A Nutrient-limited Screen Unmasks Rifabutin Hyperactivity for XDR Acinetobacter baumannii

Abstract

Industry screens of large chemical libraries traditionally have relied on rich media to ensure rapid bacterial growth in high-throughput testing. We used eukaryotic, nutrient-limited growth media in a compound screen that unmasked a previously unknown hyper-activity of the old antibiotic, rifabutin (RBT), against highly resistant Acinetobacter baumannii. In nutrient-limited, but not rich media, RBT was 200-fold more potent than rifampin (RIF). RBT was also substantially more effective in vivo. The mechanism of enhanced efficacy was a Trojan horse-like import of RBT but not RIF through FhuE, only in nutrient-limited conditions. These results are of fundamental importance to efforts to discover antibacterial agents.

Discovery of new antibiotics has traditionally occurred via high throughput screening assays. These assays use rich media, such as Mueller-Hinton II (MHII) broth, to enable rapid microbial growth to facilitate the efficiency of the screens. Unfortunately, such methods rarely identify any new candidates.^1-5,7 However, we hypothesized that modifying the screening methodology by changing the media to be more reflective of the in vivo environment could uncover new activity of drugs with previously unknown efficacy. To test this hypothesis, we designed a nutrient-limited screen using media, more reflective of the in vivo environment. We sought to identify new antibiotics active against Acinetobacter baumannii, one of the most drug resistant bacterial pathogens, which is listed as the top priority of unmet need for new antibiotics by the World Health Organization.⁶ Blood or lung infections caused by extreme drug resistant (XDR) A. baumannii cause >50% mortality,⁸ and the antimicrobial pipeline for such infections is inadequate.^9-13

The ReFRAME compound drug repurposing library¹⁴ was screened against our well-characterized XDR, hyper-virulent clinical isolate of A. baumannii HUMC1.^15-19 As the growth media for our screen, we selected Roswell Park Memorial Institute medium (RPMI), a defined eukaryotic growth medium containing glucose (2 g/L) as the only carbon source, and mineral salts. We supplemented the RPMI with serum to better simulate the bloodstream environment.^20,21 Previous studies have identified the positive benefits of screening for antimicrobial compounds in culture conditions that are more reflective of conditions found in vivo^7,22,23.

In total we screened 11,862 small molecules at 20 μM (Supplementary Data File 1). The screen resulted in a hit rate of 0.52%; 62 putative hits were identified for follow-up confirmation in a serial dose response screen. About half of the compounds (32/62) were validated to inhibit 50% of bacterial growth (IC₅₀) at a concentration of less than 20 μM. These compounds were also assessed for half maximal mammalian cell cytotoxicity (CC₅₀), and 4 compounds had a favorable specific antibacterial activity (IC₅₀ <20μM) and high selectivity (CC₅₀/IC₅₀ > 10); of these rifabutin (RBT) was the most potent (i.e., had the lowest IC₅₀s, Fig. 1A).

Figure 1-. Summary of the ReFRAME library screen used for the identification of rifabutin.

11,862 small molecules were screened against A. baumannii ATCC 17978 (antibiotic-sensitive) and HUMC1 (extensively drug-resistant strains). A) Specificity (mammalian cell cytotoxicity / inhibition of bacteria growth; CC₅₀ / IC₅₀) of the top 4 compound screen hits against A. baumannii HUMC1 and HEK293T and HepG2 mammalian cells. B) The IC₅₀ was determined for 8 rifamycins. C) Specificity (mammalian cell cytotoxicity / inhibition of bacteria growth; CC₅₀ / IC₅₀) of 8 rifamycins were tested against HEK293T mammalian cells and A. baumannii ATCC17978 andHUMC1 in RPMI with and without 50% serum. D) The fold change of MICs, when bacteria were cultured in RPMI or MHII, were determined for RBT and RIF against a panel of ESKAPE clinical isolates. One biological replicate of high throughput assays (A-C) were done in 3 technical replicates. MIC assays were repeated twice in duplicate.

We confirmed the superiority of RBT IC₅₀ values in RPMI media, compared to a panel of 8 rifamycins, using a more sensitive, 11-point dose response curve (Fig. 1B). RBT was also confirmed to be the most potent and the least toxic rifamycin tested (Fig. 1C).

The minimum inhibitory concentration (MIC) of each drug was determined against A. baumannii HUMC1, a hyper-virulent, carbapenem-resistant, XDR strain. In nutrient-limited RPMI, the MIC of RBT (0.0156 μg/mL) was 200-fold lower as compared to RIF (3.125 μg/mL) (Fig. 1E). However, there was no difference in MIC when the bacteria were cultured in nutrient-rich MHII (MIC=3.125 μg/mL for both RBT and RIF) (Fig 1D). In contrast to A. baumannii, we did not find a universal improvement of RBT potency against other bacterial species tested (Fig. 1D, Extended Data Fig. 1).

We sought to understand the drivers responsible for the change in RBT potency in RPMI versus MHII media. Modulating the growth rate (temperature), carbon metabolism (glucose supplementation), or activity of cellular efflux pumps (efflux pump inhibitors) did not affect the MIC shift in nutrient-rich vs. -limited media. (Table 1). To distinguish if the discrepant RBT MICs in RPMI vs. MHII media was due to inhibition of drug’s activity by MHII or enhancement of its activity by RPMI, we mixed RPMI and MHII media in equal parts together and used this media to test the MIC. The MIC of RBT was 8 μg/mL in the hybrid media, similar to the rich media and much higher than the nutrient-limited media (Table 1). Thus, the rich MHII media antagonized the hyper-activity of RBT.

Table 1: MICs were performed against HUMC1 to determine which component of MHII inhibits the activity of RBT.

The MIC for CCCP and thioridazine alone are 8 and 64 μg/mL respectively.

	RBT MIC (μg/mL)
	MHII	RPMI
Baseline	4.00	0.02
Temperature
25 °C	4.00	0.05
Glucose Supplementation
MHII + 200 mg/L glucose	2.00	-
MHII + 2,000 mg/L glucose	2.00	-
Modified RPMI
5X RPMI	-	0.06
RPMI Supplemented with MHII	-	8.00
Efflux Pump Inhibitor
CCCP (2 μg/mL)	1.56	-
thioridazine (8 μg/mL)	1.56	-
Verapamil (50 μg/mL)	1.56	-
Fractionated MHII Spiked Into RPMI
10-30 kDa MHII Fraction	-	8.00
<10 kDa MHII Fraction	-	8.00
Acetonitrile Treated <10 kDa MHII Fraction
Aqueous Layer	-	0.03
Organic Layer	-	4.00
Organic Layer, Proteinase K Digested	-	4.00
Organic layer, Sodium Periodate Treated	-	4.00
Amino Acid Supplementation RPMI
Non-Essential Amino Acid Solution	-	0.78
Essential Amino Acid Solution	-	1.56
L-Arginine (0.5 mM)	-	0.05
Glutamic Acid (0.5 mM)	-	0.05
Glycine (0.5 mM)	-	0.05
Leucine (1 mM)	-	0.05
L-Histidine (0.5 mM)	-	1.56
L-Tryptophan (0.125 mM)	-	1.56
Iron Supplementation
+ 400 μM iron citrate	3.13	3.13

To identify the inhibitory agent(s) in rich media, we separated MHII fractions and tested each by adding them to RPMI and repeating the MIC assay (Table 1). We were unable to remove the antagonizing component in rich media by proteinase K or sodium periodate treatments which suggested that the antagonizing factor was not a polypeptide or carbohydrate species (Table 1). Interestingly, we did identify that free, hydrophobic amino acids, alone and in combination, could reduce RBT activity in nutrient-limited media (Table 1).

We also modulated iron levels in the media since RPMI is highly iron-depleted compared to MHII ^24,25. We found that increasing the iron content in RPMI antagonized RBT and increased the MIC. In contrast, lowering the iron content in MHII media increased the potency of RBT and lowered the MIC (Fig. 2A, Extended Data Fig. 2).

Fig. 2. The role of fhuE and sensitivity to RBT.

FhuE is predicted to be involved in iron transport. A) Iron was removed from MHII medium by the addition of transferrin or chelex to make the iron-depleted MHII formulation. The addition of 400 μM iron citrate to RPMI resulted in a greater MIC. All conditions (n=3 independent replicates) were tested against A. baumannii HUMC1. Statistical comparisons were determined by one-way ANOVA as compared to the MHII group. B) An overnight culture of A. baumannii HUMC1 was subcultured in fresh RPMI, with or without supplemented amino acids (AA). The intracellular concentration of RBT was measured by mass spectrometry and normalized to 1E8 CFUs (n=3 independent replicates); and C) The number of viable cells was determined by plating serial dilutions on agar plates at 0, 1, 8, and 24 hours and enumerating CFUs (n=3 independent replicates). D) A GFP expressing reporter strain of A. baumannii ATCC17978 was cultured in RPMI, with or without supplemented amino acids, and challenged with RBT. The addition of amino acids (dashed lines) resulted in less inhibition of GFP fluorescence which is consistent with the higher MICs observed in these same conditions. The data is representative of 2 experiments. E) Expression of fhuE (n=3 independent replicates) was measured by qPCR (normalized to the housekeeping gene rpoD) in various culture conditions. F) fhuE^HUMC1 was cloned from A. baumannii HUMC1 (hypersensitive strain) in pVRL2Z under the control of an arabinose inducible promoter. MICs (n=3 independent replicates) were done with and without arabinose induction in MHII media. Median and interquartile ranges are plotted for panels B, C, E, and F.

We hypothesized that inhibition of RBT’s hyper-activity was related to reduction in uptake of the antibiotic into the bacterial cell in rich but not nutrient-limited media. We therefore sought to measure the accumulation of RBT in the bacteria by LC/MS (Fig. 2B). CFU density and RNA transcription were inversely correlated with intracellular drug accumulation, consistent with the known mechanism of rifamycin antibacterial effect (inhibition of RNA transcription). We found that adding amino acids to nutrient-limited media reduced RBT entry into the bacterial cell, to levels similar to those seen in rich media, enabling a greater number of surviving bacteria and increased transcription (Fig. 2B-D).

Since free, hydrophobic amino acids were mediating the antagonism of RBT activity in RPMI, we hypothesized that RBT, and not RIF, was entering the bacterial cell through amino acid transporters. We evaluated single gene transposon-disrupted mutants of known amino acid transporters, including apoP, tryP, hisM, mtr, hisP, and hisJ, but we did not observe a difference in MIC (Extended Data Fig. 3).

Therefore, to identify the gene(s) responsible for the hypersensitivity phenotype in the RPMI media, we selected for HUMC1 spontaneous mutants that lost their hyper-sensitivity phenotype to RBT. We passaged bacteria on RPMI agar plates supplemented with 1 μg/mL RBT, and then sequenced genomes of the resulting mutants (Supplementary Data File 2). To determine the effect of the most commonly identified gene mutations, individual transposon mutants were obtained from the Acinetobacter baumannii mutant library²⁶ and RBT MICs were determined (Extended Data Fig. 4). Only fhuE disrupted mutants displayed a shift of the RBT MIC from 0.05 μg/mL of the parent strain to 3.13 μg/mL of the mutant (Extended Data Fig. 4). A second, independent mutant selection experiment also identified resistant mutants with fhuE mutations (Extended Data Fig. 5). Defined fhuE mutants were also generated to confirm the observed phenotype (Extended Data Fig. 6).

FhuE has been previously described as an iron-regulated, outer membrane protein transporter that is involved with iron acquisition in bacteria.²⁷ We found that fhuE expression was upregulated in low iron conditions (RPMI and iron-depleted MHII), and expression was downregulated in high iron conditions (RPMI + iron citrate, and MHII) (Fig. 2E). The addition of hydrophobic amino acids did not alter fhuE expression. The lack of change of expression of fhuE but inhibition of uptake of RBT mediated by amino acids even in low iron conditions (in which fhuE is expressed) indicated that amino acids worked by competitive inhibition of RBT uptake via the expressed transporter, rather than by altering its expression. Thus RBT hyperactivity requires both low iron and low free amino acid conditions.

Of 43 A. baumannii clinical isolates tested in vitro, 65% had a hyper-sensitive phenotype in nutrient-limited media, while the remainder had MICs similar between nutrient-limited and rich media (i.e., the rifabutin remained active, still exhibiting microbial killing at concentrations achievable in vivo, even for non-hypersusceptible strains). We sought to determine if differences in the amino acid sequences of the FhuE protein in A. baumannii accounted for these susceptibility differences. Using a panel of A. baumannii clinical isolates from the CDC and FDA Antibiotic Resistance Isolate Bank, we created a phylogenetic clustering based on the amino acid sequence of the FhuE protein. The outer ring was represented predominantly by isolates that were hypersensitive (MIC < 0.05 μg/mL) to RBT in RPMI. The inner ring was made up of isolates that were found to have a higher MIC (>3.13 μg/mL) to RBT in RPMI (Extended Data Fig. 7). Hence, FhuE amino acid sequences correlated with the hyper-sensitive phenotype.

To confirm that expression of the fhuE isoform from HUMC1 (fhuE^Humc1) was sufficient to reproduce the RBT hypersensitivity phenotype, we created a fhuE^Humc1 overexpression construct under the control of an inducible promoter in a strain disrupted of fhuE or its wild-type parent. We showed that overexpression of fhuE^Humc1 resulted in increased ATCC17978 sensitivity to RBT and was able to complement the ΔfhuE mutant (Fig 2F, Extended Data Fig. 8). Overexpression of fhuE^Humc1 in LAC-4, a strain that encodes a different fhuE isoform that clusters with the non-hypersensitive group, also increased sensitivity to RBT (Fig 2C). Thus, fhuE ^Humc1 expression was sufficient to recapitulate the hyper-sensitivity phenotype.

The discrepancy between RBT activity against A. baumannii in nutrient-limited RPMI vs. rich MHII media led to a critical question: which media was better predictive of in vivo efficacy of RBT? In an immunocompromised neutropenic lung infection model, we observed a significantly greater reduction in CFUs in mice infected with A. baumannii UNT091-1 (wild-type) and treated with RBT as compared to RIF (Fig 3A). Furthermore, the effect of RBT was blunted in mice infected with the UNT091-1::ΔfhuE strain (Fig 3A) supporting the role of fhuE as mediating RBT sensitivity both in vitro and in vivo.

Fig 3. Efficacy of RBT in vivo.

Neutropenic, female CD-1 mice (n= 5) were inoculated intranasally (t = 0 hr) with equivalent titers (6.90 and 6.93 log₁₀ CFU) of A) A. baumannii UNT091-1 wildtype and A. baumannii UNT091-1 ΔfhuE mutant. The treatment (single IV dose) was administered at 2-hour post-infection and bacterial burden was reported at 26 hours by determining CFU/lung. B) Immunocompetent C3H mice (n=5) were infected (oral aspiration pneumoniae model) with A. baumannii HUMC1 (carbapenem-resistant strain) and mice were treated with 10 mg/kg/daily for 3 days with RIF or RBT by oral gavage. Mice were monitored and euthanized when moribund (percent survival). At 24 hrs, blood was collected and there was a significant reduction in CFUs and improved survival in the RBT treated group as compared to the PBS control group (Log-rank, p = 0.003; Kruskal-Wallis, p = 0.02). CFUs were significantly lower in RBT vs RIF treated groups (Kruskal-wallis, p=0.01). C) The experiment was repeated with a larger group of mice (n=8) and the dose was decreased to 5 mg/kg/daily for 3 days with RIF or RBT by oral gavage. Mice were monitored and euthanized when moribund (percent survival). Mice that received the RBT treatment had significantly lower CFUs compared to the PBS group (Kruskal-Wallis, p=0.0001) and the RIF treated group (Kruskal-Wallis, p=0.02) in the lung at 24 hrs post infection and survival was also significantly improved (Log-rank, p=0.005. D) C3H mice (n=8) were infected via tail vein injection with A. baumannii HUMC1. Mice were treated with 10 mg/kg/daily RIF or RBT for 3 days by oral gavage. RBT treatment improved survival (Log-rank, p<0.0001) and significantly reduced CFUs in the blood compared to PBS (Kruskal-Wallis, p<0.0001) and RIF groups (Kruskal-Wallis, p=0.047) at 1 hrs. The dotted grey lines signify a static, 1-log, or 2-log CFU reduction and the limit of detection (LOD). E) C3H mice (n=8) were infected with HUMC1 or F) LAC-4 IV and treated with RIF or RBT by oral gavage. Mice were monitored and euthanized when moribund (percent survival). Means and standard deviations are shown for panel A. Median and interquartile ranges are shown for panels B-F. P-values were determined using a Kruskal-Wallis and Dunn’s multiple comparison test.

We next tested the RBT efficacy in a normal, immunocompetent A. baumannii pneumonia infection model. Mice were infected with A. baumannii HUMC1 and treated with 10 mg/kg/day RIF or RBT. At 24 hours, RBT-treated mice had a 7-log reduction in the median CFUs in the blood as compared to RIF-treated mice. At the 10 mg/kg dose, RBT also improved survival vs. RIF, although RIF was partially effective vs. placebo at that dose. (Fig. 3B).

We then lowered the dose of both drugs to 5 mg/kg/daily to determine if RBT enhanced potency would be unmasked. At the lower dose, RBT-treated mice had significantly reduced lung CFUs and markedly improved survival vs. RIF (Fig. 3C). No difference was observed between the RIF and PBS control group.

Next, we tested if RBT was more effective than RIF in an immunocompetent intravenous model of infection. Mice were infected IV with HUMC1 and treated with 10 mg/kg/day RIF or RBT. RBT-treated mice had significantly lower CFUs and improved survival (Fig. 3D). Lastly, we found that RBT was fully protective when dosed at 300x lower than a RIF dose that was minimally protective (Fig. 3E). Similarly, we also observed that RBT was protective in a G. mellonella infection model at significantly lower doses as compared to RIF (Extended Data Fig. 9A).

However, there was no difference in efficacy of RBT or RIF in mice or in G. mellonella infection models against A. baumannii LAC-4, a strain which has a distinct FhuE isoform and no difference in RBT/RIF MICs in nutrient-depleted media (Fig. 3F, Extended Data Fig 9B, C).

RBT would likely be used clinically in combination with a second antibiotic, a practice that is standard with the use of rifamycin antibiotics. We measured the spontaneous frequency of resistance emergence to RBT in vitro by culture A. baumannii HUMC1 overnight in MHII or RPMI and plating bacteria on agar plates supplemented with RBT alone, colistin (COL) alone, or RBT + COL. We were unable to select for any spontaneous mutants in the double selective plates (Fig 4 A, B). Additionally, we found that RBT and COL interacted synergistically in vitro (Extended Data Fig. 10). We determined the spontaneous frequency of resistance emergence to RBT in vivo by infecting mice and then harvesting blood and kidneys at 16.5 hours post infection and then plating samples on agar plates supplemented with RIF alone or RBT alone. The frequency of spontaneous RBT resistance emergence was <1.7E-9 and <3.7E-8 CFUs in vivo (Table 2). These values were about 10-fold less as compared to the frequency of spontaneous RIF resistance emergence which were 8.4E-8 and 1.2E-7 CFUs in the blood and kidneys, respectively. Furthermore, we found that the treatment combination of RBT + COL was syngergistic in vitro, even with the less sensitive ΔfhuE mutant (Extended Data Fig. 10), and that the combination of RBT + COL is more effective in vivo (Fig 4C).

Fig. 4. RBT and COL combination therapy.

Selection of antibiotic resistant mutant by high inoculum plating. A. baumannii HUMC1 (n=3 independent replicates) was cultured in A) MHII or B) RPMI media overnight and mutants were selected by plating bacteria on drug plates containing 8 μg/mL of rifabutin, 16 μg/mL of colistin or the combination of both antibiotics. The addition of colistin is able to suppress the emergence of rifabutin resistance as compared to the PBS control group on RPMI agar (Kruskal-Wallis, p=0.046). Median and interquartile ranges are plotted. C) C3H mice (n=10 per group) were infected IV with A. baumannii HUMC1 IV and treated with subtheraptic doses of PBS, colistin (0.005 mg/kg) alone, rifabutin (0.05 mg/kg) alone, or the combination of colistin + rifabutin. Mice were treated once daily for 3 days. Only the RBT + COL treatment group was significantly improved as compared to the PBS control group (Log-Rank, p=0.02).

Table 2-. Mutation Frequencies.

C3H mice were infected with 2.7 x 10⁷ CFUs of A. baumannii HUMC1. Mice did not receive RIF or RBT treatments. Blood and kidneys were collected 16.5 hrs post infection and samples were plated on MHII or RPMI agar plates alone or supplemented with 8 μg/mL RIF or RBT.

RIF and RBT Mutation Frequency
		BLOOD		Kidney
Strain	Media	RBT [8 μg/mL]	RIF [8 μg/mL]	RBT [8 μg/mL]	RIF [8 μg/mL]
A. baumannii HUMC1	MHII	< 1.7E-9	8.4E-8	<3.7E-8	1.2E-7
A. baumannii HUMC1	RPMI	<8.3E-9	4.1E-7	3.7E-7	1.6E-6

Thus, these collective experiments confirmed that RBT has an excellent activity in vivo and is more potent when compared to RIF, and that in vitro MIC results in nutrient-limited media better predicted in vivo efficacy than in vitro MIC results in rich media. The superiority of RBT over RIF is highly significant due to a recent, randomized, controlled clinical trial that found a non-significant trend to improved clinical cure, with a significant improvement in microbiological eradication when adding rifampin to colistin for treatment of XDR A. baumannii infections ²⁸. Given that RIF had some effect clinically, and that RBT appears far more potent and effective in pre-clinical models, we believe that the addition of adjunct RBT, as compared to RIF, could serve as a critically needed therapeutic option for patients with infections caused by XDR A. baumannii, potentially resulting in superior survival compared to the current standard of care. Thus, studies with RBT are underway to evaluate clinical utility of this promising drug for the treatment of A. baumannii infections where there are currently limited treatment options.

Finally, high throughput screens of small molecules have been a standard practice for the pharmaceutical industry for decades. These libraries, which may contain millions of compounds, have been screened many times; however, new screens of the same library rarely identify novel compounds ^7,29. Our results demonstrate that using more physiologically relevant, nutrient-limited media has the potential to identify many additional treatments for antibiotic-resistant bacterial infections which have been missed to date due to screens of these large libraries based on rich culture media ^7,30,31. Thus, repeat screening of these large libraries in nutrient-limited media may offer a promising, critically-needed approach to identifying antimicrobial agents with which to combat the crisis of antibiotic-resistant infections.

Materials and Methods

Ethics statement

All animal work was conducted following approval by the Institutional Animal Care and Use Committee (IACUC protocol 20882) at the University of Southern California, in compliance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Infected mice develop weight loss, ruffled fur, poor appetite, decreased ambulation, huddling behavior, and low body temperature. Mice that display huddling behavior and are poorly mobile will be weighed 1x daily. Weight loss of greater than 15% body weight will trigger euthanasia. Mice were monitored at least twice daily for seven days. Soft bedding and other enrichment devices were provided as recommended by the vet staff. Nutritional supplements, such as the hydrogel packs were provided as needed.

Source of bacterial isolates

The complete list of bacterial isolates tested is listed in Data File 1. Strains were obtained from the CDC Antimicrobial Resistance Isolate Bank, the She lab, the Spellberg lab, and BioVersys.

Bacteria Culture

Working solutions of bacteria were prepared using frozen stocks of Acinetobacter baumannii strains as previously published,¹⁶ or by inoculating a fresh overnight culture in Tryptic Soy Broth (TSB) and incubating at 37°C/200 rpm. The overnight culture was diluted 1:100 and then subcultured in MHII at 37°C/200 rpm until the culture reached an OD₆₀₀ of 0.5. For some assays, bacteria were also subcultured in Roswell Park Memorial Institute (RPMI) 1640 Medium (Gibco, 11875119).

High Throughput Compound Screening

Compounds from the 10 mM ReFRAME library stock ¹⁴ were acoustically transferred using the Echo 555 Liquid Handler (Labcyte Inc.) at 20 μM final assay concentration into 1536-well plates. Log-phase growth A. baumannii HUMC1 was diluted in assay media (MHII or RPMI with and without 50% normal mouse serum) and dispensed into assay plates using the MultiFlo FX Multi-Mode Dispenser (BioTek). Bacterial viability was assessed 24 hours later using the BacTiter-Glo Microbial Cell Viability Assay (Promega) and the PHERAstar microplate reader (BMG Labtech). Positive controls included 10 μM colistin and doxycycline. Assay was normalized to neutral and positive controls and putative hits were selected based on 50% reduction in viability. Putative hits were re-tested in single point triplicate and upon reconfirmation were further tested in an 8-point 1:3 dose response and counter-screened against mammalian HepG2 and HEK293T cell lines using a 72-hour CellTiter-Glo Luminescent Cell Viability Assay (Promega). For the mammalian cytotoxicity counter screen 40 μM puromycin (Sigma) was used as the positive control and data was normalized as for the primary assay. All data were uploaded to and analyzed in Genedata Screener, Version 13.0.1-Standard. Replicate data were analyzed using median condensing. Dose response curves were fitted with the four parameter Hill Equation.

Antibiotic Preparation

RIF (Sigma, R3501-1G) and RBT (Sigma, R3530-25MG) were dissolved in Dimethyl sulfoxide (DMSO). The working solution of antibiotic was prepared 2X of the desired starting well final drug concentration. The antibiotic working solution dilutions were prepared in the respective media used for the MIC, MHII or RPMI.

MIC Protocol

Unless otherwise indicated, the standard broth microdilution method was used to determine MICs.⁵ The medium used for the minimum inhibitory concentration (MIC) assays performed in this study was either MHII or RPMI 1640.

Briefly, 100 μL of media, RPMI or MHII, was added to the wells in columns 2-10. Column 11 served as a positive growth control and contained only bacteria and media. Column 12 served as the negative control and contained only culture media without bacteria. Next, 200 μl of a 2X rifabutin or rifampin working solution was added to the wells in column 1. Two-fold serial dilutions of the antibiotic were performed through column 10. Next, 100 μl of a 1×10⁶ CFU/mL working solution of bacteria was added to each of the wells in columns 1-11. The inoculum concentration was confirmed by plating serial dilutions on TSA plates. MICs plates were incubated at 35±2 °C without shaking and results were recorded at 24 hours.

To test the effect of the individual components of MHII, MHII fractions (size separation, acetonitrile extracted, proteinase K digested, or sodium periodate oxidized), 10 μL of the purified fraction was added to the appropriate wells.

As indicated, amino acids were used to supplement the media for MIC testing. Mixed amino acids were tested by adding 10 μL of Gibco® MEM Amino Acids Solution (Thermo Scientific, #11130051) and Gibco® MEM Non-Essential Amino Acids (Thermo Scientific, #11140050) to each well in the MIC assay. Additionally, the effect of individual amino acids on the MIC was tested by the addition of purified amino acids at the same concentration contained in the Gibco mixed amino acid solutions listed above. Amino acids were prepared fresh and filter sterilized solutions prior to use.

For some MICs, efflux pump inhibitors were added to the media. Efflux pump inhibitor MICs were performed to identify subinhibitory concentrations of the efflux pump inhibitors. The final concentration of inhibitors used were as follows: verapamil (50 μg/mL), thioridazine (8 μg/mL), and CCCP (2 μg/mL).

For MICs involving the fhuE^Humc1 overexpression strains, the respective stains were cultured overnight and subcultured in low salt LB supplemented with 250 μg/mL zeocin at 37 °C/200 rpm to maintain the overexpression plasmid. MIC assays were done in MHII or RPMI media supplemented with 20 μg/ml of zeocin to maintain the plasmid. Expression of fhuE was induced by the addition of arabinose at the final concentrations of 8%, 4%, 1%, or no arabinose as a control.

Fractional inhibitory concentration index(FICI)

Drug -drug interaction between rifabutin or rifampin and colistin in MHII The drug-drug interaction were evaluated by calculating the fractional inhibitory concentration index(FICI). FICI=FIC_A+ FIC_B=(C_A/MIC_A)+(C_B/MIC_B), in which C_A and C_B are drug concentration of drug A and drug B in combination and MIC_A and MIC_B are the MIC of drug A and drug B alone. Synergy was defined as FICI≤0.5, no interaction was defined as FICI >0.5–4.0 and antagonism was defined as FICI >4.0.

Fractionation of MHII:

Size fractionation

10X MHII was used for the fractionation to maximize the concentration of the MHII components. The media was filtered through a 0.22 μM filter and then the media was run through 10 and 30 kDa molecular weight cut-off (MWCO) centrifugal filtration columns at 12,000 g for 20 minutes. The >30 kDa fraction was collected and reserved for experimentation. The flow through was collected and transferred to a 10 kDa MWCO column. The centrifugation step was repeated as previously stated. The <10 kDa flow through was collected from this column. The 10<X<30 kDa fraction was collected as well.

Acetonitrile Extraction

The organic and inorganic layers were separated with a liquid-liquid extraction by mixing the purified MHII fraction 1:1 with 100% acetonitrile. The sample was vortexed thoroughly and centrifuged at maximum speed for 10 minutes. The aqueous and organic layers were transferred to clean microcentrifuge tubes. To ensure the removal of any residual acetonitrile, the extracted MHII sample was dried using a SpeedVac and then resuspended in the original volume using sterile molecular grade H₂O.

Proteinase K Digestion

30 μl of proteinase K (Invitrogen, #46-7603) was added to 1 mL of the MHII <10 kDa fraction (organic extract or non-extracted as a control). The media was incubated at 65°C for 1 hour. To inactivate the proteinase K, the sample was then incubated at 80°C for 15 minutes.

Sodium Periodate Oxidation

Sodium periodate oxidation was done as previously described.³² Briefly, to oxidize the carbohydrates in the organic layer of the medium, sodium periodate (NaIO4) was added to the <10 fraction at a final concentration of 10 mM. The sample was incubated at room temperature for 30 minutes. Following incubation, the sodium periodate was quenched using 0.1 mL of 50% glycerol for every 1 mL of reaction. The sample was incubated at room temperature for 1 hour before downstream application.

Time Kill Assay

Time kill assays were performed in a 96-well U-bottom plate using the same plate set up as the MIC assay. We measured viable cell counts at 1, 8, and 24 hours. At each time point, the contents of an individual well were collected and CFUs were determined by plating serial dilutions on TSA plates and incubating overnight at 37°C.

As a control, the remaining wells of the 96-well plate were returned to the incubator after each sample collection and an MIC was determined as described above.

Mass Spectrophotometry

Extractions and measurements were done as previously described ^33,34. Log-phase A. baumannii culture was incubated 0.79 or 0.38 μg/mL RBT in the presence or absence of amino acid mixture at 37 °C. Bacteria were harvested 0, 1, 8, and 24 hours and CFUs were determined by plating serial dilutions on agar media. The cell free supernatant was collected by filtration through a 0.22 μm filter. RBT were extracted by adding LC-MS grade acetonitrile:methanol:water (40:40:20) solution that was precooled to − 40 °C. Liquid chromatography mass spectrometry (LC-MS) differentiation and detection of RBT was performed using a Cogent Diamond Hydride Type C column (Microsolve Technologies) with an Agilent Accurate Mass 6230 TOF coupled with an Agilent 1290 Liquid Chromatography system as previously published.^33,35 An isocratic pump was used for continuous infusion of a reference mass solution to allow mass axis calibration. Detected RBT ion was validated based on unique accurate mass-retention time identifiers for masses. RBT level was analyzed using Agilent Qualitative Analysis B.08.00 (Agilent Technologies) with a mass tolerance of <0.005 Da. The intracellular RBT was calculated as the [RBT]_{drug only control} – [RBT]_filtrate. 3 biological replicates were tested per group.

Selection of spontaneous RBT-resistant mutants and WGS:

The RBT hypersensitive strain HUMC1 was cultured overnight in 50 mL of RPMI. Bacteria were plated on RPMI agar plates supplemented with 1μg/mL of RBT. Individual colonies were selected and then counter screened in RPMI agar plates supplemented with 25 μg/mL RBT.

Colonies that grew on the RPMI agar plates supplemented with 1 μg/mL, but not the 25 μg/mL RBT, were processed for whole genome sequencing. Sequencing libraries for the selected mutants and parent strain were prepared at the USC Molecular Genomics Core. Libraries were simultaneously prepared from extracted genomic DNA using the Illumina Nextera XT library prep kit according to the manufacturer’s protocol (Cat#FC-131-1024, Illumina). Prepared libraries were sequenced on the Illumina Miseq V2 at 2x150 cycles. Assembled reads were aligned to the previously published A. baumannii HUMC1 sequence (NCBI Reference Sequence: NZ_LQRQ01000007.1). The FASTQ files were uploaded to PartekFlow® software, version 6.0 (Partek, Inc., St. Louis, MO) on a Linux based High Performance Computing system at Pennsylvania State University College of Medicine, adapter-trimmed, and remapped to HUMC1 reference sequence(NZ_LQRQ01000007.1) using alignment algorithm (BWA-MEM) with a few modifications (mismatch penalty 4, gap open penalty 6, clipping penalty 5, and alignment score cutoff 30) for short read mapping. After alignment, annotated variants were saved.

FhuE phylogenetic tree

Maximum-Likelihood phylogenetic tree of 43 fhuE aminoacid sequences, constructed using the CLC Genomics Workbench 12 (Qiagen) software. Nucleotide sequences were generated by assembling SRA reads of the 41 clinical isolates from the CDC and FDA Antibiotic Resistance Isolate Bank and aligning to the HUMC1 or LAC4 fhuE template. Numbers at the base of nodes correspond to bootstrap (10.000) probability >90%. The scale indicates the number of amino acid substitutions per site.

fhuE^Humc1 overexpression construct

The open reading frame of the fhuE gene was cloned into the pVRL2Z plasmid under the control of the arabinose indcubible promoter. HUMC1 genomic DNA and pVRL2Z plasmid DNA were extracted using the Quick-DNA Fungal/Bacterial Kits (Zymo, #D6005) and the GeneJET Plasmid Miniprep Kit (Thermo Fisher, #K0502), respectively. The open reading frame of the fhuE gene from HUMC1 was cloned by PCR using oligonucleotide primers (5’-taa gca GAA TTC GCA TCG AGG GTT GCAT TTC C-3’ / 5’-taa gca GCG GCC GCT CTA CTT CAC CCT TGC GGC T-3’) and purified using PureLink® PCR Purification Kit (Thermo Fisher Scientific, #K310001). The Q5® High-Fidelity 2X Master Mix (NEB, M0492S) was used for all cloning related PCRs. The vector and insert were both digested with NotI/EcoRI (NEB, #R0189S/NEB, #R0101S), digested products were purified, and the vector/insert were ligated using T4 ligase (NEB, #M0202S).

Transformation of A. baumannii AB5674, ATCC 17978, and LAC-4 was performed by electroporation using freshly prepared competent cells. Transformants were selected on low salt LB agar supplemented with 250 μg/ml of zeocin. Transformants were confirmed by reisolating the plasmid by miniprep, digestion of the plasmid DNA using NotI/EcoRI, and visualizing the digested products on an agarose gel.

Electrocompetent cells were prepared as by subculturing bacteria to log-phase growth in TSB at 37°C/200 rpm/2.5 hrs. Bacteria were collected by centrifugation and the cell pellet was washed in 1 mL of ice-cold 10% glycerol by pipetting up and down. Cells were collected by centrifugation at 16,000 rcf/5 min/4°C. Bacteria were washed 5X and resuspended in 300 μl of 10% glycerol. 100 μL of electrocompetent cells were used for each transformation.

RNA extraction

Bacteria were cultured overnight in different media conditions (Fig. 2B) at 37 °C/200 rpm. Bacteria were subcultured to log-phase at 37°C/200 rpm/3 hrs and RNA was extracted with the RiboPure™ RNA Purification Kit per manufacturer protocol (Thermo Fisher Scientific, #AM1925).

Quantitative fhuE gene expression

cDNA was synthesized by using LunaScript™ RT SuperMix Kit (NEB, #E3010L) after normalizing RNA template. Targets were amplified as using the Luna® Universal qPCR Master Mix (NEB, #M3003S) per manufacturer protocol and the following primers. HUMC1 fhuE: 5’-GGC GTT AGA GAA ACA CCG GA-3’ / 5’-GGC GCA CTG GGA TAA TGA GA-3’; rpoD: 5’-GCG TGA AAT GGG TAC AGT AGA A-3’/ 5’-TTG GCC AGT AAG CGA TTG AG-3’; LAC4 fhuE: 5’- CCC ATT GCG AAC CAT TAG CG-3’ /5’-ACA CAA GAT GGG CGA GTA CG-3’. Expression of fhuE was normalized to the housekeeping gene rpoD using the ΔΔCT method.

fhuE gene deletion

fhuE was knocked out using a two-step recombination method previously described ³⁶. DNA regions (700 bp) flanking fhuE were PCR amplified using the following primers. fhuE-up: 5’-AGA ATT GAG GCC TCT CGA GGA ATT CTT GAT GCC ATG TAC TCG C-3’ / 5’-CCA TAA AAA GGT TGG CAT AAC AAA TCG C-3’; fhuE-down: 5’-TTA TGC CAA CCT TTT TAT GGT GCC CCA G-3’ / 5’-CCT GCA GGC TCT AGA CAT CGA GGG TTG CAT TTC-3’. The fhuE up- and downstream fragment were cloned in the knockout platform pVT77 previously digested with EcoRI/XbaI using NEBuilder HiFi DNA assembly (NEB). The cloned fhuE knockout plasmid was conjugated in A. baumannii and genomic plasmid integration was selected on LB agar plates containing 100 μg/ml sodium tellurite. Obtained clones were transferred on LB agar plates containing 1 mM isopropyl-β-D-1-thiogalactopyranoside and 200 μg/ml 3’-azido-3’-deoxythymidine to select for plasmid removal from the genome. Scarless deletion of fhuE was screened with primers 5’-TAC TTC ACC CTT GCG GCT AC-3’ / 5’-AAA GAT ACG ACC AAC CGC CC-3’ and confirmed by DNA sequencing (Microsynth AG, Balgach, Switzerland).

GFP expression

Transformation of A. baumannii ATCC 17978 was performed by electroporation using freshly prepared competent cells. Transformants were selected on TSA agar supplemented with 10 μg/ml of gentamicin (Sigma, #G1272). Transformants were confirmed by GFP expression using a fluorescent plate reader.

Transformants were then subcultured in RPMI supplemented with 10 μg/ml of gentamicin with 4% arabinose. 1×10⁵ CFU bacteria and the appropriate concentration of RBT were added to RPMI supplemented with 10 μg/ml gentamicin and 4% arabinose, up to the total volume of 200μl. 10 μl of AA was added in appropriate samples as well. GFP expression was measured using flow cytometry analysis every 30mins.

G. mellonella infection model

Galleria mellonella larvae (10 per group) were infected with logarithmic phase growing (OD₆₀₀ = 0.5) bacteria resuspended in PBS to reach the specified cfu/larvae with a 10 μl injection in the right second proleg. The infected larvae were subsequently treated with rifabutin or rifampicin 1 hour after infection with a 10 ul injection of the specified antibiotic dose in the left second proleg. The injected larvae were incubated at 37 °C in 90-mm plastic Petri dishes and monitored for their survival for 72 h. Larvae were considered dead when they did not move in response to stimulus with a pipette tip.

In Vitro Mutant Selection

Spontaneous rifabutin and colistin resistant mutants were selected by high inoculum plating on selective agar plates supplemented with 8 μg/mL of rifabutin, 16 μg/mL of colistin or the combination of both antibiotics. Overnight cultures of A. baumannii HUMC1 were grown in 20 ml MHII or RPMI at 37°C / 200 rpm. For bacteria cultured in MHII, 100 μL of culture was plated directly on selective plates to enumerate the antibiotic-resistant population and serial dilutions were plated on non-selective plates to enumerate the total population. Because the bacteria grow to a lower density in RPMI, the bacteria were first concentrated by 10x by centrifugation prior to plating on non-selective and selective agar plates as described above. All experiments were done in triplicate.

Mouse Studies

Mice were randomly assigned to treatment groups and treatment groups were not blinded.

Intravenous (IV) infection

A. baumannii HUMC1 frozen stock was prepared as described in previous work.¹⁶ Frozen stocks of HUMC1 were thawed and diluted in PBS to adjust the bacterial density as needed for infection. Male C3HeB/FeJ mice, 8 to 12 weeks old, were infected with 2x10⁷ CFUs via tail vein injection and the inoculum bacterial density was confirmed by plating serial dilutions on TSA plates and incubating overnight at 37°C.

Oral aspiration (OA) infection

Single colonies of A. baumannii HUMC1 grown on TSA were used to inoculated TSB and bacteria were cultured overnight at 37°C / 200 rpm. The following day, the bacteria was subcultured by diluting the overnight 1:100 in fresh TSB and cultured for 3hrs at 37°C / 200rpm. The subculture was washed with PBS three times and adjust to optical density (OD₆₀₀) equal to 0.5. The inoculum was concentrated to 2x10⁹ CFUs/ml and 9 to 10 weeks old male C3HeB/FeJ mouse is infected with 50 μL (1x10⁸ CFUs) of inoculum via oral aspiration. The inoculum CFUs was confirmed by plating on TSA plates and incubating overnight at 37°C.

Neutropenic pneumonia infection

Female CD-1 mice were made neutropenic by the administration of 150 & 100 mg/kg of cyclophosphamide, intraperitoneally (IP), at 4 and 1 day prior to infection. Frozen stock of A. baumannii BioVersys UNT091-1 and UNT091-1::ΔfhuE were plated on TSA and incubated overnight (ON) at 37°C. The plate growth was suspended into TSB to 1.0e+09 CFU/mL (OD 1 @ 600 nm), and then 10-fold serial diluted in fresh TSB and this log10 dilution was used as the infecting inoculum. The infecting inoculum was approx. 6-7 log₁₀ CFU/animal. Additionally, the OD adjusted plate suspension was 10-fold serially diluted and spot plated onto BHI/charcoal plates to determine input CFU.

Mice were anesthetized by IP injection 0.2 ml of a Ketamine HCL (40 mg/kg b.w.) + Xylazine (6 mg/kg b.w.) mixture. Anesthetized mice were intranasally (IN) inoculated with 0.05 ml of the designated inoculum. For IN inoculation, drops were placed onto the eternal nares and waited for inhalation. After inoculation, each mouse was placed back into their cage with heat pads for recovery. Plate counts were performed to confirm the exact CFU input for each strain (targeting 6.5 log₁₀ CFU/mouse).

Mice were euthanized by CO₂ inhalation at 26 hrs post-infection, lungs aseptically removed, placed in cold PBS, homogenized, serially diluted and spot plated on BHI + charcoal for the determination of bacterial lung titers (log₁₀ CFU/lung). Colony counts were performed on the agar plates. The number of colonies was converted to CFU/lung by multiplying the number of colonies by the volume of the lung homogenate spotted and the dilution at which the colonies were counted (5-50 colonies/spot). All count data were transformed into log₁₀ CFU/lung for calculation of means and standard deviations. The limit of detection was Log₁₀ = 2.35

Mice were inoculated intranasally with equivalent titers (6.75 and 6.90 log₁₀ CFU) of each of the two A. baumannii strains and the treatment (single IV dose) with rifabutin and rifampicin was administered at 2-hour post-infection.

Antibiotic treatments

RIF (Sigma, R3501-1G) and RBT (Sigma, R3530-25MG) were dissolved in Dimethyl sulfoxide (DMSO). The working solution of antibiotics are prepared fresh daily. The appropriate concentration of antibiotic working solution was prepared in PBS with 10% DMSO and administered by oral gavage. The control mice received the same volume of PBS with 10% DMSO without drug. RIF, RBT, and the control were administered once a day for three days starting the day of infection.

Blood CFUs

50 - 100 μL of blood was collected by tail nick at the indicated time points post OA or IV infection. Blood samples were serially diluted in PBS and plated on TSA plates. Agar plates were incubated overnight at 37°C and CFUs were counted the next day.

Lung CFUs

At 18 hrs post infection, lungs were harvested, weighed, and homogenized in sterile PBS. Lung homogenates were serially diluted in PBS and plated on TSA plates. The plates are incubated overnight at 37°C and CFUs were counted the next day.

Mutant Selection

Male C3HeB/FeJ mice were infected with 2.7 x 10⁷ CFUs of A. baumannii HUMC1. Mice did not receive RIF or RBT treatments. Blood and kidneys were collected 16.5 hrs post infection and samples were plated on TSA or RPMI agar plates alone or supplemented with 8 μg/mL RIF or RBT.

Statistics

Bacterial burden was compared using the Mann Whitney test. Time to death was compared using the Log Rank test. P values < 0.05 were considered significant.

Data Availability

Screening data are available on ReFRAMEdb.org. Genome sequencing data are available at NCBI Accession PRJNA629056. Source data for the figures are provided with the paper.

Extended Data

Extended Data Fig. 1. MIC of rifabutin in RPMI + 10% FCS on non-A. baumannii Gram-negative ESKAPE species.

Extended Data Fig. 2. MIC of rifabutin in RPMI + 10% FCS supplemented with increasing amount of ammonium iron(III) citrate.

Extended Data Fig. 3-. Rifabutin MICs against A. baumannii AB5075 transposon disruption mutants.

Mutants were deficient in amino acid transport genes. MIC assay was done in both MHII and RPMI media. AB5075-UW is the parent strain for the transposon mutants.

Extended Data Fig. 4 -. Rifabutin MICs against A. baumannii AB5075 transposon disruption mutants.

MIC assay was done by culturing the bacteria in either MHII or RPMI media.

Extended Data Fig. 5. MIC of rifabutin and rifampicin on HUMC1 and the serial passaged mutants, and WGS of mutants with increased rifabutin MIC.

Extended Data Fig 6. MIC of rifabutin and rifampicin on the fhuE deleted mutants and their parental strains.

Extended Data Fig. 7. Clustering of fhuE by amino acid sequence.

MICs were done for 43 A. baumannii clinical isolates to determine if the isolates were hypersensitive to RBT in RPMI medium. 28 of the 33 isolates in the outer rings exhibit the hypersensitive phenotype (open circles) in RPMI (RBT MIC = 0.05 μg/mL). 9 of 10 isolates in the inner ring are not hypersensitive (filled-in shapes). Of the 41 carbapenem-resistant isolates tested, 26 of those are hypersensitive to RBT.

Extended Data Fig 8. MIC of rifabutin on the plasmid mediated fhuE expressing ATCC-17978 strains.

Extended Data Fig. 9. RPMI MIC predicts in vivo response to treatment.

Galleria mellonella larvae (10 per group) were infected with A. baumannii. A) Larvae were infected with strain HUMC1 at 1.6 x 10⁴ cfu/larvae and treated with rifabutin (plain lines) or rifampicin (dashed lines) at 0.1 mg/kg (blue lines), 1 mg/kg (red lines) and 10 mg/kg (green lines) and survival was measured over 72 hours. B) G. mellonella larvae were infected with A. baumannii LAC-4 and treated with RBT or C) RIF. Consistent with the RPMI MIC data, there was no difference in outcomes based on treatment.

Extended Data Fig 10-. Drug -drug interaction between rifabutin or rifampin and colistin in MHII The drug-drug interaction were evaluated by calculating the fractional inhibitory concentration index (FICI).

FICI=FIC_A+ FIC_B=(C_A/MIC_A)+(C_B/MIC_B), in which C_A and C_B are drug concentration of drug A and drugB in combination and MIC_A and MIC_B are the MIC of drug A and drug B alone. Synergy was defined as FICI≤0.5, no interaction was defined as FICI >0.5–4.0 and antagonism was defined as FICI >4.0.