Mechanisms of Resistance to Ceftolozane/Tazobactam in Pseudomonas aeruginosa: Results of the GERPA Multicenter Study

Resistance mechanisms of Pseudomonas aeruginosa to ceftolozane/tazobactam (C/T) were assessed on a collection of 420 nonredundant strains nonsusceptible to ceftazidime (MIC > 8 μg/ml) and/or imipenem (>4 μg/ml), collected by 36 French hospital laboratories over a one-month period (the GERPA study). Rates of C/T resistance (MIC > 4/4 μg/ml) were equal to 10% in this population (42/420 strains), and 23.2% (26/112) of the isolates were resistant to both ceftazidime and imipenem.

ABSTRACT

Resistance mechanisms of Pseudomonas aeruginosa to ceftolozane/tazobactam (C/T) were assessed on a collection of 420 nonredundant strains nonsusceptible to ceftazidime (MIC > 8 μg/ml) and/or imipenem (>4 μg/ml), collected by 36 French hospital laboratories over a one-month period (the GERPA study). Rates of C/T resistance (MIC > 4/4 μg/ml) were equal to 10% in this population (42/420 strains), and 23.2% (26/112) among the isolates resistant to both ceftazidime and imipenem. A first group of 21 strains (50%) was found to harbor various extended-spectrum β-lactamases (1 OXA-14; 2 OXA-19; 1 OXA-35; 1 GES-9; and 3 PER-1), carbapenemases (2 GES-5; 1 IMP-8; and 8 VIM-2), or both (1 VIM-2/OXA-35 and 1 VIM-4/SHV-2a). All the strains of this group belonged to widely distributed epidemic clones (ST111, ST175, CC235, ST244, ST348, and ST654), and were highly resistant to almost all the antibiotics tested except colistin. A second group was composed of 16 (38%) isolates moderately resistant to C/T (MICs from 8/4 to 16/4 μg/ml), of which 7 were related to international clones (ST111, ST253, CC274, ST352, and ST386). As demonstrated by targeted mass spectrometry, cloxacillin-based inhibition tests, and gene bla_PDC deletion experiments, this resistance phenotype was correlated with an extremely high production of cephalosporinase PDC. In part accounting for this strong PDC upregulation, genomic analyses revealed the presence of mutations in the regulator AmpR (D135N/G in 6 strains) and enzymes of the peptidoglycan recycling pathway, such as AmpD, PBP4, and Mpl (9 strains). Finally, all of the 5 (12%) remaining C/T-resistant strains (group 3) appeared to encode PDC variants with mutations known to improve the hydrolytic activity of the β-lactamase toward ceftazidime and C/T (F147L, ΔL223-Y226, E247K, and N373I). Collectively, our results highlight the importance of both intrinsic and transferable mechanisms in C/T-resistant P. aeruginosa. Which mutational events lead some clinical strains to massively produce the natural cephalosporinase PDC remains incompletely understood.

INTRODUCTION

Pseudomonas aeruginosa is considered by the World Health Organization as a top priority pathogen against which new therapeutic solutions are eagerly awaited. In this context, the recently released β-lactam/β-lactamase inhibitor combination ceftolozane-tazobactam (C/T) appears as a valuable option to treat severely ill patients infected by multidrug-resistant P. aeruginosa strains (1), with the same efficacy as that of meropenem (2). Because of a bulky group grafted at the 3-position side chain of the β-lactam ring, which provides a steric hindrance, ceftolozane better resists the attack of class C β-lactamases than does its parent compound ceftazidime (3). Corroborating these molecular properties, an ever-growing literature has provided evidence that C/T has good in vitro and in vivo activities on P. aeruginosa mutants overproducing the chromosomally encoded class C β-lactamase PDC (4–6). In addition, ceftolozane does not seem to be significantly impaired by active efflux mechanisms such as the Mex pumps (7). However, contrasting with its good stability vis-a-vis class C enzymes, ceftolozane offers no particular advantage over ceftazidime regarding most of the extended-spectrum β-lactamases (ESBLs) and carbapenemases encountered in P. aeruginosa clinical strains (8–11). While many studies have focused on the determination of rates of susceptibility to C/T in various countries and clinical contexts, little is yet known about the various mechanisms that may lead to C/T resistance in P. aeruginosa.

RESULTS

Resistance rates to C/T are highest among P. aeruginosa strains resistant to both ceftazidime and imipenem.

To retrospectively assess the mechanisms of resistance of P. aeruginosa to C/T before the marketing of this new drug combination, we analyzed a collection of 420 sequential nonredundant strains isolated over a one-month period (October 2015) in 36 French public hospitals participating in the GERPA epidemiological survey. Because C/T is intended to be used against multidrug (MDR) or extensively drug resistant (XDR) P. aeruginosa as an alternative to carbapenems, only those strains showing a nonsusceptibility to ceftazidime (MIC > 8 μg/ml, n = 134), imipenem (MIC > 4 μg/ml, n = 174), or both (n = 112) according the EUCAST 2019 criteria were considered in this study. These non-cystic-fibrosis bacteria were isolated from respiratory samples (n = 171; 40.7%), urine samples (n = 101; 24.0%), wounds (n = 42; 10.0%), abscesses/internal organs (n = 35; 8.3%), blood (n = 21; 5%), catheters (n = 21; 5%), feces (n = 5; 1.2%), and various other body sites (n = 24; 5.7%) from both pediatric and adult patients, 33% of whom were hospitalized in intensive care units (Table S1 in the supplemental material). Colistin (98.6% bacterial susceptibility), C/T (90.0%), amikacin (81.7%), ceftazidime-avibactam (CZA; 80.7%), tobramycin (76.2%), and aztreonam (69.0%) were the most efficient antibiotics in vitro on this collection, while the other molecules tested, including imipenem and meropenem, were efficacious against less than 55% of bacteria. Resistance rates to C/T were the highest among the isolates exhibiting a nonsusceptibility to both ceftazidime and imipenem (n = 26/112; 23.2%), lower in the ceftazidime-resistant, imipenem-susceptible population (n = 14/134; 10.4%), and minimal in imipenem-resistant, ceftazidime-susceptible strains (n = 2/174; 0.1%). Of note, 90.2% and 82.9% of imipenem-resistant strains remained susceptible to C/T (n = 258/286) and CZA (n = 237/286), respectively, while susceptibility rates of 86.7% and 77.3% were recorded for C/T (n = 202/233) and CZA (n = 180/233) among the meropenem-resistant population (MIC > 2 μg/ml). Altogether, our data confirm the potential interest of using C/T and CZA against these hard-to-treat organisms. Comparison of the performance of these two drugs revealed that 55.6% (n = 45/81) of CZA-resistant strains were susceptible to C/T, whereas only 14.3% (n = 6/42) of C/T-resistant isolates were CZA susceptible, thereby demonstrating a better in vitro activity of C/T on this MDR population. Though no pandrug resistance was noticed, seven (1.7%) P. aeruginosa strains were susceptible to only one agent out of the 13 tested (n = 5 to colistin and n = 2 to tobramycin). Susceptibility to two or three agents only included colistin systematically (14/14 and 10/10, respectively) and, for some isolates, aztreonam (6/14 and 3/10), C/T (2/14 and 4/10), and CZA (0/14 and 3/10). Of note, 15 (3.6%) of the bacteria exhibited a pan-β-lactam resistance.

A first group of C/T-resistant strains produces ESBLs and/or carbapenemases.

A number of Ambler’s class A, B, or D ESBLs and carbapenemases are known to strongly impair the activity of C/T toward P. aeruginosa (8–11). The mechanisms of the 42 (10%) C/T-resistant isolates of the GERPA collection were analyzed by phenotypic and molecular biology methods, as described in the Materials and Methods. Thus, a first group of 21 strains was found to produce various ESBLs (OXA-14: n = 1; OXA-19: n = 2; OXA-35: n = 1; GES-9: n = 1; and PER-1: n = 3), carbapenemases (GES-5: n = 2; IMP-8: n = 1; and VIM-2: n = 8), or both (VIM-2/OXA-35: n = 1 and VIM-4/SHV-2a: n = 1) previously identified in French isolates of P. aeruginosa (12, 13) (Table 1). Ten strains were isolated from pulmonary samples, four from urine, four from superficial sites, including catheters, two from abscesses, and one from a blood culture.

TABLE 1.

Characteristics of group 1 strains harboring ESBLs and/or carbapenemases

Strain	Serotypec	STc	MIC (μg/ml)c				β-lactamases
			C/T	CAZ	FEP	CZA	PDC (amt)a	Transferable
AMI03	O:11	ST235	64	512	32	>64	PDC-35 (97.5)	OXA-19
AMI14	O:11	ST235	32	256	32	>64	PDC-35 (51.2)	OXA-19
BESA02	O:11	ST235	32	256	32	>64	PDC-35 (21.4)	OXA-14
BICH30	O:11	ST235	>64	128	>64	>64	PDC-35 (16.6)	VIM-4 + SHV-2a
BICH32	O:4	ST175	>64	64	32	>64	PDC-1 (6.2)	VIM-2 + OXA-35
BICH33	PA	ST244	8	2b	64	1	PDC-1 (3.9)	OXA-35
BOR14	O:2	ST348	>64	512	>64	>64	PDC-53 (1.7)	IMP-8
CAE01	O:5	ST244	>64	32	16	64	PDC-1 (2.0)	VIM-2 + OXA-10
CAE13	PA	ST111	>64	64	32	64	PDC-3 (223)	VIM-2
CALE03	O:11	ST235	16	16	16	4	PDC-35 (<1)	GES-5
CALE04	PA	ST235	16	8	>64	32	PDC-35 (1.6)	VIM-2 + OXA-10
HEGP13	O 11	ST235	>64	256	64	16	PDC-19 (34.0)	GES-9
LYON12	O:11	ST235	8	16	8	4	PDC-35 (3.0)	GES-5
NANT08	O:11	ST235	>64	16	8	32	PDC-35 (40.8)	VIM-2 + OXA-2
NEC02	O:10	ST654	64	256	64	>64	PDC-3 (<1)	VIM-2
NEC15	O:12	ST111	>64	512	64	>64	PDC-3 (2.7)	VIM-2
POIT02	O:11	ST235	64	256	32	64	PDC-35 (<1)	PER-1 + OXA-2
ROUE14	NA	ST235	>64	512	64	64	PDC-35 (649)	PER-1
STRA09	NA	ST534	>64	512	32	64	PDC-35 (41.7)	PER-1
TOUL13	O:4	ST175	>64	16	8	64	PDC-1 (<1)	VIM-2
TOUL20	O:4	ST175	>64	256	>64	>64	PDC-1 (67.0)	VIM-2

^aRelative amounts of PDC variants as determined by targeted mass spectrometry, by reference to PDC-1 production in mutant PAO1ΔdacB (arbitrarily set at 1,000).

^bValues underlined indicate bacterial susceptibility according to EUCAST breakpoints for 2019.

^cPA, polyagglutinable with specific antisera; NA, nonagglutinable with specific antisera; ST, sequence type; C/T, ceftolozane-tazobactam; CAZ, ceftazidime; FEP, cefepime; CZA, ceftazidime-avibactam.

In agreement with global epidemiological data (14), these ESBL and carbapenemase producers all belonged to international “high-risk” clones, namely, ST111 (n = 2; serotype O:12), ST175 (n = 3; O:4), CC235 (ST235: n = 11 and ST534: n = 1; O:11), ST244 (n = 2; O:5), ST348 (n = 1; O:2), and ST654 (n = 1; O:10), respectively, and with few exceptions were highly resistant to almost all the antibiotic molecules tested except colistin (Table 1 and Table S1). Susceptibility to carbapenems in this group was less than 10%, while colistin (100%) and aztreonam (52%) were the most efficient agents in vitro (Fig. 1). As observed previously (8, 15, 16), avibactam restored the activity of ceftazidime (MIC ≤ 8μg/ml) in two clonally related ST-235 isolates expressing class A carbapenemase GES-5, but failed to prevent the hydrolysis of the antibiotic by the other β-lactamases identified, including class A ESBLs PER-1 and GES-9, a G243S variant of avibactam-susceptible enzyme GES-1 (17).

FIG 1.

Susceptibility rates of C/T-resistant isolates to antibiotics. Black bars of the histogram represent group 1 strains, and gray bars group 2 and 3 strains. EUCAST breakpoints for 2019 have been used for strain categorization (the values are indicated in the Materials and Methods).

To assess the possible contribution of the natural cephalosporinase PDC to C/T resistance in this first group of strains, PCR sequencing experiments were carried out which showed none of the bacteria encoded PDC variants with increased hydrolytic activity against ceftolozane (18). Variants PDC-1, PDC-3, and PDC-35 are common in clonal complexes ST175 (and ST244), ST111, and ST235, respectively. When analyzed by targeted liquid chromatography-tandem mass spectrometry (LC-MS/MS), 17 out of 21 strains appeared to produce detectable amounts of PDC, ranging from low (relative value [rv] = 1.6) to moderate (rv = 649) compared to those of the PDC-1 enzyme in wild-type strain PAO1 (rv = 3.3), and its overproducing mutants PAO1ΔampD (rv = 32.3) and PAO1ΔdacB (rv = 1,000) (Tables 1 and 2). In contrast to most ESBLs and carbapenemases, these PDC levels are not expected to provide a significant resistance to C/T or CZA (see mutants PAO1ΔampD and PAO1ΔdacB in Table 2). Thus, the unusual resistance profile of strain BICH33 to C/T (MIC = 8 μg/ml), cefepime (64 μg/ml), and ceftazidime (2 μg/ml) would depend on the production of ESBL OXA-35. However, while this β-lactamase is known to have a better hydrolytic activity on cefepime than on ceftazidime (19), its capacity to degrade ceftolozane requires confirmation.

TABLE 2.

Characteristics of group 2 strains exhibiting a high production of intrinsic β-lactamase PDC

Strain	Serotype	STc	MIC (μg/ml)c				PDC (amt)a	Alterations
			C/Td	CAZ	FEP	CZA
Reference strains
PAO1	O:5	ST549	1	2	2	2	PDC-1 (3.3)
PAO1ΔampD	O:5	ST549	1	8	4	2	PDC-1 (32.3)	AmpD
PAO1ΔdacB	O:5	ST549	2	16	8	2	PDC-1 (1,000)	DacB
PAO1ΔampD,dacB	O:5	ST549	4	64	16	4	PDC-1 (3,772)	AmpD, PBP4
Group 2 strains
AIX06	O:3	ST3318	8 (1)	256	32	64	PDC-303 (6,782)	AmpD
BESA09	O:11	ST532	8 (1)	256	32	>64	PDC-59 (9,517)	AmpR
BESA18	O:11	ST532	8 (2)	256	64	>64	PDC-59 (11,202)	AmpR
BOR21	O:3	ST569	8 (2)	512	64	64	PDC-53 (5,229)	Mpl, PBP3
CAE12	O:12	ST111	8 (2)	64	32	32	PDC-3 (4,767)b	PBP4
GAR02	O:6	ST3315	16 (1)	256	64	64	PDC-3 (8,747)	AmpR
HEGP08	O:10	ST253	8 (2)	64	32	16	PDC-34 (221)	PBP4
LILL04	O:6	ST3316	8 (2)	64	16	16	PDC-338 (5,607)	Mpl
LILL05	O:6	ST3317	8 (2)	256	64	32	PDC-3 (2,701)
LYON07	O:3	ST386	8 (2)	512	64	64	PDC-8 (3,152)	PBP4
MET04	O:3	ST268	16 (2)	256	32	32	PDC-24 (11,987)	AmpR
MONT08	O:6	ST792	8 (2)	64	32	16	PDC-5 (8,425)	AmpD, PBP2, PBP4
POIT11	O:1	ST3319	8 (2)	64	32	64	PDC-8 (5,525)	Mpl
ROUE01	O:1	ST3314	8 (1)	64	32	16	PDC-133 (416)	PBP4
ROUE09	O:3	ST268	8 (1)	64	32	8	PDC-24 (6,007)	AmpR
ROUE19	O:4	ST389	8 (2)	64	32	16	PDC-1 (8,891)	AmpR

^aRelative amounts of PDC variants as determined by targeted mass spectrometry, by reference to PDC-1 production in mutant PAO1ΔdacB (arbitrarily set at 1,000).

^bThis strain also produced penicillinase PSE-1.

^cST, sequence type; C/T, ceftolozane-tazobactam; CAZ, ceftazidime; FEP, cefepime; CZA, ceftazidime-avibactam. Underlined value refers to bacterial susceptibility according to EUCAST breakpoints 2019.

^dValues in brackets have been determined in the presence of 2,000 μg/ml cloxacillin (PDC inhibitor).

A second group strongly overproduces wild-type variants of cephalosporinase PDC.

Phenotypic tests and genome sequencing failed to detect ESBLs and carbapenemases in the 16 strains of this second group, of which 7 (44%) belonged to known epidemic clones or clonal complexes such as ST111 (n = 1), ST253 (n = 1), ST268 (n = 2; CC274), ST386 (n = 1), and ST532 (n = 2). Compared at the core genome level (data not presented), two ST532 isolates from the same institution (BESA09 and BESA18), turned out to differ by only two single nucleotide polymorphisms (SNPs), and as thus can be considered almost identical. The nine remaining strains were related to sporadic genotypes represented by a few or single occurrences in the Pseudomonas aeruginosa MLST database (https://pubmlst.org/paeruginosa/).

Group 2 strains were moderately resistant to C/T with a modal MIC of 8 μg/ml (range from 8 to 16 μg/ml), while this value was >64 μg/ml for group 1 (Table 2). It should be noticed that 8 (50%) of these strains were nonsusceptible to all the β-lactam molecules tested. In this second group, as in group 1, CZA retained an activity on a minority of strains (6.25% versus 14.3%), but the rates of susceptibility to imipenem (50% versus 9.5%), meropenem (43.75% versus 9.5%), ciprofloxacin (56.25% versus 0%), tobramycin (93.75% versus 4.8%), and amikacin (56.25% versus 9.5%) were notably higher. No pandrug resistance was recorded despite the fact that one strain was colistin resistant (LILL04).

Consistent with an elevated production of cephalosporinase PDC being the major driver of C/T resistance in this group, a complete (1 μg/ml) or partial (2 μg/ml) recovery of wild-type susceptibility to C/T was obtained in the presence of Ambler’s class C inhibitor cloxacillin at 2,000 μg/ml. A similar reduction of MIC values of C/T (up to 16-fold) and other cephalosporins (up to 128-fold) was obtained when the bla_PDC gene was deleted in four strains amenable to genetic manipulations (GAR02, HEGP08, MET04, and ROUE19) (Table S3). To quantitatively assess the production of PDC, group 2 bacteria were grown in drug-free Mueller-Hinton broth (MHB) and then lysed by ultrasounds and submitted to LC-MS/MS analysis after trypsin digestion (see Materials and Methods). These experiments confirmed the presence of extremely large amounts of cephalosporinase in most of the strains (rv from 221 to 11,987). These levels largely exceeded those of a single ΔampD mutant of PAO1 (rv = 32.3), and were superior to those of ΔdacB (reference value of 1,000) and ΔampD,dacB (rv = 3,772) mutants in 14 (87%) and 12 (75%) strains, respectively (Table 2). However, while being more resistant than PAO1ΔampD,dacB to cephalosporins, four clinical strains displayed similar (LYON07) or lower (HEGP08, LILL05, and ROUE01) PDC levels, a result which suggests that the cephalosporinase synergized with undetermined resistance mechanisms in at least this subset of bacteria.

The primary structure of PDC was inferred from the genomic sequences of the bacteria and subsequently confirmed by PCR amplification and Sanger sequencing. We noted that only one strain encoded the prototype enzyme PDC-1 of strain PAO1, while the 14 other variants exhibited common sequence polymorphisms, namely, PDC-3 (n = 3), PDC-5 (n = 1), PDC-8 (n = 2), PDC-24 (n = 2), PDC-34 (n = 1), PDC-53 (n = 1), PDC-59 (n = 2), PDC-303 (n = 1), and PDC-338 (n = 1) (Table 2 and Table S2). A previous study from our laboratory demonstrated that these polymorphisms do not modify the resistance profile conferred by the overexpressed enzyme in comparison with PDC-1 (18). Finally, the remaining strain of that group, ROUE01, appeared to harbor a novel PDC variant (PDC-133) carrying 10 amino acid substitutions, of which four do not seem to have been reported in the literature so far: Q204R, P274Q, S306A, and K396A (note that the amino acid numbering used includes the 26 residues of the signal peptide of precursor protein). According to the recently proposed SANC (structural alignment-based numbering of class C) system that refers to the mature form of the AmpC enzyme from Enterobacter cloacae P99 (20), the aforementioned variations lie at the consensus positions 177, 247, 279, and 369, respectively. Whole-genome sequence comparisons with the NCBI database revealed that ROUE01 is related to a small group of strains isolated in Europe, Australia, and the USA from human, animal, and environmental samples (data not shown). This group is phylogenetically distinct from the PAO1, PA14, and PA7 phyla, and exhibits a clone-specific PDC sequence identical to or showing more than 99% identity with PDC-133. In parallel, our sequence alignments revealed that two group 2 strains were deficient in the DNA-mismatch repair system protein MutL because of a frameshift (GAR02) and a premature stop codon in the mutL coding sequence (LILL04), respectively. Hypermutability is a well-known cause of bacterial evolution and adaptation to antibiotics (21), including novel β-lactam/β-lactamase inhibitor combinations (22, 23). Thus, it is tempting to assume that, in some clinical situations, a hypermutator background predisposes the strain to the emergence of mutations leading to very high PDC levels. Of note, none of the strains investigated in this work exhibited inactivating mutations in the MutS-encoding gene.

To get an insight into the causative events responsible for such a high cephalosporinase production in group 2 strains, the sequences of a subset of chromosomal genes belonging to the resistome of P. aeruginosa were analyzed and compared with that of reference strains PAO1, PA14, and LESB58 by using CLC Genomics Workbench package (Qiagen). It is worth mentioning that the cystic fibrosis epidemic strain LESB58 harbors an R504C mutation in the essential penicillin-binding protein PBP3, which is predicted to affect the binding of β-lactams, including ceftazidime, to the enzyme transpeptidase domain (24) (Table S2). Considering the complex sequence polymorphisms occurring in clinical strains, only genes showing disruptive mutations (e.g., frameshifts, insertion sequences, or premature stop codons) or single point variations previously associated with β-lactam resistance were considered significantly impaired. The published data used to assess the functionality of the genes and their products are listed in the supplemental material (Table S2). As a consequence, a number of amino acid substitutions not characterized by mechanistical studies were not taken into account in this analysis. For instance, strain LILL05 harbored sequence variations in several proteins of interest such as AmpD, AmpDh3, DacB, and AmpO, but no notable disruption of the screened genes. Evident alterations were found in the gene bla_PDC regulator AmpR (D135N/G in 6 strains) and various key elements of the peptidoglycan recycling pathway, whose inactivation was demonstrated elsewhere to upregulate bla_PDC expression, such as the N-acetyl-muramyl-l-alanine amidase AmpD (n = 2), penicillin-binding protein PBP4 (n = 5), and UDP-N-acetyl-muramyl-l-alanyl-γ-d-glutamyl-meso-diaminopimelate ligase MPL (n = 3) (for a recent review see reference 25) (Table 2). According to Caille et al. (26), the D135N substitution turns AmpR into a constitutive transcriptional activator of bla_PDC. Highly conserved among AmpR regulators of Gram-negative ESKAPE pathogens, the aspartate residue at position 135 is located in the effector-binding domain of these LysR proteins, next to the “gatekeeper” residue R133 that enables access of suppressor and activator muropeptides to the effector-binding site (27). In vitro emergence of a D135N AmpR mutant resistant to C/T (MIC of 8 μg/ml) was reported to occur during evolution experiments performed with a hypermutable PAO1 strain exposed to ceftazidime (28). A similar mutant expressing fully derepressed levels of expression of the gene bla_PDC (>600-fold the level of uninduced strain PAO1) was selected under ceftazidime treatment from a cystic fibrosis strain in a murine model of chronic lung infection (29). In addition, several XDR strains from various geographical origins have also been reported to harbor a D135N or D135G mutation, thereby emphasizing the somewhat overlooked role of AmpR in β-lactam resistance acquisition (30–33).

In one of our isolates (BOR21), the PBP3 protein harbored the same mutation R504C as in LESB58 (whose resistance level to ceftazidime and C/T is equal to 16 μg/ml and 2 μg/ml, respectively), whereas the MutL-defective strain MONT08 exhibited a deletion of six amino acids in the nonessential penicillin-binding protein PBP2, a target for some β-lactam molecules (34) but whose suppression has minimal consequences on β-lactam MICs (35). Additional polymorphic variations of unclear significance were noted in some of the proteins screened, including AmpD and PBP4/DacB (Table S2), as well as in various others involved in peptidoglycan homeostasis (CreBCD, AmpE, AmpG, AmpO, AmpP, and SltB1) (data not shown). On the whole, and according to our criteria, relatively few strains appeared to simultaneously contain gross defects in two (BOR21) or three (MONT08) proteins previously recognized as involved in β-lactam resistance. Dual alteration of AmpD/PBP4 was ascertained only in MutL mutant MONT08. Though other cellular alterations or mechanisms can potentially contribute to the development of C/T resistance in P. aeruginosa, as suggested by in vitro evolution experiments (22, 36), this study highlights the predominant role played by PDC in clinical strains. Ceftolozane is not a good substrate for the efflux pumps of P. aeruginosa, and thus is not expected to have its activity compromised significantly by upregulated Mex systems (7). Further ruling out a possible contribution of the major pump MexAB-OprM to C/T resistance, targeted mass spectrometry analysis established that none of these second group isolates overproduced this system significantly (Table S4).

Isolates of the third group encode extended-spectrum PDC variants.

The third group was composed of five strains isolated from two respiratory samples, a blood culture, an abscess, and a wound, respectively. Comparison of their core genomes showed that two isolates belonging to the international clonal complex ST308 were distant from each other by 131 SNPs, while the others were related to sporadic strains or locoregional clones (e.g., ST564). All the bacteria of group 3 except one (POIT06) displayed an elevated resistance to ceftazidime (≥128 μg/ml). As indicated in Table 3, their susceptibility levels to C/T (from 8 to >64 μg/ml) and to other β-lactams were quite variable. Colistin remained active in vitro on all these bacteria, while imipenem, tobramycin, and amikacin were active on four, ciprofloxacin on three, and meropenem on two of them, respectively (Table S1). No ESBL- or carbapenemase-encoding genes that could account for the observed resistance profiles were detected in the genomes of these strains. In contrast, four of the bacteria appeared to determine PDC sequence variants harboring mutations known to increase the β-lactamase activity toward ceftazidime and C/T, namely, F147L, N373I, and E247K (18), corresponding to SANC positions 121, 346, and 219, respectively. The remaining isolate, POIT06, coded for a new variant of PDC that lacked four amino acids (ΔL223–Y226; SANC positions 196 to 199) in the Ω loop of the enzyme, a part of the catalytic site often targeted by single point mutations (18). Of note, amino acid deletions of different lengths in this region have been associated with the development of C/T resistance in treated patients (37–40).

TABLE 3.

Characteristics of group 3 strains producing PDC variants with increased activity against C/T

Strain	Serotype	STd	MIC (μg/ml)d				PDC		Alterations
			C/T	CAZ	FEP	CZA	PDC (amt)a	Mutations
BICH11	O:7	ST3320	8 (8)	64	16	16	PDC-264 (1,484)	F147L	PBP4, Mpl
MONT13	O:11	ST308	16 (1)	>512	>64	>64	PDC-87 (8,217)b	N373I	Mpl
MONT14	O:7	ST671	>64 (>64)	512	16	32	PDC-337 (347)	E247K	Mpl
POIT06	O:11	ST308	16 (32)	16	8c	8	PDC-339 (511)	ΔL223-Y226	Mpl
ROUE12	O:9	ST564	64 (128)	256	16	8	PDC-340 (5.7)	E247K

^aRelative amounts of PDC variants as determined by targeted mass spectrometry, by reference to PDC-1 production in mutant PAO1ΔdacB (arbitrarily set at 1,000).

^bThis strain also produced penicillinase TEM-2.

^cValues underlined refer to bacterial susceptibility according to EUCAST breakpoints 2019.

^dST, sequence type; C/T, ceftolozane-tazobactam; CAZ, ceftazidime; FEP, cefepime; CZA, ceftazidime-avibactam.

PDC production as assessed by targeted mass spectrometry varied greatly within group 3 strains, from a low basal (ROUE12) to a very high level (MONT13) (Table 3). Consistent with previous observations (18), cloxacillin failed to restore susceptibility to C/T in 4 out of 5 strains harboring extended-spectrum PDC variants. Thus, gene deletion experiments were carried out to further characterize the strains. Inactivation of gene bla_PDC was successful in two isolates, BICH11 and ROUE12. As predicted, these deletions resulted in a strong decrease in the MICs of cephalosporins, including C/T (2 and 1 μg/ml, respectively) (Table S3).

Analysis of the β-lactam resistome highlighted various mutations in the PBP4- and Mpl-encoding genes in strains BICH11, MONT13, MONT14, and POIT06, but none in ROUE12, consistent with our mass spectrometry data (Table S2). Though previous findings demonstrated that ceftazidime is a weak inducer of bla_PDC expression in P. aeruginosa (41), our results suggest that this induction might be much more pronounced in some genetic backgrounds, like ROUE12. Supporting this notion, scatter colonies were observed in the inhibition zone of a C/T gradient strip (Fig. S1). Similar results were obtained when several of these colonies were subcultured in drug-free MHB, consistent with an adaptive and transient mechanism of resistance (data not shown).

DISCUSSION

This study shows that three different mechanisms expressed by multidrug-resistant strains of P. aeruginosa can compromise the clinical efficacy of C/T. Two of these mechanisms are associated with known or yet-undetermined mutational events that generate either a massive production of PDC or various structural changes in the β-lactamase, enabling bacteria to inactivate ceftolozane more efficiently. Acquisition of transferable ESBLs or carbapenemases represents another way by which epidemic clones (so-called “high risk clones”) become refractory to C/T. The impact of all these resistance mechanisms on bacterial virulence was not addressed specifically in this work, but the fact that all the GERPA strains were considered infectious and not mere contaminants by clinicians argues against major pathogenicity deficiencies. Since these strains were collected before the release of C/T and CZA in France (year 2015), it is evident that none of these two-drug combinations, but rather older antipseudomonal β-lactams such as ceftazidime, cefepime, or piperacillin-tazobactam, selected for C/T resistance. Interestingly, recent reports highlighted the emergence under C/T or CZA treatment of resistant mutants that produce OXA-type ESBL or PDC variants (23, 37–40, 42–46). Cross resistance of many of these mutants to all cephalosporins and penicillins supports the notion that P. aeruginosa is able to adapt preexisting mechanisms in order to resist novel β-lactam molecules and inhibitors.

Not surprisingly, we observed that C/T-resistant strains producing ESBLs and/or carbapenemases (group 1) belong to international clones prone to collect resistance genes from various bacterial reservoirs (14, 47). Against these XDR bacteria, very few therapeutic options exist, and C/T provided no particular advantage in vitro over older molecules, including ceftazidime. CZA was superior in vitro to C/T against producers of class A carbapenemase GES-5 only. One of the objectives of the GERPA survey was to determine the prevalence of ESBLs and carbapenemases among multidrug-resistant French P. aeruginosa isolates. Our data show that this prevalence is still low (23/420; 5.5%) and has remained rather stable over the last decade (12, 13). Among the 378 C/T-susceptible strains, one encoded the ESBL OXA-35 (C/T MIC = 2 μg/ml) and one the carbapenemase GES-5 (MIC = 4 μg/ml) (data not shown).

As exemplified by most group 2 strains (16/420; 3.8%), a clinically significant resistance to C/T (MIC ≥ 8 μg/ml) may result from an extremely high production of cephalosporinase PDC. Mutational dysregulation of the peptidoglycan recycling pathway is a well-known cause of PDC-dependent β-lactam resistance in clinical strains of P. aeruginosa (48). AmpR, the master regulator of gene bla_PDC expression can be activated as a result of an imbalance between various cytoplasmic muropeptide species acting as AmpR ligands (25), or by mutations occurring in or close to AmpR functional domains (26). How far the impact of these mutations on PDC expression is influenced by strain-specific traits remains poorly understood (41). For instance, Zamorano et al. demonstrated that dramatically different levels of bla_PDC expression and β-lactam resistance can result from the deletion of genes ampD and dacB in nonclonal clinical strains (49), thereby supporting the notion that still-undetermined factors modulate the effects of such inactivations. Since a wild-type susceptibility to cephalosporins was restored in group 2 bacteria upon cloxacillin exposure and/or gene bla_PDC deletion, such factors would logically have a direct or an indirect impact on PDC expression. In this study, only clear-cut gene alterations were considered significant (i.e., frameshifts, premature stop codons, sequence disruption by IS); therefore, the contribution to C/T resistance of other defects in known and/or uncharacterized genes of the β-lactam resistome was overlooked and cannot be ruled out (22, 50). In reference strain PAO1, and in the absence of structural modification of PDC, mutations in at least two genes such as ampD and dacB are required to increase the C/T MIC up to 4 μg/ml (Table 2) (28). As reported previously, multiple and complex gene alterations in line with a hypermutator phenotype tend to accumulate in in vitro-evolved mutants, reaching a C/T resistance level of ≥8 μg/ml (22, 28). Interestingly, two of our group 2 strains were defective in the DNA mismatch repair system (MutL mutants) and thus would be more prone to mutationally adapt to antibiotics (28). However, none of these two appeared to contain more than one knockout gene among those considered potential sources of β-lactam resistance. Besides, a single strain carrying intact mutS and mutL genes exhibited a dual ampD-dacB inactivation (MONT08). Why such double mutants seem to be infrequent among clinical strains (51) remains unclear, since the virulence of P. aeruginosa is not expected to be impaired by such mutations (52).

Our study also confirms the important role played by mutations at position D135 of AmpR in promoting a high PDC-dependent resistance to cephalosporins such as ceftazidime and cefepime, and a moderate, albeit therapeutically significant, resistance to the more stable molecule ceftolozane. Alterations D135N/G have been identified in multidrug-resistant strains from both cystic fibrosis (CF) and non-CF patients (31–33, 43, 53). Like the R86C (33, 42, 43) and G154R (54) variations, mutations D135N/G are believed to turn MexR into a strong activator of bla_PDC expression (26).

We observed that several group 2 isolates carried disrupted Mpl-encoding genes as the sole alteration found with relation to their resistance phenotype. Since previous work has shown that strain PAO1 susceptibility to ceftazidime and cefepime remains almost unchanged upon gene mpl inactivation (55), a simple explanation would be that these GERPA strains contain other defects impairing the peptidoglycan recycling process, which synergize with the effects of mpl mutations. From a clinical perspective, the possibility that single genetic events might promote the development of C/T resistance and cause treatment failures needs to be investigated more closely. In addition, we noticed that three group 2 bacteria (HEGP08, LILL05, and ROUE01) produced PDC at low or moderate levels while displaying a clinically significant resistance to the drug combination. Ongoing experiments are being performed to determine whether these levels are induced in the presence of C/T.

Group 3 illustrates another means by which P. aeruginosa can escape antipseudomonal cephalosporins. Indeed, mutational adaptation of the PDC structure to the large C/T molecule has been documented on several occasions in treated patients (37–40, 42–45). Such adaptation actually requires at least two distinct genetic events, one resulting in PDC overproduction, and another introducing appropriate changes in the enzyme structure. Group 3 strains (5/420; 1.2%) displayed rather common amino acid variations in the omega loop, which is part of the PDC active site (18, 56). All these bacteria except one overproduced PDC at various levels and exhibited mutations in at least peptidase Mpl. While being highly resistant to C/T and ceftazidime, the remaining isolate ROUE12 showed a very low basal level of variant PDC-340, suggesting that these antibiotics might behave as inducers of gene bla_PDC expression in the ROUE12 background. Possibly accounting for its peculiar phenotype, this strain was the only one of groups 2 and 3 harboring a mutation in a putative binding site of DNA-bending, histone-like integration host factor (IHF), which partially overlaps the ampR promoter in the ampR-bla_PDC intergenic region (57) (data not shown). Reminiscent of the heterologous expression of C/T resistance in ROUE12, heteroresistance to cefepime was recently described in clinical P. aeruginosa from China, and linked to an intrastrain differential expression of PDC (58). This finding, together with our data, underscores the complex regulation of the cephalosporinase in clinical strains of P. aeruginosa, and the difficulty in linking specific mutations with precise levels of PDC and β-lactam resistance. In this study, group 2 and 3 strains were in general more susceptible to aminoglycosides and fluoroquinolones than those of group 1 (Fig. 1), likely because most ESBLs and carbapenemases are encoded by mobile genetic elements carrying multiple antibiotic resistance determinants. Whether the increasing use of C/T will promote the emergence of novel types of P. aeruginosa mutants in addition to those reported here needs to be carefully monitored.

MATERIALS AND METHODS

Bacterial strains, growth conditions, and O serotype determination.

The 420 clinical P. aeruginosa strains of the GERPA collection were isolated in 36 French public hospitals located in Europe and overseas over a one-month period (October 2015). All the isolates showing nonsusceptibility to ceftazidime (MIC > 8 μg/ml) and/or imipenem (MIC > 4 μg/ml) according to the EUCAST 2019 breakpoints (https://www.eucast.org/clinical_breakpoints/) were prospectively collected on a one strain-per-patient basis and sent to the bacteriology department of the Besançon teaching hospital (National Reference Center for Antimicrobial Resistance, NRC-AR) for analysis. The clinical samples from which the GERPA strains were isolated are indicated in Table S1 in the supplemental material, and mentioned in the text for the ceftolozane-tazobactam (C/T)-resistant bacteria. Wild-type reference P. aeruginosa strain PAO1 and its PDC-overexpressing mutants PAO1ΔampD, PAO1ΔdacB, and PAO1ΔampD,dacB were kindly provided by Antonio Oliver (4). Strain Escherichia coli CC118λpir (Δ[ara-leu] araD ΔlacX74 galE galK phoA20 thi-1 rpsE rpoB argE(Am) recAl; lysogenized with phage λpir) was used as a host for the propagation of pKNG101-based plasmids in gene deletion experiments (59). Bacterial cultures were routinely performed in Mueller-Hinton broth (MHB) (Becton, Dickinson, Grenoble, France) or on Mueller-Hinton agar plates (MHA) (Bio-Rad, Marnes-la-Coquette, France) at 35°C ± 1°C, supplemented with appropriate antibiotics for maintenance of plasmid vectors as follows: 50 μg/ml kanamycin and 50 μg/ml streptomycin for E. coli strains; 2,000 to 10,000 μg/ml streptomycin for P. aeruginosa strains. Pseudomonas aeruginosa strains resistant to C/T were serotyped according the Habs scheme by slide agglutination with specific antisera from Bio-Rad unless genomic sequences were available for in silico determination of serotypes (see below).

Drug susceptibility testing.

MICs of selected antibiotics were determined by the broth microdilution method with microplates containing customized drug panels (Thermo Fisher Scientific, Illkirch-Graffenstaden, France). The microplates were inoculated with ca. 10⁵ CFU/ml in MHB and incubated aerobically at 35°C ± 1°C for 18 h ± 2 h, as recommended. Strains were categorized as “susceptible,” “intermediate,” or “resistant” according to the EUCAST 2019 breakpoints as follows: ticarcillin (TIC: S ≤ 16 μg/ml; R > 16 μg/ml); piperacillin plus tazobactam (TZP: ≤16/4 μg/ml; >16/4 μg/ml); aztreonam (ATM: ≤16 μg/ml; >16 μg/ml); ceftazidime (CAZ: ≤8 μg/ml; >8 μg/ml); cefepime (FEP: ≤8 μg/ml; >8 μg/ml); ceftolozane plus tazobactam (C/T: ≤4/4 μg/ml; >4/4 μg/ml); ceftazidime plus avibactam (CZA: ≤8/4 μg/ml; >8/4 μg/ml); imipenem (IPM: ≤4 μg/ml; >4 μg/ml); meropenem (MEM: ≤2 μg/ml; >8 μg/ml); ciprofloxacin (CIP: ≤0.5 μg/ml; >0.5 μg/ml); tobramycin (TOB: ≤4 μg/ml; >4 μg/ml); amikacin (AMK: ≤8 μg/ml; >16 μg/ml); and colistin (CST: ≤2 μg/ml; >2 μg/ml). Ceftazidime MICs > 32 μg/ml were determined more precisely by using Etest strips (bioMérieux, Marcy-l’Etoile, France). Liofilchem gradient strips (Roseto degli Abruzzi, Italy) were used to assess the MIC of C/T in the presence of 2,000 μg/ml cloxacillin.

Screening and identification of β-lactamase genes.

GERPA strains resistant to imipenem were all phenotypically screened for carbapenemase production by using (i) a double-disk synergy test (DDST) with zinc chelator EDTA (homemade disks loaded with 10 μl of 0.5 mM solution, pH 8.0) and β-lactam molecules imipenem and ceftazidime (Bio-Rad disks each loaded with 10 μg),  and (ii) the combined disk tests (CDT) proposed by Rosco Diagnostica (kit KPC/MBL in P. aeruginosa/Acinetobacter, v2) that includes disks of imipenem, imipenem plus cloxacillin, imipenem plus EDTA, imipenem plus dipicolinic acid, and imipenem plus boronic acid. GERPA strains resistant to ceftazidime were also screened for the presence of extended-spectrum β-lactamases (ESBLs) when their culture on MHA supplemented with 2,000 μg/ml of β-lactamase PDC inhibitor cloxacillin failed to restore susceptibility to ceftazidime (Bio-Rad disk at 10 μg). DDSTs included ceftazidime (10 μg disk) and clavulanic acid (Bio-Rad disk loaded with 10 μg plus 20 μg amoxicillin), cefepime and clavulanic acid, ceftazidime and imipenem, cefepime and imipenem. In all cases, MHA plates were inoculated by swabbing with a fresh suspension of bacteria adjusted to 0.5 McFarland as recommended by the EUCAST protocol for antibiograms (https://www.eucast.org/ast_of_bacteria/), and were incubated for 18 h ± 2 h at 35°C ± 1°C. For DDSTs, enlargement of the inhibition zone around β-lactam disks of >3 mm in the direction of inhibitor was considered a potentially significant result. CDTs were interpreted according to the manufacturer’s instructions. For example, less than 5 mm of difference between the diameters of inhibitory zones around imipenem and imipenem-cloxacillin disks was evocative of carbapenemase production, while a difference of ≥5 mm between imipenem and imipenem-dipicolinic acid disks (and/or ≥8 mm between imipenem and imipenem-EDTA disks) was suggestive of metallo-β-lactamase production. Strains with positive results were next submitted to specific PCRs targeting the genes encoding ESBLs and carbapenemases as reported in P. aeruginosa. The primers and PCR conditions used are listed in Table S5. In brief, bacterial DNA was extracted with the Qiagen DNA minikit (Qiagen France, Courtaboeuf). PCR mixtures consisted of 3 μl DNA extract, 0.06 μM each primer, 10 μl MyTaq Red Mix (Bioline, France), 1.25 U MyTaq Red DNA polymerase (Bioline), 3 μl dimethyl sulfoxide (DMSO), and 27 μl distilled water in a final volume of 50 μl. Amplicons were purified with High Pure PCR Product purification kit (Roche Diagnostics, Meylan, France) and sequenced on both strands with the BigDye Terminator kit (Thermo Fisher Scientific, Illkirch-Graffenstaden, France) on a 3500 DNA sequencer (Thermo Fisher Scientific). The resultant sequences were aligned with those deposited in GenBank (https://blast.ncbi.nlm.nih.gov/Blast.cgi). Gene bla_PDC was amplified from clinical strains by PCR and sequenced as described previously (18).

Deletion of gene bla_PDC in clinical strains.

PDC-negative mutants were obtained by combining PCRs, cloning experiments, and recombination/excision events in the P. aeruginosa genome, as previously described (60, 61). Briefly, the chromosomal regions flanking gene bla_PDC-1 in strain PAO1 were amplified by PCR using the following pairs of primers, respectively: RD1/RD2 (5′-GAGGTGTCGAAGACGATGCT-3′; 5′-TCACCCGAATCTGGTATCGCGCATGA-3′) and RD3/RD4 (5′-CCAGATTCGGGTGAAGATCGCCTAC-3′; 5′-ATCCGCCTCGACGAACTG-3′). The resultant amplicons were then used as the templates to generate a single fragment by overlapping PCR with external primers RD1/RD4. This mutagenic DNA was cloned into plasmid pCR-Blunt (ccdB lacZα; Km^r) according to the manufacturer’s instructions (Thermo Fisher Scientific), and subcloned into E. coli CC118λpir on suicide vector pKNG101 (sacBR mobRK2 oriR6K; Sm^r) as an ApaI/SpeI fragment (61). The recombinant plasmid pKNGΔampC was next transferred from E. coli CC118λpir to P. aeruginosa GERPA strains by conjugation, with subsequent selection of transconjugants on Pseudomonas isolation agar (Becton, Dickinson, France) plates containing from 2,000 μg/ml to 10,000 μg/ml streptomycin, depending on the strains. Excision of the pKNG101 backbone was forced by culture on M9 minimal medium supplemented with 5% sucrose. The lack of the 1,107-bp fragment in the coding sequence of gene bla_PDC was checked by PCR amplification and sequencing experiments.

Genomic analyses.

The whole DNA of selected GERPA strains was extracted and purified with PureLink genomic DNA minikit (Thermo Fisher Scientific). This material was then sequenced by the P2M platform at the Pasteur Institute, Paris (Pasteur International Bioresources network PIBnet). In brief, Nextera XT DNA Library preparation kit (Illumina, Paris, France) was used for library construction. The pooled libraries were sequenced on an Illumina NextSeq 500 instrument using paired-end 150-bp runs. The fastq files were generated and demultiplexed with the bcl2fastq Conversion Software v2.20 (Illumina). The final average sequencing depth was 50×. The reads were assembled with Spades v3.13.0. De novo contigs were analyzed using the CARD Resistance Gene Identifier 5.1.0 available online (https://card.mcmaster.ca/analyze/rgi). Further data analyses were performed with BioNumerics 7.6.3 software (Applied Maths, bioMérieux) and CLC Genomics Workbench 10.0.1 software (Qiagen) for data extraction and sequence comparisons, respectively. In silico determination of strain serotypes was performed by uploading de novo contigs to the Center for Genomic Epidemiology (CGE) service platform (https://cge.cbs.dtu.dk/services/PAst-1.0/) (62).

Strain characterization by multilocus sequence typing.

Determination of sequence types (ST) of selected GERPA isolates was carried out according to the protocol of van Mansfield et al. (63), and by uploading the housekeeping gene sequences to the P. aeruginosa multilocus sequence typing (MLST) web site (https://pubmlst.org/paeruginosa/).

LC-MS/MS assessment of PDC and MexAB-OprM production.

(i) Bacterial cell lysis and trypsin digestion. For both assay development and strain analysis, series of 6 samples were prepared by pipetting 200 μl of thawed bacterial cell suspensions prepared in water (Abs_600nm = 1) into 1.5-ml microtubes filled with approximately 70 μl of 150 to 212 μm glass beads (Sigma-Aldrich, Saint-Quentin Fallavier, France). Fifty microliters of trypsin solution (porcine pancreas grade, Sigma-Aldrich) prepared at 1 mg/ml in 150 mM NH₄HCO₃ (Sigma-Aldrich) were added to the bacterial cell aliquots. Bacterial cell disruption and protein trypsin digestion were concomitantly carried out in a Bioruptor ultrasonicator (Diagenode, Lièges, Belgium) with 10 cycles of 30 s and 30 s of pause between 2 cycles. A heater option was incorporated into the commercial ultrasonicator so as to maintain a bath temperature of 50°C. Digestion was stopped by adding 5 μl of formic acid (Sigma-Aldrich) and the microtubes were centrifuged at 9,600 × g for 5 min. Aliquots (150 μl) of supernatants were finally transferred into screw cap tubes (2 ml, Labbox, Rungis, France) endowed with a glass insert.

(ii) Targeted mass spectrometry analysis. Multiple reaction monitoring (MRM)-based assay development and strain analysis were performed on a hybrid triple quadrupole/linear ion trap mass spectrometer 6500 QTRAP (AB Sciex, Toronto, Canada) equipped with an ESI Turbo V ion source operating at 550°C and using an ion spray voltage of 5,500V. The tandem mass spectrometer (MS/MS) was hyphenated to a liquid chromatography (LC) setup comprising two binary pumps (1290 series, Agilent technologies, Les Ulis, France) for performing on-line solid-phase extraction (SPE) prior to peptide separation. Instrument control, data acquisition, and processing were performed using Analyst 1.6.2 software. The curtain gas and nebulizing gas were adjusted to 50 and 60 lb/in², respectively. For each run, 80 μl of the trypsin digest was injected and submitted to on-line SPE on a PLRP-S, 12.5 mm × 2.1 mm column (Waters SAS, Saint Quentin-en-Yvelines, France) connected to a Rheodyne 10-injection port valve. After injection, peptides were desalted with 100% water containing 0.1% formic acid vol/vol (LC-MS grade, Fisher Scientific, Illkirch, France) for 3 min at a flow rate of 100 μl/min. After the valve switching, peptides were separated on a Waters BEH C₁₈ column (100 mm × 2.1 mm; 3.5 μm particle size) using the following gradient: 98% water solvent A (LC-MS grade, Fisher Scientific) containing 0.1% formic acid (vol/vol), and 2% acetonitrile solvent B (LC-MS grade, Fisher Scientific) containing 0.1% (vol/vol) formic acid for 3 min followed by a 13.5 min linear gradient to reach 40% of solvent B.

(iii) MRM assay development and choice of peptide surrogates. Tryptic peptide surrogates of the proteins of interest were selected combining in silico prediction and systematic analysis of P. aeruginosa reference strains. The 50S-L13 ribosomal protein (used for normalizing the cell number) and antibiotic resistance proteins, including all listed variants of PDC, MexA, MexB, OprM, MexXY, and MexY were digested in silico using Skyline software (https://skyline.ms) (64). All doubly and triply charged precursor ions for peptides containing between 6 and 25 amino acids were considered putative candidates, excluding peptide sequences containing a cysteine residue because proteins were digested without the classical reduction alkylation process. A first Skyline method was built tracking three y ions with m/z above m/z value of the precursor ion. Effective detectability of peptide surrogates was then assessed by building a second MRM method where all transitions were monitored. At this stage, theoretical optimal collision energies were predicted by Skyline equations for the 6500 QTRAP instrument. The three best responding peptide candidates were finally retained in the final assay for MexA, MexB, OprM, MexX, and MexY, while 15 peptides were necessary for ensuring PDC detection with at least 3 peptides considering the combination of variants (Table S6). Ultimately, the mass spectrometer parameters and fragment ion selection for defining the transitions were finely tuned with synthetic forms of peptides in order to warrant the optimal lowest limit of detection for each protein target. Hence, a final scheduled-MRM assay was built for monitoring 33 peptides tracked with 99 transitions with a cycle time of 0.575 s (MexXY data not shown).

(iv) Peptide and protein detection validation. MultiQuant TM 2.1 software (AB Sciex, Les Ulis, France) was used for peak detection and surface integration (MQ4 algorithm). Peak integration parameters were set as follows: a Gaussian smooth width of 2.0 point, a retention time half window of 35 s, a baseline subtraction window of 2 min and a 50% noise level for baseline. A transition value area (TVA) was calculated for each synthetic peptide detected, which served as a validation criterion for confirming peptide detection across the 84 clinical strain analyses. Only the peptides with a TVA standard deviation below 20% relative to the mean TVA were ultimately validated (Equation 1).

$$\hspace{0.17em}\mathit{TVA}\hspace{0.17em}=\frac{\mathit{Peak}\hspace{0.17em}\mathit{area}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{the}\hspace{0.17em}\mathit{most}\hspace{0.17em}\mathit{intense}\hspace{0.17em}\mathit{transition}}{\mathit{Sum}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{the}\hspace{0.17em}\mathit{peak}\hspace{0.17em}\mathit{areas}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{the}\hspace{0.17em}\mathit{three}\hspace{0.17em}\mathit{transitions}}$$ (1)

A second validation criterion was applied for confirming the protein presence based on the conserved ratio between the relative intensity of peptide surrogates of the same protein. Pairwise correlations were calculated from the ratio between the sum of transitions of peptide surrogates of each protein (Equation 2) which were detected in the 84 clinical strain analyses. It should be noted that the panel of 15 peptide surrogates enabled the detection of all PDC variants with at least three peptides. Protein presence was validated only when the relative standard deviation of the three peptide surrogate ratio (PSR) was less than 30% relative to the mean of PSR calculated from the 84 analysis.

$$\mathit{PSR}\hspace{0.17em}=\frac{\mathit{Sum}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{transition}\hspace{0.17em}\mathit{areas}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{peptide}\hspace{0.17em}x}{\mathit{Sum}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{transition}\hspace{0.17em}\mathit{areas}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{peptide}\hspace{0.17em}y}$$ (2)

Low protein expression levels could result in the detection of only two peptides out of three. Having passed the TVA-filtering criteria, the protein was considered present whether or not the two PSR fell within the confidence interval. For three strains, PDC was validated with only TSAADLLR peptide quantifier due to very weak expression but in line with the genomic analysis.

(v) Protein quantification. It is assumed that the MS1 signal and the most intense transitions of the three best flying peptides per protein correlate with protein abundance (65) and therefore can be used as a proxy for label-free quantification. To take into account the putative intrinsic decrease in sensitivity of the instrument and slight variations in bacterial suspension densities, quantification of the proteins related to resistance signals was expressed relative to ribosomal protein 50S-L13 using Equation 3.

$$\mathit{Quantification}\hspace{0.17em}=\frac{\mathit{Sum}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{the}\hspace{0.17em}3\hspace{0.17em}\mathit{transitions}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{peptide}\hspace{0.17em}\mathit{quantifier}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{resistan}\mathit{ce}\hspace{0.17em}\mathit{protein}\hspace{0.17em}}{\mathit{Sum}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{the}\hspace{0.17em}3\hspace{0.17em}\mathit{transitions}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{peptide}\hspace{0.17em}\mathit{quantifier}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{protein}\text{\hspace{0.17em}50S-L13}}$$ (3)

Data availability.

The full shotgun sequencing project has been deposited at NCBI GenBank under the accession numbers JABUGF000000000 to JABUGY000000000 and JABUOD000000000.