s exact test, n = 200 and 194 for within MOA and between MOA, respectively. Source data are provided as a Source Data file.