Therapeutically relevant antibiotics display a wide range of uptake in M. abscessus

To assess the relative efficiency of antibiotic accumulation in M. abscessus, we developed a liquid chromatography-mass spectrometry (LC-MS) method to simultaneously measure an arrayed panel of 20 antibiotics used to treat mycobacterial infections [19,20] in the type strain of M. abscessus subspecies abscessus (ATCC19977) (Fig 1A). 19 of these antibiotics were detectable over linear ranges that enabled relative quantitation of their uptake in M. abscessus (S1 Fig). We measured relative accumulation of these drugs in M. abscessus, which represents the ratio of drug that was either taken up or tightly bound to the bacterium after four hours of incubation compared to the amount of drug that was initially in the media. This accumulation measurement also includes drug metabolism; any compound that is taken up but then degraded or modified by the bacterium will not appear to accumulate in the cell. Strikingly, these antibiotics displayed a wide range of accumulation, with greater than 1000-fold variation between the highest and lowest accumulating compounds (Fig 1B and S1 Table). Interestingly, for antibiotics with known activity against intracellular targets in M. abscessus, there was a statistically significant negative correlation (Pearson r = −0.791 with p-value = 0.012 and Spearman r = −0.803 with p-value = 0.11) between intracellular antibiotic accumulation and the average minimum inhibitory concentrations for those drugs (Fig 1C). In contrast, for drugs with no known activity against intracellular targets in M. abscessus, there was no significant correlation between antibiotic accumulation and drug potency (Fig 1D). The anti-correlation between accumulation and minimum inhibitory concentration was more pronounced (Pearson r = −0.99) when comparing antibiotics with the same mechanism of action (Fig 1E), suggesting that intracellular accumulation could explain the differential efficacy of similarly acting drugs. Together, these data suggest that poor intracellular accumulation may be a major determinant of antibiotic efficacy for those compounds.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. TnSeq reveals differential requirement of numerous membrane transporters with linezolid exposure.

a, Schematic of antibiotic accumulation assay. b, LC-MS measurement of relative accumulation of indicated mycobacterial antibiotics in M. abscessus ATC19977 reference strain. Values represent intracellular level of antibiotic after 4 hr incubation normalized to initial antibiotic levels in media prior to incubation. All values are normalized to internal standard and are represented as individual values along with mean ± s.d. n = 3 biological replicates. c, Historical half-maximal minimum inhibitory concentration (MIC₅₀) values of antibiotics with known activity against intracellular enzymes in M. abscessus plotted against the relative antibiotic accumulation values as determined in (b). Each individual data point represents the mean relative accumulation +/- standard deviation for a unique antibiotic, with the x-value for each point determined from historical values. Pearson correlation coefficient (r) = -0.791, with p-value from a two-tailed test = 0.012. Spearman correlation coefficient = -0.803, with p-value from a two-tailed test = 0.011. d, Historical half-maximal minimum inhibitory concentration (MIC50) values of antibiotics with no known activity against intracellular enzymes in M. abscessus plotted against the relative antibiotic accumulation values as determined in (b). Each individual data point represents the mean relative accumulation +/- standard deviation for a unique antibiotic, with the x-value for each point determined from historical values. Pearson correlation coefficient (r) = -0.091, with p-value from a two-tailed test = 0.81. Spearman correlation coefficient = -0.195, with p-value from a two-tailed test = 0.59. e, Historical half-maximal minimum inhibitory concentration (MIC50) values of macrolides plotted against the relative antibiotic accumulation values as determined in (b). Pearson correlation coefficient (r) = -0.999, with a p-value from a two-tailed test = 0.011. Spearman correlation coefficient = -1.00, with p-value from a two-tailed test = 0.33. f, Schematic of TnSeq in M. abscessus clinical isolate upon linezolid exposure. g, log2-fold ratio of transposon insertion counts plotted against significance with linezolid treatment versus no drug. P values derived from two-sided permutation test and are displayed without multiple hypothesis correction. a, f Created with BioRender.com, released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license.

https://doi.org/10.1371/journal.ppat.1013027.g001

We found that linezolid accumulated to a particularly low level (Fig 1B), suggesting that uptake is a potential hurdle to its efficacy. Linezolid is a bacterial protein synthesis inhibitor that binds to the bacterial 23S ribosomal RNA of the 50S subunit [21]. Linezolid is currently in use as part of an effective regimen for treating multi-drug resistant Mycobacterium tuberculosis [22,23], other opportunistic nontuberculous mycobacteria (NTM) pathogens [20,24], and many gram-positive pathogens [21], but it is only moderately effective against M. abscessus in liquid culture [25]. However, linezolid has the potential to be an effective co-treatment for M. abscessus infections [26,27], as linezolid synergizes with frontline M. abscessus drugs, amikacin and clarithromycin [28,29], and several studies have demonstrated improved clinical outcomes when linezolid was included in therapeutic regimens [2,30]. Notably, despite multiple inquiries into clinical populations with linezolid resistance, M. abscessus fails to show any resistance-associated point mutations in ribosomal linezolid binding sites, as have been observed in other bacteria [31]. However, some clinical isolates have been shown to have increased transcriptional expression of efflux pumps [26], and in vitro resistance selection studies have revealed mutations in essential genes involved in drug efflux [32], suggesting that linezolid efficacy could be augmented by improving its intracellular accumulation.

Linezolid treatment imposes a requirement for numerous membrane transporters

To identify genes that are necessary for M. abscessus to survive linezolid and, therefore, likely involved in intrinsic resistance, we performed a genetic screen using transposon mutagenesis and sequencing (TnSeq) [33,34] (Fig 1F and S2 Table) in a clinical isolate, M. abscessus subspecies massiliense BWH-F [35] that displays moderate linezolid resistance (S2A Fig). 25 genes were significantly more required (P value < 0.0005) to survive linezolid treatment compared to the untreated condition (Fig 1G and S3 Table), and these genes represent a variety of shared functional categories (Table 1) [36]. This includes several cell wall synthesis genes, such as MAB_1915 (fadD), a mycolic acid synthesis gene, which has been previously characterized to play a role in intrinsic antibiotic resistance to multiple antibiotics, including linezolid [37]. However, we were particularly intrigued by the large subset of the genes differentially required upon linezolid treatment that encode membrane transporters with diverse annotated functions across 3 of the 5 superfamilies of bacterial transporters (Figs 1G, S2B, and S2F and Table 1). Given the diversity and likely independent functions of these annotated membrane transporter genes, we reasoned that they play a role in the efflux or membrane permeability of linezolid and thus impact intrinsic resistance in M. abscessus. These genes share significant sequence homology with numerous other membrane proteins across mycobacteria (S4 Table) [36]. However, this degree of sequence identity is common for structurally related membrane transporters and does not necessarily indicate functional homology. Given the uncertainty about the functions of these proteins, we hypothesized that they may function to form a permeability barrier or efflux mechanism that limits linezolid accumulation in M. abscessus.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. List of genes most significantly required upon linezolid treatment.

https://doi.org/10.1371/journal.ppat.1013027.t001

Membrane transporter knockdown increases susceptibility to linezolid

With the exception of MAB_4116c, which is involved in the export of cell envelope glycopeptidolipids [38], the functions of these putative membrane transporters and their roles in drug resistance have yet to be characterized in M. abscessus. The M. tuberculosis homolog of MAB_2807c has been implicated in drug resistance [39], and the M. smegmatis and M. tuberculosis homologs of MAB_1963 can impact both population heterogeneity and drug resistance [40]. Thus, we first chose to examine the requirement for each of these transmembrane protein encoding genes upon linezolid treatment by utilizing an anhydrotetracycline-(ATc) inducible CRISPR interference (CRISPRi) system in the M. abscessus reference strain to generate transcriptional knockdowns [34,41] (S5 Table). Knockdown of each of these genes results in increased susceptibility to linezolid in both solid and liquid media (Figs 2A–I, S3A-I and S3L), suggesting that each gene plays some role in determining linezolid efficacy. For the two mmpL genes identified in our genetic screen, MAB_2303 and MAB_4116c, knockdown of their cognate mmpS membrane genes, MAB_2302 (S3C, S3J and S3L Fig) and MAB_4117c (S3H, S3K and S3L Fig), also led to increased linezolid sensitivity. These genes did not appear as significant hits in the original screen, most likely due to their small size, which results in too few potential transposon insertion sites. We also examined the effect of knockdown of MAB_4115c (MmpL4b), an MmpL known to function in conjunction with MAB_4116c (MmpL4a) [42]. We found that MAB_4115c knockdown also results in increased sensitivity to linezolid (S3I and S3L Fig), consistent with our results and previous literature [38], which suggests that MAB_4116c, with its cognate MmpS and MmpL proteins, likely contributes to linezolid resistance. Together these data suggest that each of these membrane transporters plays a role in establishing intrinsic linezolid resistance.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Knockdown of membrane transporters increases sensitivity to linezolid.

a, Proliferation rates of M. abscessus ATCC19977 CRISPRi strains pre-depleted by treatment with 500 ng mL⁻¹ ATc for 24 hours and then treated with 0.5 μg mL⁻¹ linezolid or vehicle for 48 hours. Proliferation rates calculated from optical density of cultures over time. b, Ratio of proliferation rates of pre-depleted M. abscessus ATCC19977 CRISPRi strains treated with 0.5 μg mL⁻¹ linezolid or vehicle along with ATc for 48 hours. Data are represented as individual values along with mean ± s.d. n = 3 biological replicates. c-i, Half-maximal minimum inhibitory concentration (MIC₅₀) dose responses of pre-depleted M. abscessus ATCC19977 strains as measured by reduction of a colorimetric dye after treatment with indicated concentrations of linezolid along with ±500 ng mL⁻¹ ATc for 24 hours. Values normalized to vehicle only control per strain. All data are represented as individual values along with mean ± s.d. n = 3 biological replicates. NT = non-targeting sgRNA. KD = knockdown. ATc = anhydrotetracycline. LZD = linezolid.

https://doi.org/10.1371/journal.ppat.1013027.g002

Candidate genes impact both general cell permeability and efflux

We next interrogated the mechanistic role of each of these proteins in mediating linezolid susceptibility. Several of the knockdown strains display baseline growth defects (Figs 2A, 2B, S3A and S3H), suggesting that they play an important role in the normal biology of the bacterium. However, the knockdown of these genes does not induce any gross alterations to general cell morphology (S4A-G Fig), indicating that these genes’ effects on linezolid susceptibility are not mediated by substantial cellular deformities. Instead, we posited that these membrane proteins may have specific effects on drug uptake. We first tested if these knockdown strains have altered general permeability to chemicals by measuring the accumulation of calcein AM, a hydrophobic, non-fluorescent molecule that can passively diffuse through cells; once inside cells, calcein AM is cleaved by host esterases into calcein, a fluorescent product, that remains intracellular [43]. Knockdown of MAB_2303, MAB_2624c, or MAB_2807 does not induce increased intracellular calcein accumulation, suggesting that knockdown of these genes does not alter the general permeability of the cell (Figs 3A and S5A-H). However, MAB_1963 and MAB_4116c knockdowns display increased intracellular calcein accumulation, suggesting that these genes play a role in limiting the general permeability of the cell (Figs 3A, S5B and S5G). Furthermore, these results are consistent with previous literature that links MAB_4116c with the production of glycopeptidolipids, which have been implicated in the establishment of the barrier function of M. abscessus [38,44]. However, what role MAB_1963 may play in general cell permeability is currently unknown.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Membrane transporters disrupt general cell permeability or efflux.

a, Rate of calcein accumulation as measured by calcein fluorescence in CRISPRi strains pre-treated with 500ng mL⁻¹ ATc for 24 hours prior to addition of calcein AM. Values normalized to vehicle only control per strain. b, Rate of ethidium accumulation as measured by fluorescence in live membrane transporter CRISPRi strains pre-treated with 500ng mL⁻¹ ATc for 24 hours prior to addition of ethidium bromide in. c, Rate of ethidium accumulation as measured by fluorescence in strains pre-treated with 500ng mL⁻¹ ATc for 24 hours prior to addition of ethidium bromide in live and fixed PBP-lipo CRISPRi strain. d, Rate of ethidium accumulation as measured by fluorescence in live and fixed M. abscessus ATC19977 wildtype treated with 50μM CCCP. e, Rate of ethidium accumulation as measured by fluorescence in live and fixed membrane transporter CRISPRi strains pre-treated with 500ng mL⁻¹ ATc for 24 hours prior to addition of ethidium bromide. All values normalized to vehicle only control per strain. Data for all graphs are represented as individual values along with mean ± s.d. n = 3 biological replicates. Statistical significance was calculated with two-way ANOVA; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. ns = not significant. NT = non-targeting sgRNA. ATc = anhydrotetracycline. CCCP = carbonyl-cyanide m-chlorophenylhydrazone.

https://doi.org/10.1371/journal.ppat.1013027.g003

To further examine general permeability as well as efflux, we performed an ethidium accumulation assay. Ethidium is an efflux substrate that increases in fluorescence when intercalated with DNA [45]. Either high cell permeability or low efflux rate will result in ethidium accumulation within the cell, leading to increased fluorescence intensity over time [45]. Knockdown of MAB_1963, MAB_2303, MAB_2624c, and MAB_4116c all lead to significantly increased rates of ethidium accumulation (Figs 3B and S5I-P), suggesting that these genes play a role either in establishing cell permeability or efflux of ethidium.

To distinguish between these possibilities, we compared ethidium accumulation in live and fixed cells. We reasoned that fixation would preserve the gross architecture of the cell wall but cease metabolism and thus eliminate the influence of active efflux on ethidium accumulation. Indeed, upon knockdown of PBP-lipo, a cell wall synthesis gene previously shown to increase membrane permeability to hydrophobic compounds [34], both live and fixed cells exhibited significantly increased rates of ethidium accumulation (Figs 3C, S5S and S5T), demonstrating that disruption to membrane permeability is preserved in fixed cells. Further, fixed wild-type cells display increased accumulation of ethidium compared to unfixed cells (S5Q and S5R Fig), demonstrating that fixation abrogates the ability of the cell to prevent ethidium accumulation. In contrast, efflux pumps that do not disrupt general membrane permeability should only impact ethidium accumulation in live cells but not fixed. Consistent with this model, total ethidium accumulation was significantly increased in live cells treated with carbonyl-cyanide m-chlorophenylhydrazone (CCCP), a protonophore that disrupts the proton gradient and thus hinders efflux activity [46], but was unchanged in fixed cells treated with CCCP (Figs 3D, S5Q and S5R). Thus, genes that impact ethidium uptake in live but not fixed cells are likely involved in efflux.

Using this methodology, we examined whether each membrane protein of interest is involved in efflux, general permeability, or both. Knockdown of MAB_4116c and MAB_2807 still resulted in increased rates of ethidium accumulation after fixation (Figs 3E and S5U-AB), suggesting that general cell wall permeability is likely compromised in those strains, allowing for greater ethidium accumulation despite the cessation of cell metabolism. In contrast, the knockdown of MAB_2303 and MAB_2624c did not lead to increased rates of calcein accumulation (Fig 3A) and only altered rates of ethidium accumulation in live cells (Fig 3C and 3E), suggesting that these genes may play a role in active efflux.

MAB_2303 is an efflux pump that plays a direct role in linezolid efflux

Given the substantial increase in linezolid susceptibility induced by MAB_2303 knockdown (Fig 2A, 2B and 2E), we chose to characterize this gene further. Based on the calcein and ethidium accumulation assays (Fig 3A, 3B and 3E), we posited that MAB_2303 may directly efflux linezolid, leading to drug resistance. To rule out the possibility that knockdown of MAB_2303 disrupts membrane or cell wall biology through indirect alteration of protein complex formation, we generated an enzymatically dead mutant of MAB_2303 through a tyrosine-856 to histidine (Y856H) mutation predicted to abrogate function [47]. The resulting protein is expressed at similar levels to wild type (S7A Fig), and, therefore, should participate in protein complexes but fail to perform its enzymatic function. Expression of a sgRNA-resistant construct containing both MAB_2302 and MAB_2303 to ensure proper expression of both members of the operon was able to rescue linezolid sensitivity of the MAB_2303 knockdown (Fig 4A and 4B). In contrast, the Y856H mutant was only partially able to rescue linezolid sensitivity (Fig 4A and 4B), indicating that the enzymatic function of MAB_2303 is required to fully counteract the effects of linezolid.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. MAB_2303 effluxes linezolid.

a, Half-maximal minimum inhibitory concentration (MIC₅₀) dose responses of M. abscessus with non-targeting sgRNA (NT) or sgMAB_2303 complemented with either wildtype (WT) or mutant (Y856H) sgRNA-resistant MAB_2302-2303 treated with indicated concentrations of linezolid in the presence or absence of 500 ng mL⁻¹ ATc for 48 hours. Values normalized to vehicle only control for each strain. b, Ratio of proliferation rates of pre-depleted MAB_2303 knockdown strains complemented with either wild type (WT) or mutant (Y856H) sgRNA-resistant MAB_2302-2303 treated with 0.5 μg mL⁻¹ linezolid in the presence or absence of 500ng mL⁻¹ ATc for 48 hours. Proliferation rates calculated from optical density of cultures over time. c, LC-MS measurement of cell-associated linezolid accumulation in pre-depleted MAB_2303 knockdown strains complemented with either wild type (WT) or mutant (Y856H) sgRNA-resistant MAB_2302-2303. Values are normalized to initial OD₆₀₀ measurements. d, LC-MS measurement of cell-associated linezolid accumulation in paraformaldehyde-fixed (+ Fixation) versus unfixed (-Fixation) MAB_2303 knockdown strains pre-treated for 18 hr with 500ng mL⁻¹ ATc (+ ATc) or with vehicle (-ATc) prior to fixation, then treated with 20 µM linezolid for 4 hr in the presence or absence of 500ng mL⁻¹ ATc. Relative accumulation indicates the ratio of normalized peak area of cell-associated linezolid after 4 hr incubation to normalized peak area of linezolid in the initial culture medium. All data are represented as individual values along with mean ± s.d. n = 3 biological replicates. Statistical significance was calculated with a Student’s T-test; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. ATc = anhydrotetracycline. NT = non-targeting sgRNA. KD = knockdown. LZD = linezolid.

https://doi.org/10.1371/journal.ppat.1013027.g004

To examine directly whether linezolid accumulation is affected by MAB_2303, we measured the intracellular accumulation of linezolid in the induced knockdown mutants by liquid chromatography-mass spectrometry (LC-MS). MAB_2303 knockdown increases linezolid accumulation in cells, and that accumulation is rescued by constitutive expression of the MAB_2302-2303 operon (Fig 4C and S6 Table), supporting the role of MAB_2303 as a linezolid efflux pump. To examine whether these effects are specific to the function of MAB_2303 rather than the general effects of knocking down membrane proteins, we depleted two different MmpL proteins, MAB_1134c and MAB_0987c, which were not significant hits in our TnSeq screen (Fig 1G and S3 Table). Knockdown of either of these proteins was not sufficient to increase linezolid accumulation (S6A Fig), demonstrating that the observed effects of MAB_2303 knockdown are likely specific to that protein’s function rather than general effects.

To further test the mechanistic role of MAB_2303 in linezolid accumulation, we examined whether MAB_2303 enzymatic function is required to prevent linezolid retention. Indeed, expression of the MAB_2302-2303 Y856H operon fails to reduce linezolid accumulation (Fig 4C and S6 Table), consistent with its inability to rescue linezolid susceptibility and supporting the conclusion that MAB_2303 plays a direct and active role in the efflux of linezolid. To orthogonally test whether MAB_2303 is an efflux pump or whether it impacts linezolid uptake by altering cell envelope structure, we examined whether cells fixed after knockdown of MAB_2303 display increased linezolid accumulation. Consistent with its potential role as an efflux pump, MAB_2303 knockdown does not result in any increase in linezolid accumulation in fixed cells (Fig 4D), suggesting that MAB_2303 depletion does not increase linezolid uptake through changes to cell envelope structure that are preserved after fixation. Interestingly, treatment of M. abscessus with 50 μM CCCP, a dose sufficient to increase ethidium uptake (Fig 3D), does not impact linezolid accumulation in the presence or absence of MAB_2303 knockdown (S6B Fig), suggesting that linezolid accumulation may be less sensitive to alterations to the proton gradient compared with ethidium. Together, these results are consistent with the possibility that MAB_2303 acts as an efflux pump for linezolid.

MAB_2303 alters susceptibility of some compounds chemically similar to linezolid

If MAB_2303 is responsible for exporting linezolid, its knockdown might alter susceptibility to other antibiotics with similar chemical structures. To identify compounds that might be exported by MAB_2303, we clustered 529 antibiotics based on their chemical similarity both by atom pair similarity (Fig 5A and S7 Table) and a fragment-based approach (Fig 5B and S8 Table). These plots are dimensional reductions that preserve the pairwise distance between each compound, such that the linear distance between two antibiotics on each plot represents their pairwise dissimilarity by the given metric. We sought to test whether compounds with more chemical similarity to linezolid are more likely to be effluxed by MAB_2303, as many efflux pumps display specificity for multiple drugs [14]. Further, linezolid is not likely to be the optimal substrate for MAB_2303, as the M. abscessus reference strain was never exposed to linezolid, indicating that MAB_2303 likely did not evolve to export linezolid. As a result, we hypothesized that there might be a chemical space over which drugs are more likely to be an efflux substrate of MAB_2303 that includes linezolid but is likely not centered on linezolid. Based on this hypothesis, we identified the clinically relevant antibiotics pretomanid, chloramphenicol, clofazimine, ofloxacin, moxifloxacin, and bedaquiline as chemically similar compounds that might be subject to efflux by MAB_2303. Accumulation measurements of 19 structurally diverse antibiotics in M. abscessus indicated that MAB_2303 limits accumulation of only pretomanid, chloramphenicol, and trimethoprim in addition to linezolid (Fig 5C). The substantial change in pretomanid accumulation suggests that the ideal substrate of MAB_2303 might fall closer to pretomanid in chemical space than linezolid. To test this model, we examined whether the antibiotics closest to pretomanid by both atom pair similarity and by fragment based similarity were significantly enriched for drugs that appear to be effluxed by MAB_2303. Indeed, by both atom pair similarity and by fragment based similarity, compounds that are more similar to pretomanid are statistically more likely to build up in cells upon MAB_2303 knockdown (S9 and S10 Tables), demonstrating that antibiotic efflux follows predictable chemical patterns and that MAB_2303 may provide resistance to antibiotics with a similar structure to linezolid and pretomanid.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. MAB_2303 effluxes compounds that share similar chemical properties.

a-b, 529 antibiotics represented by (a) multidimensional scaling of atom pair similarity and (b) UMAP dimensional reduction of fragment-based similarity. Panel of mycobacterial antibiotics from Fig 1 are highlighted in orange and labeled. c, LC-MS measurement of intracellular accumulation of indicated antibiotics in pre-depleted (+ATc) sgMAB_2303 CRISPRi M. abscessus strain incubated for 4 hr with antibiotics. Values normalized to internal standard, initial antibiotic levels in media prior to incubation, and to a non-depleted strain control (-ATc). Data are represented as individual values along with mean ± s.d. n = 3 biological replicates. d-f, Half-maximal minimum inhibitory concentration (MIC₅₀) dose responses of pre-depleted MAB_2303 knockdown strain as measured by reduction of a colorimetric dye after treatment with indicated concentrations of (d) chloramphenicol, (e) clarithromycin, and (f) rifampicin in the presence or absence of 500 ng mL⁻¹ ATc for 24 hours. Values normalized to vehicle only control per drug. Data are represented as individual values along with mean ± s.d. n = 3 biological replicates. ATc = anhydrotetracycline.

https://doi.org/10.1371/journal.ppat.1013027.g005

To test this hypothesis, we examined whether MAB_2303 knockdown alters chloramphenicol efficacy, as M. abscessus is not responsive to pretomanid [48] or trimethoprim [49], precluding determination of susceptibility to those compounds. Consistent with the model, knockdown of MAB_2303 increases susceptibility to chloramphenicol (Fig 5D), but does not affect susceptibility to structurally dissimilar compounds such as clarithromycin and rifampicin (Figs 5E, 5F, and S7B-F). Further, the lack of effect on susceptibility to ribosome-targeting antibiotics, including clarithromycin (Fig 5E) and amikacin (S7B Fig), suggests that MAB_2303 does not have direct effects on the ribosome. Of interest, MAB_2303 knockdown does result in a modest increase in susceptibility to fluoroquinolone antibiotics (S7D and S7E Fig) without increasing their accumulation, suggesting that MAB_2303 may play other roles in intrinsic antibiotic resistance beyond direct drug efflux. This result could also provide some insight into the role that MAB_2303 may play in biology beyond efflux. Additionally, we also observed that the knockdown of MAB_2303 led to a decrease in antibiotic accumulation of several compounds. We speculate that MAB_2303 knockdown might lead to an upregulation of other drug efflux pumps or other permeability barriers that could alter the accumulation of other antibiotics. Together, these results indicate that MAB_2303 contributes to the M. abscessus permeability barrier within a relatively specific chemical space and argue that drug impermeability and lack of accumulation are likely driven by the overlapping effects of numerous such membrane proteins.

Strains

All experiments were performed in the Mycobacterium abscessus subspecies abscessus type strain (ATCC19977) unless otherwise indicated. Clinical isolate M. abscessus subspecies massiliense BWH-F was isolated from a skin biopsy [35]. All plasmid construction was performed in DH5α Escherichia coli.

Mycobacterial culturing conditions

M. abscessus liquid cultures were grown in Middlebrook 7H9 broth (271310, BD Diagnostics) with 0.2% (v/v) glycerol (GX0185, Supelco), 0.05% (v/v) Tween-80 (P1754, MilliporeSigma), and 10% (v/v) oleic acid-albumin-dextrose-catalase (OADC) (90000-614, VWR) or 10% (v/v) albumin-dextrose-catalase (ADC) composed of 50 g/L albumin, 0.03 g/L catalase, 8.5 g/L NaCl, and 20 g/L dextrose. Cultures were shaken at 150 r.p.m. at 37 °C.

Mycobacterial transformations

M. abscessus strains were grown to an optical density (OD₆₀₀) of 0.8, and washed thrice with sterile 10% glycerol by pelleting at 5000 x g for 7 minutes at 22 °C. After final wash, cells were resuspended in 1% of the initial culture volume in 10% glycerol. 50 µL of electrocompetent mycobacteria were mixed well with 100 ng plasmid in 2 µL water and then transferred to a 2 mm electroporation cuvette (89047-208, VWR). The cells were electroporated at 2500 V, 125 Ω, 25 μF using an ECM 630 electroporator (45-0651, BTX). 1 mL 7H9 +OADC broth was added to the electroporated cells, and cells recovered for 4 hours shaking at 150 r.p.m. at 37 °C. 100 µL of recovered cells were spread on 7H10 + 0.5% (v/v) glycerol + 10% (v/v) OADC agar plates with 50 µg mL⁻¹ kanamycin sulfate using 4 mm borosilicate glass beads. Plates were incubated at 37 °C for 4 days or until colonies were visible.

Generation of CRISPRi and CRISPRi-resistant mutant strains

CRISPRi plasmids were constructed as previously described [41] using Addgene plasmid 166886. Briefly, plasmid 166886 was digested overnight at 55 °C with BsmBI-v2 (R0739L, New England BioLabs) and then gel purified (T1020, New England BioLabs). Three sgRNAs were selected using Pebble sgRNA Design Tool (Rock Lab, Rockefeller University) to target three different locations of the non-template strand of each gene of interest. Each individual sgRNA with appropriate overhangs was annealed and ligated using T4 ligase (M0202M, New England BioLabs) into three separate BsmBI-v2 digested backbones. To generate a triple CRISPRi plasmid with all three sgRNAs, SapI-based Golden Gate cloning site 3′ to the first sgRNA scaffold was used as previously described [41]. CRISPRi NT was constructed in a similar manner but with scrambled, non-targeting sgRNAs. Successful plasmid construction was verified using whole plasmid sequencing with Oxford Nanopore Technologies (ONT) (Plasmidsaurus). Triple CRISPRi plasmids were transformed into ATCC19977 as described above and selected on 7H10 + 0.5% (v/v) glycerol + 10% (v/v) OADC agar plates containing 50 µg mL⁻¹ kanamycin sulfate.

MAB_2302-MAB_2303 rescue plasmids were constructed by introducing synonymous mutations at the protospacer adjacent motif (PAM) and seed sequences for all sites of sgRNA targeting. Additionally, a second version of the MAB_2302-MAB_2303 rescue plasmid was constructed with the same CRISPRi-resistant synonymous mutations as well as a nonsynonymous mutation (Y856H) at the proposed catalytic site in the mmpL (MAB_2303). Gene fragments (Azenta) containing these synonymous mutations and/or a nonsynonymous mutation in the catalytic site with NdeI (R0111S, New England Biolabs) and XhoI (R0146S, New England Biolabs) overhang sites were restriction enzyme digested and then assembled into Tweety-integrating zeocin marked MOP plasmids using the Gibson Assembly standard protocol (E5510, New England Biolabs). Successful plasmid construction was verified using whole plasmid sequencing with Oxford Nanopore Technologies (ONT) (Plasmidsaurus). Plasmids transformed into the CRISPRi sgMAB_2303 strain using transformation protocol described above and selected on 7H10 + 0.5% (v/v) glycerol + 10% (v/v) OADC agar plates containing 50 µg mL⁻¹ zeocin and 50 µg mL⁻¹ kanamycin sulfate.

Minimum inhibitory concentration determination

M. abscessus strains were grown until mid-log phase (OD₆₀₀ of 0.6–0.8). Strains were induced for knockdown 18–24 hours prior to start of the assay with 500 ng μL⁻¹ ATc. Cultures were then diluted to OD₆₀₀ of 0.003 and 200 μl aliquots were plated in biological triplicate in wells (3370, Corning) containing specified antibiotics as well as fresh 500 ng μL⁻¹ ATc or vehicle when relevant. Antibiotic stocks were made as follows: 20 mg mL⁻¹ linezolid (PZ0014, Sigma-Aldrich) in DMSO, 10 mg mL⁻¹ clarithromycin (C9742, Sigma-Aldrich) in DMSO, 10 mg mL⁻¹ amikacin disulfate salt (A1774, Sigma-Aldrich) in water, 20 mg mL⁻¹ rifampicin (R3501, Sigma-Aldrich) in DMSO, and 10 mg mL⁻¹ moxifloxacin (SML1581, Sigma Aldrich) in DMSO. The cells were then incubated at 37 °C with shaking at 150 r.p.m. for 24 hours. 0.002% resazurin (R7017, Sigma Aldrich) in ddH2O was spiked into each well and plates were incubated for an additional 24 hours at 37 °C with shaking at 150 r.p.m. MIC determination was conducted using a Tecan Spark 10M plate reader (Mannedorf, Switzerland) by measuring absorbance at 570nm and 600nm and normalizing the ratio to background and no drug control. Nonlinear regression was used to fit a sigmoid curve to the dose-response data and calculate the half-maximal minimum inhibitory concentration (MIC₅₀) using GraphPad Prism. Half-maximal minimum inhibitory concentrations (MIC₅₀) for Fig 1C were derived from prior literature: clofazimine [59], azithromycin [60], bedaquiline [61], clarithromycin (this study), erythromycin (this study), linezolid (this study), moxifloxacin (this study), chloramphenicol [62], SQ109 [63], ofloxacin (this study), ethambutol [4], pretomanid [64], metronidazole [64], cycloserine [65], clindamycin [60], rifampicin (this study), isoniazid [66], and trimethoprim [49].

Growth curve

CRISPRi strains were grown until mid-log phase (OD₆₀₀ of 0.6–0.8) and then pre-depleted with ATc at 500 ng mL⁻¹ for 18-24 hours. Cultures were then back-diluted at final OD₆₀₀ of 0.02 and 200 µl of diluted cells were added in biological triplicates with DMSO or 0.5 µg mL⁻¹ linezolid and fresh ATc at 500 ng mL⁻¹ when relevant. Growth was determined by continuous OD₆₀₀ measurement in 15-minute intervals in a Spark 10M plate reader for 48 hours at 37 °C with continuous shaking at 1000 rpm. Growth curve data were analyzed using Microsoft Excel 365 and GraphPad Prism 9.

Transposon library production

The BWH-F transposon mutant library (69.5% TA insertion coverage) was previously generated [35]. Briefly, the M. abscessus BWH-F strain was transduced with temperature-sensitive φMycoMarT7 phage carrying the Himar1 transposon. After selection with 100µg mL⁻¹ kanamycin, mutant libraries with >150,000 individual bacterial mutants were harvested and stored in aliquots with 7H9 + 10% (v/v) glycerol at −80 °C.

Transposon library growth conditions and selection

Transposon mutant BWH-F library was inoculated in biological triplicates at 2.1x10⁷ CFU mL⁻¹ into 7H9 + 0.5% (v/v) glycerol + 10% (v/v) OADC either incubated with 16 µg mL⁻¹ linezolid or DMSO. After 11 doublings, cultures were pelleted at 5,000 x g for 5 min at 22 °C, resuspended in 7H9 + 0.2% (v/v) glycerol + 0.05% (v/v) Tween-80 + 10% (v/v) OADC, mixed equal volume with 50% glycerol, and frozen at −80 °C. The harvested replicates were then titered by plating on 7H10 + 0.5% (v/v) glycerol + 10% (v/v) OADC agar plates supplemented with 100 µg mL⁻¹ kanamycin sulfate. 150,000 cells of each replicate were plated on 7H10 + 0.5% (v/v) glycerol + 10% (v/v) OADC + 0.005% Tween 80 + 100 µg mL⁻¹ kanamycin sulfate on six 245 mm x 245 mm plates (431111, Corning). Colonies were grown for 4 days at 37 °C. Each replicate was combined via scraping into a 50 mL conical tube containing 5 ml 7H9 + 0.2% (v/v) glycerol + 0.05% (v/v) Tween-80 + 10% (v/v) OADC and 5 mL 50% glycerol. 2 mL aliquots of the libraries were stored at −80 °C for genomic DNA extraction.

Genomic DNA isolation

gDNA was isolated using an established protocol with minor adaptations [35,67]^28,59. The post-selection mutant libraries were thawed, pelleted at 5,000 x g for 5 min at 22 °C and then resuspended in TE Buffer (10mM Tris HCl pH 7.4 and 1mM EDTA pH 8). Cell suspensions were transferred to 2 mL tubes containing 0.1 mm silica beads (116911500, MP Biomedicals) and 700 µL of 25:24:1 phenol:chloroform:isoamyl alcohol (P3803, MilliporeSigma). Bacteria were lysed utilizing a Bead Bug 3 Microtube Homogenizer (D1030, Benchmark Scientific, Sayreville, NJ, USA) four times at 45 second intervals at 4000 r.p.m. Samples were chilled on ice for 45 seconds between cycles. Post-homogenization, cell debris was pelleted at 21,130 x g for 10 minutes at 22 °C. The aqueous phase was combined with an equal volume of 25:24:1 phenol:chloroform:isoamyl alcohol and incubated on a rocker for 1 hour at 22 °C. The mixtures were then transferred to pre-pelleted MaXtract High-Density phase-lock tubes (129065, Qiagen), followed by centrifugation at 1500 x g for 5 minutes at 4 °C. ½ volume of chloroform (193814, MP Biomedicals) was added to upper aqueous phase and centrifuged at 1500 x g for 5 minutes at 4 °C.The upper aqueous layers were transferred to new MaXtract High-Density phase-lock tubes and incubated with RNase A at 25 µg mL⁻¹ (EN0531, Thermo Fisher Scientific) at 150 r.p.m. for 1 hour at 37 °C. Samples were re-extracted with an equal volume of 25:24:1 phenol:chloroform:isoamyl alcohol, followed by centrifugation at 1500 x g for 5 minutes at 4 °C. A second extraction with ½ volume chloroform was performed, and centrifuged at 1500 x g for 5 min at 4 °C. The aqueous phase containing the DNA of each sample was transferred to a fresh conical tube prepared with 1/10th volume of 3M sodium acetate pH 5.2 (3032-16, VWR) and 1 volume of isopropanol (3032-16, VWR), followed by an overnight incubation at 22 °C. The DNA pellets were washed thrice with 5 mL 70% ethanol at 5000 x g for 10 min, dried for 10 minutes to eliminate residual ethanol, and resuspended in 500 µL nuclease-free water. DNA concentration of the samples was quantified with Qubit Fluorometer (Q33238, Thermo Fisher Scientific) using the Broad Range assay kit (Q33266, Thermo Fisher Scientific).

Transposon sequencing, mapping, and analysis

Sequencing libraries were generated from the isolated genomic DNA by amplifying chromosomal-transposon junctions, following an established protocol outlined by Long et al. 2015 [68] and sequenced on an Illumina NextSeq 500 sequencer. The resultant reads were aligned to the BWH-F genome (SRA project number PRJNA840944, accession SRX15416547). Analysis of the data was conducted employing TRANSIT [69]. Insertion counts at each TA site were subjected to trimmed total reads normalization and were then averaged across replicates. The resampling analysis in TRANSIT was applied for the comparative assessment of essentiality between genes in the no drug and linezolid exposed strains. TA site insertions at the 5% C- or N-terminus of each gene were trimmed. P values were derived from a permutation test, and adjusted P values for multiple tests were derived from the Benjamini-Hochberg method.

Calcein accumulation assay

CRISPRi strains were grown until mid-log phase (OD₆₀₀ of 0.6–0.8) and then back diluted to pre-deplete target proteins with ATc 500 ng mL⁻¹ or DMSO for 18-24 hours. Bacteria were washed twice with 1x PBS pH 7.4 (10010049, Thermo Fisher) + 0.05% Tween80 (PBST). After final wash, cells were resuspended in 1x PBST for final OD₆₀₀ of 0.4 and added to black-bottomed 96-well plate (07-200-722, Thermo Fisher). Plates were incubated at 37 °C with shaking at 150 r.p.m. for 30 minutes. 1 µg mL⁻¹ Calcein AM (C3100MP, Thermo Fisher) was then spiked into each well. Plates were incubated at 37 °C with shaking in a Tecan Spark 10M plate reader with ex/em 488nm/520nm fluorescence every minute for 40 minutes.

Ethidium accumulation assay

CRISPRi strains were grown until mid-log phase (OD₆₀₀ of 0.6-0.8) and then back diluted to pre-deplete target proteins with ATc 500 ng mL⁻¹ or DMSO for 18-24 hours. Bacteria were back diluted to an OD₆₀₀ of 0.2 and half of the total volume was transferred directly to black-bottom 96-well plates (07-200-722, Thermo Fisher) or fixed with 2% (v/v) paraformaldehyde (sc-281692, ChemCruz) for 1 hour. Following fixation, cells were centrifuged at 5000 x g for 5 minutes and resuspended in the original volume of 7H9 media and transferred to black-bottom 96-well plates. Ethidium bromide (E7637, Sigma-Aldrich) was then added to a final concentration of 1 µg mL⁻¹. The plate was shaken manually and incubated at 22 °C for 10 minutes. Ethidium fluorescence was then measured at 530nm/600 nm ex/em in 1 minute cycles for 1 hour at 37 °C using a Tecan Spark 10M plate reader.

Plate-based growth assay

CRISPRi strains were grown until mid-log phase (OD₆₀₀ of 0.6-0.8) and then back diluted to pre-deplete target proteins with 500 ng mL⁻¹ ATc or DMSO for 24 hours. Culture OD was matched for each strain +/- ATc, and 10-fold serial dilutions of each strain were plated on 7H10 + 0.5% (v/v) glycerol + 10% (v/v) OADC agar plates with 2 μg/mL linezolid, with or without 500 ng mL⁻¹ ATc as indicated. Plates were then incubated 4 days at 37 °C. Each assay was repeated in biological triplicate, and images are representative of those replicates.

Microscopy

CRISPRi strains were grown until mid-log phase (OD₆₀₀ of 0.6-0.8) and then back diluted to pre-deplete target proteins with 500 ng mL⁻¹ ATc or DMSO for 24 hours. Strains were fixed with 2% (v/v) paraformaldehyde (sc-281692, ChemCruz) for 1 hour. Following fixation, cells were centrifuged at 5000 x g for 5 minutes and resuspended in 1x PBS pH 7.4 (10010049, Thermo Fisher) + 0.05% (v/v) Tween-80 (P1754, MilliporeSigma) in 1/10^th the original volume and then seeded onto a 2.0% agarose pad. Phase contrast images were obtained at 100x magnification using an inverted Nikon TI-E microscope.

Arrayed antibiotic extraction

Wild type or CRISPRi strains were grown until mid-log phase (OD₆₀₀ of 0.6–0.8) and then diluted back and cultured for 18–24 hours to pre-deplete target proteins with either 500 ng mL⁻¹ ATc, DMSO, or no treatment for CRISPRi + ATc, -ATc, and wild type, respectively. Cultures were pelleted at 3200 x g for 10 minutes at room temperature then washed in 1X volume blood bank saline (89370-096, VWR), and resuspended in 7H9 + 0.5% (v/v) glycerol + 10% (v/v) ADC at a final OD₆₀₀ of 15 with 500 ng mL⁻¹ ATc or equivalent volume of DMSO added to appropriate cultures. For fixation experiments, cells were instead first resuspended in 4% paraformaldehyde, then incubated at 22 °C for 1 hr prior to resuspension in 7H9 + 0.5% (v/v) glycerol + 10% (v/v) ADC. For CCCP treatment, cells were either pre-treated with 50 μM CCCP or vehicle for 10 minutes at 22 °C, then 50 μM CCCP was included in the culture media for +CCCP conditions. Cells were added in biological triplicate to 96-well plates with 7H9 + 0.5% (v/v) glycerol + 10% (v/v) ADC containing a final concentration of 20 µM antibiotic in each of 20 wells resulting in a final OD₆₀₀ = 7.5. 50 µL of media was immediately collected from each well and pooled, then filter sterilized through a 0.22 µm PVDF filter (SE1M179M6, Millipore). Media was stored at −80 °C until extraction. Cells were incubated with antibiotics for 4 hr at 37 °C with shaking at 150 rpm. After incubation, cultures were combined in a conical tube, then pelleted at 3200 x g for 10 minutes at 4 °C and washed twice with pre-chilled blood bank saline to remove excess extracellular drug. Pellets were resuspended in 0.8 mL 3:1:0.004 acetonitrile:methanol:formic acid + 10 nM verapamil, then transferred to 2 mL tubes containing 0.1 mm silica beads (116911500, MP Biomedicals). Bacteria were lysed utilizing a Bead Bug 3 Microtube Homogenizer (D1030, Benchmark Scientific, Sayreville, NJ, USA) three times at 45 second intervals at 4000 rpm. Samples were chilled on ice for 2 minutes in between cycles. Samples were pelleted at 21,130 x g for 10 minutes at 4 °C. 50 µL of media was extracted with 450 µL 3:1:0.004 (v/v/v) acetonitrile:methanol:formic acid + 10 nM verapamil (V105, Millipore Sigma) as an internal standard. Media samples were vortexed for 1 minute at 22 °C, then pelleted for 10 minutes at 17,000 x g at 4 °C. Supernatant from both media and cell pellet extractions was dried using a speedvac concentrator (Eppendorf 5305) 1 hr at 45 °C, resuspended in 40 µL 3:1:0.004 (v/v/v) acetonitrile:methanol:formic acid, vortexed 1 minute at 22 °C, and pelleted at 17,000 x g at 4 °C. 35 µL supernatant was transferred to 9 mm plastic vials (Thermo C4000-11) with screw caps (Thermo 03-376-481) and stored at −80 °C until LC-MS analysis.

Linezolid extraction

CRISPRi strains were grown until mid-log phase (OD₆₀₀ of 0.6–0.8) and then back diluted to pre-deplete target proteins with ATc 500 ng mL⁻¹ or DMSO for 18–24 hours in biological triplicates in 50mL total volume each. Each replicate was normalized to an OD₆₀₀ of 0.8 and the pelleted at 4000 x g for 10 minutes at room temperature then resuspended in 1/5^th the total volume with fresh 7H9 + 0.5% (v/v) glycerol + 10% (v/v) OADC. Fresh 500 ng mL⁻¹ ATc or DMSO was added to appropriate cultures and then 20 µg mL⁻¹ linezolid was added to each culture. Cultures were incubated for 2 hours at 37 °C at 150 r.p.m. After incubation, cultures were pelleted at 4000 x g for 10 minutes at 4 °C and washed once with pre-chilled 1x PBS. Pellets were stored at −80 °C until metabolite extraction. Pellets were resuspended in pre-chilled 1mL of 2:2:1 acetonitrile (34851, Millipore Sigma) + methanol (439193, Millipore Sigma) + water with 2 µg mL⁻¹ linezolid-d3 (25038, Cayman Chemicals) and 2 µg mL⁻¹ verapamil (V105, Millipore Sigma). Cell suspensions were transferred 2 mL tubes containing 0.1 mm silica beads (116911500, MP Biomedicals). Bacteria were lysed utilizing a Bead Bug 3 Microtube Homogenizer (D1030, Benchmark Scientific, Sayreville, NJ, USA) for four times at 45 second intervals at 4000 r.p.m. Samples were chilled on ice for 2 minutes in between cycles. Post-homogenization, cell debris was pelleted at 21,130 x g for 10 minutes at 22 °C.

Liquid chromatography-mass spectrometry

The LC-MS analysis for linezolid was following a reported method as described [70,71]. We applied an Agilent 1260 HPLC coupled with an Agilent 6120 quadrupole mass spectrum for compound accumulation analysis. Metabolites were separated using a 10-µL injection volume on a Chromolith SpeedRod RP-18 column (Sigma Aldrich) with a gradient of H₂O (solvent A) and acetonitrile (solvent B) acidified with 0.1% formic acid. The gradient was as follows: 0 min, 10% B; 2 min, 10% B; 10 min, 100% B; 12 min, 10% B. Data were analyzed using Agilent ChemStation software to measure levels of the linezolid and d3-linezolid [M+H]⁺ ion with an accuracy of ± 20 ppm.,Signal intensity was quantified by standard curve of ratio of authentic linezolid versus d3-linezolid, and normalized by OD value at time of bacteria harvest.

LC-MS analysis for the antibiotic panel was performed using a QExactive+ orbitrap mass spectrometer (ThermoFisher) with a heated electrospray ionization (HESI) probe, coupled to a Dionex Ultimate 3000 UPLC system. 4 µL of extracted sample was injected into a Kinetex 2.6 µm EVO C18 column (150 x 2.1 mm), with the autosampler and column held at 4 °C and 30 °C, respectively. The chromatographic gradient consisted of 0.1% formic acid (solvent A) and 0.1% formic acid in acetonitrile (solvent B). The gradient was run as follows: 0–5 min: 1% solvent B, flow rate 0.3 mL/min; 5–15 min: linear gradient from 1–99% solvent B, flow rate 0.3 mL/min; 15–20 min: 99% solvent B, flow rate 0.3 mL/min; 20–25 min: 99% solvent B, flow rate 0.4 mL/min; 25–30 min: 1% solvent B, flow rate 0.3 mL/min. The mass spectrometer was operated in full scan, positive mode. The MS data acquisition was performed in a range of 100–1500 m/z, with the resolution set to 70,000, the AGC target at 1e6, and the maximum injection time at 50 msec.

After LC-MS analysis, antibiotic identification was performed with XCalibur 3.0.63 software (Thermo Fisher Scientific) using a 5ppm mass accuracy and a 0.5 min retention time window. Standards were used for assignment of antibiotic peaks at given m/z for the [M+H]⁺ ion and retention time and were compared to extraction buffer, media alone, and cells alone blanks. For metronidazole measurements, intracellular measurements represent hydroxymetronidazole. Linear range of detection was determined by analyzing four 10-fold dilutions of antibiotic standards containing 2 nmol, 200 pmol, 20 pmol, and 2 pmol of each antibiotic. Linear range was examined by ensuring that all data points fall within the 95% confidence interval of the linear regression, and by a Wald-Wolfowitz runs test to check deviation from linearity. Peak areas were normalized to verapamil internal standard and culture density. Relative antibiotic accumulation was calculated by normalizing intracellular antibiotic peak area after 4 hours of incubation to antibiotic peak area in media prior to incubation with M. abscessus.

Western blotting

MAB_2303 CRISPRi strains expressing either wild type MAB_2303 or MAB_2303 Y856H were grown until mid-log phase (OD₆₀₀ of 0.6–0.8) and then diluted back and cultured for 18–24 hours with 500 ng mL⁻¹ ATc. Protein was isolated by pelleting these bacteria 3200 × g for 10 minutes at 4 °C, resuspending in Tris-buffered saline (TBS) (28358 Thermo Fisher Scientific) + protease inhibitor (11873580001, MilliporeSigma), transferring to 2 mL tubes with 0.1 mm silica beads (116911500, MP Biomedicals), and homogenizing using a Bead Bug 3 Microtube Homogenizer 4 × 45 seconds at 4000 rpm with 2 minutes of incubation on ice between rounds of homogenization. SDS was added to homogenized samples to a final concentration of 0.5%, then samples were incubated with agitation 1 hr at 37 °C. Samples were then pelleted 21,130 × g for 5 minutes at 4 °C, and the supernatant was heat-killed by incubation at 80 °C for 20 minutes. Protein abundance was quantitated by absorbance at 280 nm using a Nanodrop 1000 spectrophotometer (Thermo Fisher Scientific), and samples were normalized to 6 mg/mL by dilution in TBS. After normalization, remaining DNA was digested by addition of TURBO DNase buffer (AM2238, Thermo Fisher Scientific) (final concentration of 10%) and TURBO DNase (final concentration of 2%) followed by incubation at 37 °C for 15 minutes. Samples were mixed with 4X LDS NuPage sample buffer to a final concentration of 1X. Samples were incubated at 70 °C for 10 minutes, then 35 μg of protein along with PageRuler Prestained ladder 10 kDa to 180 kDa (26616, Thermo Fisher Scientific) was loaded on a NuPage 4–12% gradient Bis-Tris pre-cast SDS-PAGE gel (NP0321, Thermo Fisher Scientific), which was electrophoresed at 200 V for 45 minutes. Proteins were transferred to a PVDF membrane (1704156, Bio-Rad Laboratories) using TransBlot Turbo Transfer System (Bio-Rad) on the Mixed MW setting. Membranes were blocked by incubating in TBS + 0.1% Tween 20 (TBST) + 5% bovine serum albumin 1 hr at 22 °C, and then were incubated with anti-His tag antibody (MA1-21315, Invitrogen) diluted 1:1,000 in TBST + 5% bovine serum albumin for 18 hr at 4 °C. Membranes were washed 3x in TBST to remove unbound antibody, then incubated 1 hr at 22 °C with anti-mouse antibody (ab97023, abcam), followed by 3x washes in TBST and developing using Azure Radiance Plus Detection Reagent (AC2103, Azure Biosystems). Membranes were imaged using the chemiluminescence detector of a c300 Gel Imaging System (Azure Biosystems). After blotting, total protein was detected by staining membranes with SYPRO Ruby Protein Blot Stain (S11791, Thermo Fisher Scientific) according to manufacturer’s instructions. SYPRO Ruby staining was imaged using the Epi Blue setting of the c300 Gel Imaging System.

Antibiotic chemical similarity

529 antibiotics were selected for chemical similarity analysis by manually curating components of a commercially available antibiotic library (HY-L067, MedChemExpress, Princeton, NJ, USA). Atom pair similarity was calculated using Tanimoto coefficients, which is the single highest-performing similarity scoring approach for chemical comparisons [72] and represented using multidimensional scaling with ChemMineR v3.54.0 [73]. Chemical fingerprints were calculated using FragFP in DataWarrior v06.02.01 [74] and represented as a 2-dimensional UMAP with nearest neighbors set to 100, minimum distance set to 0.5, and Euclidean distance.

Statistical methods

For all datapoints, error bars represent the standard deviation of the y-variable on the graph. p-values for correlation analyses were derived from two-tailed comparisons for both Pearson and Spearman correlations. p-values for chi-squared tests were derived from two-sided comparisons. All statistical tests used are indicated in figure legends.