Antibiotic Hybrids: the Next Generation of Agents and Adjuvants against Gram-Negative Pathogens?

The OM is an asymmetric bilayer (Fig. 1) with an inner leaflet consisting solely of phospholipids and an outer leaflet that contains an abundance of lipopolysaccharides (LPS). The polymeric LPS is composed of three domains: the hydrophobic lipid A, the hydrophilic core oligosaccharides, and the hydrophilic O antigen. Lipid A is responsible for forming a lipid bilayer with the inner leaflet. Core oligosaccharides and O antigen, which extend outwards to the extracellular environment, are responsible for cellular recognition and virulence (among other functions). The presence of the OM makes Gram-negative bacteria intrinsically resistant to many antibiotics, especially those with a high molecular weight and hydrophobicity. For instance, the LPS structure renders the bacterial OM more restrictive to hydrophobic antibiotics than the IM (19). It has been argued that the hydrophilic carbohydrate component of LPS creates a hydration sphere that restricts the movement and passage of hydrophobic molecules across the membrane (20). The efficient packing of lipid A, due to its molecular organization and lower unsaturated fatty acid content than that of a normal phospholipid bilayer, results in lower OM fluidity (21–23), thus limiting the membrane permeation of hydrophobic agents. Integral membrane proteins that interact directly with LPS, such as outer membrane protein H (OprH) in P. aeruginosa (24) and the Tol-Pal complex in Escherichia coli (25), further augment stability and, therefore, the impermeability of the membrane. Experimental evidence shows 50- to 100-fold-lower hydrophobic probe permeation rates in lipid bilayers that contain lipopolysaccharide (reflective of the OM) than in bilayers that consist of phospholipids only (reflective of the IM) (26). Structural variabilities and modifications in the LPS, especially the lipid A portion, result in significant differences in drug permeation rates among Gram-negative organisms (27). It is therefore clear that the OM constitutes a major hurdle for drug uptake in Gram-negative bacteria, especially for P. aeruginosa, which has 12- to 100-fold reduced outer membrane permeability relative to that of E. coli (28).

The glycopeptide antibiotic vancomycin, with a molecular mass of 1449.3 g/mol, lacks antibacterial activity against most clinically relevant Gram-negative bacteria. Vancomycin inhibits peptidoglycan synthesis by sequestering peptidoglycan precursors that ultimately prevent glycan cross-linking. In Gram-positive organisms, vancomycin exerts its antibacterial activity uninhibited, as its target is located at the cell membrane. However, it must traverse the OM and reach the periplasmic space to elicit its function in Gram-negative organisms, a feat which vancomycin is incapable of achieving due to the protective membrane barrier. The loss of antibacterial activity due to membrane impermeability is true for almost all clinically used glycopeptide antibiotics (29, 30). Most antibiotics and biomolecules with molecular masses of >600 g/mol are incapable of traversing the OM (31), with few exceptions, including polybasic amphiphiles, such as polymyxins and antimicrobial peptides, and nonbasic energy sources, such as maltohexaoses (32, 33).

Charged (cationic or zwitterionic) or noncharged small hydrophilic molecules are typically able to enter the periplasmic space via nonspecific protein channels called porins. Examples of such antibiotics with porin-dependent uptake include β-lactams, fluoroquinolones, and sulfonamides. These β-barrel-structured integral protein channels allow water-soluble molecules to traverse the restrictive hydrophobic membrane through their water-filled cavity (34–36). However, porins impose molecular sieving properties, as only small molecules, typically ≤600 g/mol, are believed to pass through their narrow channels (37–39). Drug permeation through porins may also vary among Gram-negative bacteria. For instance, P. aeruginosa possesses lower outer membrane permeability than E. coli, as it expresses a more selective outer membrane protein F (OprF) porin (28, 39). The OprF porin constitutes the majority of the porins present in P. aeruginosa (40). Porins that are present in relatively smaller amounts in P. aeruginosa include OprB, OprC, OprD, OprE, OprF, OprG, OprH, and others (41, 42). The OprF porin has been shown to also allow solute diffusion much more slowly than classical porins as a consequence of its structural conformation (39, 43). For example, the monosaccharide l-arabinose was found to diffuse 50 times slower in the OprF porin in P. aeruginosa than in the OmpF porin channel of E. coli (44). Low intracellular drug concentrations due to slow porin-mediated influx in P. aeruginosa are exacerbated by the abundance of multidrug efflux pumps and resistance-encoding genes (28, 43, 45).

However, several antibiotics with a high molecular mass (>600 g/mol) are able to pass through the OM in a mode of uptake independent of porins or passive diffusion. These compounds are mostly cationic and are hydrophilic (such as aminoglycosides [46]) or amphiphilic (such as polymyxins) in nature (46). They are able to transit the OM via a “self-promoted” uptake mechanism, which is characterized by the initial displacement of divalent cations (Ca²⁺ or Mg²⁺) that results in OM destabilization (47). Electrostatic interactions between the positively charged divalent cations and the negatively charged phosphate groups on lipid A stabilize the LPS structure (48–50). It is perceived that the subsequent localized OM disruption from divalent cation displacement facilitates the penetration of the antibiotic into the periplasmic space (51, 52). It was widely documented in early years that the antibacterial activities of aminoglycosides and polymyxins are antagonized by the exogenous addition of Mg²⁺ and Ca²⁺ cations (53–56). This observation was later attributed as being a hallmark of the self-promoted uptake mechanism, that it entails the displacement of divalent cation LPS bridges and that exogenous supplementation of the divalent cations immediately arrests the process (57). The physicochemical requirements for a molecule to display self-promoted uptake are yet to be fully understood, although the propensities of a molecule to be protonated (to effectively carry one or more positive charges) under physiological conditions and to strongly interact with LPS may be necessary characteristics.

The phospholipid bilayer that comprises the IM greatly limits the diffusion of hydrophilic molecules. Compared to the OM, hydrophobic molecules easily traverse the IM through passive diffusion. However, charged solutes such as sodium cations and hydrophilic nutrients such as glucose enter freely into the cytosol once they traverse the OM. Their uptake is achieved through the use of solute-specific energy-dependent transporter proteins (58). Some weakly charged or neutral amphiphilic compounds may also enter the cytosol by utilizing proton motive force (PMF) (59, 60). Bacterial PMF is governed by the proton gradient, ΔpH, and membrane potential, ΔΨ. The ΔpH is believed to facilitate the diffusion of weakly charged molecules through charge neutralization, while the ΔΨ is perceived to stimulate electrochemical interactions that lead to molecular uptake (61, 62). For instance, the cytoplasmic uptake of the sulfonamide (63) and tetracycline (64) classes of antibiotics is ΔpH dependent, while the uptake of the aminoglycoside class of antibiotics appears to be ΔΨ dependent (65).

Once a drug makes its way intracellularly, it could be effluxed out before it mediates its antibacterial action. Efflux pumps are membrane proteins that can expel their substrates from the cytosol into the periplasm or from the periplasm into the external environment. So far, all studied Gram-negative organisms are known to express at least one multidrug efflux pump (66). Bacterial efflux systems have been extensively reviewed in the literature (see references 67–72). However, it should be noted that drug efflux affects the intracellular concentration of a therapeutic agent and that the overexpression of multidrug efflux pumps confers intrinsic antibiotic resistance on the pathogen.

It is evident that drug permeability in Gram-negative bacteria is more challenging for antibiotics with cytosolic targets, as they must transit two protective lipid bilayers (73). One approach to overcome this involves appending or tweaking different functional groups on a lead structure, in the hope of generating more-amenable derivatives with enhanced biological activity and cellular permeation. “Rule-of-thumb” knowledge in structural modification is used by medicinal chemists to make rational decisions for drug optimization. Lipinski's rule of five for favorable drug oral bioavailability (74) has become a popular in silico guideline for the development of therapeutic agents that are able to cross intestinal epithelial cells. To possess good pharmacokinetics in the human body, Lipinski's rule proposes that an agent may not have (i) more than five hydrogen bond donors, (ii) more than 10 hydrogen bond acceptors, (iv) a molecular mass of >500 g/mol, and (v) a lipophilicity factor (log P) of >5, as measured by the octanol-water partition coefficient (74, 75). Unfortunately, Lipinski's metrics do not hold true for antibacterial agents that require bacterial membrane penetration. Molecular passage through the OM appears to be governed by a different set of physicochemical rules that are orthogonal to the IM (18). Compounds that are optimized solely to traverse the OM most likely would not be able to cross the IM and vice versa. A widely acceptable set of membrane permeation rules for antibacterial agents appears to be nonexistent (76). An attempt was previously made to formulate a guideline by binning all antibiotics in the pipeline and in clinical evaluation to correlate discernible physicochemical parameters with antibacterial activity (77). Compounds were binned into three categories, namely, compounds that have activity against only Gram-positive organisms, compounds that have activity against Gram-negative organisms, and compounds that are antipseudomonal. A high polarity (for porin uptake) and a reasonable level of lipophilicity (to ensure lipid membrane penetration) were observed to be ideal for agents against Gram-negative organisms (77). By exploiting the ideal physicochemical properties revealed from binning antibiotics, an effort to optimize the activity of oxazolidinones against Gram-negative bacteria was recently described (78). Members of the oxazolidinone class of antibiotics, such as linezolid, do not have potent activity against most Gram-negative organisms, presumably due to permeation impediments across the OM and/or efflux (79). The most active prepared oxazolidinone analog demonstrated only a modest enhancement of activity against E. coli, and those authors noted that a fully realized set of permeation guidelines is direly needed for optimizing lead compounds (78). A recent article emphasized the inherent hurdles in developing agents that are able to permeate Gram-negative membranes and suggested that the binning process should be further refined to include the route(s) of cellular entry for each antibacterial agent (18). However, the development of reliable methods that allow data mining for cellular entry and the accumulation of antibiotics is necessary to realize such a suggestion. At this point, several experimental protocols may hold the key to tackling this proposition. For instance, the elucidation of bacterial uptake mechanisms and the subsequent quantification of cytoplasmic accumulation that utilize techniques such as tandem liquid chromatography-mass spectrometry (LC-MS) (80, 81), Raman spectroscopy (82), and microspectroscopy (83) have been reported. A proof-of-concept study that utilized LC-MS for quantification and several correlation programs for analysis of 10 sulfonyladenosine-containing agents in terms of their membrane permeation in E. coli, Bacillus subtilis, and Mycobacterium smegmatis was reported (84). This systematic approach successfully delineated the relationship between the physicochemical properties and the cytoplasmic accumulation of sulfonyladenosines. For instance, the cytoplasmic accumulation of the 10 sulfonyladenosine-containing compounds in E. coli was positively correlated with hydrophobicity but negatively correlated with polarity (84). This platform is envisioned by those authors (84) to be applicable to a larger diverse panel of chemical agents and other bacterial organisms and may therefore be utilized for formulating a set of antibacterial permeation rules in the future.

The impermeability of Gram-negative bacterial membranes greatly limits our ability to develop new antibiotics, and there is an apparent void in the fundamental understanding of physicochemical properties necessary for an agent to overcome the double barrier of protection in Gram-negative bacteria. However, recent advances have shed some light on this hurdle (84, 85). It was recently shown that for small molecules to accumulate in the Gram-negative bacterium E. coli, they must contain an amine group (a primary amine is preferred over a secondary or tertiary amine), be amphiphilic, be rigid, and have low globularity (defined as the spatial parameter of the molecule) (85). By applying these rules, the natural product deoxynybomycin, which targets DNA gyrase and which is active against Gram-positive bacteria only, was converted into an antibiotic with activity against a diverse panel of multidrug-resistant Gram-negative pathogens, excluding P. aeruginosa (85).

It may also be argued that instead of spending great efforts and resources on the development of intracellularly targeting antibacterials, one may focus on exploring membrane targets that are easily accessible (86). For instance, the membranolytic function of antimicrobial peptides and amphiphilic agents (87–90) or the inhibition of essential outer membrane proteins (91, 92) may be exploited. Looking forward, we foresee the materialization of the essential paradigm to predict bacterial membrane penetration. But for how long? Only time will tell.

The development of agents that are not bactericidal but that indirectly inhibit the molecular pathway responsible for bacterial communication is a viable strategy to address the problem of antibiotic resistance (98–100). This therapeutic approach is based on the purported delayed development of bacterial resistance, as it is perceived that such agents exert reduced evolutionary selective pressure (101). On the other hand, agents that challenge bacterial survival by directly inhibiting a molecular target may result in higher rates of resistance development. For example, blocking bacterial quorum sensing may be a feasible approach. Quorum sensing is characterized by the production, release, and group-wide detection of autoinducer molecules by bacteria as a mode of communication of bacteria with their neighbors (102). This network of communication is triggered by environmental factors within the microbial community, such as differences in bacterial density or the presence of environmental challenges (either physical or chemical) (103, 104). Once these signaling molecules are detected, cascades of physiological and metabolic changes occur by orchestrated alterations in bacterial gene expression, resulting in the secretion of biomolecules needed for biofilm formation and virulence (105). Therefore, hindering quorum sensing may result in the pathogen not being able to cause harm to the host. Extensive discussions of bacterial quorum sensing and the development of agents that are able to quench this bacterial process have been reported (see references 103 and 105–108). Several agents that block quorum sensing are in preclinical development. For example, the synthetic agent meta-bromothiolactone (mBTL) has been reported to curb the production of the virulence factor pyocyanin and biofilm formation in P. aeruginosa by affecting the regulation of the Las and Rhl quorum-sensing systems (109). Moreover, in vitro protection of human lung epithelial cells and in vivo protection of Caenorhabditis elegans against P. aeruginosa by mBTL were described (109). A follow-up report detailed the optimization of mBTL for enhanced stability, as the thiolactone ring is susceptible to chemical and enzymatic hydrolysis (110). Other anti-quorum-sensing agents (111–114) have also been reported to exhibit similar promising in vitro and in vivo results. However, this paradigm was recently challenged (115, 116), and several clinical isolates have been reported to be resistant to established anti-quorum-sensing agents (117). Anti-quorum-sensing agents are yet to reach clinical trials.

Combination therapy has been well received by the scientific and medical communities and has existed for more than 3 decades (118). Clinicians often prescribe two or more antibiotics concomitantly during empirical treatment to ensure the coverage of all possible bacterial pathogens and resistance profiles. It was later realized that the use of multiple antibiotic agents in a therapeutic cocktail may limit the development of resistance in vitro in comparison to drug monotherapy. The overall expected clinical outcome for this strategy is lower patient mortality rates. However, combination therapy is not limited to antibiotic agents but includes therapeutic interventions that may use bioactive helper molecules, also known as adjuvants, to enhance the efficacy of a primary antibiotic. In fact, it has been argued that the adjuvant-antibiotic combination approach offers a more attractive option for the treatment of drug-resistant bacterial infections than the use of multiple antibiotics (119). Here, we discuss combination therapy as (i) an antibiotic-adjuvant approach and (ii) an antibiotic-antibiotic approach.

Considering the “success” of several antibiotic-antibiotic combinations in the past few decades, the strategy remains fallible, as several important pharmacological questions remain unanswered. For instance, other than for tuberculosis, there is no clinical evidence to support the notion that antibiotic resistance is suppressed by antibiotic-antibiotic combinations (162). This is a tough concept to prove, as clinical studies are usually designed not to measure the emergence of antibiotic resistance but to prevent or treat it. These data are often extrapolated from in vitro studies and in animal models and may be different from what is obtainable in human hosts. The clinical translatability of drug synergy in vitro and in animal models to humans is also debatable (162). The limited available clinical evidence suggests a statistically insignificant difference between antibiotic-antibiotic combination therapy and monotherapy for the treatment of Gram-negative bacterial infections in terms of mortality rates, which is the postulated clinical outcome for synergistic drug combinations (163–165). For instance, a systematic study reported no appreciable improvement for the combination of a β-lactam–aminoglycoside over β-lactam monotherapy for the treatment of endocarditis caused by a Gram-positive bacterium (e.g., Staphylococcus aureus), even though the combination shows synergism in vitro (166). Another study showed that the combination of a β-lactam and either an aminoglycoside or a fluoroquinolone in comparison to β-lactam monotherapy imposes no benefit for patient mortality for the treatment of infections caused by the Gram-negative organism P. aeruginosa (164). However, it should be noted that the supporting evidences for these studies were based on meta-analyses of a small amount of available clinical data that included those during early years when drug resistance was not as prevalent. Caution should therefore be taken in the extrapolation of data to fit the current landscape, where incidences of MDR and XDR bacterial infections are much higher. Currently, there seems to be an apparent consensus that combination therapy is preferred for the treatment of MDR pathogen infections of severely ill patients and for empirical therapy (167). A lack of pharmacokinetic complementarity between different drugs might indeed contribute to the discrepancies between in vitro data and clinical observations (168). Each drug component may be absorbed or distributed in the human body to different degrees. Pharmacokinetic variances, but also the patient's overall condition, would certainly impose a challenge in fine-tuning the dosages of administered drugs to replicate their observed in vitro synergy, as both drugs are required to be located at the site of infection at their optimal concentrations simultaneously. Noncomplementary absorption and distribution rates may therefore be circumvented by fusing both pharmacophoric molecules together to make a single hybrid antibacterial agent. Here, the fundamental concept behind antibiotic hybrids is discussed.

A mutual prodrug, as defined by the IUPAC, is the covalent attachment of two drugs to form a unique molecule that undergoes biotransformation (such as bond cleavage) to exhibit its pharmacological effects (179). To achieve this for antibiotic hybrids, a prodrug strategy necessitates the use of a cleavable linker/tether with bacterium-specific lability. The linker/tether must deliver both constituent molecules to their site of action (bacterial compartment) before being degraded. The resulting antibiotic hybrid prodrug may or may not possess biological activity by itself.

Hybrid prodrugs in the literature are comprised mostly of β-lactams linked to other therapeutic agents. The foundation of antibiotic hybrid prodrugs is premised on extensive mechanistic studies of the β-lactam core structure. In 1965, it was discovered that some molecular substituents adjacent to carbon 3 of the cephalosporin (a type of an unsaturated β-lactam/penem) backbone was displaced following β-lactam ring hydrolysis (182). Mechanistically (Fig. 3), it was perceived that the liberated secondary amine (from β-lactam ring hydrolysis) would delocalize its electrons on carbon 2 to form an imine and consequently displace the unsaturated bond (carbon-carbon double bond) along carbon positions 2 and 3. Electron delocalization would ultimately release the molecular substituent or leaving group (LG) adjacent to carbon 3. This knowledge was then used to construct cephalosporin-based hybrid prodrugs where an antibiotic was designed to be the leaving group adjacent to carbon 3. Notably, most investigational antibiotic hybrid prodrugs are cephalosporin based (169, 183). Approximately 3 decades later, it was discovered that the installation of an S-aminosulfenimine moiety at carbon 6 of penicillin (a type of saturated β-lactam/penam) resulted in the rapid intramolecular displacement of the S-amino substituent following β-lactam ring hydrolysis (Fig. 4) (184). Nuclear magnetic resonance (NMR) studies of this phenomenon (185) led to the proposition that the release of the substituent at penicillin position 6 is due to the formation of a disulfide bond-characterized intermediate that further decomposes to yield several by-products (Fig. 4A). Interestingly, a vinyl ester linkage (186) was reported to be a viable alternative to the S-aminosulfenimine linkage, as it was also found to be displaced following β-lactam hydrolysis (Fig. 4B). However, we have yet to see any prospective β-lactam-based hybrid prodrug with either a cleavable S-aminosulfenimine or vinyl ester linkage in the literature.

β-Lactam-based prodrugs function via a two-step process: (i) the β-lactam antibiotic elicits its activity by inhibiting the enzyme transpeptidase via acylation (ester bond formation) of the active-site serine residue at the hydroxyl side chain, which consequently results in β-lactam ring amide bond hydrolysis, followed by (ii) the release of another functional antibiotic to elicit its own antibacterial activity. In the presence of serine-based β-lactamases that confer enzymatic drug resistance to β-lactam antibiotics, the β-lactam-containing prodrug could serve as a sacrificial adjuvant to inhibit the resistance enzyme. The drug appended to the hydrolyzed sacrificial adjuvant can then be released to elicit its biological action. Inhibition of serine-based β-lactamases occurs via acylation of the active site of serine at the hydroxyl side chain. Unfortunately, β-lactam-based prodrugs do not have any benefit against metallo-β-lactamases, as this resistance enzyme utilizes zinc ions that activate water molecules to hydrolyze the β-lactam ring and therefore would not be inhibited by acylation. It would be interesting to see the future development of antibiotic hybrid prodrugs that contain zinc-sequestering agents, such as the recently reported agent aspergillomarasmine A (137), to combat metallo-β-lactamase-producing Gram-negative bacteria, as this enzyme is credited as being the main cause of pandrug resistance worldwide (187).

Several advantages and disadvantages are discussed above. In addition, achieving synergistic antibacterial activity through the hybrid prodrug approach is a possibility. As the prodrug reaches its target bacterial compartment, the hybrid entity is expected to be cleaved into two functioning drugs that may have synergistic interactions. Through the prodrug strategy, prediction of in vivo drug synergism from in vitro combination studies may be easier, as the problem of noncomplementary pharmacokinetics, a concern in combination therapy, as mentioned above, is bypassed. However, this may hold true only for drug combinations with a 1:1 optimal concentration ratio. Nonetheless, it may be ideal to consider any advantageous synergistic relationship of the antibiotics sought to be fused together.

The major challenge with this approach is the difficulty in designing a bacterium-specific cleavable linker, as the human body is a complex biological system comprising specific and nonspecific enzymes capable of degrading various covalent bonds. If the cleavable linker of the prodrug is degraded before it reaches the bacteria, then the design would no longer be different from that of combination therapy. Therefore, it is crucial to select a linker that is bacterium selective and capable of withstanding human metabolic enzymes. The stability of the cleavable linker in the human body and the drug permeability impediments in Gram-negative bacteria are the major limitations of the hybrid prodrug approach. However, like the mechanistic findings for β-lactam hydrolysis, more and more data on the bacterium-specific molecular interplay become unraveled as we progress into the future, and there is great optimism that human ingenuity will pave the way for the next generations of hybrid prodrugs.

The majority of hybrid prodrugs in the literature consist of a non-β-lactam antibiotic fused to the adjacent carbon 3 of the cephalosporin core structure (second antibiotic) (Fig. 5). The earliest hybrid prodrug, reported in 1976, was a cefamandole (cephalosporin) derivative linked to omadine at carbon 3 (hybrid prodrug 1) (188). Omadine, also known as 2-mercaptopyridine-N-oxide or pyrithione, inhibits bacterial ATP synthesis and is known as a metal chelator (189). Hybrid prodrug 1 was shown to be active against a panel of Gram-negative pathogens with a geometric mean MIC of 0.5 to 16 μg/ml (188). The mode of action of hybrid prodrug 1 was hinted to be mainly that of a cephalosporin, with some contributions from its omadine pharmacophore, as its MIC was reduced by 4- to 32-fold in bacterial strains that express β-lactamases. As noted by those authors, that study validated the concept of a hybrid prodrug but was of no significant therapeutic relevance, as omadine displays undesirable systemic toxicity. Ten years later, a desacetylcephalothin (cephalosporin) linked to a chloroalanyl dipeptide inhibitor of alanine racemase (hybrid 2) was reported to be active against E. coli (MIC of 7.05 to 14.1 μg/ml) (190, 191). The cephalosporin pharmacophore serves as an essential structural component for most hybrid prodrugs, and some of the resulting hybrid entities advanced to preclinical/clinical trials. For instance, NB2001 (hybrid 3) by NewBiotics Inc., which consists of a desacetylcephalothin fused with a chlorine-containing phenolic triclosan, was reported to possess potent broad-spectrum bacteriostatic activity against Gram-positive and Gram-negative bacteria (192). However, NB2001 (hybrid 3) was inactive (MIC of >32 μg/ml) against P. aeruginosa (192), which was not surprising due to the organism's restrictive OM and abundance of multidrug efflux pumps. No data were reported for A. baumannii. The antibacterial activity of NB2001 was attributed to the release of its triclosan component, as the intact hybrid exhibited significantly reduced in vitro binding to transpeptidases relative to that of cephalothin, suggesting a diminished mode of action of cephalosporin (193). Moreover, the intact hybrid was unable to inhibit the enzyme enoyl-acyl carrier protein reductase FabI (target enzyme for triclosan) (193). NB2001 (hybrid 3) entered preclinical evaluation for bacterial and nosocomial infections but was discontinued in 2005 for undisclosed reasons. Another worthy example of a β-lactam-based hybrid prodrug is Ro 23-9424, a story of high hopes but dashed dreams.

In 1989, Roche pharmaceuticals announced the development of a cephalosporin-fluoroquinolone ester hybrid prodrug called Ro 23-9424 (hybrid 4) (Fig. 5) with potent broad-spectrum bactericidal activity against Gram-positive and Gram-negative organisms (194). The hybrid prodrug contains a desacetylcefotaxime (a cephalosporin) covalently linked to a fleroxacin (a fluoroquinolone) via a cleavable ester linkage adjacent to carbon 3, which is consequently cleaved following enzymatic hydrolysis of the β-lactam ring structure (Fig. 6). Its mode of uptake was proposed to be porin mediated (195, 196) (even though it has a modestly high molecular mass of 764.7 g/mol). Ro 23-9424 (hybrid 4) displayed only limited activity against P. aeruginosa (194), which may be attributed to reduced OM permeability (due to the selective OprF porin that is majorly expressed) and/or an abundance of multidrug efflux pumps of the organism. Cephalosporins inhibit peptidoglycan synthesis by acetylating the active site of transpeptidases, while fluoroquinolones inhibit DNA synthesis via the inhibition of DNA gyrase and topoisomerase IV. Ro 23-9424 (hybrid 4) acts initially as a cephalosporin, where the hydrolysis of the β-lactam ring results in a fluoroquinolone secondary mode of action (197). The intact hybrid prodrug exhibits only minimal DNA synthesis inhibition (195). As conceptualized, the antibacterial activity was retained in E. coli strains that are resistant to either β-lactams, fluoroquinolones, or both agents (196). The in vitro half-life of the hybrid prodrug in human serum was reported to be 6.3 h (198), suggesting an adequate stability of the cleavable ester linkage toward nonspecific enzymatic degradation. pH-dependent stability in an aqueous phosphate buffer solution was also described, where half-lives of 6.9 and 3.0 h were observed at pH 6.5 and 7.4, respectively (198). Ro 23-9424 (hybrid prodrug 4) was also shown to enhance the induction of LPS-stimulated tumor necrosis factor alpha (TNF-α), yet it reduced the production of the proinflammatory cytokine interleukin-1β (IL-1β) in human monocytes (199), suggesting a potential immunomodulatory benefit in reducing the possibility of LPS-induced septic shock. Promising preclinical in vivo pharmacokinetic parameters and tolerability were described for mouse, rat, dog, and baboon models with single- or multiple-dose intravenous administration (200). Excellent in vivo efficacy was reported for systemic mouse infection models of Gram-positive and Gram-negative bacterial infections, including strains that are resistant to cefotaxime and fleroxacin (198, 201). For instance, subcutaneously administered Ro 23-9424 (hybrid 4) was more active (50% effective dose [ED₅₀] of 17 mg/kg) than cefotaxime (ED₅₀ of 50 mg/kg) and fleroxacin (ED₅₀ of >100 mg/kg) in a murine meningitis model of infection by the Gram-positive organism S. pneumoniae (201). The hybrid prodrug resulted in an efficacy (ED₅₀ of 13 mg/kg) that was similar to that of fleroxacin (ED₅₀ of 9 mg/kg) but better than that of cefotaxime (ED₅₀ of >100 mg/kg) in a murine meningitis model of infection by the Gram-negative organism K. pneumoniae (201). These promising preclinical data advanced the hybrid prodrug to phase 1 clinical trials for bacterial infections. Unfortunately, the trial was discontinued around the mid-1990s for undisclosed reasons. Several explanations were speculated (202), such as the fact that the in vivo drug stability in humans did not replicate the initial observations in vitro and in animal models, i.e., that the ester linkage connecting the cephalosporin and fluoroquinolone fragments was degraded by nonspecific enzymes present in humans. Moreover, in contrast to the expected multimodal suppression of drug resistance evolution, development of resistance (16- to 128-fold increase) to Ro 23-9424 (hybrid prodrug 4) was described for several bacterial strains after 2 weeks of serial passage at subinhibitory concentrations (203). Resistance in E. coli was attributed to two factors: (i) decreased Ro 23-9424 outer membrane permeation due to altered porin uptake and (ii) impeded fluoroquinolone activity as demonstrated by a replicative DNA biosynthesis assay in toluene-permeabilized cells (204). However, caution in interpreting the reported mechanism of resistance against Ro 23-9424 (hybrid 4) is advised, as efflux-mediated resistance was not fully recognized at the time of publication of that study. Fluoroquinolone resistance via efflux pumps was first reported in 1994 (205) and has since been known to be a major mechanism of resistance against this class. Sadly, Ro 23-9424 (hybrid 4) presented the antithesis to the idea of delayed drug resistance generation with antibiotic hybrids. This story exemplified the ingenuity of bacteria in coping with chemical assault by restricting their cellular entry. It may be beneficial to design future antibiotic hybrids as agents than can enter bacterial cells via mechanisms independent of porins, as these protein channels can be easily modified genetically by the pathogen to confer drug resistance.

Classical antibiotic hybrid drugs can be distinguished from prodrugs through the stability of their linker/tether. In the typical hybrid drug approach, the participating therapeutic agents are covalently linked by a robust noncleavable molecular linker that can withstand enzymatic and nonenzymatic assaults throughout its time course in the body. Upon entering the pathogen, a prototypical hybrid drug is expected to elicit its antibacterial action by utilizing either of its pharmacophoric domains or both domains simultaneously. However, it should be noted that the development of a hybrid drug that is able to simultaneously inhibit both drug targets by utilizing only a singular molecular entity at the same time is a difficult feat to achieve. Ideally, the molecular targets of the conjoined pharmacophores should be in close proximity, while the linker/tether should be of an appropriate spatial length to efficiently anchor them. This imposes a conceptual challenge that is difficult to be satisfied. For example, a single hybrid molecule that has two different pharmacophoric components—one that inhibits DNA synthesis and another that inhibits protein synthesis—will not be able to inhibit both targets at the same time due to the spatial separation of the target compartments. For now, antibiotic hybrid drugs in the literature are believed to interact with only one of the possible targets at a given time (170). However, this does not nullify the concept of multimodal mechanisms for hybrid antibiotics as long as they retain the interactions of both individual antibacterial pharmacophores.

The antibiotic hybrid drug approach is considerably more popular in the literature than the prodrug approach. This is attributed mainly to the limited number of accessible bacterium-specific cleavable linkers for an effective prodrug delivery approach. Either way, it should be noted that both hybrid strategies require tremendous synthetic efforts, as therapeutic agents typically possess dissimilar molecular stabilities and reactivities under different preparative conditions. Designing a specific linker that is expected to be biotransformed by a bacterium-specific enzyme under specific physiological conditions therefore makes hybrid prodrugs relatively more difficult to prepare than classical hybrid drugs. Nevertheless, the characteristic high molecular weight (>600 g/mol) of hybrid drugs makes it challenging to generate agents that are able to permeate the dual membrane of Gram-negative bacteria. Emerging reports, however, project a good prognosis for this strategy, as several hybrid drugs (discussed below) that are capable of eradicating MDR Gram-negative bacteria and presumably able to delay the onset of drug resistance are already in preclinical/clinical evaluation (http://www.pewtrusts.org/en/multimedia/data-visualizations/2014/antibiotics-currently-in-clinical-development).

Similar to the advantageous pharmacokinetic properties of hybrid prodrugs, a hybrid drug is expected to remain a unimolecular entity as it travels to the site of infection and traverses the bacterial membrane into inner compartments (periplasmic and/or cytosolic space). However, differences in the two therapeutic approaches lie in how they elicit their biological function. A hybrid prodrug is subjected to enzymatic biotransformation, as it enters the bacterial cell, to yield two functional therapeutic entities, while a hybrid drug would remain a single entity throughout its time course in the body. Therefore, hybrid drugs may be advantageous in terms of their kinetics (drug metabolism and elimination), as they are expected to be cleared from the host as a single molecule. A hybrid prodrug, on the other hand, is designed to be cleaved into two separate functional molecules that may possess different pharmacokinetics after intracellular biotransformation. Drug metabolism and excretion are important factors that influence dosing regimens, as bioaccumulation above a certain threshold may result in toxicity. This, along with the above-mentioned advantages and disadvantages, should be taken into consideration when designing an antibiotic hybrid.

A substantial number of hybrid drugs that are in development or have entered clinical trials possess limited antibacterial activity against Gram-negative pathogens (178, 206, 207). This is attributed mainly to impeded bacterial cellular uptake due to a high molecular mass of >600 g/mol and physicochemical properties that render the hybrid drug unable to penetrate both outer and inner membranes. Drug uptake through nonselective porin channels in bacteria is restricted to low-molecular-mass (typically <600 g/mol) and highly polar compounds. With this in mind, we performed a literature search for hybrid drugs that are potent against Gram-negative bacteria. Here, we highlight select antibiotic hybrids from recent literature (2010 to early 2017).

The opportunistic pathogen P. aeruginosa is the leading cause of nosocomial infections in immunocompromised patients and is abundantly found on many medical devices in the hospital. P. aeruginosa, alongside other ESKAPE pathogens, is a major threat to public health for which effective therapy is rapidly becoming elusive (3, 254, 255). In 2017, the WHO ranked carbapenem-resistant P. aeruginosa as critical (priority 1) in its list of the world's most dangerous superbugs that pose a serious threat to human health (http://www.who.int/mediacentre/news/releases/2017/bacteria-antibiotics-needed/en/). Since the major impediment that antibiotics face against P. aeruginosa is low penetration/uptake across its impermeable outer membrane; the first tobramycin-based hybrid analog, hybrid 11, was initially conceptualized to take advantage of the self-promoted uptake mechanism of aminoglycosides to deliver a second antibiotic into the periplasm of the cell. Moreover, aminoglycosides possess a pleiotropic mechanism of antibacterial action. They interact with rRNA to inhibit protein translation at lower concentrations (≤4 μg/ml), while they are known to disrupt the bacterial membrane at higher concentrations (≥8 μg/ml) (256). Tobramycin is a particularly attractive aminoglycoside, as it is currently one of the most active and effective agents for the treatment of P. aeruginosa infections (257).

It was shown previously that tobramycin-based hybrid drugs that lose innate in vitro antibacterial activity (Table 2) are able to enhance the potency of other antibiotic classes (except aminoglycosides and carbapenems) against P. aeruginosa (Table 3 and Fig. 11) (247). These synergistic effects are exclusive to the hybrids, as their composing antibiotic fragments do not exhibit this property. For example, a tobramycin-ciprofloxacin hybrid (hybrid 11) that displayed weak antibacterial activity as a stand-alone agent (MIC of ≥16 μg/ml) was found to restore the efficacies of ciprofloxacin and moxifloxacin (FICI of 0.03 to 0.38) against ciprofloxacin-resistant MDR or XDR P. aeruginosa clinical isolates (247). It is noteworthy that some of these isolates were also resistant to colistin and carbapenems, our last line of defense. Importantly, ciprofloxacin-susceptible (MIC of ≤1 μg/ml) or -intermediate (MIC of 2 μg/ml) CLSI breakpoints were reached for most of the fluoroquinolone-resistant clinical isolates in the presence of ≤8 μg/ml (6 μM) of hybrid 11 (247). Measureable in vivo potency of this hybrid was demonstrated by using the Galleria mellonella larva infection model. In spite of the inherent limitations of this model, including a lack of adaptive immune responses in insects and the ease of manipulation, the remarkable similarities between the innate immune response of G. mellonella and those of vertebrates make this model desirable for studying the virulence of bacteria and the efficacy of antimicrobial agents (258). Upon injection of the larvae with a lethal load of XDR P. aeruginosa strain 101856, a 1:1 concoction of hybrid 11 and moxifloxacin (37.5 mg/kg each) administered as a single dosage at 2 h postinfection resulted in 86% survival of the larvae after 24 h. In comparison, single-dose monotherapy with either hybrid 11 or moxifloxacin alone (50 mg/kg each) was ineffective and resulted in 100% killing within the same time frame. In vitro biochemical assays revealed that the DNA gyrase A- and topoisomerase IV-inhibitory activities of the ciprofloxacin domain of hybrid 11 were retained (3- to 5-fold higher than those of ciprofloxacin), whereas the protein translation-inhibitory properties of the tobramycin domain were lost (156- to 1,290-fold lower than those of tobramycin) (247). These in vitro biochemical observations are consistent with the activities of other aminoglycoside-fluoroquinolone hybrids, e.g., neomycin B-ciprofloxacin hybrids 8 and 9, that were previously reported by another group (219). Mechanistic studies correlated the observed adjuvant effect of tobramycin-ciprofloxacin hybrid 11 to its ability to perturb the OM of P. aeruginosa in a dose-dependent manner, thus facilitating the influx (and bioaccumulation) of antibiotics that are typically unable to cross the OM, such as rifampin, novobiocin, vancomycin, and erythromycin (247). Since one of the main mechanisms of resistance to fluoroquinolones in P. aeruginosa is the overexpression of multidrug efflux pumps composed of a tripartite protein assembly spanning the inner and outer membranes (28), it is possible that the perturbation of the membrane's lipid composition, particularly the lipids localized around the transmembrane protein, could compromise the integrity of efflux pumps and restore the potency of fluoroquinolones. This hypothesis was tested by assessing the interaction(s) between fluoroquinolones and known permeabilizers in P. aeruginosa. Colistin, cetrimonium bromide, benzethonium chloride, and C₁₆-(Dab)₄-NH₂ (259) were unable to potentiate moxifloxacin in XDR P. aeruginosa 96918. This suggests that tobramycin-ciprofloxacin hybrid 11 either possesses another biological mechanism aside from membrane permeabilization or exhibits an augmented membrane interaction relative to those of common permeabilizers. Indeed, fluoroquinolones are known to be rarely potentiated by classical adjuvants against MDR Gram-negative bacteria, especially P. aeruginosa (260). Tobramycin-ciprofloxacin hybrid 11 was relatively nontoxic to human epithelial breast (JIMT1) and prostate (DU145) cancer cell lines (the effect was comparable to those of ciprofloxacin and moxifloxacin), as determined by a 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium (MTS) assay (261), and was nonhemolytic to human erythrocytes (<10% at 1,000 μg/ml) (247). Ciprofloxacin and moxifloxacin are known to exhibit modest cytotoxicity toward these cell lines (262) and were thus used to assess the cytotoxicity of hybrid 11.

Interestingly, amphiphilic aminoglycosides composed of aliphatic hydrocarbons attached to tobramycin (e.g., tobramycin-C₁₄ tether fragment of the hybrids) were found to possess beneficial immunomodulatory functions that may be exploited therapeutically. These amphiphilic tobramycin analogs selectively induced the chemokine IL-8 in macrophages but not growth-related oncogene (Gro-α), proinflammatory cytokines (TNF-α and IL-1β), or the IL-1 antagonist IL-1RA (263). IL-8 is a potent neutrophil chemotactic factor responsible for the migration and activation of monocytes, lymphocytes, basophils, and eosinophils necessary for the resolution of infections (264). Moreover, the production of the LPS-induced proinflammatory cytokine TNF-α was abrogated by amphiphilic tobramycin via a mechanism that is independent of LPS interaction/neutralization, perhaps through an alteration of intracellular signaling downstream of pattern recognition in macrophages (263). This finding was of importance because the significant production of proinflammatory cytokines induced by endotoxins such as LPS has been implicated in septic shock (265, 266). Tobramycin by itself is not known to possess these immunomodulatory properties, but a tobramycin-copper complex has been reported to possess anti-inflammatory properties (267). Host defense peptides (HDPs) such as cathelicidin LL-37 and indolicidin are also known to be potent inducers of IL-8 (268–270), but they can also induce the production of other chemokines (such as monocyte chemoattractant protein 1 [MCP-1] and Gro-α) (268, 271), suggesting a selective chemoattractant ability of amphiphilic tobramycins. Relative to tobramycin, the tested amphiphilic tobramycin analogs exhibited negligible to <10% cytotoxicity against human monocytic THP-1 cells (ATCC TIB-202) at working concentrations (263).

Enthused by the unexpected biological properties of the tobramycin-ciprofloxacin hybrid, the core (and least possible) structural fragment necessary for the adjuvant effect of this scaffold was investigated. It was reasoned (and data later confirmed) that the tobramycin fragment might be critical to the hybrid's scaffold, due to its well-known pleiotropic mechanism of action and/or perhaps its self-promoted uptake mechanism. A nonclassical structure-activity relationship study was therefore instituted to replace the ciprofloxacin domain of tobramycin-ciprofloxacin hybrid 11 with other pharmacophoric fragments (Fig. 9 and 10).

A recent review aptly described the role of proton-dependent processes in the propagation of bacterial infections (298). Membrane-active agents are known to disrupt vital bacterial processes by perturbing the well-ordered membrane constituents, resulting in the loss of transmembrane protein integrity and function but also the loss of transmembrane potential (ion/proton balance) (86, 299). The mechanistic action of membrane disruption may stem from membrane perforation via the formation of transient pores (300), lipid disintegration leading to lysis, and/or the segregation of phospholipids (301). These concentration-dependent effects are usually driven by the amphiphilic nature of the compounds and could range from mild to fatal (302–307). Since the bacterial membrane is central to a host of its vital, survival, and adaptive mechanisms such as energy production via the respiratory chain, quorum sensing, and efflux pumps, etc., agents that alter the transmembrane protein environment (such as membrane charge, fluidity, and thickness) and/or steric hindrance of membrane-embedded proteins can theoretically lead to cell death via the inhibition of signaling cascades. For instance, swimming motility is a flagellum-dependent bacterial movement that is governed by the respiratory chain on the cytoplasmic membrane (273). When cytoplasmic membrane potential or PMF is disrupted, electron transfer across the respiratory chain is inhibited, resulting in a reduction of ATP synthesis, which is essential for flagellar function (308). These flagella (and pili) are surface appendages that serve as the major means of motility (chemotaxis) as well as the anchors that facilitate initial binding to the asialylated glycolipid (asialoGM1) receptor on the host's epithelial cells prior to the destruction the protective glycocalyx (309).

Membrane depolarization can therefore result in the resensitization of MDR pathogens to antibiotics, as the inhibition of signal transduction coordinated by ATP machinery of the respiratory chain could inflict a crippling fitness cost on the bacteria. Some HDPs are endowed with excellent membrane-disruptive capabilities for selectively interacting with bacterial membranes (310) but have severe limitations of protease cleavage and systemic cytotoxicity, etc. (311). Other polycationic agents such as the pentabasic polymyxin analogs that increase OM permeability in Gram-negative bacteria are also relatively toxic (312). Reducing the number of basic groups in polymyxins may represent a path forward to reduce potential nephro- and/or ototoxicity, as evident in preliminary studies with tribasic SPR741 (150). The reduction in basic groups could, however, compromise the ability of the adjuvant to sensitize certain pathogens, such as the inability of SPR741 to synergize antibiotics in P. aeruginosa (149).

The mechanism of action of tobramycin-based hybrids is quite consistent with that of membrane-active agents that target cellular energetics of prokaryotes. The cellular uptake and mode of action of this scaffold, particularly in but not limited to P. aeruginosa, involve the following (Fig. 12).

Competitive displacement of divalent cation cross-bridges that stabilize the LPS structure on the outer leaflet of the OM, in a fashion analogous to that of the self-promoted uptake mechanism of aminoglycosides and amphiphilic aminoglycosides (90, 224, 312–322), results in OM destabilization. This destabilization facilitates the permeation of other antibiotics into the periplasm and may explain the observed sensitization of MDR Gram-negative bacteria to membrane-impermeable antibiotics such as rifampin, novobiocin, and vancomycin, etc.

Insertion into the OM, presumably promoted by the hydrophobic aliphatic hydrocarbon linker, allows tobramycin-based hybrids to reach the periplasm. This may further augment the entry of other membrane-impermeable antibiotics, as reflected by their synergistic relationship with the hybrids.

Perturbation of transmembrane efflux protein domains prevents the relay of signaling cascades required to elicit conformational changes necessary to extrude substrate molecules (such as antibiotics). The loss of protein integrity may be induced by the hybrid directly, via hydrogen bonding with exposed residues, or indirectly, by altering the lipid composition surrounding the protein. This may in part explain the observed enhancement of the intracellular concentration (synergy) of efflux-susceptible antibiotics such as fluoroquinolones.

Dissipation of the membrane potential (ΔΨ) component of PMF leads to compensatory transmembrane pH gradient (ΔpH) adjustments to maintain a constant PMF. PMF is composed of an electrical component (ΔΨ) and a proton component (ΔpH), which are complementary to one another. If one component is altered, the bacteria compensate for and ensure a stable PMF by adjusting the other complementary component (59). This phenomenon was previously described as being the basis for the potentiating effects of the antidiarrheal drug loperamide (276). Tobramycin hybrids dissipate ΔΨ (at sub-MIC values), thereby prompting the pathogen to raise its transmembrane ΔpH. This view is supported by the following findings: (i) the swimming motility (a ΔΨ-controlled process) of P. aeruginosa was severely constrained in the presence of the hybrids; (ii) the hybrids could not potentiate the aminoglycoside class of antibiotics, as they require an optimal ΔΨ component for cytoplasmic uptake; and (iii) the strong synergy between tobramycin-based hybrids and minocycline (and other tetracyclines) is indicative of an increased ΔpH component. Tetracyclines are known to penetrate bacterial cells in a ΔpH-dependent fashion (64). Moreover, the inability of tobramycin-based hybrids to potentiate meropenem, which is exclusively taken up by active transport, may indirectly suggest that their effects on PMF not only affect efflux but also may reduce active energy-dependent transport.

A combination of any or all of the above-described mechanisms, simultaneously or sequentially, is most likely responsible for the adjuvant effects of tobramycin-based hybrids. Although these hybrids target bacterial membrane energetics, some FDA-approved nonantibiotics have also been shown to target and disrupt the physical integrity of this membrane (59, 274, 276). This suggests that these membrane-active compounds may similarly be tolerated in humans, as evident in pilot studies with G. mellonella larvae and human epithelial cancer cells.

Indeed, the characteristic mechanism of action of this tobramycin hybrid scaffold is what sets it apart from those of other hybrid antibiotics discussed above, and we encourage everyone involved in antimicrobial drug discovery to partake in this exciting area of investigation.