Adaptation-coupled CRISPRi identifies antibiotic-sensitizing genes leading to a novel plasmid selection marker

Abstract

Antibiotic resistance through horizontal gene transfer is a major challenge in microbiology and medicine. We describe a CRISPR-based screening system that couples spacer adaptation with gene repression to identify bacterial genes that sensitize bacteria to antibiotics. We used this screening system in Escherichia coli, on seven different antibiotics to select spacers that conferred resistance to them. The screen identified both known and previously unrecognized genes that modulate antibiotic susceptibility. We defined this collection of genes that sensitize bacteria to antibiotics as the ‘sensitasome’. Among the identified targets were the essential genes gyrA and gyrB, whose repression increased resistance to multiple antibiotics. Some sensitization genes exhibited a clear trade-off: their repression conferred resistance to certain antibiotics while increasing susceptibility to others. Sensitasome-targeting spacers were also effective in Shigella and Salmonella, indicating that the resistance mechanism is conserved across species. Finally, we repurposed a selected spacer as a CRISPR-based antibiotic selection marker using the compact Cas12m effector. Together, these findings establish adaptation-coupled CRISPRi as a simple platform for discovering antibiotic sensitization genes and provide an environmentally safer alternative to conventional antibiotic resistance markers.

Graphical Abstract

Graphical Abstract.

Introduction

Antibiotic resistance remains one of the most pressing challenges in global health [1]. As resistance to last-line antibiotics continues to rise, there is growing interest in strategies that sensitize bacteria to existing antimicrobials, thus restoring or enhancing drug efficacy through genetic or chemical perturbations [2, 3]. Identifying genes whose repression increases antibiotic resistance (and vice versa) offers a promising avenue toward this goal, with potential applications in biotechnology [4].

Girgis and colleagues demonstrated that knocking out certain genes increases Escherichia coli’s intrinsic antibiotic tolerance, underpinning the idea that bacteria have a built-in set of sensitivity factors [5]. Another study has shown that CRISPR-based interference (CRISPRi) can be coupled with CRISPR adaptation to enable genome-wide functional screening without the need for pre-designed libraries [6]. This study used an adaptation-linked CRISPRi system in Staphylococcus aureus to identify sensitization genes for gentamicin, relying on plasmid-based adaptation to feed new spacers into CRISPR arrays. However, that prior platform was limited to a single antibiotic and required complex steps (such as electroporating pre-designed spacers), which could hinder its scalability and accessibility.

Here, we describe a streamlined genome-wide screening platform in E. coli that leverages spontaneous CRISPR adaptation from genomic and plasmid DNA [7], coupled with CRISPR interference (CRISPRi) [8], to enrich for spacers that confer resistance by silencing endogenous genes. This system allows bacteria to acquire multiple spacers against each gene (including essential genes) through spontaneous adaptation, and it relies on straightforward phenotypic selection, eliminating the need for any pre-made guide library. Using this platform across multiple antibiotics, we uncovered a broad network of antibiotic-sensitizing genes, which we term the ‘sensitasome’.

Our results reveal both known and novel contributors to antibiotic susceptibility, including genes involved in electron transport and essential cell functions. We identify gyrA and gyrB as sensitization targets, where their CRISPRi-mediated repression increased resistance to several antibiotics. Several sensitization genes displayed a resistance trade-off, where their repression conferred resistance to several antibiotics but also increased susceptibility to others. We further show that three out of three tested sensitization spacers are conserved across Salmonella and Shigella in conferring antibiotic resistance. We finally show that these spacers can be repurposed as nondegradative, CRISPR-based plasmid selection markers. These markers may reduce the use of antibiotic-degrading enzymes like beta-lactamases or aminoglycoside-modifying proteins, which contribute to environmental dissemination of resistance [9, 10]. Together, this work establishes a scalable, selection-driven approach to uncover sensitization networks and provides new tools for antimicrobial research and genetic engineering.

Materials and methods

Reagents, strains, oligonucleotides, and plasmids

Luria–Bertani (LB) medium (20 g/l tryptone, 15 g/l yeast extract, 20 g/l NaCl) and 2× YT medium (soya peptone 16 g/l, yeast extract 10g/l, NaCl 5 g/l) were from Formedium and agar was from Oxoid. Kanamycin, streptomycin, and ampicillin were obtained from Calbiochem, nalidixic acid, gentamicin sulphate and trimethoprim were from Sigma Merck, carbenicillin was from Novagen, hygromycin B was from InvivoGen, ciprofloxacin 98% was from Thermo Scientific – ACROS, polymyxin B sulphate and azithromycin dihydrate 99% were from Angene. L-arabinose was from GoldBio and L-rhamnose monohydrate was from Apollo Scientific. M-toluic acid was from Sigma MERCK. Restriction enzymes, ligation enzymes, DNA modification enzymes were from New England Biolabs. Phanta Flash Master High-Fidelity and Rapid polymerase chain reaction (PCR) Master Mix (Cheetah) were from BioGate. Isopropyl-β-D-thiogalactoside (IPTG) and 5-bromo-4-chloro-3-indolyl β-D-galactopyranoside (X-gal) were from Bio-lab. The strains, plasmids and oligonucleotides used in this study are listed in Supplementary Table S1.

Plasmid and strain construction

Supplementary Table S2 lists the entire set of plasmids and strains constructed in this study using the indicated primers and templates. Plasmids and strains were constructed using standard molecular biology techniques. DNA segments were amplified by PCR. Standard digestion of the PCR products and vector by restriction enzymes was done according to the manufacturer’s instructions.

Preparation of a genome-wide library using CRISPR adaptation

Genome-wide CRISPRi libraries were generated in E. coli K-12 based on findings in our earlier study [11]. We used strains derived from E. coli K-12 (RG8610), lacking the cas3 gene. The strain encodes T7 RNA polymerase under an arabinose-induced promoter. We introduced plasmids pIY347 encoding the Cascade and CRISPR array consisting of a leader and a single repeat under rhamnose-inducible promoter, and pIY325 encoding Cas1^V76L variant with enhanced adaptation activity [12] and Cas2 under an arabinose-inducible promoter. The control strain harbored the same plasmids except that the dCas1^D221A variant [lacking adaptation activity [7, 13] replaced the Cas1^V76L. The cultures were grown overnight in LB supplemented with 50 μg/ml ampicillin and 50 μg/ml streptomycin at 37°C, respectively. Thereafter, two cycles of adaptation were carried out. The overnight cultures were diluted 1:30 in 10 ml of 2× YT supplemented with 50 μg/ml streptomycin, 50 μg/ml ampicillin and 0.2% (w/v) L(+)-arabinose. The cultures were incubated at 37°C overnight for spacer acquisition. The adaptation efficiency was then determined by PCR analysis using primers MG7R and TM137F primers resolved on a 2% (w/v) agarose gel. The second cycle of adaptation was carried out by diluting the overnight cultures from the first cycle of adaptation 1:30 in 10 ml of 2× YT supplemented with 50 μg/ml streptomycin, 50 μg/ml ampicillin and 0.2% (w/v) L(+)-arabinose. These cultures were incubated at 37°C overnight for spacer acquisition. The adaptation efficiency was then determined by PCR analysis using primers MG7R and TM137F primers resolved on a 2% (w/v) agarose gel.

Spacer libraries for screening repressors of lacZ and cas3 were generated following the procedure described above with minor modifications. RG8610-derived strains were engineered to carry a kanamycin resistance cassette either replacing lacZ (strain MG5538) or downstream of the cas3 promoter (strain MG5555). Plasmid pIY327, encoding Cascade and the CRISPR repeat under an L-arabinose-inducible promoter, was introduced together with plasmid pIY326, encoding the Cas1^V76L variant and Cas2 under an L-rhamnose-inducible promoter. Plasmid pIY329, harboring the dCas1^D221A, served as a control.

Preparation of adapted library PCR product for deep sequencing

The DNA from the second cycle of the adapted library was amplified using 2× Phanta Flash master mix. In this PCR reaction, 25 μl of 2 × Master Mix master mix, 2 μl of 10 μM forward and reverse primers with adapters (Supplementary Table S2), 2 μl of bacterial lysate, and 19 μl of double-distilled water were used. The PCR started with 30 s at 98°C followed by 35 cycles of 10 s at 98°C, 5 s at 55°C and 5 s at 72°C. The final extension step at 72°C was performed for 1 min. Then, 50 ul of the amplified PCR reaction was loaded on a 2% (w/v) agarose gel and electrophoresed for 35 min at 120 V. Following gel separation, the expanded band of ∼240 bp was excised from the gel and purified using the FavorPrep™ Gel/PCR Purification Mini Kit. The PCR product described above was used for the preparation of the library and was sequenced using (150PE) MGI- the DNBSEQ-G400 machines according to the manufacturer’s instructions. The reads were then mapped on the reference sequence of the E. coli strain K-12 genome (NC_000913.2) using the Bowtie (version 1.0) [14]. For the genomic mapping of unique protospacers and PAM distribution, identical protospacer sequences were retained only once. The protospacers were assigned to 10-kb genomic bins according to the midpoint of their mapped coordinates, and density was expressed as the number of unique protospacer sequences per kilobase. The PAM density was determined by counting all overlapping AAG motifs and their reverse-complement CTT motifs within each genomic bin.

Minimum inhibitory concentration determination

The adapted and control libraries were serially diluted tenfold, and 7.5 μl aliquots of each dilution were spotted on 1.5% (w/v) LB agar plates supplemented with antibiotics at various concentrations. The minimum inhibitory concentration (MIC) for each antibiotic was calculated and is reported in Supplementary Table S3. The MIC for the library was determined at the concentration that suppresses the control but still allows the adapted library to grow. This calibrated MIC was used for the selection and enrichment of spacers for each antibiotic.

The MICs for the strains carrying plasmids with dominant spacers and an empty vector were determined on LB agar plates overlaid with soft agar [0.7% (w/v)], using the ETEST^® method (Biomeriueux) and MIC Test Strip (Liofilchem^® MTS™), following the manufacturer’s instructions. 12 different spacers cloned in pIY331 were transformed to RG8610 with pAH071. The control was an empty pIY331 with the pAH071 plasmid in RG8610. These strains were grown in LB at 37°C for 16 h, supplemented with ampicillin and streptomycin. Each bacterial culture was then diluted 50-fold in 2× YT supplemented with 0.2% arabinose (w/v), ampicillin and streptomycin, and the diluted culture that reached an OD₆₀₀ of ∼0.6 within 3–4 h of growth at 37°C was used for subsequent steps. Volumes of the cultures corresponding to OD600 units of 1.0 were mixed with 7 ml of molten soft LB agar and overlaid onto LB agar plates supplemented with 0.2% arabinose (w/v), ampicillin and streptomycin. The plates were allowed to dry for 1.5 h before placing ETEST strips on the soft agar overlay [412 311 ETEST^® Ciprofloxacin (CI), 412 257 ETEST^® Azithromycin (AZ), 412 368 ETEST^® Gentamicin (GM), 412 450 ETEST^® Rifampicin (RI) from Biomeriueux and LIO-92034 Kanamycin (K), LIO-92132 Nalidixic acid (NA), LIO-92037 Trimethoprim (TM), and LIO-92004 Polymyxin B (PB) from Liofilchem^® MTS™]. Plates were then incubated lid-up at 37°C for 16 h, and the MICs were recorded according to the border of bacterial growth around the ETEST strip and according to the EUCAST Clinical Breakpoint Tables v. 16.1 for Enterobacterales (24 June 2026).

Selection and enrichment for spacers conferring antibiotic resistance

Selection of spacers conferring resistance was carried out on seven antibiotics: kanamycin, gentamicin, azithromycin, hygromycin, nalidixic acid, rifampicin, and trimethoprim at their MIC. Overnight cultures from the second cycle of adaptation (described above) were diluted in a ratio of 1:30 in 2× YT supplemented with 50 μg/ml streptomycin and 0.2% (w/v) L(+)-rhamnose monohydrate and grown for 2 h till the OD₆₀₀ reached 0.6 at 37°C. Then, 0.6 ml of the culture was spotted on each of the 1.5% (w/v) LB agar plates containing the respective antibiotic at its MIC. The plates were incubated at 37°C overnight. The colonies from each antibiotic plate were then counted and collected into 15 ml tubes for plasmid extraction.

The first cycles were carried out directly on the colonies that underwent adaptation or on the control. The subsequent cycles were carried out after plasmid transformations. The plasmids isolated from the adapted and control libraries were re-transformed into naive RG8610 harboring the pIY347 strain and grown for 1 h for recovery at 37°C. The transformants were supplemented with 0.2% (w/v) L(+)-rhamnose monohydrate after 1 h and incubated for another 1 h for induction in 0.4 ml of 2× YT. The cultures were then spread on dried plates supplemented with the respective antibiotics at their MIC and grown overnight at 37°C. PCR analysis using primers MG7R and TM137F was then carried out on DNA extracted from these colonies. The enrichment cycles were repeated until mostly or fully adapted bands were observed in 2% (w/v) agarose gel.

For identifying the repressors of the lactose operon or Cas3, overnight cultures from the second adaptation cycle were diluted in a ratio of 1:30 into 2× YT supplemented with streptomycin and 0.2% (w/v) L(+)-arabinose and grown at 37°C for 2 h until reaching an OD₆₀₀ of 0.6. A 0.6 ml aliquot of each culture was then spotted onto 1.5% (w/v) LB agar plates supplemented with streptomycin + kanamycin + arabinose (kanamycin 50 µg/ml for strain MG5538 and 10 µg/ml for strain MG5555). Plates were incubated overnight at 37°C. The following day, colonies were counted and harvested into 15 ml tubes for plasmid extraction. Purified plasmids were re-transformed into MG5538 and MG5555, followed by 1 h recovery at 37°C. Cultures were then supplemented with 0.2%(w/v) L(+)-arabinose and incubated for an additional hour for induction in 0.4 ml 2 × YT. After induction, the cultures were plated on 1.5% (w/v) LB agar plates supplemented with streptomycin + kanamycin + arabinose and spread using glass beads. Plates were incubated overnight at 37°C.

Sanger sequencing and deep sequencing analysis of enriched spacers on different antibiotics

After each cycle of enrichment of spacers on different antibiotics, a colony PCR was carried out using primers MG7R and TM137F. The colonies which contained new spacers were Sanger sequenced. The spacer sequences were aligned to the reference sequence of the E. coli strain K-12 genome (NC_000913.2). cydA spacer in kanamycin, cyoA spacer in gentamicin and gyrB spacer in azithromycin were detected during the first enrichment cycle through Sanger sequencing (Table 1). Spacers aligned to sequences on genes or upstream genes were considered to repress these genes. Plasmids from single colonies conferring resistance were extracted and re-transformed into RG8610. The antibiotic resistance at the MIC was tested again to validate the activity of the spacer. Further validation of the spacer activity is described below.

Table 1.

Sensitasome genes conferring antibiotic resistance

Antibiotic at MIC	Dominant spacers
Kanamycin	cydA*, cyoA, cyoB, dusB, dxs, ubiB, ubiD, ubiE, ubiJ
Gentamicin	cyoA*, dxs, ubiB, ubiE, ubiJ
Hygromycin	dxs, ubiB, ubiD, ubiE
Azithromycin	gyrA, gyrB*
Rifampicin	dusB
Trimethoprim	dusB, fusA, gyrA
Nalidixic acid	fusA

* Identified using Sanger sequencing. All others were identified using high-throughput sequencing.

For deep sequencing, the colonies from the first enrichment cycle on the MIC were grown overnight on LB at 37°C. Plasmids were then isolated from the culture, and the enriched spacers were amplified using 2× Phanta Flash master mix. In this PCR reaction, 25 μl of 2 × Master Mix master mix, 2 μl of 10 μM forward and reverse primers (Supplementary Table S2), 2 μl of bacterial lysate and 19 μl of double-distilled water were used. The PCR started with 30 s at 98°C followed by 35 cycles of 10 s at 98°C, 5 s at 55°C and 5 s at 72°C. The final extension step at 72°C was performed for 1 min. Then, 50 μl of the amplified product was run on 2% (w/v) agarose gel. The single amplified band of 171–173 bp of new spacers was cut from the gel and deep-sequenced. The premium PCR sequencing was performed by Plasmidsaurus using Oxford Nanopore Technology. The samples consisted of linear, double-stranded DNA of 171–173 bp. For this service level, up to 3000 reads were generated per sample. The FASTQ file generated from deep sequencing raw data was used to extract the spacer sequences by locating the two direct repeats of 28 bp flanking them. This was done using the Python (3.14.4) script, and its link is provided in the data availability statement. The script accounts for mutations on the repeat sequences up to three indels and ranks the spacers flanked by the two repeats based on the number of times it appears in the FASTQ file, along with their size and cumulative percentage (Supplementary Table S4). The spacers extracted were aligned to the reference sequence of the E. coli strain K-12 genome (NC_000913.2) using the Bowtie (version 1.0) and the annotated spacers are listed for each of the antibiotics (Supplementary Table S5 and at https://doi.org/10.5281/zenodo.22061709).

Analysis of lacI promoter spacers was performed by aligning the promoter region sequence to the deep sequencing data. The alignment parameters permitted a single mismatch, gap, or insertion for every eight bases.

Validation of the spacers’ activity

Spacer cloning was performed by ligation of phosphorylated oligonucleotide pairs and the parent vector, pIY331 (harboring only leader and repeat under l-arabinose promoter), as indicated in Supplementary Table S2. The clones were selected on ampicillin plates at 50 μg/ml and verified for having the insertion by using MR45F and IY1077 primers. The spacers cloned were transformed with pAH071 (harboring Cascade under l-arabinose promoter) into RG8610 strain. The transformants were selected on streptomycin and ampicillin 50 μg/ml plates. The control was an empty pIY331 plasmid along with pAH071 transformed into RG8610. The colonies were grown in LB overnight at 37°C supplemented with streptomycin and ampicillin 50 μg/ml. Next day the culture was diluted 1:30 and grown in 2× YT supplemented with streptomycin, ampicillin and 0.2% (w/v) L(+)-arabinose. The culture was then grown for 2 h and serially diluted tenfold followed by spotting on different antibiotics at their MIC. Only spacers showing resistance after this procedure were listed in Table 1.

Determining cross-resistance or sensitivity of each validated spacer on different antibiotics

The spacers listed in Fig. 3 were cloned in pIY331. They were transformed along with pAH071 into RG8610 for determining the cross-resistance or sensitivity on various antibiotics. The control included empty pIY331 and pAH071 in RG8610. The transformed RG8610 bacteria were grown overnight in 3ml LB supplemented with 50 μg/ml streptomycin and 50 μg/ml ampicillin at 37°C. The cultures were then diluted in a ratio of 1:30 in 3 ml 2× YT supplemented with 50 μg/ml streptomycin, 50 μg/ml ampicillin and 0.2%(w/v) L(+)-arabinose for 2 h till the OD₆₀₀ reached 0.6. Each of the cultures was serially diluted 10-fold in a 96-well plate. Then, 7.5 μl of the serially diluted cultures were spotted on 1.5% (w/v) LB agar plates supplemented with 50 μg/ml streptomycin, 50 μg/ml ampicillin, 0.2% (w/v) L(+)-arabinose, and 11 different antibiotics at their respective MIC indicated in Supplementary Table S6. The plates were incubated at 37°C overnight. Sensitivity/resistance was then determined by counting the CFU on each dilution.

Figure 3.

Cross-resistance and sensitivity profiles of validated sensitasome spacers across different antibiotics. A heat map highlighting spacer-specific resistance and sensitivity profiles across antibiotic classes. Validated spacers targeting the indicated genes were cloned individually and tested for their effects on bacterial survival across different antibiotics at their respective MICs. Values represent the log₁₀ change in CFU relative to the control strain carrying an empty spacer array.

Conservation of resistance conferring across species

Spacers repressing cydA, dxs and dusB cloned in pIY331 were transformed along with pAH071 into Salmonella enterica serovar arizonae (MG4510) and Shigella sonnei ATCC 9290 (MG4727) for determining conservation of resistance conferring across species. The cultures were grown in 10 ml of 2× YT supplemented with 0.2%(w/v) L(+)-arabinose, streptomycin and ampicillin overnight at 37°C. After 2 h the cultures were serially diluted tenfold and 7.5 μl of the culture was spotted on 1.5% (w/v) LB agar plates supplemented with kanamycin, gentamicin and hygromycin at their respective MIC indicated in Supplementary Table S6. The plates were incubated at 37°C overnight and were counted the next day.

Construction of mutant gyrB^I78A K-12 using pORTMAGE

We constructed the mutant gyrB^I78A K-12 strain using pORTMAGE [15]. We selected the mutants gyrB^I78A based on their resistance to novobiocin and their reduced ATPase activity used to supercoil relaxed plasmid DNA. This mutant maintained a normal growth rate in comparison to other mutants i.e. gyrB^D73E and gyrB^G77A [16]. Escherichia coli K-12 was transformed with pORTMAGE-Ec1. Oligonucleotides for MAGE were designed using the MODEST software [17]. Primers were ordered from SIGMA Merck which were of 90 bp phosphorothioate bonds connecting the two 5-most bases [18, 19]. The freshly streaked culture of the host strain with the pORTMAGE plasmid was grown overnight at 37°C, with kanamycin 50 μg/ml on plate. A single colony was inoculated into a 15 ml tube, containing 3 ml LB aliquot, supplemented with kanamycin. It was incubated overnight in a shaking incubator at 37°C, 250 revolutions per minute (rpm). From the stationary phase culture, the culture was diluted 100-fold in 25 ml LB, supplemented with kanamycin in a 150 ml glass flask. The culture was incubated in a shaking incubator at 37°C, 250 rpm. Upon reaching an OD_600nm 0.3–0.4, 1 mM of M-toluic acid was added and incubated in a shaking incubator at 37°C, for 30 min at 250 rpm. The cells were placed immediately on ice and chilled for at least 5 min. Electrocompetent cells were prepared by washing and pelleting them twice in 10 ml ice-cold ddH₂O at 4000 rpm for 10 min in a refrigerated centrifuge with swing-out rotor at 4°C. For electroporation, 50 μl of competent cells were mixed with 1 μl 100 μM MAGE oligo (resuspended in molecular grade water). The cells were electroporated in a pre-chilled 2 mm gap cuvette. Immediately after electroporation, 1 ml room temperature 2× YT was added. The cell suspension was transferred immediately to 4 ml pre-warmed (30°C) 2× YT medium, in a 150 ml glass flask. Recovering was allowed for 60 min at 37°C at 250 rpm. Then, 5 ml room temperature LB medium was added and supplemented with kanamycin. The cells were grown overnight at 37°C, 250 rpm. The culture was then diluted 100-fold in 25 ml LB, supplemented with kanamycin in a 150 ml glass flask for a new cycle. After the second cycle, the cells were serially diluted and 50 μl were plated from 10⁻⁴–10⁻⁶ dilutions on LB agar plates with kanamycin. Screening of 16–20 single colonies was done by PCR, and desired clones were selected following Sanger sequencing.

The mutant gyrB^I78A K-12 and K-12 were grown overnight in LB at 37°C. They were then diluted in a ratio of 1:30 and grown in 3 ml LB for an additional 3 h. The cultures were serially diluted 10-fold in a 96-well plate. Then, 7.5 μl of the serially diluted cultures were spotted on 1.5% (w/v) LB agar plates supplemented with antibiotics at their respective MIC. The plates were incubated at 37°C overnight and the CFU was determined the next day.

Construction of CRISPR-based selection markers

For the construction of a CRISPR-based selection marker, the cydA spacer was cloned under the cas12 m-modified plasmid. In this construct, the cas12m was placed under the constitutive J23119 promoter and the conventional antibiotic resistance marker was replaced with lacZ-α fragment amplified from puC19. For the positive control, a conventional plasmid having kanamycin resistance marker was used, into which the lacZ-α fragment was ligated. Similarly, for the negative control, the modified cas12m plasmid lacking a spacer was used into which the lacZ-α fragment was ligated in place of the resistance marker. The ligation reaction was carried out using Quick ligation kit (NEB#M2200L), maintaining a vector to insert molar ratio of at least 1:4. The ligation mixtures were cleaned up and transformed to chilled NEB5α competent cells and recovered at 30°C for 24 h with shaking. The transformants were selected on dried 1.5% (w/v) LB agar plates supplemented with kanamycin 15 μg/ml, 40 μg/ml X-gal, and 1 mM of IPTG. The plates were incubated at 37° for 24 h. For the CRISPR-based selection construct containing the cydA spacer, incubation was extended to 36 h to allow sufficient colony formation. Blue colonies were quantified for both the positive control and the Cas12m-cydA construct.

Results

Establishing an adaptation-linked CRISPRi assay to identify repressors

We initially aimed to develop an assay for identifying transcriptional repressors by coupling CRISPR adaptation with CRISPR interference (CRISPRi). The strategy relied on generating a spacer library from the E. coli genome. The randomly acquired spacers silenced random genes, including repressors. We then selected spacers that inhibited repressors, thus restoring expression from a repressed promoter of interest. This promoter controlled a kanamycin resistance gene (kanR), so its derepression conferred resistance, allowing identification of spacers that target the corresponding repressor.

To implement this, we constructed an E. coli strain encoding Cas1–Cas2 (adaptation plasmid) and a chromosomally integrated kanR under the control of the lac promoter (Fig. 1A). The strain also carried a plasmid expressing the Cascade complex and a CRISPR array for CRISPRi, while cas3 was deleted from the genome. In the absence of Cas3, the E. coli type I-E CRISPR system binds target genes without cleaving them, and therefore acts as a transcriptional repressor [8]. We hypothesized that the system would isolate spacers targeting lacI, the known repressor of the lac promoter.

Figure 1.

An adaptation-coupled CRISPRi assay to identify repressors. (A) Schematic representation of an E. coli strain encoding a chromosomally integrated kanamycin resistance gene under the control of the lac promoter. The strain carries plasmids encoding the CRISPR-cascade complex for CRISPRi and cas1–cas2 for adaptation. To avoid DNA cleavage, cas3 was deleted. (B) Plasmids extracted upon selection on kanamycin were retransformed to naïve cells and plated under three conditions (i) no kanamycin (control), (ii) kanamycin + IPTG (enables resistance regardless of CRISPRi), (iii) kanamycin + rhamnose (enables CRISPRi-dependent resistance). Spacer analysis from the surviving colonies after two selection cycles revealed three unique spacers targeting lacI. The table indicates their sequence and the number of reads that were identified in next-generation sequencing (NGS). Bars represent mean ± standard deviation (SD) of three independent experiments. ns, nonsignificant; *P <.05.

We induced the adaptation plasmid in this strain to enable spacer acquisition from the genome. As a negative control, we used bacteria encoding a catalytically inactive Cas1 mutant (Cas1^D221A). These strains were plated on kanamycin, and plasmids from the surviving colonies were extracted and transformed into naïve recipient bacteria. Transformants were then plated under three conditions: without kanamycin, with kanamycin + IPTG, and with kanamycin + rhamnose. IPTG derepresses the lac promoter by inactivating LacI and served as a positive control. Rhamnose induces CRISPRi by activating expression of Cascade and the CRISPR array, enabling spacer-guided gene repression. As an additional positive control we used bacteria encoding a CRISPRi construct with a designed spacer targeting lacI.

Similar colony counts were observed on plates without antibiotics or with kanamycin and IPTG, regardless of the plasmid source. However, on plates with kanamycin and rhamnose, plasmids extracted from the strain with the functional adaptation enabled growth, with ∼0.2% of colonies surviving (Fig. 1B). No colonies were recovered from the Cas1^D221A control, indicating that survival depended on spacer adaptation. Spacer analysis from the surviving colonies revealed three unique spacers targeting lacI. The positive control with the spacer targeting lacI also supported growth on kanamycin–rhamnose plates, as expected. These results confirmed that repression of lacI restored promoter activity, enabling kanamycin resistance and validating the functionality of the adaptation-linked CRISPRi screen.

Discovery of spacers targeting antibiotic-sensitizing genes

Encouraged by the initial success of our selection system, we sought to extend it to identify repressors of cas3, a key gene of the type I-E CRISPR-Cas system in E. coli, for which no natural repressor has yet been described as opposed to Cascade genes, controlled by the H-NS repressor [20]. To this end, we cloned the kanR downstream the cas3 promoter region and applied the same selection strategy described above. However, no resistant colonies carrying specific spacers targeting a putative repressor were recovered. We hypothesized that the strength of the cas3 promoter driving kanR expression was insufficient to confer resistance under high kanamycin concentrations even if its repressor is silenced. To test this, we reduced the kanamycin concentration to the minimal inhibitory concentration (MIC), which then enabled the isolation of colonies encoding unique spacers. We sequenced these spacers and identified that they targeted and repressed cydA and ubiB (Fig. 2A).

Figure 2.

Identifying spacers targeting antibiotic-sensitizing genes. (A) A schematic representation of identifying repressors of cas3 in a strain encoding a kanamycin resistance cassette downstream of the cas3 promoter. Colonies appeared to grow on low concentration of kanamycin. Selection and enrichment enabled the isolation of colonies encoding unique spacers targeting cydA and ubiB genes. (B) The assay was repeated on a strain which does not encode a kanamycin resistance cassette. Spacers with similar functions were revealed upon selection and enrichment on low concentration of kanamycin. Bars represent mean ± SD of three independent experiments. ns, nonsignificant; *P <.05; **P <.01.

Literature analysis revealed that disruption of the ubiquinone biosynthetic pathway, and specifically mutations in ubiD and ubiF, have been previously associated with increased resistance to aminoglycosides such as streptomycin and gentamicin, likely through impairment of the proton motive force and reduced antibiotic uptake [21]. We therefore suspected that the resistance phenotypes we observed were due to ubiB being part of this pathway and cydA gene, which encodes a subunit of the cytochrome bd oxidase complex, also participates in respiratory electron flow. We hypothesized that the spacers confer kanamycin resistance regardless of the kanR gene cloned downstream the cas3 promoter. We thus repeated the assay in a strain that does not encode a kanR gene. Indeed, spacers targeting the same ubiquinone biosynthetic pathway, specifically in ubiB and ubiE, were isolated conferring kanamycin resistance even in the absence of the kanR gene. Here too, the strain with the dead Cas1 produced no colonies (Fig. 2B).

Although originally aimed at identifying a cas3 repressor, the experiment thus unexpectedly revealed a powerful secondary application: uncovering genes whose repression increases antibiotic resistance. We subsequently redirected the system to this broader goal, and termed the collection of the genes we wished to identify as the “sensitasome”. While a similar concept had been explored for a single antibiotic using type II systems in S. aureus by electroporating exogenous genomes into it and thus generating a diverse library of spacers [6], we sought to expand the analysis using the type I-E CRISPR–Cas system in E. coli, using its endogenous genome, across a wide array of antibiotics.

Spacer selection reveals antibiotic sensitization targets

To systematically identify genes whose repression enhances bacterial resistance to antibiotics, we used an E. coli strain lacking resistance markers in the genome but harboring plasmids encoding Cascade, a naïve CRISPR array, and Cas1–2. We induced spacer acquisition to allow the strain to spontaneously acquire spacers from its own genome, generating a diverse spacer library. To evaluate the coverage of this library before selection, we deep-sequenced this naive library after the second cycle of adaptation. The acquired spacers mapped across the genome, and their density qualitatively correlated with the distribution of the PAM motifs (AAG and CTT) (Supplementary Fig. S1). We inoculated bacteria from this library on kanamycin and individually on six additional antibiotics (nalidixic acid, gentamicin, azithromycin, trimethoprim, hygromycin, and rifampicin), each at its MIC (MICs used are indicated in Supplementary Table S6; the standard working antibiotic concentrations, for reference, are indicated in Supplementary Table S7). The surviving colonies were collected and re-plated three times under the same antibiotic regime to enrich for clones carrying spacers that confer resistance.

CRISPR arrays from resistant colonies were amplified by PCR and sequenced, either through individual colony Sanger sequencing (marked with asterisk in Table 1) or by pooled high-throughput sequencing (all others). We identified both previously known and also novel genes whose repression conferred resistance. For example, on kanamycin-selection plates, several enriched spacers mapped to genes known to affect kanamycin sensitivity (e.g. cydA and ubiB), validating the selection strategy. Some of these genes were also identified on other antibiotics (e.g. ubiB and ubiD identified also on hygromycin). In addition, we identified novel sensitization targets, including dxs, gyrA, gyrB, and fusA whose repression enhanced antibiotic resistance (Table 1).

To validate these findings, we cloned the spacers into a naïve array on a plasmid and introduced them into E. coli strains expressing Cascade. Repression of all the genes listed in Table 1 reproducibly increased the respective antibiotic resistance, confirming their roles in antibiotic sensitivity (Supplementary Fig. S2). Collectively, these results revealed that a wide range of unrelated bacterial genes can modulate antibiotic susceptibility when repressed.

We further measured the MICs of these strains using ETEST strips to validate our screen (Supplementary Table S8). In most cases, the MICs successfully validated the resistance trends, showing fold-increases that are highly consistent with the increased survival and resistance profiles shown in Fig. 3. For example, gyrA and gyrB knockdowns showed a three-fold increase in azithromycin MIC. The fusA knockdown showed an over five-fold increase in nalidixic acid MIC. Similarly, the dusB knockdown showed a four-fold increase in MIC for both azithromycin and trimethoprim. We also validated the cydA, cyoA, cyoB, and dxs knockdowns which showed a 2- to 2.5-fold increase in kanamycin MIC, and ubiD and ubiE knockdowns which showed a four- to five-fold increase. While most hits were quantitatively validated, a few targets like ubiB and ubiJ knockdowns on kanamycin or fusA knockdown on trimethoprim did not show a MIC shift. Overall, these shifts confirm that the majority of our screen’s hits represent stable, robust changes in susceptibility that match the resistance profiles of these knockdowns.

To characterize the physiological effects and potential fitness costs associated with gene silencing, we monitored the growth profiles of all spacer-mediated repression strains (Supplementary Fig. S3). Under noninducing conditions, most strains exhibited growth curves identical to the wild-type control strain harboring an empty spacer array (Supplementary Fig. S3A). Upon induction of CRISPRi with L-arabinose, we observed varying growth rates and moderate fitness costs among the strains, with some targets showing reduced growth rates or slightly lower final cell densities (Supplementary Fig. S3B). This variation is physiologically expected when repressing important metabolic or essential genes. Importantly, all strains remained viable and successfully reached stationary phase, indicating that the repression is sub-lethal and nontoxic. While these varied growth profiles demonstrate that the repression does not cause acute lethality, they leave open the possibility that sub-lethal bottlenecks in core synthetic pathways could influence drug susceptibility through complex physiological feedbacks, a point we address further in the Discussion.

Multidrug sensitizations and resistances

Each of the validated sensitasome genes was discovered on the indicated antibiotic and then screened on others to reveal its effect on them. Notably, all sensitasome genes appear to mediate resistance to more than one antibiotic when silenced (Fig. 3).

Some genes were identified in screens on more than one antibiotic (e.g. ubiB, ubiD, ubiE, dusB). The fact that some genes were not identified initially on more than a single antibiotic suggests that the screen’s coverage is yet incomplete, so additional screening may reveal more genes. Additionally, some known genes such as ubiF, known to confer kanamycin resistance when repressed [21], have not been detected in our screens, further supporting the screens’ incompleteness. Nevertheless, the partial redundancy indicated that we have neared saturation. Among the most striking findings was the identification of spacers targeting gyrA and gyrB, both essential genes encoding the DNA gyrase. Downregulating them conferred resistance to azithromycin, and upon further examination, also to rifampicin, trimethoprim, nalidixic acid, and ciprofloxacin (Fig. 3). To our knowledge, downregulation of gyrA and gyrB as a mechanism of resistance to these antibiotics has not been previously reported. Repression of gyrA/gyrB, however, increased sensitivity to novobiocin, a known inhibitor of GyrB [22]. This outcome was expected, since novobiocin acts competitively to reduce DNA gyrase activity and thus amplifies the effect of gyrB knockdown [23]. Notably, the gyrB spacer was accompanied by an additional spacer, rhaR, which regulates the promoter used to express the spacers. This spacer probably reduced the repression of the essential gyrB gene, by tuning down the repression. However, we confirmed that the gyrB-targeting spacer by itself can confer a similar resistance phenotype (Table 1 and Supplementary Fig. S2).

Spacer analysis further revealed distinct resistance profiles. For example, spacers against cydA and dxs mediated broader resistance to multiple aminoglycosides, including kanamycin, hygromycin, and gentamicin (Fig. 3). Their repression, on the other hand, sensitized the bacteria to trimethoprim, novobiocin, and ciprofloxacin. Similarly, dusB repression conferred an even broader resistance profile, providing increased tolerance to kanamycin, gentamicin, hygromycin, and azithromycin. In contrast, its repression increased sensitivity to novobiocin, ciprofloxacin, and nalidixic acid. These contrasting phenotypes highlight the complex, and sometimes antagonistic, interplay between bacterial resistance and sensitivity mechanisms (Fig. 3). The unique resistance/sensitization profiles and trade-offs of downregulation of a specific gene may explain why certain resistance pathways are not constitutively active in bacteria, as selective pressure from one antibiotic may simultaneously expose vulnerability to another.

Validation of gyrB repression effect by a gyrB mutant

GyrB repression, as indicated above, conferred resistance to azithromycin, and increased sensitivity to novobiocin. We wondered whether the reduced activity of GyrB would phenocopy the resistance phenotype. We therefore constructed a gyrB mutant (I78A) with decreased ATPase activity, a substitution known to perturb the ATP/novobiocin binding site [24]. We first compared the growth of this strain to a control wt strain on LB and on novobiocin. This experiment confirmed that GyrB^I78A confers resistance to novobiocin (Fig. 4A), as reported [12].

Figure 4.

Validation of the gyrB repression effect on resistance by a gyrB mutant. (A) Escherichia coli K-12 gyrBI78A growth on high concentrations of novobiocin (220 μg/ml) compared to the isogenic strain with wild type gyrB. (B) The mutant gyrBI78A also shows resistance, compared to the isogenic strain with wild type gyrB, to azithromycin, nalidixic acid, ciprofloxacin, and rifampicin—all antibiotics to which the spacer-repressed gyrB shows resistance. (C) The mutant strain was also tested on antibiotics to which no resistance was observed due to the spacer repression of gyrB. The bar chart represents normalized colony forming units (CFU), calculated as the ratio of CFU (CFU/ml) on the antibiotic at its MIC to the total CFU (CFU/ml) on LB supplemented with streptomycin and ampicillin. The statistical significance between samples was calculated using an unpaired, two-tailed Student’s t-test. Data are shown as the mean ± SD; n = 3. ns, nonsignificant; *P <.05; **P <.01; ***P <.001.

We then tested this mutant growth on on all antibiotics to which the spacer-repressed gyrB showed resistance (i.e. azithromycin, nalidixic acid, ciprofloxacin, rifampicin, and trimethoprim), and compared it to the wt. We also tested the mutant resistance on three antibiotics to which the spacer-repressed gyrB did not provide resistance (kanamycin, gentamicin, and polymyxin). Remarkably, we found that GyrB^I78A conferred significantly higher resistance against all antibiotics to which the spacer provided resistance (Fig. 4B), except trimethoprim, but not on those that the spacer did not provide resistance (Fig. 4C). These findings further substantiate that reducing GyrB’s activity, and specifically its ATPase function, confers the antibiotic resistance phenotype.

Conservation of sensitization targets across species

To determine whether repression of the identified sensitasome genes confers antibiotic resistance across bacterial species, we introduced the CRISPRi system into S. enterica serovar Arizonae and S. sonnei strains (ATCC 9290). We selected representative spacers, regulating cydA, dxs, and dusB to evaluate the activity of gene repression in these species. We then inoculated serial dilutions of these bacteria on plates with three antibiotics to which the E. coli strain was resistant when these spacers were expressed (kanamycin, gentamicin, and hygromycin). We observed, as shown in Fig. 5, that all of these spacers conferred resistance for most tested antibiotics in both S. enterica serovar Arizonae and S. sonnei (excluding dusB for kanamycin and gentamicin in Salmonella and cydA on hygromycin in Shigella).

Figure 5.

Sensitization is across several bacterial species. Three selected spacers—cydA, dxs, and dusB—were tested for conferring resistance to kanamycin, gentamicin, and hygromycin at their respective MICs in both (A) S. enterica serovar Arizonae and (B) S. sonnei. The bar chart represents normalized CFU, calculated as the ratio of CFU (CFU/ml) on the antibiotic at its MIC to the total CFU (CFU/ml) on nonselective media supplemented with streptomycin and ampicillin plates. The statistical significance between samples was calculated using one-way analysis of variance (ANOVA) with Dunnett multiple comparisons correction. Data are shown as the mean ± SD; n = 3. ns, nonsignificant; *P <.05; **P <.01; ***P <.001; ^****P <.0001.

These observations suggest that the sensitization network is conserved across phylogenetically diverse members of the Enterobacteriaceae. These findings also suggest that CRISPRi-based repression of sensitizing genes may be broadly applicable for modulating antibiotic susceptibility in diverse bacterial hosts.

Sensitasome-derived spacers as CRISPR-based selection markers

Lastly, we evaluated the practical utility of sensitasome-derived spacers as alternatives to classical antibiotic resistance cassettes for plasmid selection. Unlike conventional resistance genes, which often encode antibiotic-degradative enzymes and can disseminate through horizontal gene transfer, CRISPR-based repression offers a nondegradative, genetically compact mechanism for selective growth on media containing a MIC of antibiotic. To implement this, we combined a representative sensitasome spacer, repressing cydA, with the compact CRISPR effector Cas12m [25, 26] and replaced the antibiotic resistance gene on a standard plasmid with the Cas12m–spacer cassette (Fig. 6A).

Figure 6.

A sensitasome-derived spacer as a CRISPR-based selection marker. (A) Schematic representation of the plasmid construct in which the conventional antibiotic resistance marker was replaced by a sensitasome spacer targeting cydA, together with the compact CRISPR effector Cas12m, on a standard plasmid backbone. The gene of interest, lacZ α complementation peptide in this case, was ligated into either the Cas12m–spacer cassette compared to a conventional plasmid carrying the kanR marker. Upon selection on kanamycin at its MIC, the two constructs yielded comparable numbers of colonies carrying the insert. (B) Bar chart showing the percentage of blue colonies obtained after selection on LB plates supplemented with kanamycin (15 µg/ml), IPTG (1 mM), and X-gal (40 µg/ml). The statistical significance between samples was calculated using one-way ANOVA with Dunnett multiple comparisons correction. Data are presented as mean ± SD, n = 3. ns, not significant; ^****P <.0001.

We then ligated a gene of interest, lacZ-α, into either the conventional plasmid bearing the kanR marker or the modified plasmid containing Cas12m and the cydA sensitizing spacer or to a negative control—a plasmid encoding the Cas12m without the cydA spacer. Electroporation of the ligation products yielded comparable percentage (>60%) of blue colonies in both plasmids with the Cas12m-cydA and the kanR marker, demonstrating that CRISPR-based repression provides a robust and reliable selection mechanism. In contrast, the plasmid without the cydA spacer resulted in 0% blue colonies, as expected (Fig. 6B). Sequencing of a few representatives of the blue colonies revealed the correct insertion of lacZ- α in the Cas12m-cydA cassette. This strategy offers a reduced genetic footprint, eliminates the need for antibiotic-degrading genes, and supports effective selection at lower antibiotic concentrations. Taken together, our results highlight the potential of sensitasome-based CRISPRi markers as scalable, environmentally friendly alternatives to traditional antibiotic-resistance cassettes, with broad efficacy across species and utility in both laboratory and industrial settings.

Discussion

Our study uncovered a broad set of bacterial genes whose repression increases antibiotic resistance, defining a distributed, selectable sensitization network that we term the ‘sensitasome.’ While previous studies focused on specific pathways or antibiotic classes, our screen across seven antibiotics revealed several known and novel genes, validating the approach and expanding the repertoire of sensitization targets. A recent study by Li et al. [27] investigated what they termed the genetic landscape of antibiotic sensitivity in S. aureus, using a type II CRISPR system and spacer acquisition from exogenous DNA sources [27]. In contrast, our work used a type I system in E. coli, with spacer acquisition occurring from the endogenous genome through self-targeting. The systems differ not only in the bacterial host and CRISPR type but also in the source of DNA sampled and in the downstream validation strategy. While Jiang et al. relied on enrichment scores to infer sensitization, we empirically validated all identified genes. In addition, we confirmed the activity of representative spacers in Shigella and Salmonella, and explored their application as nondegradative selection markers. These differences underscore the complementarity between the two studies in both method and outcome.

Recent studies have highlighted the power of CRISPRi screens for uncovering previously unrecognized determinants of antibiotic susceptibility. In Mycobacterium tuberculosis, genome scale CRISPRi libraries have been used to identify essential genes, antibiotic vulnerabilities, and pathways that modulate drug potency and resistance across diverse genetic backgrounds [28, 29]. More recently, CRISPRi screens have revealed strain-specific and collateral vulnerabilities associated with drug resistance [30]. Together with the current study in E. coli, as well as the recent analysis of the genetic landscape of antibiotic sensitivity in S. aureus [27], these studies establish CRISPRi as a powerful and rapidly emerging platform for dissecting complex resistance networks and uncovering previously unknown determinants of antibiotic susceptibility.

Our study revealed an example of trade-offs in resistance: repression of dusB, for example, conferred resistance to aminoglycosides and azithromycin, while sensitizing to novobiocin, ciprofloxacin, and nalidixic acid. The dusB encodes dihydrouridine synthase B, an enzyme involved in transfer RNA modification [31]. Notably, deletion of this gene was shown to confer rifampicin resistance [32]. The gene fis is located downstream of dusB, is co-transcribed with it, and is translationally regulated by sequences within the 5′ end of the dusB mRNA [33, 34]. Given that Fis has a pronounced effect on the expression of numerous genes across the genome, it is likely that the resistance and sensitization phenotypes observed upon dusB repression are mediated, at least in part, through reduced fis expression [35–37]. Our results extend its sensitizing role to antibiotics other than rifampicin, through a yet unknown mechanism.

We also identified gyrA and gyrB as sensitization genes for several antibiotics. GyrB is the ATPase subunit of bacterial DNA gyrase (a type II topoisomerase essential for introducing negative supercoils into DNA to facilitate replication and transcription) [38]. The ATPase activity of GyrB drives conformational changes necessary for the enzyme’s function [39]. Interestingly, despite both novobiocin and nalidixic acid targeting the DNA gyrase complex, only GyrB repression increased sensitivity to novobiocin, whereas GyrA repression did not increase sensitivity to nalidixic acid. This apparent discrepancy likely stems from differences in the antibiotics’ mechanisms of action. Novobiocin acts as a competitive inhibitor of the ATP-binding site on GyrB, reducing the enzyme’s catalytic activity [23]. When GyrB is already partially repressed, novobiocin further compromises the limited pool of functional gyrase complexes, exacerbating the deficit in supercoiling activity and leading to increased sensitivity. In contrast, nalidixic acid stabilizes GyrA–DNA cleavage complexes and prevents re-ligation, causing double-stranded breaks [40, 41]. Fewer functional gyrase complexes due to GyrA repression result in fewer cleavage complexes and thus fewer DNA breaks induced by nalidixic acid. This reduction in toxic intermediates may account for the observed lack of sensitization. Such mechanistic distinctions illustrate how different modes of antibiotic action interact with gene repression to shape resistance outcomes. To our knowledge, this is the first report that downregulating gyrA/gyrB can increase bacterial resistance to azithromycin and other antibiotics. Accordingly, a mutant gyrB, with reduced DNA gyrase activity, specifically showed resistance to azithromycin and to the other antibiotics to which the spacer provided resistance. These results demonstrate that the decreased activity indeed results in resistance to these antibiotics.

Interestingly, the gyrB^I78A mutant did not fully reproduce all phenotypes observed following gyrB repression. In particular, resistance to trimethoprim was observed with the gyrB targeting spacer but not with the ATPase mutant. This discrepancy suggests that some resistance phenotypes associated with gyrB repression may arise from broader physiological consequences of reduced gyrase abundance or altered DNA topology, rather than from reduced ATPase activity alone, as exhibited by the mutant.

The system described here has the unique capacity to identify essential genes that modulate antibiotic susceptibility. In contrast to conventional CRISPRi screening approaches that require the design, synthesis, and cloning of large guide libraries, the present system generates its own spacer library through CRISPR adaptation, substantially reducing the experimental complexity of genome-scale screening. This advantage arises from the coupling of CRISPR adaptation with interference, which enables genome-wide sampling of spacers without pre-selection or design constraints. As a result, a vast diversity of targets, including essential genes, is explored. Unlike other platforms that exclude essential genes, our approach permits partial repression, revealing sensitization roles that would otherwise remain undetected. Several essential genes, including gyrB, were identified precisely because of this inherent breadth and resolution. A notable example of this capability was the isolation of a gyrB-targeting spacer accompanied by a second spacer targeting rhaR, a transcriptional activator of the rhamnose promoter, which mitigated CRISPRi expression and consequently the over-repression of gyrB. This outcome demonstrates the strength of our selection system in fine-tuning the level of gene repression required to confer resistance while still allowing bacterial survival.

Our results may also relate to how antibiotic susceptibility depends on the metabolic and physiological state of the cell [42, 43]. Because our CRISPRi system causes moderate, sub-lethal gene repression instead of lethal knockouts, it allows us to study essential genes like gyrA and gyrB. However, because some induced guides moderately slow down growth (Supplementary Fig. S3B), these sub-lethal knockdowns might trigger suppressive drug-gene interactions. As shown by Bollenbach et al. [44], a sub-lethal bottleneck in one pathway can sometimes balance a bottleneck in another pathway, restoring growth in the presence of an antibiotic. Under normal conditions, blocking a single cellular process with an antibiotic may cause a toxic imbalance because other unchecked pathways keep running at full speed, creating a lethal buildup of cellular intermediates. By using CRISPRi to slow down a parallel pathway, such as DNA replication (e.g. gyrB), we create a physiological bottleneck. This bottleneck acts like a traffic jam that slows down the cell’s assembly lines, and may prevent toxic build-up caused by another antibiotic, thus helping the cell survive. Several of our phenotypes may reflect these complex, network-level metabolic feedback that mitigate drug toxicity rather than classic, structural resistance mechanisms [42].

Our findings open an avenue for novel therapeutic strategies. For example, one could pair standard antibiotics with small-molecule inducers that elevate the expression of these sensitization genes; such induced overexpression would sensitize bacteria to the drugs, making treatment more effective.

We also demonstrated the scalability of this screening approach across species. Representative spacers targeting sensitasome genes conferred resistance in both S. enterica and S. sonnei, indicating that the underlying sensitization mechanisms are conserved and that our platform can be applied to diverse bacterial hosts. One potential limitation of the approach is its application in bacterial hosts that encode endogenous CRISPR-Cas systems. The strains used in this study, including E. coli, S. enterica subspecies arizonae, and S. sonnei, encode complete or partial CRISPR systems, and all encode cas3. In the primary E. coli screening strain, the chromosomal cas3 gene was deleted to prevent CRISPR mediated DNA cleavage and ensure that the introduced type I-E system functions solely as a transcriptional repression platform. In contrast, no such modification was introduced into the Salmonella and Shigella strains used for cross-species validation. Although endogenous CRISPR systems could in principle interact with the introduced machinery, productive interference would generally require compatibility between the endogenous Cas3 protein and the introduced Cascade–spacer complex. The successful implementation of the platform in both Salmonella and Shigella suggests that such interactions do not substantially interfere with its function. Nevertheless, compatibility between endogenous and introduced CRISPR components should be considered when extending the approach to additional bacterial hosts.

Lastly, we demonstrated that sensitasome-derived spacers can be repurposed as environmentally friendly genetic selection markers. While the development of a sensitasome-based selection marker requires prior identification of a suitable target gene and spacer, the approach offers several advantages over conventional antibiotic resistance cassettes. First, it avoids the use of enzymes that directly inactivate antibiotics, thereby reducing the dissemination of hazardous resistance determinants through horizontal gene transfer. Second, the cassette is relatively small, particularly when combined with compact effectors such as Cas12m. Finally, because sensitization genes are often conserved, as demonstrated here for Salmonella and Shigella, the same selection strategy may be transferable across related bacterial species. These features may make the system particularly attractive from an environmental safety perspective in applications where minimization of conventional resistance markers is desirable and simple alternatives are actively sought. This approach thus offers a scalable, sustainable alternative for plasmid selection in synthetic biology and industrial applications.

Supplementary data

Supplementary data are available at NAR online.

Conflict of interest

None declared.

Funding

H2020 European Research Council (grant numbers 101248381 and 818878). Funding to pay the Open Access publication charges for this article was provided by ERC.

Data availability

The NGS data of the spacers generated in this study are available at the NCBI SRA with the BioProject accession – PRJNA1460245. The Python script used for spacer extraction is available via Zenodo: https://doi.org/10.5281/zenodo.22061709. The corresponding GitHub repository is available at: https://github.com/anushreehaldar002/Sensitasome/tree/v1.1 .The repository comprises source code and usage instructions.