Antibiotics accumulate in Klebsiella pneumoniae liver abscesses but fail to eliminate antibiotic-tolerant populations

Significance

Klebsiella pneumoniae liver abscesses are serious infections that often relapse or require surgery, even when treated with appropriate antibiotics. In this study, we developed a mouse model and used advanced imaging techniques to investigate why antibiotics may fail to clear these infections. Our findings show that despite adequate drug penetration to the site of infection, K. pneumoniae can adopt a phenotypic state that renders it tolerant to antibiotics. This tolerance is distinct from traditional antibiotic resistance and could contribute to treatment failure. By shifting the focus beyond resistance alone, this work highlights the need to better understand and target tolerance mechanisms, an important step as hypervirulent strains of K. pneumoniae continue to spread globally.

Abstract

Liver abscesses caused by hypervirulent Klebsiella pneumoniae (hvKp) can lead to severe metastatic complications, with mortality rates ranging from 5 to 40%. Even in the absence of antibiotic resistance, hvKp liver abscesses often respond poorly to treatment, sometimes requiring surgical resection. The reason for these poor outcomes remains unknown. Here, we established an hvKp wound model in outbred immunocompetent mice, which progresses to systemic infection and hepatic abscesses that were intractable to antibiotic therapy. Using a combination of quantitative and infrared matrix-assisted laser desorption electrospray ionization imaging mass spectrometry, we found that antibiotics fail to kill K. pneumoniae in liver abscesses, independently of resistance or spatial distribution of antibiotics. Our results show that antibiotic concentrations detected in the liver are sufficient to eradicate hvKp under standard in vitro conditions, but the continued presence of viable bacteria in vivo indicates that hvKp adopts an antibiotic-tolerant state within the liver. Notably, the inadequate antibiotic efficacy observed in our mouse studies mirrors clinical outcomes. These findings underscore the urgent need to elucidate the mechanism underlying hvKp tolerance, which could inform the development of novel therapeutic approaches.

Classical Klebsiella pneumoniae (cKp) isolates typically cause hospital-acquired infections, such as pneumonia, bacteremia, and urinary tract infections (UTIs) in immunocompromised individuals (1). Over the past three decades, a distinctive pathotype of tissue-invasive hypervirulent K. pneumoniae (hvKp) has gained attention for its capacity to induce severe disease, including intractable liver abscesses in immunocompetent individuals. The frequency of hvKp infections is highest among populations in Asian countries, yet global prevalence is rising and hvKp is emerging as a significant cause of pyogenic (pus filled) liver abscesses in the United States and Europe (2). HvKp expresses a range of virulence factors including siderophores, capsule, and a hypermucoviscous (HMV) phenotype which facilitate community-acquired infection of healthy individuals (2–4). Capsule is an outer polysaccharide layer that facilitates immune evasion (5, 6). HMV is capsule dependent, but is phenotypically distinct, and has been shown to impede interactions with immune cells (4, 7, 8). In vitro, the HMV phenotype is characterized by a viscous filament >5 mm in length when stretched from a colony (3, 4). Strains of K. pneumoniae that cause liver abscesses in patients are typically associated with HMV production (7, 9).

K. pneumoniae asymptomatically colonizes the mucosal epithelium of the gut and the oropharynx (1). Typically, liver abscesses originate when bowel leakage allows hvKp to migrate from the gut through the portal vein to the liver (10). However, liver abscesses can also arise from other primary sources including wound or lung infections (11). Once a liver abscess is established, these infections become challenging to treat with antibiotics and are at risk of progressing to disseminated infections such as endophthalmitis or meningitis, resulting in vision loss or death (3, 12). Clinical studies have reported mortality rates for hvKp liver abscesses ranging from 5 to 40% (13–15). Despite the clinical significance of these infections, very little is known about the cause of treatment failure of KLAs.

Bacteria can survive antibiotic treatment through phenotypic adaptations that do not involve genetic resistance (16, 17). These adaptations often involve entering a slow-growing or nonreplicating state, allowing them to endure high concentrations of antibiotics. While antibiotic tolerance is known to contribute to the treatment failure of Staphylococcus aureus and Escherichia coli infections in patients (18, 19), its role in KLAs remains unknown. Several conditions known to promote antibiotic tolerance in other bacterial species, including nutrient limitation (20, 21), oxidative and nitrosative stress (22–25), hypoxia (26), and acidic pH (21, 27), have been detected in liver infections or abscess environments (27–30). Abscesses have also been shown to limit antibiotic penetration (27), which could contribute to treatment failure. However, whether treatment failure in KLAs is driven primarily by limited drug penetration, phenotypic tolerance, or both, has not yet been investigated. Identifying the dominant mechanism underlying treatment failure is a critical step toward improving therapeutic strategies to eradicate abscesses effectively.

Previous attempts to study K. pneumoniae liver colonization have employed intravenous or intraperitoneal mouse infection models (5, 31–33). These models have shown that Kupffer cells play a primary role in eliminating K. pneumoniae from the liver sinusoids and that although there are over 77 distinct capsule (K) types identified through phenotypic studies, K1 and K2 capsule types are most resistant to clearance by Kupffer cells (5, 33). These findings correlate with the clinical observation that KLAs are almost exclusively caused by K1 or K2 serotypes (4, 34, 35). Despite these important findings, a limitation of these studies is that infection with K1 or K2 serotypes of hvKp typically results in death of the mice within 1 to 2 d postinfection (PI) which is prior to the formation of macroabscesses (5, 33). In this study, we have developed an hvKp wound infection model that results in systemic infection including the formation of hepatic abscesses. Importantly, mice do not succumb to infection until at least 4 d PI, making it a useful model for examining abscess development and the efficacy of frontline antibiotics against KLAs. Here, we examine the contribution of antibiotic penetration and tolerance to treatment failure of KLAs for three distinct clinically relevant antibiotics.

Results

HvKp Wound Infection Progresses to Liver Abscesses.

To examine a disseminating infection model, a full-thickness excision wound was created on the back of SKH-1 outbred immunocompetent hairless mice (Fig. 1A). Outbred mice were chosen for their increased genetic diversity, which better emulate the genetic diversity of human populations (36). The wounds were infected with hvKp strain KPPR1S. KPPR1S has been used extensively in the literature to study colonization and virulence in gut and pneumonia infection models (37, 38). This strain was selected as it displays the HMV phenotype (conferred by the rmpD gene) which is a characteristic of most hvKp strains and is associated with KLAs in humans (7, 9). Most mice succumbed to the infection on day 5 PI (Fig. 1B). Mice were euthanized according to the approved procedure by the University of North Carolina at Chapel Hill Institutional Animal Care and Use Committee (IACUC), the burden of KPPR1S in the wound had expanded compared to the inoculum (Fig. 1C) and the bacteria had disseminated to the liver, kidney, and spleen (Fig. 1 D–F). Importantly, HMV-filaments were visible during excision of the wound (Fig. 1A), indicating that HMV was produced in this infection model. Although bacteria had disseminated to the liver, kidney, and spleen (Fig. 1 C–F), abscesses were clearly observed on the liver (Fig. 1G) but were not immediately obvious in the other organs. Similar trends were observed with the hvKp liver abscess strain SGH10 (39, 40) (Fig. 1 C–F).

Fig. 1.

HMV is required for virulence in a murine K. pneumoniae infection model. (A) Schematic representing a wound infection model which leads to systemic infection. Briefly, a 4-mm full-thickness circular wound was created on the back of 10- to 14-wk-old SKH-1 immunocompetent outbred mice. The wound was infected with ~5 × 10⁷ CFU of hvKp KPPR1S (HMV+), SGH10 (HMV+), KPPR1S rmpD (HMV−), or the classical K. pneumoniae isolate INF168 (HMV−). The red arrow indicates HMV string characteristic of KPPR1S. (B) Mortality of mice infected with KPPR1S (C–F) On day 4 PI, mice were euthanized, and the bacterial burden was enumerated in (C) wound, (D) liver, (E) kidney, and (F) spleen. (G) Photograph of excised liver containing abscesses outlined by white boxes. Statistical significance was determined using Kruskal–Wallis one-way ANOVA with Dunn’s multiple comparison test (n=5 mice/group). *, **, ns, denotes P < 0.0.5, P < 0.005, not significant, respectively. LOD, limit of detection.

To determine whether HMV was required for virulence in this model, we infected the mice with two strains that do not express HMV; a KPPR1S rmpD mutant (7) and a classical K. pneumoniae INF168, originally isolated from a patient wound infection (41). At 4 d PI, the rmpD mutant did not expand beyond the initial inoculum in the wound and did not display significant organ dissemination (Fig. 1 C–F). Similarly, the classical K. pneumoniae isolate INF168 was avirulent in this model (Fig. 1 C–F). Together, these data indicate that HMV is a major mediator of KLA in this model, emulating clinical findings (35).

Since liver abscesses were visible on day 4 PI, we examined histological sections of liver tissue using a hematoxylin-eosin (H&E) stain over a time course of infection (Fig. 2). On day 1 PI, liver sections were characterized as normal. On day 2 PI, microabscesses, mixed inflammation (neutrophils, macrophages), and bacteria were occasionally observed in these foci (Fig. 2). On day 3 PI, macroabscesses were identified in the liver sections, and coagulation necrosis was evident that is consistent with a thrombotic event (e.g., bacterial emboli; fibrin thrombosis; pigment laden Kupffer cells) (Fig. 2). For comparative analysis, the KPPR1S rmpD mutant (HMV-negative control), and uninfected control had no notable histological findings in the liver (Fig. 2).

Fig. 2.

Hypervirulent K. pneumoniae strain KPPR1S forms liver abscesses. Mouse wounds were infected with ~5 × 10⁷ CFU of KPPR1S or the isogenic rmpD mutant (HMV-negative strain, control). (A) Representative images of microscopic liver findings on days 1 to 3 PI, paraffin-embedded liver sections were stained with a H&E stain. The images from left to right are increasing magnifications of boxed regions in the group (100×, 400×, and 800×, respectively). No abscesses were observed in the wild type KPPR1S on day 1 PI or KPPR1S∆rmpD on day 4 PI. Microabscesses (white arrowheads) characterized by foci of fragmented neutrophils and macrophages were associated with individual hepatocyte necrosis or more extensive coagulation necrosis (* denotes necrotic hepatocytes), were evident in the wild type on day 2 and 3 PI. Day 3 PI, macroabscesses were also present within the liver parenchyma, often associated with bacteria (black arrow) rimmed by fragmented neutrophils and macrophages with extensive acute coagulation and hepatocyte necrosis (*). (B) Microabscess stained by H&E and by immunohistochemistry (IHC) (400× magnification). IHC for neutrophils (Ly6G) demonstrates positively staining cells (brown) primarily localized to the center of the microabscess as well as encircling the central core. IHC for macrophages (IBA1) demonstrates a dense band of positively staining cells (brown) around the margins of the microabscess. Kupffer cells, which line the sinusoids, stain with IBA1, and are evident at the margins of the image. (C) This table demonstrates the average histological finding scores from n = 3 biological replicates.

K. pneumoniae in Liver Abscesses Are Tolerant to Frontline Antibiotics.

Next, we aimed to determine whether the mortality observed on day 5 PI could be prevented or delayed by treating the mice with antibiotics on day 4 PI (Fig. 3). Levofloxacin (a fluoroquinolone), ceftriaxone (a third generation cephalosporin), and ertapenem (a carbapenem) were selected as clinically relevant antibiotics for invasive infections caused by hvKp, including KLAs (12). Following 24 h of antibiotic therapy, the wound and liver were harvested and surviving bacteria were enumerated. Antibiotic treatment was effective in sustaining mouse survival until day 5 PI, and all three antibiotics significantly reduced the bacterial burden in the wound (Fig. 3 A and B and SI Appendix, Fig. S1A). Levofloxacin was the most effective in reducing the bacterial burden in the wound by 96%, followed by ceftriaxone (89%) and ertapenem (88%) (Fig. 3B). In contrast, none of the antibiotics significantly reduced the bacterial burden in the liver (Fig. 3C), despite dosing regimens chosen to mimic pharmacokinetics seen in humans with favorable time above the minimum inhibitory concentrations (MIC) in plasma (42, 43) (SI Appendix, Table S1). Importantly, the reduced antibiotic efficacy observed in the liver was not associated with antibiotic resistance (Fig. 3D).

Fig. 3.

KLAs are recalcitrant to multiple days of frontline antibiotic therapy. Mouse wounds were created and infected with KPPR1S. Bacterial burdens were quantified in the wounds and livers on day 4 PI (baseline CFU in untreated mice), on day 5 PI following 1 d of antibiotic treatment and on day 7 PI following 3 d (days 4 to 6 PI) of treatment with 120 mg/kg ceftriaxone (CTX) by IP injection q12, 90 mg/kg levofloxacin (LEV) by IP injection q12 or 50 mg/kg ertapenem (ERT) by subcutaneous injection (SC) q6. Surviving bacteria in the (A and E) wound and (C and G) liver were enumerated before and after antibiotic challenge. (B and F) The % bacterial survivors in the wound (annotated on top of bars) were calculated by dividing the median survivors after each antibiotic treatment by the median of the untreated group. (D and H) Resistant bacteria were enumerated in the liver after antibiotic challenge. Data from (A–D) n =11 untreated mice (UNT), n = 12 LEV mice, n = 8 CTX or n = 12 ERT or (E–H) n = 13 untreated mice (UNT), n = 10 LEV mice, n = 10 CTX or n = 8 ERT mice/group are shown. The horizontal line indicates the median. To account for the unequal variances between groups, statistical significance was determined using a Brown–Forsythe and Welsh ANOVA multiple comparison test on the raw (non-log-transformed) CFU data. *, **, ***, **** denotes P < 0.05, P < 0.005, P < 0.0005, and P < 0.0001, respectively. q12, every 12 h; q6, every 6 h, ns, not significant. LOD, Limit of detection. See also SI Appendix, Fig. S1.

Abscesses consist of an accumulation of bacteria, neutrophils, and necrotic debris which may be encapsulated (depending on chronicity) by fibrous connective tissue and a rim of leukocytes (27). We reasoned that once abscesses have established, multiple days of antibiotic therapy may be required to significantly reduce the bacterial burden in the liver. Therefore, we examined the efficacy of a 3-d antibiotic regimen initiated on day 4 PI, when macroabscesses are present (Fig. 2). Interestingly, 3 d of levofloxacin treatment significantly reduced the bacterial burden in the wound by over 99% (Fig. 3 E and F). In contrast, levofloxacin efficacy in the liver was inconsistent: levofloxacin cleared the infection in 4 out of 10 mice (Fig. 3G). In the 6 mice where levofloxacin did not work, the infection was equivalent to that in the untreated animals. Thus, in the liver, levofloxacin is either highly effective or entirely ineffective (Fig. 3G). Strikingly, we observed similar trends in mice treated for 3 d with ceftriaxone and ertapenem (Fig. 3G and SI Appendix, Fig. S1). Importantly, the limited efficacy of each antibiotic in the liver was not associated with the development of antibiotic resistance (Fig. 3H). To effectively eliminate a bacterial infection, antibiotics must achieve a sufficiently high concentration and exert activity at the site of infection (44). We reasoned that if the antibiotics are on the cusp of reaching an efficacious concentration in the liver, this may explain their inconsistent efficacy.

Antibiotics Are Detected Within Abscesses.

Next, we aimed to quantify the concentrations of levofloxacin, ceftriaxone, and ertapenem in the plasma and the liver tissue by liquid chromatography tandem mass spectrometry (LC–MS/MS) (Fig. 4). After 3 d of antibiotic therapy, levofloxacin, ceftriaxone, and ertapenem were detected at 24-, 29-, and 2,600-fold higher than the MBC in the liver tissue, respectively (Fig. 4B and SI Appendix, Table S1). To examine whether these antibiotic concentrations were sufficient to kill hvKp, we challenged broth cultures with each of the three antibiotics in vitro. Ceftriaxone and ertapenem sterilized hvKp when cultured to exponential phase but had limited activity against cultures in stationary phase (Fig. 4 C and D). This is not unexpected as β-lactam antibiotics, such as ceftriaxone and ertapenem, target cell wall synthesis, and thus are more effective against growing cells that are actively synthesizing cell wall (45). Fluoroquinolones such as levofloxacin are most effective against actively dividing bacteria as they prevent the religation step during DNA synthesis leading to double strand breaks and cell death (46). However, fluoroquinolones retain some activity against stationary-phase cells, particularly in Gram-negative bacteria (46, 47). Accordingly, levofloxacin eradicated hvKp more rapidly in exponential-phase cultures, but still sterilized cultures grown to stationary phase over 3 d (Fig. 4 C and D). Together, these data indicate that antibiotic concentrations detected in the liver are sufficient to kill hvKp under standard in vitro conditions. However, the persistence of viable bacteria within the liver suggests that hvKp may enter an antibiotic-tolerant state in vivo.

Fig. 4.

Antibiotics achieve therapeutic levels in liver tissue. (A) Mouse wounds were infected with KPPR1S. On day 4 PI, mice were treated with 90 mg/kg levofloxacin (LEV) by IP q12, 120 mg/kg ceftriaxone (CTX) by IP q12 or 50 mg/kg ertapenem (ERT) by SC q6 for 3 d (days 4 to 6). Two hours after the last antibiotic dose, mice were euthanized. Whole blood was collected by cardiac puncture for plasma isolation, and livers were harvested. (B) LEV, CTX, and ERT in plasma and liver tissue were quantified by LC–MS/MS and displayed as fold increase above the minimum bactericidal concentration (MBC) displayed in SI Appendix, Table S1 (LEV MBC 0.16 µg/mL; CTX MBC 0.08 µg/mL; ERT MBC 0.02 µg/mL). The bar graph depicts mean with SD (from n = 3 mice/group). KPPR1S grown to (C) exponential phase or (D) stationary phase in LB broth culture, prior to addition of antibiotics at concentrations that were detected in the liver tissue: 3.9 µg/mL LEV (24× MBC), 2.3 µg/mL CTX (29× MBC), or 52.2 µg/mL ERT (2,600× MBC). At indicated time points, aliquots were washed and surviving CFU were enumerated by plating. The average of n = 6 biological replicates and SD are shown. q12, every 12 h; q6, every 6 h; CTR; control, LOD; limit of detection.

Although the drugs reached therapeutic levels in the bulk liver tissue, it is possible that the drugs are excluded from the abscesses. This hypothesis is supported by previous studies which demonstrate differential penetration of antibiotics into abscesses (25, 27). Fluoroquinolones effectively penetrate human abscesses in various locations (48), although liver abscesses were not included in these studies (27, 49). Additionally, some studies suggest that β-lactams have restricted access into brain and odontogenic abscesses (27, 50–52). However, it is unknown how the frontline antibiotics used to treat KLAs effectively penetrate the site of infection.

To assess drug accumulation within an abscess, the pus is typically drained and analyzed. However, draining the abscess fluid does not allow for differentiation of antibiotic levels across the various layers of the abscess (27). To address this, we employed infrared matrix-assisted electrospray ionization (IR-MALDESI) and mass spectrometry imaging (MSI) to visualize the antibiotic distribution into liver sections containing abscesses. IR-MALDESI MSI is an atmospheric pressure ionization approach that discretely samples the chemical composition from a tissue cryosection at high spatial resolution, allowing mapping of the spatial distribution of targeted analytes (53–55). We employed IR-MALDESI MSI to visualize the distribution of levofloxacin or ceftriaxone, which served as a representative β-lactam in these experiments, throughout cross-sections (n = 3) of liver samples containing abscesses. A full mass spectrum was evaluated at 100 μm increments over the entirety of each liver section, providing simultaneous collection of chemical information for each antibiotic as well as the control ions cholesterol, heme, and bile acid. These spatial distributions were coregistered with H&E images from subjacent tissue sections from each sample to evaluate the distribution of each analyte relative to KLAs (Fig. 5 A and B and SI Appendix, Fig. S2 A and B). When compared to bulk homogenate LC–MS/MS concentrations, the mean antibiotic MSI signal abundance across each section from the same tissue samples increased proportionally for ceftriaxone but not for levofloxacin (SI Appendix, Fig. S2 C and D). While levofloxacin and ceftriaxone exhibited relatively uniform spatial distributions throughout liver cross-sections, we observed two key exceptions. First, areas with high antibiotic signal intensity, particularly levofloxacin, were closely associated with bile acid (SI Appendix, Fig. S2 E and F), suggesting elimination via bile secretion. This antibiotic enrichment in bile is expected since both levofloxacin and ceftriaxone are eliminated through the kidneys and liver. Levofloxacin is primarily cleared by the kidneys, while ceftriaxone undergoes dual renal and biliary excretion (56, 57). Second, regions with lower levofloxacin signal abundance corresponded to areas of elevated cholesterol (Fig. 5A), which has previously been used as a marker of inflammation and has been linked to necrotic tissue (58). The mean levofloxacin signal abundance in areas of elevated cholesterol is reduced by ~43% of the signal in the rest of the tissue (Fig. 5A). Immunohistochemical analysis of adjacent tissue sections, supported by pathologist annotations, confirmed that these regions with reduced levofloxacin aligned with microscopic findings of necrosis and abscessation in which bacteria were readily apparent (Fig. 5A). Together, this suggests that although there is reduced levofloxacin detected within abscesses, it is detected at therapeutic levels (~10-fold higher than the MBC) within abscesses. These regions of levofloxacin penetration heterogeneity may explain the differences observed with LC–MS/MS homogenate concentrations (SI Appendix, Fig. S2C).

Fig. 5.

Levofloxacin and ceftriaxone are detected within abscesses. Mouse wounds were infected with KPPR1S. On day 4 PI, mice were treated with 90 mg/kg levofloxacin (LEV) by IP q12 or 120 mg/kg ceftriaxone (CTX) by IP q12 for 3 d (days 4 to 6). Two hours after the last antibiotic dose, mice were euthanized, and liver tissue was harvested and scrutinized with IR-MALDESI and MSI. (A and B) X, Y-axes reflect distance in units in mm. Color scale reflects measured signal abundance of cholesterol, CTX, LEV, heme, and bile acid (taurocholic acid) ions. The arbitrary units are proportional to the number of ions that can be correlated to a concentration with calibration. The resulting MSI image data were then coregistered with H&E-stained sections to evaluate distribution of analyte relative to abscess populations (annotated with a blue outline). q12, every 12 h. n = 1 biological sample. Additional replicates are in SI Appendix, Fig. S2.

In contrast to levofloxacin, the average ceftriaxone MSI response within necrotic tissue regions was not reduced compared to surrounding tissue (Fig. 5B). Linking this homogeneous IR-MALDESI MSI signal abundance with quantitative LC–MS/MS from the same tissues suggests that ceftriaxone is present at concentrations that are ~30 fold higher than the MBC present within the abscesses (SI Appendix, Fig. S2D). These findings strongly suggest that although therapeutic drug levels are achieved at the infection site, antibiotic treatment was insufficient to eradicate KLAs. As bactericidal concentrations are achieved but bacterial death does not occur, liver abscesses appear to demonstrate antibiotic tolerance.

Our imaging data revealed that levofloxacin concentrations were diminished in necrotic regions of KLAs, which also showed marked cholesterol enrichment (Fig. 5). To test whether cholesterol contributes to antibiotic tolerance, we preexposed exponential-phase hvKp cultures to cholesterol before treating with levofloxacin or ceftriaxone at concentrations matching those observed in infected liver tissue. Despite its enrichment in the necrotic core, cholesterol had no effect on bacterial survival during antibiotic challenge compared to the surfactant vehicle control (SI Appendix, Fig. S3A). The surfactant used to solubilize cholesterol, methyl-β-cyclodextrin (MβCD), did increase tolerance to levofloxacin (SI Appendix, Fig. S3 A–C). Cyclodextrins are known to extract lipids from bacterial membranes (59), and this membrane disruption could potentially trigger a protective stress response or tolerant state in hvKp. Although MβCD is a synthetic compound not found in vivo, it may mimic the effects of naturally occurring surfactant-like host lipids or other membrane-active molecules that accumulate in necrotic tissue and alter bacterial susceptibility. The observation that neither bile acid nor heme, additional endogenous compounds identified in our MSI data, altered antibiotic susceptibility underscores the specificity of this effect (SI Appendix, Fig. S3 B and C). While the mechanism of host-induced antibiotic tolerance remains to be elucidated, our findings highlight the complex interplay between host factors and antibiotic activity in infected tissues, which will be further explored in follow-up studies.

Discussion

Even in the absence of antibiotic resistance, treatment of KLAs is challenging, frequently resulting in relapsing infections, metastatic infection or the need for surgical liver resection. KLA treatment usually combines systemic antibiotic therapy and abscess drainage (12). Antibiotic therapy is typically based on in vitro antibiotic susceptibility testing, which is poorly predictive of efficacy in vivo (60, 61). To date, there have been no randomized clinical trials to compare the efficacy of different antibiotic therapies against KLAs. As a result, the optimal choice of antibiotic therapy remains a subject of ongoing debate (62–64).

First-generation cephalosporins, alone or in combination with aminoglycosides, were traditionally favored for KLAs due to their cost-effectiveness, efficacy, and the susceptibility of most hvKp strains (63, 65). However, although aminoglycosides enhance bactericidal activity of cephalosporins against some Gram-negative infections, studies have not shown superior clinical outcomes over cephalosporin monotherapy for KLAs (62–64). Additionally, the risk of nephrotoxicity often outweighs potential benefits, especially in critically ill patients. More recently, third-generation cephalosporins such as ceftriaxone have become the preferred treatment due to their enhanced β-lactamase stability, longer half-life allowing for once-daily dosing, and dual hepatic-renal elimination, which benefits patients with liver or kidney impairment (66–68). However, even with appropriate antibiotic therapy (based on in vitro susceptibility testing), clinical outcomes remain variable. In some clinical cases, extended intravenous therapy with ceftriaxone combined with abscess drainage was insufficient to reduce abscess size or eliminate fever, even in patients without underlying conditions, necessitating the surgical removal of an infected portion of the liver (69–71). Conversely, other studies have reported successful resolution of hvKp liver abscesses with ceftriaxone therapy alone, without the need for liver resection (72–74). Importantly, these contrasting clinical outcomes are also reflected in this study’s mouse model.

A recent study from Taiwan found that patients with hvKp liver abscesses that were treated with fluoroquinolones had a shorter hospital stay and shorter intravenous antibiotic duration than those treated with β-lactams (75). Human and animal pharmacokinetic data indicate that levofloxacin rapidly penetrates liver tissue and can effectively penetrate abscesses in various anatomical locations (27, 49, 76). Our study corroborates these findings using high-resolution MSI, which uniquely enabled direct visualization of antibiotic distribution within abscesses. However, despite therapeutic levels of levofloxacin in the liver, treatment failure still occurred in 60% of animals, suggesting that the phenotypic state of the bacterial population likely plays a role in treatment failure. Most conventional antibiotics are effective at killing actively growing bacteria as they typically target active cellular processes such as cell wall synthesis or DNA replication (44, 77). Antibiotic tolerant cells are phenotypic variants in a nongrowing, metabolically indolent state that can survive lethal doses of antibiotics without acquiring genetic resistance mechanisms (16, 17). While the specific stressors that induce antibiotic tolerance in K. pneumoniae during infection remain unexplored, various conditions known to promote antibiotic tolerance in other bacterial species, such as nutrient or oxygen deprivation (20, 21, 26), oxidative or nitrosative stress (22–25), and acidic pH (21, 27), are detected during liver infection or within the abscess environment (27–30). Our study indicates that antibiotic tolerance is a key contributor to treatment failure of KLAs. A strength of this study is the use of outbred mice, which offer greater genetic diversity and may better reflect variability in human responses. The reason for the striking contrast in treatment outcomes, where 40% of the mice cleared the infection with antibiotic therapy while the remaining 60% showed no efficacy, is not yet known, but may be attributed to host-specific factors which could impact the induction of tolerant populations or bacterial clearance in the presence of antibiotics (22–25).

One limitation of our study is that the duration of antibiotic treatment in humans is typically longer than is feasible in a mouse model. Despite these differences, our study remains valuable as it demonstrates treatment failure patterns that closely mirror those observed in clinical cases. This alignment suggests that our findings are relevant and can provide insights into the challenges of treating hvKp infections and paving the way for improved therapeutic strategies.

Previous studies have implicated tolerant cells as reservoirs for the evolution of resistance, both in vitro and in patients with S. aureus bacteremia (18, 78, 79), underscoring the importance of understanding the precise cause of treatment failure for each antibiotic. As convergent K. pneumoniae strains that display hypervirulence and antibiotic resistance continue to spread (80–84), it is increasingly important to define the limitations of current therapies and identify strategies to improve treatment outcomes for KLA infections. Future research will focus on elucidating the mechanisms underlying antibiotic tolerance within KLAs, with the goal of informing the development of new therapeutic approaches. This may include host-directed strategies aimed at reducing the induction of tolerant bacterial populations or enhancing the efficacy of existing antibiotics. In parallel, the use of antibiotics or adjuvant compounds with specific activity against tolerant populations represents a promising avenue to improve treatment outcomes.

Materials and Methods

Bacterial Strains and Growth Conditions.

K. pneumoniae strains KPPR1S [a derivative of ATCC43816, which originated from a patient with pneumonia, K2 capsule type (4, 85, 86)], KPPR1SΔrmpD (7), SGH10 [originated from a liver abscess, K1 capsule type (39, 40)] and classical isolate INF168 [which originated from a wound infection, K107 capsule type (41)] were grown overnight in Luria Broth (LB) at 200 rpm at 37 °C.

In Vivo Studies.

SKH1-elite outbred hairless mice (strain code 477) were purchased from Charles River Laboratories and bred in a pathogen-specific free facility. Mice were housed in a controlled environment that meets all regulatory standards for animal care. They are provided with ad libitum access to food and water, ensuring continuous nourishment. The housing rooms are maintained on a 12-h light/dark cycle to support normal circadian rhythms. Temperature and humidity levels are carefully regulated to ensure animal well-being. All animal protocols were approved by the Institutional Animal Care and Use Committee at the University of North Carolina at Chapel Hill and met guidelines of the US NIH for the humane care of animals. Mice were euthanized before the planned time point if they underwent ≥20% weight loss compared to the weight at the start of the experiment and/or severe lethargy or any other behavior indicative of undue stress.

Mouse Infections.

Ten- to fourteen-week-old male and female mice were used for animal infection models. The wound model was modified from our previous study (87). Briefly, general anesthesia was induced using 4% isoflurane at a flow rate of 1 L/min in 95%oxygen/5% CO₂. 1 to 3% isoflurane was used to maintain anesthesia while a 4-mm full thickness circular excision wound was created on surgically prepped skin on the left hind back (over the abdomen). Wounds were infected with ~5 × 10⁷ CFU of K. pneumoniae and covered with an occlusive dressing. For pain reduction, 5 mg/kg meloxicam (Covetrus) was administered once daily for 3 d by subcutaneous (SC) injection. Where indicated, mice were treated with antibiotics 90 mg/kg levofloxacin (Cayman) (via intraperitoneal (IP) injection q12), 120 mg/kg ceftriaxone (MedChemExpress) (via IP injection q12) or 50 mg/kg ertapenem (MedChemExpress) (via SC injection q6). Ceftriaxone and ertapenem doses were selected to emulate pharmacokinetics seen in humans with favorable time above the MICs in plasma (42, 43). Levofloxacin dosing was chosen based on a previous study (88). At 1 to 7 d postinfection, mice were humanely euthanized, and the wound tissue, liver, kidneys, and spleen were harvested and homogenized. Serial dilutions were plated on LB agar to enumerate bacterial survivors.

Mice treated with levofloxacin or ceftriaxone on days 4 to 6 PI and were euthanized on day 6 (2 h after the last antibiotic dose), whole blood was collected by cardiac puncture for plasma extraction. Livers were collected and bisected for LC–MS/MS and MALDESI MSI.

To ensure antibiotic treatment failure was not due to the generation of spontaneous antibiotic resistance, organ homogenates from all antibiotic-treated mice were plated on agar plates with and without antibiotics. Resistant mutants grow on media containing antibiotics plates while bacteria surviving due to poor drug penetration or antibiotic tolerance will not. Bacterial burden was enumerated for CFU/g. The number of animals and the statistical analyses used to generate figures are described in the figure legends. Blinding or randomization was not necessary as outputs (CFU/g tissue) are objective.

Histopathology and Microscopy of the Liver.

On days 1 to 3, mice with infected wounds, or uninfected control were humanely euthanized, and livers were collected and placed in 10% neutral buffered formalin for 24 h. Tissues were transferred to the Pathology Services Core Facility at the University of North Carolina, Chapel Hill, NC where they were routinely processed to paraffin, embedded, sectioned to 5 µm, and stained using H&E or by IHC (400x magnification) for neutrophils (Ly6G) and macrophages (IBA1). Stained sections of the liver (n = 3 per group) were microscopically evaluated by a board-certified veterinary pathologist at the Pathology Services Core at UNC and qualitatively evaluated on a scale of 0 to 5: 0 = no finding; 1 = minimal finding; 2 = mild finding; 3 = moderate finding; 4 = marked finding; 5 = severe finding or P = the finding is present (not graded). When present usually only one large abscess was identified in the submitted sample and was marked as present rather than graded. Microabscesses were characterized as small foci of neutrophils and macrophages with or without associated hepatocyte necrosis. Scoring was based on distribution and qualitatively assessed number of microabscesses. Coagulation necrosis was evaluated similarly.

MIC Assays.

MIC assays were determined using a microdilution method. ~5 × 10⁵ CFU of K. pneumoniae isolates in LB broth were incubated with various concentrations of levofloxacin, ceftriaxone, or ertapenem in a 96-well plate. The plate was covered with a Breathe-Easier sealing strip (Sigma) and incubated at 37 °C for 24 h. The MIC was determined to be the lowest concentration of the antibiotic which prevented bacterial growth. Assays were performed in biological triplicate.

MBC Assays.

MIC assays were performed as described above and after 24 h of incubation, the dilution representing the MIC and three of the more concentrated wells were serially diluted and plated onto LB agar and enumerated to determine viable CFU/mL. MBC was determined as the lowest concentration that demonstrated a 99.9% reduction in CFU/mL compared to the input CFU. Assays were performed in biological triplicate.

LC–MS/MS.

Plasma was separated from whole blood by centrifugation. Ceftriaxone was extracted from plasma samples by protein precipitation with methanol containing the stable, isotopically labeled internal standard, ceftriaxone-d3. Samples were vortexed, centrifuged, and mixed 1:10 with water prior to analysis. Tissue samples were initially homogenized in 70:30 acetonitrile:water. Resulting homogenates were mixed with methanol containing ceftriaxone-d3, vortexed, centrifuged, and diluted 1:10 with water prior to analysis. Plasma and tissue extracts were both analyzed by LC–MS/MS under gradient conditions on a Waters Atlantis T3 (50 × 2.1 mm, 3 mm particle size) analytical column with 10 mM ammonium acetate (mobile phase A) and methanol (mobile phase B) with detection on an AB Sciex API-5000 triple quadrupole mass spectrometer. The calibration ranges were 500 to 50,000 ng/mL (plasma) and 25 to 12,500 ng/mL (tissue homogenates) with calibration standard and QC values within 15% of nominal concentrations.

Levofloxacin was extracted from plasma samples by protein precipitation with methanol containing the stable, isotopically labeled internal standard, 13C, 2H3-levofloxacin. Samples were vortexed, centrifuged, and mixed 1:9 with water prior to analysis. Tissue samples were initially homogenized in 70:30 acetonitrile:water. Resulting homogenates were mixed with 50:50 methanol:water containing 13C, 2H3-levofloxacin, vortexed, centrifuged, and transferred to a 96-well plate for analysis. Plasma and tissue extracts were both analyzed by LC–MS/MS under gradient conditions on a Waters Atlantis T3 (50 × 2.1 mm, 3 mm particle size) analytical column with 0.1% formic acid in water (mobile phase A) and methanol (mobile phase B) with detection on an AB Sciex API-6500 triple quadrupole mass spectrometer. The calibration ranges were 5.00 to 5,000 ng/mL for both plasma and tissue homogenates with calibration standard and QC values within 15% of nominal concentrations.

Ertapenem was extracted from plasma samples by protein precipitation with methanol containing the stable, isotopically labeled internal standard, 2H4-ertapenem. Samples were vortexed, centrifuged, and mixed 1:1 with water prior to analysis. Tissue samples were initially homogenized in 70:30 acetonitrile:water. Resulting homogenates were mixed with methanol containing 2H4-ertapenem, vortexed, centrifuged, and mixed 1:1 with water prior to analysis. Plasma and tissue extracts were both analyzed by LC–MS/MS under gradient conditions on a Waters Atlantis T3 (50 × 2.1 mm, 3 mm particle size) analytical column with 0.1% formic acid in water (mobile phase A) and 0.1% formic acid in acetonitrile (mobile phase B) with detection on an AB Sciex API-5000 triple quadrupole mass spectrometer. The calibration ranges were 50.0 to 50,000 ng/mL for both plasma and tissue homogenates with calibration standard and QC values within 15% of nominal concentrations.

In Vitro Killing Assays.

KPPR1S cultures were grown overnight in Luria Bertani (LB) and back-diluted to 1:100 in fresh media and grown for 3 h (exponential phase) or 16 to 18 h (stationary phase). Starting CFU were enumerated by plating serial dilutions on LB agar. Cultures were treated with antibiotics at concentrations that were detected by LC–MS/MS in the bulk liver tissue: 3.9 µg/mL levofloxacin (24× MBC), 2.3 µg/mL ceftriaxone (29× MBC) or 52.2 µg/mL ertapenem (2,600× MBC). At indicated time points, aliquots were washed in PBS and surviving CFU were enumerated by plating. The average of n = 6 biological replicates (performed on 2 separate days) and SD are shown.

To determine whether cholesterol, bile acid, or heme influenced antibiotic efficacy, KPPR1S cultures grown overnight in Luria Bertani (LB) broth, back-diluted 1:100, and incubated for 2 h 40 min in fresh LB media (1.25× for cholesterol experiments, 1× for bile acid and heme experiments). Cultures were supplemented with 600 µL water-soluble cholesterol (final concentration 3 mg/mL; Sigma C4951), 600 µL methyl-β-cyclodextrin (MβCD) (55.46 mg/mL stock; Sigma M7439; surfactant vehicle control in the cholesterol experiments), taurocholic acid (final concentration 80 µg/mL; MedChem HY-N0545), hemin (final concentration 10 µg/mL; Sigma H9039, protected from light), or equivalent DMSO controls prior to addition of antibiotics. Starting CFU were enumerated by plating serial dilutions on LB agar. Cultures were then treated with antibiotics at concentrations detected in liver abscesses by LC–MS/MS: 1.6 µg/mL levofloxacin (10× MBC) or 2.3 µg/mL ceftriaxone (29× MBC). Surviving CFU were quantified at the indicated time points by plating serial dilutions. Data represent the average ± SD of six biological replicates performed across two independent experiments.

IR-MALDESI MSI.

Extracted murine liver tissue was bisected, with one-half flash-frozen on dry ice for 5 min and subsequently stored at −80 °C. MSI was performed on cryostat cross-sections and sequential sections were sent to the UNC Pathology core for H&E staining and abscess identification. Cryosections were thaw-mounted directly onto glass slides for MSI, performed as described previously (53, 89). MSI experiments were conducted using an IR-MALDESI source coupled with a Thermo QE+ mass spectrometer. The sample was translated on an actuated stage in 100 μm intervals between sampling events until the entirety of a tissue section has been evaluated. All acquisitions were performed in positive ionization mode at a resolution of 140,000 (at m/z 200) with a spatial resolution of 100 µm. Data were processed using MSiReader software. Calibration was performed using a series of 0.1 µL calibration spots (0 to 32 µg/mL) spiked onto blank murine liver tissue. By collecting a full mass spectrum at each sampling point, chemical information is collected simultaneously for antibiotics (levofloxacin or ceftriaxone) and control ions: cholesterol (m/z 369.3509), heme (m/z 616.1768), and putatively identified taurine-conjugated bile acid (m/z 516.299). The two-step laser desorption/ionization IR-MALDESI process caused in-source fragmentation of ceftriaxone in initial characterization using a calibrant spiked onto blank tissue, producing two predominant peaks (m/z 324.058 and 352.053). While these peaks were absent in predicted ceftriaxone fragmentation spectra from human metabolomic and DrugBank databases, they were validated as fragment transitions through direct injection of a ceftriaxone standard using the HESI ion source. Levofloxacin was primarily detected as a protonated molecular ion (m/z 362.1511). Initial IR-MALDESI analysis using an electrospray solvent with 50% methanol as the organic component provided insufficient sensitivity for levofloxacin detection at clinically relevant concentrations. However, substituting 75% acetonitrile as the organic fraction of the ESI solvent improved sensitivity more than fourfold. Optimized ceftriaxone and levofloxacin assays enabled the consistent detection of antibiotics in all dosed samples. The resulting MSI data were then coregistered with information derived from H&E-stained sections to evaluate distribution of analyte relative to abscess populations.

Quantification and Statistical Analyses.

Statistical significance was determined using Prism 10 (GraphPad Software, Inc., La Jolla, CA). Statistical tests used and the number of replicates used to generate figures are described in the figure legends.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Raw data used to generate figures are available in a persistent repository (UNC Dataverse) at https://doi.org/10.15139/S3/SMOVNJ (90).