Synergy in the Treatment of Severe Infections Caused by Extensively Drug-Resistant and Pandrug-Resistant Gram-Negative Bacteria: From the Checkerboard to the Bedside

Abstract

Extensively drug-resistant (XDR) and pandrug-resistant (PDR) Gram-negative bacteria—principally carbapenem-resistant Acinetobacter baumannii (CRAB), carbapenem-resistant Enterobacterales (CRE), and difficult-to-treat resistant (DTR) Pseudomonas aeruginosa—account for a growing share of severe nosocomial infections and rank among the pathogens of greatest public health concern. In the 2024 World Health Organization Bacterial Priority Pathogens List, CRAB and carbapenem-resistant Enterobacterales are classified as critical-priority pathogens and carbapenem-resistant P. aeruginosa as a high-priority pathogen, whereas difficult-to-treat resistance is a clinical resistance definition and not a WHO priority category. Because monotherapy is frequently inadequate or unavailable, combination regimens remain central to bedside decision-making, and antimicrobial synergy testing has moved from a niche laboratory exercise to a question with direct clinical relevance. This narrative review synthesises recent literature identified through a structured search, the preparation of which was guided by the SANRA quality criteria for narrative review articles. The synthesis is organised across four linked domains. First, the molecular mechanisms that determine which of the following combinations are rational: carbapenemase production, efflux- and porin-mediated impermeability, lipid A modification and polymyxin resistance, and iron-transport alterations underlying emerging cefiderocol resistance. Second, the laboratory methods used to detect synergy—the checkerboard assay and fractional inhibitory concentration index, time-kill assays, and dynamic pharmacokinetic/pharmacodynamic models, including the hollow-fibre infection model—and the genotype-dependence that limits their generalisability. Third, two agents whose combination behaviour is atypical: (1) cefiderocol, whose iron-transport-dependent uptake makes its synergy testing uniquely medium-dependent and for which pooled observational data in CRAB have reported lower mortality with monotherapy than with cefiderocol-containing combination regimens, although confounding by indication limits causal interpretation and (2) intravenous fosfomycin, whose unique MurA target and the risk of rapid resistance emergence during monotherapy often favour its use in combination regimens for severe MDR/XDR infections. Fourth, the clinical evidence from the ATTACK randomised trial of sulbactam-durlobactam in CRAB to large multicentre cohorts of ceftazidime-avibactam plus aztreonam in metallo-β-lactamase-producing Enterobacterales. We conclude that in vitro synergy testing retains clear value for individualising therapy when no single active agent can be identified, but that its translation to the bedside requires genotype-aware interpretation, pharmacokinetic or pharmacodynamic context, and corroboration against comparative clinical data rather than extrapolation from checkerboard results alone.

1. Introduction

Extensively drug-resistant (XDR) and pandrug-resistant (PDR) Gram-negative bacteria have moved from laboratory curiosity to routine bedside emergency. Of the roughly 97 antibacterial agents in the global clinical pipeline as of 2023 [1], only a minority target the Gram-negative organisms carrying the greatest clinical burden, and carbapenem-resistant Enterobacterales and carbapenem-resistant Acinetobacter baumannii (CRAB) remain critical-priority pathogens on the World Health Organization’s 2024 Bacterial Priority Pathogens List, in which carbapenem-resistant P. aeruginosa is classified as a high-priority rather than a critical-priority pathogen [2]. A one-year Romanian intensive care unit (ICU) surveillance study illustrates the clinical reality behind that ranking. Carbapenem-resistant Klebsiella pneumoniae (CRKP) was the second most frequently isolated ESKAPEE organism (24.5% of isolates), and a substantial share of critical-priority isolates recovered from mechanically ventilated patients met XDR or even PDR criteria [3].

Three overlapping clinical problems dominate this space: CRAB, carbapenem-resistant Enterobacterales (CRE, including metallo-β-lactamase [MBL] producers), and difficult-to-treat resistant (DTR) Pseudomonas aeruginosa. Resistance gene acquisition, rather than any single virulence trait, is the common thread linking them, and the therapeutic armamentarium remains narrow and organism-specific [4]. When an isolate resists essentially every antibiotic class with a favourable toxicity profile, three imperfect options remain: a relatively toxic last-resort agent (usually a polymyxin), a novel narrow-spectrum agent whose activity may not extend to the isolate’s specific mechanism, or a combination regimen chosen empirically or guided by in vitro synergy data. Throughout this review, the relevant terms are kept distinct and are used in their established senses. Multidrug-resistant (MDR), XDR, and PDR are phenotypic categories defined by the number of antimicrobial classes to which an isolate is non-susceptible; DTR is a clinical resistance definition applied here specifically to P. aeruginosa; and CRAB, CRE, CRKP, and carbapenem-resistant Gram-negative bacteria (CR-GNB) denote organism- or mechanism-defined groups rather than resistance phenotypes.

This review addresses the third option. Section 2 sets out the search strategy and selection criteria underlying the synthesis.

2. Materials and Methods

2.1. Rationale for a Narrative Rather than Systematic Design

We chose a narrative synthesis deliberately. The question addressed here—when, if ever, is combination therapy justified for severe MDR Gram-negative infection—spans four distinct evidence domains that cannot be pooled: pharmacodynamic theory, in vitro microbiology, pharmacokinetic modelling, and clinical outcome studies. Quantitative synthesis has already been undertaken for specific questions within this field—meta-analyses of in vitro combination efficacy against carbapenem-resistant Gram-negative bacilli [5] and of ceftazidime-avibactam-based combinations specifically [6], and a systematic review and meta-analysis of clinical outcomes with ceftazidime-avibactam in bacteraemia and pneumonia [7]—and a further pooled analysis would duplicate that existing work while excluding precisely the mechanistic and methodological material that explains why such meta-analyses can disagree with one another. A narrative structure also allows explicit discussion of methodological quality and of the assumptions built into laboratory assays, both of which a single pooled effect estimate necessarily conceals. We nonetheless applied a structured search, summarised in Section 2.2, Section 2.3, Section 2.4 and Section 2.5. SANRA (Scale for the Assessment of Narrative Review Articles) is a scale for the quality assessment of narrative review articles rather than a reporting guideline; its criteria were therefore used to guide the preparation of this review and are not presented as a reporting framework [8].

2.2. Data Sources

We searched PubMed/MEDLINE, Embase, Scopus and the Cochrane Central Register of Controlled Trials from database inception to 20 July 2026. We additionally searched ClinicalTrials.gov and the EU Clinical Trials Register for completed and ongoing trials of combination regimens, and screened conference proceedings of ECCMID/ESCMID Global and IDWeek for the years 2020–2025 to capture data not yet available in full-text form. Reference lists of all included reviews, guidelines, and primary studies were hand-searched (backward citation tracking), and forward citation tracking of the pivotal randomised trials and of the key methodological papers was performed in Scopus.

2.3. Search Strategy and Keywords

The search combined four conceptual blocks—organism, resistance phenotype, intervention, and synergy/interaction—using Boolean operators, with MeSH/Emtree terms exploded and combined with free-text terms in title/abstract fields. The following keyword strings were used, adapted syntactically for each database. They represent the conceptual blocks of the search rather than the complete database-specific syntax, which was not retained, and the search should be regarded as structured rather than formally reproducible:

Block 1—Organism: “Gram-negative bacteria” [MeSH] OR Enterobacterales OR Klebsiella OR “Escherichia coli” OR “Acinetobacter baumannii” OR “Pseudomonas aeruginosa” OR “Stenotrophomonas maltophilia”

Block 2—Resistance phenotype: “drug resistance, multiple, bacterial” [MeSH] OR multidrug-resistant OR MDR OR “extensively drug-resistant” OR XDR OR pandrug-resistant OR PDR OR “difficult-to-treat resistance” OR DTR OR carbapenem-resistant OR carbapenemase OR KPC OR OXA-48 OR NDM OR VIM OR IMP OR “metallo-beta-lactamase” OR “extended-spectrum beta-lactamase” OR ESBL OR colistin-resistant

Block 3—Intervention: “drug therapy, combination” [MeSH] OR combination therapy OR combination regimen OR dual therapy OR triple therapy OR monotherapy OR colistin OR polymyxin OR meropenem OR carbapenem OR sulbactam OR ampicillin-sulbactam OR sulbactam-durlobactam OR cefiderocol OR ceftazidime-avibactam OR meropenem-vaborbactam OR imipenem-relebactam OR ceftolozane-tazobactam OR aztreonam-avibactam OR cefepime-enmetazobactam OR fosfomycin OR tigecycline OR eravacycline OR rifampicin OR aminoglycoside

Block 4—Synergy/interaction: “drug synergism” [MeSH] OR synerg OR antagonis OR additiv OR indifferen OR “fractional inhibitory concentration” OR FICI OR FIC index OR checkerboard OR chequerboard OR “time-kill” OR “time kill” OR isobologram OR “Loewe additivity” OR “Bliss independence” OR “hollow fibre” OR “hollow fiber” OR “dynamic model” OR heteroresistance

Blocks were combined as (Block 1) AND (Block 2) AND (Block 3) AND (Block 4). A parallel, narrower search omitting Block 4 was run and restricted to randomised controlled trials, controlled clinical trials and practice guidelines, to capture clinical outcome studies that compare monotherapy with combination therapy without using synergy terminology—an important source of retrieval bias in this field, since the pivotal negative trials do not describe themselves as synergy studies.

2.4. Eligibility Criteria

Included: randomised controlled trials, prospective and retrospective comparative cohort studies, systematic reviews, and meta-analyses, in vitro and in vivo studies addressing mechanisms or methods of synergy assessment, pharmacokinetic/pharmacodynamic analyses, and society guidelines or guidance documents. Publications in English, French, Spanish, Italian or Romanian were eligible.

Excluded: single case reports without mechanistic novelty; studies exclusively addressing Gram-positive organisms, mycobacteria or fungi; studies of non-antibacterial adjuncts (e.g., immunomodulators) except where directly relevant to combination pharmacology; conference abstracts superseded by full publication; and articles for which the full text could not be obtained. No French-, Spanish-, Italian- or Romanian-language records met the eligibility criteria above after screening. The final reference list is accordingly composed entirely of English-language publications.

2.5. Study Selection and Data Handling

Titles and abstracts were screened independently by two authors (M.-E.V. and O.-E.I.). Full-text assessment of all potentially eligible records was performed by the same two authors. Data extraction was performed by M.-E.V. and verified by N.-M.V. and P.-A.D., and disagreements at any stage were resolved by discussion with a third author (R.-C.C.). Extracted data were recorded in a standardised spreadsheet capturing study design, population, pathogen and resistance mechanism, comparator, dosing regimen, definition of synergy where applicable, outcome definitions and effect estimates. The number of records identified, de-duplicated, screened at title and abstract level, assessed in full text and finally included was not recorded, and no PRISMA-style flow diagram was therefore constructed. The search is accordingly described as structured rather than systematic, and it is not reproducible in the quantitative sense required of a systematic review. Priority was given to randomised evidence for clinical questions and to mechanism-defined studies for microbiological questions. Where randomised evidence was absent, we preferentially cited multicentre cohorts with explicit adjustment for confounding by indication, and such studies are identified as observational throughout the text.

2.6. Assessment of Evidence and Limitations of the Method

We did not perform formal risk-of-bias scoring or GRADE assessment, and no protocol was registered. Both are acknowledged limitations. Narrative reviews are inherently susceptible to selection and interpretation bias, and the search—although broad—will have missed grey literature and non-indexed regional publications, which is relevant given the geographic concentration of PDR Gram-negative infections in settings that publish less in indexed journals. Publication bias in the in vitro synergy literature is also likely to be substantial and directional. Studies demonstrating synergy are far more likely to be submitted and accepted than studies demonstrating indifference, so the apparent frequency of synergy in the published literature almost certainly exceeds its true frequency. The randomised clinical evidence discussed below is less susceptible to this particular distortion and should be weighted accordingly.

3. Molecular Basis of the XDR/PDR Phenotype

Predicting whether a combination will succeed against a particular isolate requires knowing which resistance mechanism is operating. Four broad classes account for most XDR/PDR phenotypes, and they frequently co-exist within a single strain (Table 1).

Carbapenemase production is, among the mechanisms considered here, a major determinant of the therapeutic options that remain. In A. baumannii, class D oxacillinases (OXA-23-like, OXA-24/40-like, OXA-58-like, and up-regulated intrinsic OXA-51-like) predominate, compounded by CarO porin loss and efflux overexpression, with outer-membrane vesicles and transduction disseminating determinants between strains [9]. Among Enterobacterales, clonal and plasmid epidemiology itself shapes the landscape. Clone ST258 maintained KPC carbapenemases through recombination and remained hospital-endemic, whereas ST307 achieved global spread through plasmid-mediated acquisition of KPC, NDM, and OXA-48-like enzymes—two evolutionary routes converging on the same XDR endpoint [10]. Carbapenemase genes were detectable in 42.9% of Gram-negative ICU isolates in one Saudi survey, with K. pneumoniae far more likely than A. baumannii to carry one (odds ratio 4.23 versus 0.20) [11]. The same convergence appears outside the classical triad, in Proteus species [12] and the Citrobacter freundii complex [13].

Non-enzymatic mechanisms matter because they defeat combinations that enzymatic reasoning predicts should work. Efflux pumps extrude both partner agents simultaneously. A plasmid-borne efflux gene cluster (tmexCD-toprJ) conferring tigecycline resistance was recently found co-resident with blaNDM-1, blaNDM-5, blaIMP-4 and blaOXA-1 on the same plasmid backbone, genetically linking efflux and enzymatic resistance within one strain [14]. In P. aeruginosa, resistance to the newer β-lactams arises through AmpC Ω-loop mutations, acquired OXA and GES/KPC enzymes, and efflux overexpression that removes the β-lactam and its partner inhibitor together, while PBP3 and Piu/Pir iron-transport mutations drive emerging cefiderocol resistance [15]. The organism is better understood as a dynamically regulated resistance–persistence network than as a fixed resistant phenotype [16].

Polymyxin resistance deserves separate emphasis because colistin underpins most salvage combinations. Two convergent routes operate: chromosomal mutations modifying lipid A, which add cationic groups that neutralise the negative surface charge colistin exploits, and horizontally transferable mobile colistin resistance (mcr) genes [17]. In A. baumannii, PmrCAB- and LpxL-mediated lipid A modification simultaneously blunts Toll-like receptor 4 immune recognition, compounding the clinical challenge [18]. Regionally, mcr-mediated plasmid resistance now exceeds chromosomal mutation among colistin-resistant Escherichia coli in the Middle East [19].

Iron acquisition underlies both cefiderocol’s design and its erosion. Siderophore–antibiotic conjugates hijack TonB-dependent transporters to cross the outer membrane; resistance now emerges through cirA transporter mutations, envZ-ompR and baeSR regulatory changes, and β-lactamase gene amplification [20]. Treatment-emergent resistance is documented for essentially every novel agent introduced over the past decade—KPC Ω-loop mutations such as KPC-31 for ceftazidime-avibactam, PBP3, and CMY changes for aztreonam-avibactam in NDM-producing E. coli, and iron-transporter defects for cefiderocol [21].

Table 1. Molecular resistance mechanisms underlying the XDR/PDR phenotype in Gram-negative bacteria: molecular basis, organisms chiefly affected and consequences for the choice of combination partner.

Mechanism Class	Molecular Basis	Principal Organisms	Implication for Combination Choice	Refs.
Carbapenemase production	Class A (KPC, GES), class B metallo-β-lactamases (NDM, VIM, IMP), class D oxacillinases (OXA-23/24-40/58/51-like); plasmid- and clone-mediated dissemination	K. pneumoniae, A. baumannii, P. aeruginosa, Proteus spp., C. freundii	Determines which β-lactamase inhibitor is useful. Avibactam, vaborbactam, and relebactam are inactive against MBLs; aztreonam-based or siderophore-based strategies are the principal remaining options, the choice depending on the organism, susceptibility profile and drug availability.	[9,10,11,12,13]
Efflux overexpression	RND-family pumps (AdeABC, MexAB-OprM); plasmid-borne tmexCD-toprJ co-resident with carbapenemase genes	A. baumannii, P. aeruginosa, Klebsiella spp.	Both partner agents may be extruded at once. Combinations that enzymatic reasoning predicts should work may therefore fail. Efflux is frequently linked genetically to carbapenemase carriage.	[14,15]
Porin loss/reduced permeability	CarO porin loss or downregulation; intrinsically low outer-membrane permeability	A. baumannii, P. aeruginosa	Provides the rationale for permeabilisation-based combinations using a polymyxin as entry facilitator	[9,16]
Target-site mutation	AmpC Ω-loop mutations; PBP3 substitutions; KPC Ω-loop variants (e.g., KPC-31)	P. aeruginosa, NDM-producing E. coli, K. pneumoniae	Drives treatment-emergent resistance to ceftazidime-avibactam and aztreonam-avibactam; argues for combination as resistance suppression	[15,21]
Lipid A modification (polymyxin resistance)	Chromosomal pmrCAB, phoPQ, lpxL mutations adding cationic groups to lipid A; horizontally transferable mcr genes	A. baumannii, K. pneumoniae, E. coli	Undermines the most broadly active salvage class; lipid A modification also blunts TLR4-mediated immune recognition	[17,18,19]
Iron-transport alteration (cefiderocol resistance)	cirA and other TonB-dependent transporter mutations; envZ-ompR and baeSR regulatory changes; β-lactamase gene amplification	K. pneumoniae, NDM-producing Enterobacterales, P. aeruginosa	Erodes the principal remaining option for MBL producers; supports combination or rescue pairings when cefiderocol MIC rises on therapy	[20,21]

Table 1. Molecular resistance mechanisms underlying the XDR/PDR phenotype in Gram-negative bacteria: molecular basis, organisms chiefly affected and consequences for the choice of combination partner.

Mechanism Class	Molecular Basis	Principal Organisms	Implication for Combination Choice	Refs.
Carbapenemase production	Class A (KPC, GES), class B metallo-β-lactamases (NDM, VIM, IMP), class D oxacillinases (OXA-23/24-40/58/51-like); plasmid- and clone-mediated dissemination	K. pneumoniae, A. baumannii, P. aeruginosa, Proteus spp., C. freundii	Determines which β-lactamase inhibitor is useful. Avibactam, vaborbactam, and relebactam are inactive against MBLs; aztreonam-based or siderophore-based strategies are the principal remaining options, the choice depending on the organism, susceptibility profile and drug availability.	[9,10,11,12,13]
Efflux overexpression	RND-family pumps (AdeABC, MexAB-OprM); plasmid-borne tmexCD-toprJ co-resident with carbapenemase genes	A. baumannii, P. aeruginosa, Klebsiella spp.	Both partner agents may be extruded at once. Combinations that enzymatic reasoning predicts should work may therefore fail. Efflux is frequently linked genetically to carbapenemase carriage.	[14,15]
Porin loss/reduced permeability	CarO porin loss or downregulation; intrinsically low outer-membrane permeability	A. baumannii, P. aeruginosa	Provides the rationale for permeabilisation-based combinations using a polymyxin as entry facilitator	[9,16]
Target-site mutation	AmpC Ω-loop mutations; PBP3 substitutions; KPC Ω-loop variants (e.g., KPC-31)	P. aeruginosa, NDM-producing E. coli, K. pneumoniae	Drives treatment-emergent resistance to ceftazidime-avibactam and aztreonam-avibactam; argues for combination as resistance suppression	[15,21]
Lipid A modification (polymyxin resistance)	Chromosomal pmrCAB, phoPQ, lpxL mutations adding cationic groups to lipid A; horizontally transferable mcr genes	A. baumannii, K. pneumoniae, E. coli	Undermines the most broadly active salvage class; lipid A modification also blunts TLR4-mediated immune recognition	[17,18,19]
Iron-transport alteration (cefiderocol resistance)	cirA and other TonB-dependent transporter mutations; envZ-ompR and baeSR regulatory changes; β-lactamase gene amplification	K. pneumoniae, NDM-producing Enterobacterales, P. aeruginosa	Erodes the principal remaining option for MBL producers; supports combination or rescue pairings when cefiderocol MIC rises on therapy	[20,21]

4. Why Combine? Four Distinct Mechanistic Rationales

Combination therapy is not a single strategy but at least four distinct strategies, and identifying which one applies is the best available predictor of whether a pairing will work against a given genotype (Table 2). Target redundancy pairs agents binding different penicillin-binding proteins. Membrane permeabilisation uses a polymyxin to open the outer membrane for a partner drug otherwise excluded. Enzymatic protection pairs a hydrolysis-stable agent with an inhibitor of the co-produced enzyme that would otherwise inactivate it. Resistance-suppression combinations aim less at immediate killing than at preventing the emergence of resistant subpopulations during therapy.

Table 2. Four mechanistic rationales for antimicrobial combination therapy against XDR/PDR Gram-negative bacteria: underlying principle, representative regimens, and genotype dependence.

Rationale	Underlying Principle	Representative Combinations	Genotype Dependence	Refs.
Target redundancy	Two agents bind different penicillin-binding proteins, so that neither alone achieves bactericidal activity but the pair does.	Sulbactam-durlobactam + imipenem; dual carbapenem regimens	Moderate; depends on PBP3 mutational status	[22,23]
Membrane permeabilisation	A polymyxin disrupts the LPS-rich outer membrane, admitting a partner drug otherwise excluded.	Colistin or polymyxin B + meropenem, tigecycline, eravacycline, rifampicin, fosfomycin, fusidic acid	Low; synergy observed with structurally unrelated partners, including some agents with little intrinsic Gram-negative activity; effect varies by isolate and regimen	[24,25,26,27]
Enzymatic protection	A hydrolysis-stable agent is paired with an inhibitor of the co-produced enzyme that would otherwise inactivate it.	Ceftazidime-avibactam + aztreonam; aztreonam-avibactam (fixed dose)	High; effective specifically against MBL producers co-expressing a serine β-lactamase, indifferent otherwise	[28,29,30]
Resistance suppression	The combination prevents outgrowth of resistant subpopulations rather than increasing immediate killing.	Meropenem + fosfomycin; phage + meropenem; ceftazidime-avibactam + fosfomycin	Variable; demonstrated for OXA-48-carrying and KPC-producing strains	[31,32,33]

Recognising which rationale is operating explains several otherwise puzzling observations. Colistin-based combinations have shown in vitro synergy with a range of partner agents, including some agents with little intrinsic activity against Gram-negative bacteria, such as fusidic acid [26] or plant-derived phenols [27], consistent with a largely non-specific membrane-permeabilisation mechanism rather than anything particular to the partner drug. However, the effect varies by isolate and regimen, and in vitro synergy does not necessarily translate into clinical benefit. Enzymatic-protection combinations, such as ceftazidime-avibactam plus aztreonam, behave quite differently. They are exquisitely genotype-specific, showing synergy against MBL producers and indifference against others tested in the same isolate collection [28].

5. Laboratory Synergy Testing

5.1. Methods and Their Limits

Three methodological tiers are now in routine research use. They differ in what they measure and in how closely they approximate clinical exposure (Table 3). The checkerboard assay screens two-dimensional dilution combinations and yields a fractional inhibitory concentration index (FICI), conventionally interpreted as indicating synergy at FICI ≤ 0.5, indifference at 0.5 < FICI ≤ 4, and antagonism at FICI > 4. Time-kill assays track viable counts over time and can reveal suppression of regrowth that a single-timepoint checkerboard readout miss. Dynamic pharmacokinetic/pharmacodynamic (PK/PD) models, particularly the hollow-fibre infection model, simulate fluctuating human drug concentrations and can incorporate carbapenemase-mediated drug degradation explicitly—a methodological refinement that materially changes results, as shown for meropenem plus fosfomycin against carbapenemase-producing K. pneumoniae [31].

Table 3. Laboratory methods for antimicrobial synergy testing: endpoint measured, principal strengths, key limitations, and representative references.

Method (References)	Endpoint Measured	Principal Strengths	Key Limitations
Checkerboard assay (FICI) [5,6,28]	Fractional inhibitory concentration index at a single 18–24 h growth-inhibition endpoint. Synergy: FICI ≤ 0.5; indifference: 0.5 < FICI ≤ 4; antagonism: FICI > 4	Inexpensive; high throughput; screens many drug pairs against many isolates; the most widely reported synergy metric	Static concentrations; growth-inhibition rather than killing endpoint; no information on regrowth or resistance emergence; poor correlation with clinical outcome
Time-kill assay [6,31,32]	Viable count over 24 h; synergy conventionally ≥ 2-log10 CFU reduction versus the most active single agent	Dynamic killing kinetics; reveals suppression of regrowth that FICI misses; can be extended to non-antibiotic partners such as phage	Labour-intensive; still uses static drug concentrations; carbapenemase-mediated degradation of the partner drug during incubation may distort results unless modelled
Hollow-fibre/dynamic PK/PD model [31,34]	Bacterial response to simulated fluctuating human drug exposures over days	Reproduces clinically achievable concentration-time profiles; can incorporate enzymatic drug degradation; supports dose-regimen selection	Costly and low throughput; no host immune component; requires validated PK input for each agent
Mechanism-based PK/PD modelling [35,36]	Simulated probability of target attainment and synergy across dosing regimens and genotypes	Translates in vitro data directly into candidate clinical regimens; can span many isolate genotypes	Model assumptions dominate output; requires high-quality in vitro input data; not a substitute for clinical validation

A systematic review and isolate-level meta-analysis of 45 in vitro studies (734 isolates) formally applied the FICI ≤ 0.5 threshold alongside a parallel time-kill definition (≥2-log₁₀ reduction at 24 h versus the most active single agent), explicitly treating the two as complementary rather than interchangeable, and found ceftazidime-avibactam plus aztreonam highly synergistic specifically against carbapenem-resistant K. pneumoniae and E. coli [6]. Mechanism-based PK/PD modelling has taken this further, translating static findings into simulated dosing regimens: polymyxin B combined with meropenem and fosfomycin achieved mechanistic synergy in 81–98% of simulations across six genotypically diverse carbapenem-resistant K. pneumoniae isolates [35].

The interpretation of checkerboard-derived FICI values nevertheless requires caution. The FICI ≤ 0.5 threshold is conventional rather than standardised, and universally accepted interpretive criteria do not exist. Results are sensitive to inoculum size, to the antimicrobial concentration ranges and dilution steps tested, to incubation conditions and duration, and to the endpoint used to define growth inhibition, so that nominally identical experiments performed in different laboratories are not always comparable. Checkerboard and time-kill assays also answer different questions and agree only in part, since a growth-inhibition endpoint read at 18–24 h cannot capture the killing kinetics, regrowth and selection of resistant subpopulations that time-kill assays reveal. The correlation between either assay and patient outcome is weaker still, as the randomised data discussed in Section 9 show [5,6,37,38].

5.2. What the In Vitro Data Actually Show

Two patterns recur across the studies summarised in Table 4. First, synergy is strongly genotype-dependent rather than species-dependent. Against 58 XDR and PDR carbapenemase-producing K. pneumoniae and A. baumannii isolates, colistin-meropenem showed synergy in only 35.5–40.7% of isolates, and higher carbapenemase gene expression was independently associated with lower synergy rates [39]. Ceftazidime-avibactam plus meropenem was synergistic in 91–100% of KPC-producing or non-carbapenemase-producing strains, while ceftazidime-avibactam plus aztreonam was synergistic specifically in MBL producers, and combinations with eravacycline, amikacin or colistin were largely indifferent—all within the same 59-isolate collection [28].

Table 4. Representative in vitro synergy profiles of combination therapies against XDR/PDR Gram-negative bacteria, grouped by combination partner.

Combination	Organism/Genotype	Principal Finding	Refs.
Colistin + meropenem	58 XDR/PDR carbapenemase-producing K. pneumoniae and A. baumannii	Synergy in 35.5–40.7% and additive effect in 35–37% of isolates; no antagonism. Higher carbapenemase expression was independently associated with lower synergy rates	[39]
Polymyxin B + tigecycline, ceftazidime, cefepime or imipenem	10 KPC-producing CRKP, including polymyxin-B-resistant strains	Synergy in 7–9 of 10 isolates depending on partner; concentration-dependent and demonstrable at clinically attainable concentrations	[24]
Polymyxin B + tigecycline or eravacycline	40 KPC- and MBL-producing CRKP	Synergy in 35–40% overall, preserved at 70.6% among polymyxin-B-resistant isolates and significantly higher than with ceftazidime-avibactam pairings	[25]
Eravacycline + polymyxin B	42 ST11 K. pneumoniae co-producing KPC-2 and NDM-1	Time-kill synergy in dual-resistant isolates; combinations with ceftazidime-avibactam, meropenem or amikacin only additive. Superior survival confirmed in a Galleria mellonella model	[40]
Ceftazidime-avibactam + meropenem or aztreonam	59 carbapenem-resistant K. pneumoniae of varied genotype	Synergy in 91–100% of KPC-producing or non-carbapenemase-producing strains with meropenem; synergy with aztreonam confined to MBL producers; pairings with eravacycline, amikacin or colistin largely indifferent	[28]
Ceftazidime-avibactam + colistin or fosfomycin	OXA-48- and KPC-producing CRKP	Strongest synergy with colistin (78.2%) and fosfomycin (63.6%)	[41]
Ceftazidime-avibactam + aztreonam	150 CRKP of varied carbapenemase genotype	Synergy strongest in NDM-carrying strains and accompanied by reduced expression of biofilm-related genes	[29]
Fosfomycin + ceftazidime-avibactam or colistin	42 MBL-producing K. pneumoniae and P. aeruginosa	Synergy in 72.7% (ceftazidime-avibactam) and 65% (colistin); time-kill assays confirmed suppression of the regrowth seen with monotherapy	[33]
Minocycline + biapenem	23 cefiderocol-resistant NDM-producing Enterobacterales	Synergy in 22/23 isolates (median FICI 0.063) with rapid bactericidal activity; proposed as rescue therapy after cefiderocol failure	[42]
Meropenem + fosfomycin	KPC-2/KPC-3-producing CRKP	Synergy confirmed in both static time-kill and hollow-fibre models when carbapenemase-driven meropenem degradation was explicitly modelled	[31]
Polymyxin B + meropenem ± fosfomycin	6 genotypically diverse carbapenem-resistant K. pneumoniae	Mechanistic synergy in 81–98% of simulations; findings translated into candidate clinical dosing regimens	[35]
Colistin + fusidic acid	E. coli co-harbouring mcr-1 and blaNDM-5	Colistin lowered fusidic acid MIC > 1000-fold; ≥3-log10 CFU reduction confirmed in a murine thigh-infection model	[26]
Colistin + eugenol, cinnamaldehyde or carvacrol	Sequenced CRAB and carbapenemase-producing Enterobacterales	Synergy (FICI < 0.5) in 37.5–50% of strains with colistin but near-total indifference with β-lactams or gentamicin, confirming permeabilisation as the operative mechanism	[27]
Lytic phage + meropenem	12 K. pneumoniae of diverse carbapenemase genotype	Combination suppressed emergence of resistant mutants specifically in OXA-48/pOXA-48-carrying isolates	[32]

Polymyxins, by contrast, frequently demonstrate in vitro synergy with partner agents of divergent mechanism, although the effect is isolate- and regimen-dependent and has not translated consistently into clinical benefit. Tigecycline and eravacycline combined with polymyxin B achieved synergy in 35–40% of carbapenemase-producing CRKP isolates overall, a rate preserved at 70.6% even among polymyxin-B-resistant isolates and significantly higher than that achieved with ceftazidime-avibactam combinations [25]. Polymyxin B paired with tigecycline, ceftazidime, cefepime or imipenem showed synergy in 7–9 of 10 KPC-producing isolates at clinically attainable concentrations [24]. This reflects the non-specific permeabilisation mechanism described in Section 4, and it is why the polymyxin class remains central to salvage regimens despite its toxicity.

Checkerboard methodology has also been applied to isolates against which the first-line novel agents are no longer active. Against 23 cefiderocol-resistant NDM-producing Enterobacterales, minocycline plus biapenem showed synergy in 22 of 23 isolates (median FICI 0.063) with time-kill-confirmed bactericidal activity, proposed as a rescue option specifically for cefiderocol failure [42]. Fosfomycin combined with ceftazidime-avibactam achieved synergy in 72.7% of MBL-producing K. pneumoniae and P. aeruginosa isolates, with time-kill assays confirming suppression of the regrowth seen with either agent alone [33]. The same time-kill methodology has been extended to non-traditional partners, and lytic phage combined with meropenem suppressed resistant-mutant emergence specifically in OXA-48-carrying K. pneumoniae isolates [32].

6. Cefiderocol-Based Combinations

Cefiderocol is transported across the outer membrane by TonB-dependent iron transporters rather than diffusing through porins, and this single fact has three consequences for combination testing that apply to no other agent (Table 5).

First, its measured activity is medium-dependent. Reference susceptibility testing is performed in iron-depleted cation-adjusted Mueller-Hinton broth, the standardised condition that maximises siderophore-mediated uptake. Activity measured under these conditions may therefore not be reproduced wherever iron is more freely available. Among 238 MDR and XDR K. pneumoniae isolates, cefiderocol susceptibility was 97.9% under iron-depleted conditions, but activity was lost in iron-enriched medium; adding avibactam significantly restored MIC reduction even under iron-replete conditions, and a triple combination, including the outer-membrane permeabiliser polymyxin B nonapeptide reduced MICs further [43]. This observation is mechanistically important, but its extrapolation to the bedside should be cautious. Iron availability in vivo is complex and is governed by host iron sequestration and nutritional immunity, so iron-replete broth cannot be treated as a straightforward surrogate for conditions at an infected site; whether iron-rich foci such as abscesses or haematomas reproduce the behaviour observed in iron-enriched medium has not been demonstrated directly.

Second, which partner helps depends on whether the limiting step is enzymatic hydrolysis or uptake. Against 39 characterised carbapenemase producers, adding avibactam, vaborbactam or relebactam reduced cefiderocol MICs 4- to 256-fold in KPC producers and restored susceptibility in 10 of 12 cefiderocol-resistant KPC-producing K. pneumoniae, yet conferred no benefit against OXA-23 producers or most MBL producers [44]. In A. baumannii the picture is different again, and cefiderocol alone was active against only 66.7% of 21 XDR/PDR isolates, but was synergistic with ceftazidime-avibactam in 100% and with sulbactam-durlobactam in 95.2% [45], a finding mirrored from the sulbactam-durlobactam side in a companion study [46]. Non-β-lactam partners also perform respectably, and time-kill testing of 40 ICU P. aeruginosa isolates found the strongest synergy with eravacycline, meropenem, and levofloxacin, with no antagonism observed [47]. Confirmation that the uptake route itself drives the effect comes from keratitis isolates, where cefiderocol plus moxifloxacin was synergistic in 7 of 7 strains, while ceftazidime—a non-siderophore comparator—plus moxifloxacin performed distinctly worse [48].

Third, and least often acknowledged, the same combinations can select for resistance. Exposure of carbapenem-resistant Acinetobacter spp. to cefiderocol with either sulbactam or avibactam selected cefiderocol-resistant variants; whole-genome and transcriptomic analysis identified mutations in efflux pumps, transcriptional regulators and iron-homeostasis genes, with altered expression of outer-membrane proteins and β-lactamases [49]. Antagonism, not merely absence of synergy, is a documented and mechanistically explicable outcome. This sits alongside the high prevalence of cefiderocol heteroresistance, particularly in Acinetobacter spp. [50]—although reduced-susceptibility mutants carrying baeS and fepA lesions have been shown to remain responsive to combination [51].

Table 5. Cefiderocol-based combinations: in vitro synergy, antagonism and the mechanistic reasons why cefiderocol behaves unlike other β-lactams in synergy testing.

Partner Agent	Organism/Genotype	Principal Finding	Refs.
Ceftazidime-avibactam; sulbactam-durlobactam	21 XDR/PDR A. baumannii + 1 NDM-1 strain	Cefiderocol alone active against only 66.7% of strains, but synergistic with ceftazidime-avibactam in 100% and sulbactam-durlobactam in 95.2%; amikacin, doxycycline and sulbactam synergistic in >50%. Combinations also lowered cefiderocol MIC against the NDM-1 producer	[45]
Cefiderocol (from the sulbactam-durlobactam side)	22 A. baumannii (21 XDR/PDR, 1 MBL producer)	Sulbactam-durlobactam synergistic with cefiderocol, cefepime, meropenem, ceftazidime-avibactam and piperacillin-tazobactam in ≥95% of strains, with rapid and sustained bactericidal killing in time-kill assays	[46]
Avibactam, vaborbactam, relebactam, tazobactam	39 characterised carbapenemase producers	4- to 256-fold cefiderocol MIC reduction in KPC producers; susceptibility restored (MIC ≤ 2 mg/L) in 10/12 cefiderocol-resistant KPC-producing K. pneumoniae. Synergy also seen for OXA-48-like producers but absent for OXA-23 and most MBL producers	[44]
Fosfomycin; ceftazidime-avibactam; ampicillin-sulbactam	Genomically characterised CRE, CRAB and CR-P. aeruginosa	Cefiderocol + fosfomycin and cefiderocol + ceftazidime-avibactam synergistic against strains already resistant to cefiderocol; cefiderocol + ampicillin-sulbactam active against an ampicillin-sulbactam-resistant P. aeruginosa isolate	[52]
β-lactamase inhibitors ± polymyxin B nonapeptide	238 MDR/XDR K. pneumoniae, incl. dual carbapenemase producers	Cefiderocol susceptibility 97.9% under iron-depleted conditions but activity lost in iron-enriched medium. Avibactam significantly reduced MIC even in iron-replete conditions; adding an outer-membrane permeabiliser reduced it further	[43]
Eravacycline, meropenem, levofloxacin, colistin, tobramycin, ceftazidime-avibactam	40 ICU P. aeruginosa isolates, carbapenemase-genotyped	Time-kill at 1×MIC/24 h: strongest synergy with eravacycline, meropenem, and levofloxacin (4/8 strains each); colistin, tobramycin, and ceftazidime-avibactam synergistic in 3/8. No antagonism observed	[47]
Imipenem-relebactam	3 MBL-producing DTR P. aeruginosa (2 IMP-13, 1 VIM-2)	Etest FICI showed synergy in both IMP-13 producers but not in the VIM-2 producer; all three patients were successfully treated with the combination—a rare direct FICI-to-outcome pairing showing MBL-subtype dependence	[53]
Ceftazidime-avibactam (mechanism-based PK/PD)	3 sequential carbapenem-resistant Achromobacter xylosoxidans isolates	Combination achieved ≥ 2-log10 CFU reduction; the model required two subpopulations (one ceftazidime-resistant/cefiderocol-susceptible, one resistant to both), with infusion duration the critical driver at clinical exposures	[36]
Moxifloxacin (ceftazidime as non-siderophore comparator)	P. aeruginosa keratitis isolates incl. XDR 2023 outbreak strain	Time-kill synergy in 7/7 isolates; Etest additivity/synergy in 79%, falling to 60% in quorum-sensing-defective strains, whereas ceftazidime + moxifloxacin performed distinctly worse—evidence that synergy is uptake-route-dependent	[48]
Polymyxin B	P. aeruginosa keratitis isolates and reduced-susceptibility mutants	Synergy for two XDR outbreak isolates but only additive-to-indifferent across the wider panel. Mutants with reduced cefiderocol susceptibility carrying baeS and fepA mutations remained responsive to the combination	[51]
Bacteriophage	Gram-negative aerobic bacilli, planktonic, and biofilm	Phage partner produced a 2- to 64-fold cefiderocol MIC reduction with complete biofilm eradication in some isolates (isothermal microcalorimetry)	[54]
Sulbactam or avibactam (antagonism)	Carbapenem-resistant Acinetobacter spp.	Exposure to cefiderocol plus sulbactam or avibactam selected cefiderocol-resistant variants; WGS and transcriptomics identified mutations in efflux pumps, transcriptional regulators, and iron-homeostasis genes with altered outer-membrane-protein and β-lactamase expression	[49]

Table 5. Cefiderocol-based combinations: in vitro synergy, antagonism and the mechanistic reasons why cefiderocol behaves unlike other β-lactams in synergy testing.

Partner Agent	Organism/Genotype	Principal Finding	Refs.
Ceftazidime-avibactam; sulbactam-durlobactam	21 XDR/PDR A. baumannii + 1 NDM-1 strain	Cefiderocol alone active against only 66.7% of strains, but synergistic with ceftazidime-avibactam in 100% and sulbactam-durlobactam in 95.2%; amikacin, doxycycline and sulbactam synergistic in >50%. Combinations also lowered cefiderocol MIC against the NDM-1 producer	[45]
Cefiderocol (from the sulbactam-durlobactam side)	22 A. baumannii (21 XDR/PDR, 1 MBL producer)	Sulbactam-durlobactam synergistic with cefiderocol, cefepime, meropenem, ceftazidime-avibactam and piperacillin-tazobactam in ≥95% of strains, with rapid and sustained bactericidal killing in time-kill assays	[46]
Avibactam, vaborbactam, relebactam, tazobactam	39 characterised carbapenemase producers	4- to 256-fold cefiderocol MIC reduction in KPC producers; susceptibility restored (MIC ≤ 2 mg/L) in 10/12 cefiderocol-resistant KPC-producing K. pneumoniae. Synergy also seen for OXA-48-like producers but absent for OXA-23 and most MBL producers	[44]
Fosfomycin; ceftazidime-avibactam; ampicillin-sulbactam	Genomically characterised CRE, CRAB and CR-P. aeruginosa	Cefiderocol + fosfomycin and cefiderocol + ceftazidime-avibactam synergistic against strains already resistant to cefiderocol; cefiderocol + ampicillin-sulbactam active against an ampicillin-sulbactam-resistant P. aeruginosa isolate	[52]
β-lactamase inhibitors ± polymyxin B nonapeptide	238 MDR/XDR K. pneumoniae, incl. dual carbapenemase producers	Cefiderocol susceptibility 97.9% under iron-depleted conditions but activity lost in iron-enriched medium. Avibactam significantly reduced MIC even in iron-replete conditions; adding an outer-membrane permeabiliser reduced it further	[43]
Eravacycline, meropenem, levofloxacin, colistin, tobramycin, ceftazidime-avibactam	40 ICU P. aeruginosa isolates, carbapenemase-genotyped	Time-kill at 1×MIC/24 h: strongest synergy with eravacycline, meropenem, and levofloxacin (4/8 strains each); colistin, tobramycin, and ceftazidime-avibactam synergistic in 3/8. No antagonism observed	[47]
Imipenem-relebactam	3 MBL-producing DTR P. aeruginosa (2 IMP-13, 1 VIM-2)	Etest FICI showed synergy in both IMP-13 producers but not in the VIM-2 producer; all three patients were successfully treated with the combination—a rare direct FICI-to-outcome pairing showing MBL-subtype dependence	[53]
Ceftazidime-avibactam (mechanism-based PK/PD)	3 sequential carbapenem-resistant Achromobacter xylosoxidans isolates	Combination achieved ≥ 2-log10 CFU reduction; the model required two subpopulations (one ceftazidime-resistant/cefiderocol-susceptible, one resistant to both), with infusion duration the critical driver at clinical exposures	[36]
Moxifloxacin (ceftazidime as non-siderophore comparator)	P. aeruginosa keratitis isolates incl. XDR 2023 outbreak strain	Time-kill synergy in 7/7 isolates; Etest additivity/synergy in 79%, falling to 60% in quorum-sensing-defective strains, whereas ceftazidime + moxifloxacin performed distinctly worse—evidence that synergy is uptake-route-dependent	[48]
Polymyxin B	P. aeruginosa keratitis isolates and reduced-susceptibility mutants	Synergy for two XDR outbreak isolates but only additive-to-indifferent across the wider panel. Mutants with reduced cefiderocol susceptibility carrying baeS and fepA mutations remained responsive to the combination	[51]
Bacteriophage	Gram-negative aerobic bacilli, planktonic, and biofilm	Phage partner produced a 2- to 64-fold cefiderocol MIC reduction with complete biofilm eradication in some isolates (isothermal microcalorimetry)	[54]
Sulbactam or avibactam (antagonism)	Carbapenem-resistant Acinetobacter spp.	Exposure to cefiderocol plus sulbactam or avibactam selected cefiderocol-resistant variants; WGS and transcriptomics identified mutations in efflux pumps, transcriptional regulators, and iron-homeostasis genes with altered outer-membrane-protein and β-lactamase expression	[49]

The available observational clinical evidence, however, suggests a different pattern, and this is among the most important caveats in the cefiderocol literature. In a systematic review and meta-analysis of 18 studies comprising 733 cefiderocol-treated and 473 best-available-therapy-treated CRAB patients, cefiderocol was associated with lower mortality than best available therapy overall (RR 0.74, 95% CI 0.57–0.95). Yet mortality was significantly lower with cefiderocol monotherapy than with cefiderocol-containing combination regimens (RR 0.64, 95% CI 0.43–0.94; p = 0.024) [55]. Confounding by indication is one possible explanation, since combination therapy is preferentially given to sicker patients and to those with higher-MIC isolates, but the finding stands as a reason for caution against assuming that in vitro synergy justifies routine combination. Contemporary reviews reach the same conclusion for DTR P. aeruginosa. Despite in vitro synergy and enhanced activity in hollow-fibre and animal models, no clinical evidence currently supports combination therapy [56], and the mono-versus-combination decision is best driven by pathogen susceptibility, illness severity, and local epidemiology rather than by synergy data [57].

Three further caveats bear on how these data should be used. First, susceptibility testing of cefiderocol in iron-depleted medium follows the standardised laboratory procedure, and results obtained under iron-enriched experimental conditions constitute a research approach that cannot be extrapolated directly to clinical efficacy. Second, cefiderocol susceptibility testing is itself subject to methodological variability and to heteroresistance, which is particularly frequent in A. baumannii and may escape routine testing, while on-treatment emergence of resistance has been documented in clinical series [50,58]. Third, the synergy demonstrated in vitro for cefiderocol-containing combinations, including those directed against carbapenem-resistant A. baumannii, should be read as experimental evidence and not as a clinical recommendation, since the available clinical evidence in this setting remains inconclusive, and pooled analyses do not favour combination over monotherapy [55,59].

7. Intravenous Fosfomycin as a Combination Backbone

Intravenous fosfomycin has an unusually strong theoretical claim as a combination partner. It inhibits MurA (UDP-N-acetylglucosamine enolpyruvyl transferase), the enzyme that catalyses the first committed step of peptidoglycan synthesis and a target exploited by no other clinically available class, so there is no cross-resistance with β-lactams, polymyxins, aminoglycosides or tetracyclines; it also penetrates tissue, bone, and biofilm well [60,61].

A major limitation in severe MDR/XDR infections is the rapid emergence of resistance during fosfomycin monotherapy. Resistance arises through two routes: mutations in the GlpT and UhpT transporters and their regulators, which block uptake, and acquired fosA-family glutathione S-transferase genes, of which 11 variants are now described globally and which disseminate within high-risk clones, such as E. coli ST69 and ST131 and K. pneumoniae ST11 [62]. Mechanisms differ systematically by species. P. aeruginosa relies on peptidoglycan-recycling enzymes and chromosomal Fos enzymes, producing higher baseline MICs. So E. coli breakpoints should not be extrapolated to it [63]. Combination regimens are therefore often favoured in severe MDR/XDR infections, and one recent synthesis describes rational combination as a strategic necessity rather than an optimisation [60].

The clearest demonstration that the partner’s contribution is resistance suppression rather than added killing comes from a 7-day hollow-fibre infection model against three CRAB isolates. Sulbactam monotherapy produced 3–4 log₁₀ killing within 8–32 h followed by regrowth to 8–10 log₁₀ CFU/mL, whereas fosfomycin combined with continuous-infusion sulbactam produced a 2–4 log₁₀ reduction sustained across the full seven days [34]. This demonstrates that the partner’s benefit lies chiefly in prolonging suppression of regrowth rather than in increasing peak bactericidal potency, and it is the strongest mechanistic case for the combination anywhere in the fosfomycin literature.

8. Clinical Evidence by Pathogen

The clinical evidence base is summarised in Table 6 and discussed by pathogen below. The evidence levels used in Table 6 and Table 7 are descriptive labels assigned by the authors based on study design alone. No formal GRADE assessment or risk-of-bias evaluation was performed, and the labels should not be read as a formal evidence-grading system. Level 1a denotes the pre-specified primary analysis of a randomised controlled trial. Level 1b denotes a secondary, subgroup or post hoc analysis of a randomised controlled trial, which preserves randomisation only in part, is exposed to multiplicity and is typically underpowered, and should therefore be read as exploratory and hypothesis-generating rather than as carrying the evidentiary strength of the pre-specified primary randomised comparison. Level 2 denotes a prospective comparative cohort study or target-trial emulation. Level 3 denotes a retrospective comparative cohort study. Level 4 denotes a case series or single case report. In Table 7, “High”, “Moderate”, and “Low” summarise the strength of the aggregate evidence supporting the preferred approach in that clinical scenario, using the same design-based criteria, and are not GRADE certainty ratings.

The overall shape of this evidence base is as follows: randomised evidence exists chiefly for CRAB, observational evidence dominates for MBL-producing CRE, and combination therapy for DTR P. aeruginosa is generally a second-line strategy once an active novel agent is available. Current guidance is explicit on this point, and when an active β-lactam is available, routine combination therapy is recommended neither for carbapenem-resistant Enterobacterales nor for DTR P. aeruginosa, and the choice of agent is driven by the resistance mechanism identified rather than by the species alone [64].

Table 6. Clinical evidence for combination and novel-agent regimens against XDR/PDR Gram-negative pathogens, by pathogen, regimen, and study design.

Pathogen	Regimen	Design (n)	Evidence Level	Key Outcome	Refs.
CRAB	Sulbactam-durlobactam + imipenem versus colistin + imipenem	Phase 3 RCT, non-inferiority (181)	Level 1a (RCT, primary analysis)	28-day all-cause mortality 19% versus 32% (non-inferiority met); nephrotoxicity 13% versus 38%	[65,66]
CRAB	Sulbactam-durlobactam (ATTACK isolate analysis)	Post hoc microbiological analysis	Level 1b (RCT, post hoc; exploratory)	MIC correlated with microbiological outcome; PBP3 mutations identified as determinant of reduced susceptibility	[23]
CRAB	Sulbactam-durlobactam versus colistin, mono- versus polymicrobial	ATTACK subset analysis	Level 1b (RCT subgroup; exploratory)	Colistin-arm patients with monomicrobial infection fared worse than those with polymicrobial infection; no such difference in the sulbactam-durlobactam arm	[67]
CRAB	Colistin + meropenem	Secondary analysis of AIDA RCT (171; 131 CRAB)	Level 1b (RCT secondary; exploratory)	Checkerboard-confirmed in vitro synergy not associated with reduced 14-day clinical failure or 14-/28-day mortality	[37]
CR-GNB	Colistin + meropenem	Secondary analysis of OVERCOME RCT (407; 79% CRAB)	Level 1b (RCT secondary; exploratory)	Time-kill-confirmed synergy associated with lower clinical failure (55.3% versus 64.3%; OR 0.62); 28-day mortality unchanged (38.3% versus 41.4%)	[38]
CRAB	Colistin + sitafloxacin versus colistin alone	RCT (56)	Level 1a (RCT, primary analysis)	No difference in 28-day mortality (32.1% in both arms), clinical response or microbiological response despite in vitro synergy	[68]
MBL-CRE	Ceftazidime-avibactam + aztreonam	EMBARCAR subanalysis, prospective multicentre (140)	Level 2 (prospective cohort)	30-day mortality 41% overall; regimen independently associated with reduced mortality after propensity adjustment (OR 0.18)	[69]
CR-GNB	Ceftazidime-avibactam ± aztreonam	EMBARCAR full cohort, prospective (466)	Level 2 (prospective cohort)	30-day mortality 52% overall; regimen independently associated with reduced mortality among K. pneumoniae bacteraemias (OR 0.20)	[70]
MBL-CRE	Ceftazidime-avibactam + aztreonam versus other active agents	Target trial emulation, 17 hospitals (243)	Level 2 (target-trial emulation)	30-day mortality 35% versus 47% (OR 0.63); lower clinical failure and fewer adverse events (9.6% versus 22.8%)	[71]
MBL-CRE	Ceftazidime-avibactam + aztreonam (most common regimen)	Retrospective multicentre, France (156)	Level 3 (retrospective cohort)	30-day mortality 23.1%; no significant difference by regimen once ICU admission, treatment delay and source control were accounted for	[72]
NDM-CRE	Ceftazidime-avibactam + aztreonam	Single-centre retrospective, Poland (23)	Level 3 (retrospective cohort)	Microbiological eradication in all cases; mortality 17.4%, all in patients with malignancy or COVID-19	[73]
CRE/MDR P. aeruginosa	Ceftazidime-avibactam versus comparators	Systematic review and meta-analysis (24 studies)	Level 1/2 (meta-analysis)	Lower all-cause mortality (OR 0.30) and higher clinical cure in bacteraemia and nosocomial pneumonia	[7]
MDR/XDR P. aeruginosa	Ceftolozane-tazobactam versus best alternative therapy	Multicentre retrospective (206)	Level 3 (retrospective cohort)	73.3% reduction in clinical failure in lower respiratory tract infection, including ventilator-associated pneumonia	[74]
MDR/DTR P. aeruginosa, CRE	Imipenem-relebactam	Multicentre observational, Italy (29)	Level 3 (observational cohort)	Clinical success 75.9%; 30-day mortality 24.1%	[75]
XDR P. aeruginosa	Cefepime/zidebactam (compassionate use)	Case series (5)	Level 4 (case series)	Successful treatment of osteoarticular and endovascular graft infection with no drug-related adverse events	[76]
CR Pseudomonas/Acinetobacter	Cefiderocol, mono- or combination therapy	Multicentre retrospective, USA (112)	Level 3 (retrospective cohort)	Clinical success 68.8%; mortality 16.1%; on-treatment non-susceptibility emerged in 6 patients	[58]
CRAB + MBL-CRE co-infection	Sulbactam-durlobactam + aztreonam	Case report (1)	Level 4 (case report)	Sustained clinical cure after failure of meropenem, ceftazidime-avibactam and polymyxin B/eravacycline	[77]

Table 6. Clinical evidence for combination and novel-agent regimens against XDR/PDR Gram-negative pathogens, by pathogen, regimen, and study design.

Pathogen	Regimen	Design (n)	Evidence Level	Key Outcome	Refs.
CRAB	Sulbactam-durlobactam + imipenem versus colistin + imipenem	Phase 3 RCT, non-inferiority (181)	Level 1a (RCT, primary analysis)	28-day all-cause mortality 19% versus 32% (non-inferiority met); nephrotoxicity 13% versus 38%	[65,66]
CRAB	Sulbactam-durlobactam (ATTACK isolate analysis)	Post hoc microbiological analysis	Level 1b (RCT, post hoc; exploratory)	MIC correlated with microbiological outcome; PBP3 mutations identified as determinant of reduced susceptibility	[23]
CRAB	Sulbactam-durlobactam versus colistin, mono- versus polymicrobial	ATTACK subset analysis	Level 1b (RCT subgroup; exploratory)	Colistin-arm patients with monomicrobial infection fared worse than those with polymicrobial infection; no such difference in the sulbactam-durlobactam arm	[67]
CRAB	Colistin + meropenem	Secondary analysis of AIDA RCT (171; 131 CRAB)	Level 1b (RCT secondary; exploratory)	Checkerboard-confirmed in vitro synergy not associated with reduced 14-day clinical failure or 14-/28-day mortality	[37]
CR-GNB	Colistin + meropenem	Secondary analysis of OVERCOME RCT (407; 79% CRAB)	Level 1b (RCT secondary; exploratory)	Time-kill-confirmed synergy associated with lower clinical failure (55.3% versus 64.3%; OR 0.62); 28-day mortality unchanged (38.3% versus 41.4%)	[38]
CRAB	Colistin + sitafloxacin versus colistin alone	RCT (56)	Level 1a (RCT, primary analysis)	No difference in 28-day mortality (32.1% in both arms), clinical response or microbiological response despite in vitro synergy	[68]
MBL-CRE	Ceftazidime-avibactam + aztreonam	EMBARCAR subanalysis, prospective multicentre (140)	Level 2 (prospective cohort)	30-day mortality 41% overall; regimen independently associated with reduced mortality after propensity adjustment (OR 0.18)	[69]
CR-GNB	Ceftazidime-avibactam ± aztreonam	EMBARCAR full cohort, prospective (466)	Level 2 (prospective cohort)	30-day mortality 52% overall; regimen independently associated with reduced mortality among K. pneumoniae bacteraemias (OR 0.20)	[70]
MBL-CRE	Ceftazidime-avibactam + aztreonam versus other active agents	Target trial emulation, 17 hospitals (243)	Level 2 (target-trial emulation)	30-day mortality 35% versus 47% (OR 0.63); lower clinical failure and fewer adverse events (9.6% versus 22.8%)	[71]
MBL-CRE	Ceftazidime-avibactam + aztreonam (most common regimen)	Retrospective multicentre, France (156)	Level 3 (retrospective cohort)	30-day mortality 23.1%; no significant difference by regimen once ICU admission, treatment delay and source control were accounted for	[72]
NDM-CRE	Ceftazidime-avibactam + aztreonam	Single-centre retrospective, Poland (23)	Level 3 (retrospective cohort)	Microbiological eradication in all cases; mortality 17.4%, all in patients with malignancy or COVID-19	[73]
CRE/MDR P. aeruginosa	Ceftazidime-avibactam versus comparators	Systematic review and meta-analysis (24 studies)	Level 1/2 (meta-analysis)	Lower all-cause mortality (OR 0.30) and higher clinical cure in bacteraemia and nosocomial pneumonia	[7]
MDR/XDR P. aeruginosa	Ceftolozane-tazobactam versus best alternative therapy	Multicentre retrospective (206)	Level 3 (retrospective cohort)	73.3% reduction in clinical failure in lower respiratory tract infection, including ventilator-associated pneumonia	[74]
MDR/DTR P. aeruginosa, CRE	Imipenem-relebactam	Multicentre observational, Italy (29)	Level 3 (observational cohort)	Clinical success 75.9%; 30-day mortality 24.1%	[75]
XDR P. aeruginosa	Cefepime/zidebactam (compassionate use)	Case series (5)	Level 4 (case series)	Successful treatment of osteoarticular and endovascular graft infection with no drug-related adverse events	[76]
CR Pseudomonas/Acinetobacter	Cefiderocol, mono- or combination therapy	Multicentre retrospective, USA (112)	Level 3 (retrospective cohort)	Clinical success 68.8%; mortality 16.1%; on-treatment non-susceptibility emerged in 6 patients	[58]
CRAB + MBL-CRE co-infection	Sulbactam-durlobactam + aztreonam	Case report (1)	Level 4 (case report)	Sustained clinical cure after failure of meropenem, ceftazidime-avibactam and polymyxin B/eravacycline	[77]

8.1. Carbapenem-Resistant Acinetobacter baumannii

The ATTACK trial is the single most influential study in this field. In 181 patients with serious CRAB infection, sulbactam-durlobactam plus imipenem-cilastatin was non-inferior to colistin plus imipenem-cilastatin for 28-day all-cause mortality (19% versus 32%) with markedly lower nephrotoxicity (13% versus 38%) [65], a result detailed further in a companion report on the trial’s pathogen-targeted design [66]. Subsequent analyses found that sulbactam-durlobactam MIC correlated with microbiological outcome and identified PBP3 mutations as a determinant of reduced susceptibility [23], and that colistin-arm patients with monomicrobial infection fared worse than those with polymicrobial infection while sulbactam-durlobactam patients did not—suggesting durlobactam restores activity against co-pathogens that colistin does not address [67].

ATTACK paired durlobactam with imipenem deliberately, and mechanistic evidence indicates imipenem contributes complementary PBP-binding activity, providing target redundancy beyond β-lactamase inhibition alone [22]. However, ATTACK compared sulbactam-durlobactam plus imipenem-cilastatin with colistin plus imipenem-cilastatin, so that the trial establishes the efficacy of the regimen as evaluated and does not, by itself, demonstrate an additional clinical benefit of imipenem when added to sulbactam-durlobactam. The inclusion of imipenem rests on mechanistic rather than on comparative clinical grounds. Real-world experience and PK/PD modelling continue to support combining sulbactam-durlobactam with a carbapenem in critically ill patients pending dedicated monotherapy-versus-combination trials [78], while PK/PD reviews emphasise its renal safety advantage over polymyxins, tempered by its inactivity against MBL producers [79]. Two important caveats have been raised: ATTACK’s comparator arm reflects a regimen no longer recommended for CRAB pneumonia, and cefiderocol shows discordant efficacy signals across its own pivotal trials [59].

Colistin-based combinations for CRAB have performed consistently poorly when evaluated in dedicated trials. Colistin plus sitafloxacin produced no difference in 28-day mortality (32.1% in both arms) despite demonstrated in vitro synergy [68], and a separate randomised trial found colistin plus meropenem no better than colistin monotherapy for CRAB pneumonia [80]. Contemporary critical-care reviews now position sulbactam-durlobactam as at least comparable to colistin, within a landscape where high-quality randomised evidence for any single combination remains sparse [81].

8.2. Carbapenem-Resistant Enterobacterales and MBL Producers

For MBL-producing CRE, against which avibactam, vaborbactam, and relebactam are all inactive, ceftazidime-avibactam plus aztreonam is the best-supported regimen, though the evidence is observational rather than randomised. Narrative and practical management reviews position it as first-line for NDM-producing Enterobacterales, with cefiderocol as an alternative when susceptibility is confirmed [82,83]. A direct comparative evaluation concluded that fixed-dose aztreonam-avibactam offers more predictable pharmacokinetic co-exposure than the ad hoc pairing of the two agents, with observational data favouring the fixed-dose formulation in the absence of head-to-head trials [30], and broader real-world-evidence reviews position aztreonam-avibactam and cefepime-based regimens as the most defensible options for MBL producers [84].

Table 7. Practical synthesis: preferred approach and role of combination therapy and synergy testing, by pathogen and resistance genotype.

Clinical Scenario	Preferred Approach	Evidence Level	Role of Combination Therapy and Synergy Testing	Refs.
CRAB pneumonia or bacteraemia	Sulbactam-durlobactam, given with a carbapenem	High (Phase 3 RCT)	Combination is the studied regimen, not an add-on, and imipenem contributes complementary PBP binding. Colistin-based combinations show in vitro synergy but no consistent randomised benefit and higher nephrotoxicity.	[22,65,68,78]
MBL-producing Enterobacterales (NDM, VIM, IMP)	Aztreonam-avibactam where available, otherwise ceftazidime-avibactam + aztreonam	Moderate (multicentre observational)	Combination is mechanistically obligatory rather than optional, since avibactam does not inhibit MBLs and aztreonam alone is hydrolysed by co-produced serine enzymes. Fixed-dose formulation gives more predictable co-exposure	[30,69,71,83]
KPC-producing Enterobacterales	Ceftazidime-avibactam or meropenem-vaborbactam as monotherapy	Moderate (mechanistic + observational)	Routine combination not required. Watch for treatment-emergent KPC Ω-loop mutations (e.g., KPC-31); synergy testing reserved for isolates that develop resistance on therapy	[21,85]
OXA-48-producing Enterobacterales	Ceftazidime-avibactam	Low (no RCT, small series)	Monotherapy usually adequate; combination considered for high-inoculum or poorly source-controlled infection	[41,83]
DTR Pseudomonas aeruginosa	Ceftolozane-tazobactam, ceftazidime-avibactam or imipenem-relebactam per susceptibility	Moderate (retrospective + case series)	Routine combination is not recommended once an active novel agent is identified. Severe illness or immunocompromise does not, by itself, establish the superiority of combination over an appropriate active agent; any additional agent should be individualised according to susceptibility, clinical circumstances, and the limitations of the available evidence. Cefepime/zidebactam is an emerging option for refractory deep-seated infection	[74,75,76,86]
Cefiderocol-resistant or cefiderocol-failing isolate	Reassess mechanism; consider rescue pairings	Low (in vitro + case-level)	Minocycline + biapenem showed synergy in 22/23 cefiderocol-resistant NDM producers. Rising MIC on therapy should prompt repeat susceptibility testing rather than dose escalation alone	[21,42,87]
Pandrug-resistant isolate, no active single agent	Individualised synergy-guided combination	Low (in vitro synergy only)	The clearest indication for checkerboard and time-kill testing. Polymyxin-based pairings are among the most frequently synergistic in vitro across genotypes, although the effect is isolate-dependent; adjunctive bacteriophage therapy is an investigational option	[24,25,88]

Table 7. Practical synthesis: preferred approach and role of combination therapy and synergy testing, by pathogen and resistance genotype.

Clinical Scenario	Preferred Approach	Evidence Level	Role of Combination Therapy and Synergy Testing	Refs.
CRAB pneumonia or bacteraemia	Sulbactam-durlobactam, given with a carbapenem	High (Phase 3 RCT)	Combination is the studied regimen, not an add-on, and imipenem contributes complementary PBP binding. Colistin-based combinations show in vitro synergy but no consistent randomised benefit and higher nephrotoxicity.	[22,65,68,78]
MBL-producing Enterobacterales (NDM, VIM, IMP)	Aztreonam-avibactam where available, otherwise ceftazidime-avibactam + aztreonam	Moderate (multicentre observational)	Combination is mechanistically obligatory rather than optional, since avibactam does not inhibit MBLs and aztreonam alone is hydrolysed by co-produced serine enzymes. Fixed-dose formulation gives more predictable co-exposure	[30,69,71,83]
KPC-producing Enterobacterales	Ceftazidime-avibactam or meropenem-vaborbactam as monotherapy	Moderate (mechanistic + observational)	Routine combination not required. Watch for treatment-emergent KPC Ω-loop mutations (e.g., KPC-31); synergy testing reserved for isolates that develop resistance on therapy	[21,85]
OXA-48-producing Enterobacterales	Ceftazidime-avibactam	Low (no RCT, small series)	Monotherapy usually adequate; combination considered for high-inoculum or poorly source-controlled infection	[41,83]
DTR Pseudomonas aeruginosa	Ceftolozane-tazobactam, ceftazidime-avibactam or imipenem-relebactam per susceptibility	Moderate (retrospective + case series)	Routine combination is not recommended once an active novel agent is identified. Severe illness or immunocompromise does not, by itself, establish the superiority of combination over an appropriate active agent; any additional agent should be individualised according to susceptibility, clinical circumstances, and the limitations of the available evidence. Cefepime/zidebactam is an emerging option for refractory deep-seated infection	[74,75,76,86]
Cefiderocol-resistant or cefiderocol-failing isolate	Reassess mechanism; consider rescue pairings	Low (in vitro + case-level)	Minocycline + biapenem showed synergy in 22/23 cefiderocol-resistant NDM producers. Rising MIC on therapy should prompt repeat susceptibility testing rather than dose escalation alone	[21,42,87]
Pandrug-resistant isolate, no active single agent	Individualised synergy-guided combination	Low (in vitro synergy only)	The clearest indication for checkerboard and time-kill testing. Polymyxin-based pairings are among the most frequently synergistic in vitro across genotypes, although the effect is isolate-dependent; adjunctive bacteriophage therapy is an investigational option	[24,25,88]

The supporting cohorts are now substantial. In the EMBARCAR subanalysis of 140 patients with MBL-producing Enterobacterales bacteraemia, ceftazidime-avibactam plus aztreonam was independently associated with reduced mortality after propensity adjustment (odds ratio 0.18) [69]. In the full 466-patient cohort, ceftazidime-avibactam-containing regimens were associated with reduced mortality specifically among K. pneumoniae bacteraemias (odds ratio 0.20) [70]. A 17-hospital target-trial emulation of 243 patients found the combination associated with lower 30-day mortality than other active antibiotics (35% versus 47%; adjusted odds ratio 0.63) and fewer adverse events [71]. A French cohort of 156 patients recorded lower overall mortality (23.1%) but found no significant difference by regimen once ICU admission, treatment delay, and source control were accounted for [72]—a useful reminder that regimen choice is only one determinant of outcome. A Polish series of 23 NDM-producing K. pneumoniae infections achieved microbiological eradication in all cases [73], and CRAB- and CRE-directed regimens can be combined when co-infection renders either regimen alone insufficient [77].

8.3. Difficult-to-Treat Resistant Pseudomonas aeruginosa

Here the evidence centres on novel β-lactam/β-lactamase-inhibitor agents rather than on combination regimens per se, with PK/PD-guided dosing emphasised given rising MBL prevalence [85]. A meta-analysis of 24 studies found ceftazidime-avibactam associated with lower all-cause mortality (odds ratio 0.30) than comparators for bacteraemia and nosocomial pneumonia [7], and a 206-patient retrospective study found ceftolozane-tazobactam reduced clinical failure by 73.3% versus best alternative therapy in MDR/XDR P. aeruginosa lower respiratory tract infection [74]. Imipenem-relebactam achieved 75.9% clinical success in 29 critically ill patients [75], and cefepime/zidebactam—a β-lactam paired with a β-lactam enhancer rather than a classical inhibitor—succeeded in five compassionate-use cases of deep-seated XDR P. aeruginosa infection [76]. In immunocompromised transplant recipients, ceftolozane-tazobactam is preferred, and combination regimens are reserved for the most severely ill [86]. This reflects expert practice rather than demonstrated clinical benefit, since neither illness severity nor an immunocompromised state has been shown, in and of itself, to establish the superiority of combination therapy when an appropriate active agent is available. Combination therapy therefore remains second-line when an active novel agent exists, and any decision to add a further agent should be individualised according to susceptibility, clinical circumstances and the limitations of the available evidence.

9. Why In Vitro Synergy Fails to Predict Clinical Outcome

The two dedicated randomised trials of colistin-meropenem combination therapy illustrate the problem directly. The AIDA trial found checkerboard-confirmed synergy unrelated to 14-day clinical failure or 14-/28-day mortality across 171 patients [37]. The OVERCOME secondary analysis, using a time-kill-confirmed definition, found synergy associated with lower clinical failure (55.3% versus 64.3%; adjusted odds ratio 0.62) but no mortality difference [38]. Three explanations recur across the literature reviewed.

First, synergy is genotype-dependent, as Section 5.2 already shows. A pooled result across a mixed isolate collection dilutes a real effect present only in one carbapenemase background. Second, static assays use fixed concentrations that do not reproduce the fluctuating, protein-binding- and organ-function-dependent exposures of a critically ill patient, and do not capture carbapenemase-mediated degradation of the partner drug during incubation—the two limitations that dynamic PK/PD modelling, introduced in Section 5.1, is designed to address. Third, clinical outcome is multifactorial: source control, illness severity, host immune status and time to appropriate therapy each independently influence mortality, and, as the MBL-producing CRE cohorts in Section 8.2 illustrate, several of the largest studies found these factors, rather than regimen choice, to be the dominant predictors of death.

These considerations argue for a sharper distinction between three things that are often reported interchangeably: microbiological activity against the isolate in the laboratory, suppression of resistant subpopulations during exposure, and clinical effectiveness in the treated patient. A combination may achieve the first two and still leave outcome unchanged, and several of the pairings presented above as promising rest on high in vitro synergy frequencies with little or no supporting clinical evidence. Whether a given combination is worth using at the bedside depends in addition on factors that no in vitro system reproduces. The antimicrobial exposure actually achieved at the site of infection, the nature of that site and the adequacy of source control, the severity of illness and the immune status of the patient, and the additional toxicity—nephrotoxicity in particular—that a second or third agent introduces [37,38,65].

The practical conclusion is that in vitro synergy testing is best used to narrow options and generate a testable hypothesis for an individual patient’s isolate—particularly when no single agent is likely to be active—rather than as a validated surrogate endpoint capable of overriding comparative clinical data, where such data exist.

10. Practical Synthesis

Table 7, presented in Section 8, summarises how the evidence reviewed in Section 3, Section 4, Section 5, Section 6, Section 7, Section 8 and Section 9 translates into bedside decisions, organised by pathogen and resistance genotype. The evidence labels are those defined in Section 8 and reflect study design alone. Three principles cut across all scenarios. Molecular identification of the resistance mechanism should precede regimen selection wherever rapid diagnostics permit, because the same combination behaves differently against different carbapenemase classes. An active novel agent, where one exists, generally outperforms a polymyxin-based combination. Finally, synergy testing has its clearest role in the residual population for whom no single active agent can be identified.

11. Future Directions

Several priorities follow directly from the gaps identified above. Randomised trials comparing combination against monotherapy using contemporary, guideline-concordant comparators are still absent for CRAB and DTR P. aeruginosa, in contrast to the substantial observational base assembled for ceftazidime-avibactam plus aztreonam in MBL-producing CRE. Genotype-guided synergy prediction, using rapid molecular diagnostics rather than species identification alone, should be prioritised given the repeated demonstration that synergy varies by carbapenemase class. Dynamic PK/PD modelling accounting for enzymatic drug degradation deserves wider use before novel combinations advance to clinical evaluation. Stewardship strategies specifically designed to preserve ceftazidime-avibactam, aztreonam-avibactam, cefiderocol, and sulbactam-durlobactam are urgently needed, given the treatment-emergent resistance documented for each of these agents [21]. Two agent-specific priorities also follow from Section 6 and Section 7. For cefiderocol, the meta-analytic signal that cefiderocol monotherapy was associated with lower mortality than cefiderocol-containing combination therapy in CRAB (RR 0.64, 95% CI 0.43–0.94) [55] needs prospective testing rather than being dismissed as confounding by indication alone, and the behaviour of cefiderocol under iron-replete conditions deserves systematic evaluation as an experimental research approach rather than as an addition to routine clinical susceptibility reporting, since testing in iron-depleted medium remains the standardised reference method, and no iron-replete condition has yet been validated against clinical outcome [43]. For intravenous fosfomycin, the priority is methodological, and no fosfomycin combination can be rationally designed while routine automated systems misreport MICs for Klebsiella and Enterobacter [89], so validated agar-dilution-equivalent testing must become routinely available before the substantial in vitro synergy signal can be translated [90]. Finally, adjunctive bacteriophage therapy warrants continued investigation with standardised phage characterisation and outcome reporting [91].

12. Conclusions

This review examined the option of synergy-guided combination therapy in the following sequence: the search strategy and selection criteria underlying the synthesis (Section 2); the molecular mechanisms that determine which combinations are rational (Section 3 and Table 1); the four distinct mechanistic rationales underlying combination therapy (Section 4 and Table 2); the laboratory methods used to detect synergy and their limitations (Section 5 and Table 3 and Table 4); two agents whose combination behaviour is atypical and clinically contested, cefiderocol (Section 6 and Table 5) and intravenous fosfomycin (Section 7); the clinical evidence for specific regimens against each major pathogen (Section 8 and Table 6); the recurring discordance between in vitro synergy and patient outcomes (Section 9); and a practical synthesis for bedside use (Section 10 and Table 7).

Combination therapy remains central to the management of severe XDR and PDR Gram-negative infection, not because synergy testing has been validated as a clinical decision tool, but because it is frequently the only evidence-informed option once monotherapy has failed or is predicted to fail. The molecular mechanisms reviewed here explain both why certain combinations are rationally designed to succeed and why the same combination fails against a different genotype. Laboratory synergy testing has matured considerably but requires genotype-aware interpretation and, wherever possible, corroboration against comparative clinical outcome data. The strongest current evidence supports sulbactam-durlobactam-based regimens for CRAB, ceftazidime-avibactam plus aztreonam or aztreonam-avibactam for MBL-producing Enterobacterales, and novel β-lactam/β-lactamase-inhibitor agents for DTR P. aeruginosa; colistin-based combinations, despite frequently demonstrating in vitro synergy, have shown the least consistent clinical benefit under randomised evaluation. Cefiderocol and intravenous fosfomycin illustrate the two ways this translation can fail as follows: For cefiderocol, abundant and mechanistically coherent in vitro synergy contrasts with observational clinical data in which monotherapy was associated with lower mortality than combination, although confounding by indication limits causal interpretation. For fosfomycin, a compelling resistance-suppression rationale is undermined by susceptibility testing that routine laboratories cannot perform reliably. Moving this field from the checkerboard to the bedside will require continued integration of resistance genotyping, dynamic PK/PD modelling, and rigorously designed comparative trials.