In situ identification of Gram-negative bacteria in human lungs using a topical fluorescent peptide targeting lipid A

See allHide authors and affiliations

Science Translational Medicine  24 Oct 2018:
Vol. 10, Issue 464, eaal0033
DOI: 10.1126/scitranslmed.aal0033

Lung infection in real time

Lung infections are frequent causes of complications in mechanically ventilated and immunosuppressed patients. However, the diagnosis is challenging, requires risky procedures, and is time consuming. Now, Akram et al. have developed an imaging method that is able to detect Gram-negative bacteria in real time in the distal part of the human lung. Using a fluorescent probe binding to lipid A, a molecule expressed on Gram-negative bacterial membranes, in combination with an optical endomicroscope, the researchers rapidly detected Gram-negative infections in distal airways in hospitalized individuals. The results suggest that the approach could accelerate the diagnosis of bacterial lung infection and facilitate the evaluation of antibiotic treatment efficacy.


Respiratory infections in mechanically ventilated patients caused by Gram-negative bacteria are a major cause of morbidity. Rapid and unequivocal determination of the presence, localization, and abundance of bacteria is critical for positive resolution of the infections and could be used for patient stratification and for monitoring treatment efficacy. Here, we developed an in situ approach to visualize Gram-negative bacterial species and cellular infiltrates in distal human lungs in real time. We used optical endomicroscopy to visualize a water-soluble optical imaging probe based on the antimicrobial peptide polymyxin conjugated to an environmentally sensitive fluorophore. The probe was chemically stable and nontoxic and, after in-human intrapulmonary microdosing, enabled the specific detection of Gram-negative bacteria in distal human airways and alveoli within minutes. The results suggest that pulmonary molecular imaging using a topically administered fluorescent probe targeting bacterial lipid A is safe and practical, enabling rapid in situ identification of Gram-negative bacteria in humans.


Gram-negative pulmonary infection (GNPI) is a frequent and severe consequence of hospitalization, immunosuppression, and mechanical ventilation (1). Unfortunately, the accurate and rapid diagnosis of Gram-negative pneumonia is challenging. The gold standard diagnostic in situ procedure for pneumonia (defined as infection in the gas-exchanging regions of the human lung alongside a host cellular response) remains biopsy and culture (2, 3). However, biopsy is rarely performed as a diagnostic tool for pneumonia and would have safety concerns if performed in mechanically ventilated patients.

The accurate determination of the etiology of new pulmonary infiltrates on the chest x-ray (CXR) represents one of the major diagnostic challenges in mechanically ventilated patients in the intensive care unit (ICU). The clinical suspicion of nosocomial pneumonia is highly sensitive (4), often leading to inappropriate and/or overtreatment with broad-spectrum antimicrobial therapy. Current approaches to guide antimicrobial therapy for suspected pneumonia rely on the growth of bacteria from aspirated fluids or expectorated samples along with antimicrobial susceptibility testing (46). In mechanically ventilated patients and immunosuppressed lung transplant patients, bronchoalveolar lavage (BAL) is a routinely used sampling methodology. However, BAL has suboptimal specificity and sensitivity (7) and is associated with substantial delays because culture results take up to 72 hours (8). Furthermore, although BAL aims to sample the alveolar space, it is invariably contaminated by tracheobronchial organisms, resulting in low specificity (6, 9) and subsequent overtreatment. Similarly, negative culture results from distal airway sampling may lack validity as a result of concurrent antimicrobial therapy (10) or poor sampling (4, 11, 12). In addition, molecular techniques, such as polymerase chain reaction that uses amplification and sequencing of potential microbial species in expectorated or sampled fluids, are inherently oversensitive (13).

Given these limitations of existing approaches, an in situ methodology for the rapid and accurate identification of GNPI has potential benefits. Therefore, we developed and tested the feasibility of an optical molecular imaging (OMI) approach combining intrapulmonary delivery of a fluorescent probe (SmartProbe) specific for the presence of Gram-negative bacteria with optical endomicroscopy (OEM) of the distal lung.

We investigated the utility of polymyxins (PMXs), which are cationic, amphipathic, cyclic antimicrobial peptides, naturally formed by the bacterium Paenibacillus polymyxa (14), as a selective binding agent for lipid A of lipopolysaccharide (LPS) on the outer membrane of Gram-negative bacteria (15, 16). We conjugated PMX to an environmentally sensitive fluorophore, 7-nitrobenz-2-oxa-1,3-diazole (NBD), generating NBD-PMX that had excellent signal-to-noise ratios with fluorescent amplification only upon entry into the hydrophobic environment of the bacterial membrane. After in vitro and ex vivo validation in a ventilated ovine lung model, the compound underwent preclinical toxicology and was manufactured to good manufacturing practice (GMP) for a first-in-human study. Here, we used NBD-PMX with OEM in patients with bronchiectasis and in a cohort of critically ill mechanically ventilated patients in the ICU, who had suspected pneumonia, to rapidly and specifically visualize Gram-negative bacteria alongside autofluorescent cellular infiltrates in the distal lung. This method showed the potential to provide in situ molecular imaging of specific bacteria in humans that might be used as a diagnostic and monitoring tool and also as a platform to increase understanding of the pathophysiology of microbial infection.


Structure-activity relationship of NBD-PMX constructs demonstrate an optimal SmartProbe to selectively label Gram-negative bacteria with high signal-to-noise ratio in alveolar tissue

A panel of modified PMX constructs was chemically synthesized and evaluated for specificity against Pseudomonas aeruginosa and Staphylococcus aureus. PMX was modified by removal of the hydrophobic tail and two amino acid residues and replaced with various linkers attached to the fluorophore NBD (without alteration of cyclic ring component). Progressive lengthening of linker led to loss of Gram selectivity (fig. S1). The lead compound (NBD-PMX; Fig. 1A) with an amino-PEG2-carboxylate spacer showed fluorescence amplification in increasingly polar environments generated by increasing dimethyl sulfoxide (DMSO) concentration (Fig. 1B), demonstrating the required environmental reporting for fluorescence amplification upon bacterial membrane insertion in vivo, where wash steps are not possible. Moreover, the lead compound labeled P. aeruginosa in a concentration-dependent manner with high signal-to-noise ratio (Fig. 1C).

Fig. 1 NBD-PMX labels Gram-negative bacteria in a concentration-dependent manner, with fluorescence amplification in hydrophobic environments.

(A) Structure of NBD-PMX. (B) NBD-PMX fluorescence in relation to increasing concentrations of DMSO (NBD-PMX at 5 μM), n = 3. (C) Left: Fluorescence quantification of P. aeruginosa imaged on a benchtop confocal microscope in the continued presence of increasing concentrations of NBD-PMX. Images show representative images at denoted concentration of NBD-PMX. Scale bar, 5 μm. Right: Each point on the graph represents the means (±SEM) of three independent experiments (n = 3) where at least three fields of view were quantified with a single site nonlinear fit of data. (D) Representative flow histograms for unstained bacteria (gray histograms) or NBD-PMX–stained bacteria (5 μM; dashed line). Graph shows quantification of flow cytometry data for unstained bacteria (white bars, normalized) and NBD-PMX (gray bars). Bars represent means (±SEM) from n = 3. Analysis by Student’s t test, *P < 0.05, **P < 0.01. ns, not significant. (E) Bacterial panel with NBD-PMX (1 μM, green) and counterstain with Syto 82 (red). Gram-positive bacteria are indicated by the dashed box and are shown with their counterstain to demonstrate correct focal plane. Scale bars, 5 μm. (F) Quantification of bacterial panel with NBD-PMX (1 μM) with Gram-positive (white bars) and Gram-negative bacteria (gray bars). Bars show mean fluorescence (±SEM) from n = 3, where at least three fields of view were assessed. Analysis by Student’s t test, *P < 0.05, **P < 0.01, ***P < 0.001.

Gram selectivity was assessed against a broad and clinically relevant panel of organisms and demonstrated labeling selectivity for Gram-negative bacterial species P. aeruginosa, Klebsiella pneumoniae, Escherichia coli, Haemophilus influenzae, Acinetobacter baumannii, and Stenotrophomonas maltophilia but not for the Gram-positive bacterial species methicillin-sensitive S. aureus (MSSA), methicillin-resistant S. aureus (MRSA), and Streptococcus pneumoniae (Fig. 1, D to F). To demonstrate selectivity over mammalian cells, we imaged NBD-PMX in coculture with freshly isolated human peripheral blood granulocytes and mononuclear cells and also with alveolar macrophages retrieved from BAL (Fig. 2, A to C). Furthermore, there was no labeling of human lung epithelial cells demonstrated in a bacterial human lung coculture experiment (Fig. 2D), and low concentrations of NBD-PMX preferentially labeled bacteria in the presence of surfactant (fig. S2).

Fig. 2 NBD-PMX shows selectivity for bacteria over mammalian cells.

Coculture experiments of freshly isolated human neutrophils (A), freshly isolated human mononuclear cells (B), and human alveolar macrophages retrieved on BAL (C) with P. aeruginosa imaged in the continued presence of with NBD-PMX (5 μM). Green panels demonstrate NBD fluorescence, middle panels demonstrate nuclear counterstain (Syto 60), and right panels show merged images. Plot profiles corresponding to yellow arrows are shown on the right. AU, arbitrary units. (D) Confocal image of human lung tissue cocultured with bacteria and imaged after labeling with NBD-PMX (5 μM). Left: NBD and autofluorescence with excitation at 488 nm. Right: Merge with counterstain. White arrows indicate bacterial labeling, blue arrows demonstrate epithelial cells, and yellow arrows demonstrate elastin autofluorescence. All experiments are n = 3, representative images shown. Scale bars, 10 μm.

To determine the mode of action of NBD-PMX, the Burkholderia cenocepacia strain K56-2, which is highly resistant to PMX through mutations in its lipid A component (17, 18), was treated with the NBD-PMX probe. This showed a lack of binding, suggesting that NBD-PMX labeling of Gram-negative bacteria is through lipid A binding (fig. S3).

NBD-PMX demonstrated no toxicity and chemical stability

NBD-PMX demonstrated no membrane toxicity in an in vitro hemolysis assay and no toxicity or pulmonary inflammatory cell recruitment after intratracheal instillation in mice (fig. S4). Furthermore, no adverse histopathological observations were noted between the two groups at 48 hours or day 14 after administration in the lung, liver, or kidney (fig. S4). This was confirmed by a contract research organization (CRO) good laboratory practice (GLP) rat single-dose intratracheal toxicology study (with over 600 times the human dose based on lung weight), where no changes in weight, hematological, coagulation, and clinical chemistry parameters were observed (table S1). NBD-PMX was manufactured to GMP, and stability studies (24 months) demonstrated chemical stability of the aqueous drug product under frozen (−20°C) storage conditions (table S2).

NBD-PMX visualizes Gram-negative bacterial bioburden in the distal lung with topical microdosed endobronchial delivery and OEM in a large animal ex vivo lung model

An important technical prerequisite of the approach required evidence that bacteria could be rapidly detected in the distal alveolar regions with a clinically relevant limit of detection (LoD) and sampling methodology. We developed an ovine ex vivo whole lung ventilated model of distal lung bacterial burden to evaluate NBD-PMX against a diverse panel of clinically relevant pathogenic bacteria in a human size–relevant lung model (Fig. 3A). This model allowed the use of clinical equipment and the thorough testing of the technical feasibility of delivering microdoses (<100 μg) of NBD-PMX and detection of labeled bacteria in distinct bronchopulmonary segments after regional OEM. OEM enables the delineation between bronchiolar imaging and alveolar imaging through the very distinct autofluorescence patterns of elastin. The presence and distribution of bacteria were initially modeled in situ with the instillation of green fluorescent protein (GFP) S. aureus, which demonstrated a characteristic imaging pattern of punctate twinkling bacteria (movie S1). Anatomically distinct bronchopulmonary segments were instilled with bacteria or vehicle control; subsequently, microdoses of NBD-PMX were instilled and lung segments were immediately imaged using OEM. This was achieved by passing an OEM fiber into disparate bronchopulmonary segments (up to five passes per bronchopulmonary segment to capture a regional representation of fluorescence), recording images at 12 frames/s and imaging for up to 5 min. The regional exploration of the bronchopulmonary segments was intended to mimic the clinical scenario of performing sampling in bronchopulmonary segments guided by CXR infiltrates. Using this approach, we could detect Gram-negative bacteria, such as P. aeruginosa, K. pneumoniae, and E. coli, but not the Gram-positive bacteria MSSA, MRSA, and S. pneumoniae (Fig. 3, B and C). An image analysis algorithm was developed to objectively detect bacteria and analyze and quantify the videos on a frame-by-frame basis (data file S1). The analysis confirmed the ability of NBD-PMX to discriminate Gram-negative bacteria from Gram-positive and buffer control segments (Fig. 3, D to F) with bacteria visualized within minutes of NBD-PMX instillation. The imaging algorithm was developed to discriminate a clinically relevant LoD of 1 × 105 colony-forming units (CFU)/ml in BAL. To test whether the algorithm was able to distinguish various bioburdens of bacteria, we instilled different concentrations of E. coli and performed imaging as described above. Using the developed algorithm, ranges of 1 × 105 to 1 × 109 bacterial bioburden in the distal alveoli were distinguishable (fig. S5).

Fig. 3 NBD-PMX labels Gram-negative, but not Gram-positive, bacteria in situ in ex vivo ovine lungs.

(A) Experimental setup. Image demonstrating the anatomically distinct pulmonary segments of the ovine lung and timeline outlining the experimental protocol of retrieval, ventilation, bacterial instillation, and NBD-PMX instillation and OEM imaging. (B) Representative OEM images of control, Gram-positive, positive control (GFP S. aureus) and Gram-negative segments. (C) Lavage was enumerated for bacterial counts, n = 4 for all bacteria except S. pneumoniae where n = 3, analysis by one-way analysis of variance (ANOVA). (D) Analysis of entire videos for control segments (n = 7), MSSA (n = 5), MRSA (n = 4), S. pneumoniae (n = 4), P. aeruginosa (n = 4), K. pneumoniae (n = 5), and E. coli segments (n = 5) showing the proportion of frames with >80 dots per frame for the Gram-negative (gray bars) and Gram-positive segments (white bars) and control; bars represent means (±SEM). *P < 0.05, **P < 0.01 by Student’s t test. Receiver operator characteristics of image analysis videos. (E) For control (n = 7) or all Gram-negative videos (n = 14), the area under the curve was 1.0 [95% confidence interval (95% CI), 1.0 to 1.0], P = 0.0002592. (F) For Gram-negative segments (n = 14) compared to Gram-positive segments (n = 13), the area under the curve (AUC) was 0.9945 (95% CI, 0.9768 to 1.012), P < 0.001.

NBD-PMX enables immediate in situ visualization of Gram-negative bacteria in the distal lung of bronchiectasis patients

Bronchiectasis is a chronic suppurative pulmonary disease characterized by distal lung bacterial colonization and repeated infective exacerbations (19) with pathogens, often Gram-negative, that are commonly found in mechanically ventilated patients and patients who receive lung allografts. To test the clinical potential of this imaging method, GMP-manufactured NBD-PMX was topically administered as a microdose via an intrapulmonary catheter during a bronchoscopy procedure to six patients with bronchiectasis, and imaging of the distal airways and alveoli was initiated immediately after NBD-PMX administration. Four females and two males with a mean age of 65 comprised the study population, and no patient had a serious adverse event (table S3). Minor adverse events such as cough attributable to the bronchoscopy procedure were observed in three patients. BAL and sputum culture were performed after the imaging procedure, and bacterial species were identified. The imaging videos obtained after NBD-PMX demonstrated that the method was able to detect the presence (or absence) of Gram-negative bacteria (movies S2 to S7).

Furthermore, using an image analysis algorithm analogous to that used in the ovine ex vivo model, and by deploying an arbitrary cutoff value determined by the ex vivo model but adjusted for the larger field of view in the clinical system (data file S2), we distinguished positive from negative frames (Fig. 4). Negative bacterial signal was seen in three patients: D1, D3, and D4. Upon culture, patients D1 and D3 grew only a Gram-positive organism (S. aureus and S. pneumoniae, respectively), and D4 grew a PMX-resistant Gram-negative organism (Proteus mirabilis; Fig. 4 and movies S2, S4, and S5). Positive bacterial signals were seen in D2, D5, and D6. D2 and D5 grew P. aeruginosa (Fig. 4 and movies S3 and S6), whereas D6 grew two different species, S. pneumoniae and H. influenzae. Therefore, this positive signal was attributed to the presence of H. influenzae. Bacterial signals were observed in the distal alveolar regions, which conventionally are considered to be low in bacterial bioburden in chronic suppurative lung disease at postmortem (20). Bacterial aggregates were also observed in the distal bronchioles of these patients (fig. S6).

Fig. 4 NBD-PMX labels Gram-negative bacteria in vivo in humans when administered endobronchially and imaged with OEM.

(A) Representative alveolar images of baseline imaging (left) and after administration of NBD-PMX (middle) in six patients with bronchiectasis, with single-frame analysis shown on the right. (B) Graph demonstrates summary image analysis of individual frames from baseline imaging and after NBD-PMX. Mean number of frames per video analysis was 443 frames; data shown as proportion of frames with >160 dots detected per frame.

NBD-PMX and OEM enable immediate in situ visualization of Gram-negative bacteria and cellular infiltrates in the distal lung of mechanically ventilated ICU patients

We then performed the procedure in seven mechanically ventilated ICU patients with pulmonary infiltrates and suspected pneumonia to demonstrate early feasibility and assess safety and to discern in an unselected cohort whether a qualitative signature of infection in the distal lung could be observed. One patient was excluded from the analysis because no videos could be analyzed and no BAL was retrieved. In six patients, imaging and BAL analysis were performed to provide initial proof of concept (Fig. 5, table S4, and movies S8 to S13). No study-related serious adverse events were noted, and the average procedure time (including BAL) was 14.7 min with 4.8 passes in the endobronchial segments imaged. Two patients, D7 and D11 (Fig. 5 and movies S8 and S11), demonstrated a characteristic bacterial signal and an autofluorescent cellular infiltrate. BAL yielded no significant bacterial growth for patient D7, which was expected because the patient was on multiple antimicrobials including metronidazole, ciprofloxacin, vancomycin, and the antifungal fluconazole for a perforated abdominal viscus. BAL from patient D11 demonstrated growth of the Gram-negative bacteria Klebsiella oxytoca at low levels (BAL microbial culture was likely suppressed by the previous administration of intravenous temocillin; table S4) and the fungus Candida albicans. Ex vivo labeling of BAL from D11 with NBD-PMX demonstrated labeling of Klebsiella oxytoca but not of Candida albicans (fig. S7). Patients D8 and D13 demonstrated no bacterial labeling and no inflammatory cellular infiltrate in the alveoli but did demonstrate the absence of well-aerated alveoli, indicated by condensed/compressed elastin structures (Fig. 5 and movies S9 and S13) consistent with atelectasis of the segment. Growth of S. maltophilia was demonstrated for D8 in both tracheal aspirates before the procedure and on BAL at 1 × 104 CFU/ml. Given the absence of distal alveolar bacterial signal and the absence of a cellular infiltrate, this was attributed to airway colonization. Patients D9 and D12 both demonstrated no distal alveolar bacteria and no cellular infiltrate, and BAL grew bacteria below the threshold for ventilator-associated pneumonia (VAP), consistent with the absence of VAP. Patient D12 developed a pulmonary abscess before imaging, which was additionally identified on OEM by the absence of elastin autofluorescence. Although these are preliminary studies in an unselected cohort of mechanically ventilated critically ill patients with new pulmonary infiltrates on CXR, they demonstrate the feasibility of this point-of-care technology platform and the utility of a qualitative imaging signature of bacterial presence and alveolar cellular infiltrate.

Fig. 5 NBD-PMX administration and imaging in six mechanically ventilated patients in ICU demonstrates a Gram-negative bacterial signal in two patients.

Representative alveolar images of baseline imaging (left) and after administration of NBD-PMX in six patients with pulmonary infiltrates and suspected pneumonia. Cellular infiltrates are indicated by white arrowheads. Graph demonstrates the frame-by-frame analysis of each video sequence after NBD-PMX administration. Each point represents analysis of a single frame, and the vertical line represents the mean, with the percentage above the threshold of 160 dots per frame shown on the right in red.


The inexorable rise of antimicrobial resistance (AMR) has in part been driven by the indiscriminate administration of antibiotics to individuals suspected of having infections (21). Upper airway and bronchoscopic sampling and subsequent microbiological culture are routinely used in mechanically ventilated patients and in patients who have received lung allografts. However, these diagnostic approaches have limited sensitivity and specificity and significant delays in reporting (7, 8), such that empirical antibiotic therapy is frequently initiated, leading to inappropriate and/or overtreatment with broad-spectrum antibiotics, which leads to increased morbidity, exposes patients to potential drug toxicity, promotes AMR, increases the risk of antibiotic-associated infections, and wastes health service resource (2124). Moreover, there are no internationally agreed refinement/deescalation protocols for antimicrobial therapy within ICU (21). Therefore, technologies that enable bacterial detection, bacterial bioburden, and bedside qualitative measure of pulmonary infection in the distal lung may have significant utility. Molecular imaging of infection in humans has been dominated by nuclear medicine approaches, and although these do offer whole-body imaging, they are expensive, cumbersome, limited to large centers, are not point of care, offer poor specificity and variable sensitivity (25), and have not been widely adopted. Using whole-body approaches, it is impossible to delineate true distal lung alveolar infection from tracheobronchial colonization and, in the specific case of mechanically ventilated patients, adds time, safety, and often transport challenges because critically ill patients need to be moved to the scanners.

Although there have been significant technical developments in developing bacterial OMI for small animals or tissue explants (2629), clinical translation is mostly lacking. Partnering intrapulmonary microdosed fluorescent probes with OEM is a platform technology enabling molecular imaging of the distal lung at high resolution, with the potential for serial exploration of pulmonary segments. The distal airway and alveolar elastin networks allow the clear visualization of the spatial location within the lung through autofluorescence imaging at 488 nm of excitation (30). The OEM fiber is only extended through the working channel of the bronchoscope once it is wedged in the distal lung. This simple method avoids proximal tracheobronchial contamination, which is a major issue with existing methods (6). The targeting nature of the NBD-PMX coupled with selected distal lung optical imaging potentially ensures high specificity, and coupling the bacterial imaging with the imaging signature of a cellular infiltrate has the added potential to be more predictive of alveolar infection versus colonization. The true signature of infection versus colonization requires larger studies with further evaluation compared to accepted reference standards.

In clinical practice, PMX isomers are used as antimicrobial agents (31, 32); hence, we chose to use PMX as a binding ligand for OMI. PMX was linked to an environmentally sensitive fluorophore to allow signal generation only upon bacterial membrane engagement and generation of high signal-to-noise ratio when administered topically into the human lung. Salient chemical modifications of PMX enabled specific and sensitive Gram-negative bacterial molecular imaging. We retained the positively charged diaminobutyric acid residues on the cyclic ring, which are critical for overall cationic charge and bacterial binding (33, 34). Structure-activity relationship studies showed that the linker length between the PMX and NBD was crucial for binding, whereas the amphipathic properties of NBD-PMX were maintained by the hydrophobic residues that remained in positions 6 and 7 (d-phenylalanine and leucine, respectively) (16). Mechanisms of PMX resistance remain largely confined to alterations to the lipid A component of LPS (35, 36), and because PMX resistance in nosocomial Gram-negative infections is rare, even among multidrug resistance strains (32, 37, 38), NBD-PMX will have wide applicability.

To extend the possibility of using such a technology platform to monitor treatment response, bioburden of bacteria in the distal lung was distinguishable. GNPI is emerging as one of the most difficult infections to treat due to rising multidrug resistance (39, 40), and thus, the ability for repeated visualization of bacterial bioburden in the distal lung to serially monitor pathogen bioburden may help to refine/escalate and deescalate therapy. Although we have showed here that the use of OEM was safe, repeated visualization does require repeat bronchoscopy, and consequently, the associated risks of repeated use of this technology will need to be weighed against the information it could provide.

The challenge of distinguishing de novo bacterial infection versus resolving infection or colonization is an important clinical question and necessitates novel technology development. In the six bronchiectasis patients, we observed Gram-specific imaging of bacteria, and in the ICU patients, we observed qualitative signatures of infection in two patients. In this small clinical study, NBD-PMX did not detect P. mirabilis, which is known to be PMX resistant. P. mirabilis is an uncommon cause of nosocomial pneumonia (41) and opportunistic pulmonary infection in immunosuppressed patients. This observation did, however, show the potential of the platform to detect AMR.

The study has several limitations at this stage of development including the small cohort of patients, the availability of the technology for clinical applicability, and the anticipated requirement of additional fluorescent probes for polymicrobial infections. To date, the technology platform has demonstrated early-stage safety and feasibility in the ICU; however, the future use of such platforms requires large well-constructed trials to demonstrate efficacy and safety alongside health economic evaluation under the framework of accepted guidelines for performing and reporting of diagnostic accuracy studies (42). Because histopathological confirmation is not a routine test for pneumonia, given potential morbidity and mortality, this technology platform will need to be assessed against the current accepted reference standards for pneumonia diagnosis in the ICU. These multicenter studies will be essential to support the utility of this approach, distinguish colonization from infection, and improve and validate the imaging signature through further refinements of the imaging analysis algorithms.

Second, regarding clinical applicability, the imaging systems are commercially available, but the fiber optic AlveoFlex fibers are expensive, restricted to limited uses and require disinfection between uses. These implementation costs are significant and, along with the paucity of molecular imaging agents that are clinically compatible, have limited the widespread clinical use of this technology platform. However, single-use low-cost disposable fibers, which markedly reduce cost, have now been developed (43). Similarly, the cost of goods of the topically administered imaging fluorescent probes is low, and with the advent of frugal innovation and technology such as wide-field OEM (44), this approach is potentially cost effective after validation in larger trials. However, as with any novel technology platform, there is an expected time lag before widespread applicability including the time for the regulatory approval of imaging fluorescent probes.

Last, although GNPI is the predominant cause of pulmonary infections in lung transplantation (45) and in nosocomial infections in the ICU (41), pneumonia may be caused by Gram-positive or multiple bacterial species (46). Hence, in the longer term, this diagnostic platform will require a matching Gram-positive imaging agent, such as fluorescently labeled vancomycin (29) or other Gram-positive selective ligands, or alongside other bacterial and inflammation-specific probes (47, 48) because OEM has the potential to be multiplexed (25, 44). It is also feasible to exploit other properties to further distinguish bacterial labeling from intrinsic lung autofluorescence such as fluorescence lifetime imaging or a spectrally distinct fluorophore.

The proof-of-concept study was performed to determine whether Gram-negative bacteria could be labeled in situ in the distal human lung; consequently, although our ICU cohort demonstrates promise in this regard, larger studies are needed to fully address the distinction between colonization or true infection with this modality. However, the absence of a true gold standard for bacterial infection in the mechanically ventilated patient makes this challenging. Nevertheless, as demonstrated, high-resolution OEM coupled with a fluorescent probe was able to delineate bacterial presence in distinct segments of the pulmonary tree (such as the acinar gas–exchanging unit) and associated imaging parameters such as cellular infiltrates. The specificity, safety, and rapid readout of this technology platform represent potentially significant advantages over current technologies and now require further large-scale validation.


Study design

The primary objective of the clinical study was to determine whether NBD-PMX could label Gram-negative bacteria in situ in the human lung and be imaged using OEM over human autofluorescence. This was an early-stage single-arm exploratory study assessing the ability to microdose the compound topically in the lung and image in the intended manner. No randomization or blinding was undertaken because all patients received the compound, although all investigators were unaware of the final microbial results at the time of the procedure. Informed consent was taken from all patients or from a personal legal representative. Adverse events and serious adverse events were recorded. The study ( identifier NCT02491164) was approved by regional ethics committee (REC no.: 15/SS/0126).

Patient cohorts and procedure

Two cohorts were assessed: (i) an outpatient population of patients with bronchiectasis and (ii) a cohort of mechanically ventilated patients with suspected pneumonia in the ICU. Patients with known bronchiectasis were recruited from the Regional Specialist Bronchiectasis service. All patients underwent written and informed consent, underwent prebronchoscopy peripheral blood sampling and x-ray imaging, and had previously performed computed tomography (CT) imaging. Bronchoscopy was undertaken in a dedicated endoscopy suite and a flexible fiberoptic bronchoscope (Olympus BF-1T260 or Olympus BF-260), with conscious sedation using intravenous benzodiazepine and/or short-acting opiate, in keeping with routine practice in our institution. Patients were monitored continuously throughout the procedure and subsequently in a clinical environment for 4 hours, after which they were discharged home. All patients were contacted at 24 hours to assess for any adverse events. For the ICU cohort, consent was obtained from a personal legal representative because patients did not have the capacity. The patients were identified by the attending ICU consultant where there was suspected pneumonia, and the patient was due to undergo bronchoscopy and lavage. In the ICU, all procedures were done at the point of care, and the APACHE II (acute physiology and chronic health evaluation II) score (49) and associated hospital mortality probability score were calculated. The segment for NBD-PMX administration and OEM was chosen on the basis of the most affected region based on CT (for bronchiectasis patients) or guided by the CXR changes and endobronchial examination and targeting of the segment with the greatest degree of mucopurulence. If the changes were widespread, then the posterior segment of the right lower lobe would be targeted. For all patients, the bronchoscope was navigated to an affected segment where baseline (prefluorescent probe) imaging was undertaken using a clinically approved 488-nm laser excitation source OEM imaging system (Cellvizio, Mauna Kea Technologies) and a 1.4-mm diameter AlveoFlex fiber recording at 12 frames/s, through the working channel of the bronchoscope. This clinical system recorded a circular field of view of ~600 μm in diameter. GMP-manufactured NBD-PMX was instilled into the bronchopulmonary segment via a flexible catheter [1.5 mm diameter argon plasma coagulation (APC) catheter, Erbe]. After removal of the flexible catheter, the AlveoFlex was reinserted into the working channel, and images were recorded for up to 2 min. BAL was subsequently undertaken in the same segment, and bacteria were enumerated and identified. Sputum culture was also obtained, where possible, for the bronchiectasis cohort. Patients were monitored continuously in ICU after the procedure and were discharged from the study at 4 hours.

Bacterial culture

Bacterial strains were grown and counterstained as previously described (47). Strains used include P. aeruginosa [PA01, American Type Culture Collection (ATCC) 47085], P. aeruginosa (J3284, clinical isolate), A. baumannii (J3433, clinical isolate), S. maltophilia (J3270, clinical isolate), K. pneumoniae (ATCC BAA1706), E. coli (ATCC 25922), H. influenzae (clinical isolate), MRSA (ATCC 25923), MSSA (ATCC 252), S. pneumoniae (D39 NCTC 7466), B. cenocepacia (J2315 and K56-2, both clinical isolates), and GFP S. aureus (RN6390-Gfp-EryR). Bacterial cultures were reconstituted to 0.5 optical density at 595 nm (OD595nm) for confocal assays, 0.01 OD595nm for flow cytometry, or 2 OD595nm for ovine ex vivo lung experiments. Colony-forming units per milliliter were enumerated by plating serial dilution to eighth log10 dilution on broth/blood agar plates and incubation at 37°C for 16 hours (for S. pneumoniae supplemented with 5% CO2). Identification of bacterial species was confirmed through the bacterial diagnostic laboratories, Royal Infirmary of Edinburgh, Edinburgh, United Kingdom.

Neutrophil and monocyte isolation

Cells were isolated from the peripheral blood of healthy human volunteers as previously described (50). Briefly, peripheral venous blood was mixed with 3.8% citrate, centrifuged (350g for 20 min) with removal of the plasma layer, mixed with 6% dextran, and rested to allow sedimentation. The upper layer was removed and layered on discontinuous Percoll gradients of 55/68/81%; the monocyte layer was isolated from the 55/68% interface and neutrophil predominant layer from the 68/81% interface.

BAL macrophage isolation

BAL was collected from patients with pulmonary fibrosis undergoing routine bronchoscopy for a noninfective indication. The bronchoscope was navigated to the right middle lobe and wedged, a BAL was undertaken with 40 ml of instillations of 0.9% NaCl. Retrieved BALF was filtered through a 40-μm cell strainer and centrifuged at 1200 rpm for 10 min at 4°C. The cell pellet was resuspended and used in coculture experiments.

Emission spectra

The fluorescence emission of NBD-PMX solutions was measured in a Synergy H1 Multi-Mode spectrophotometer (BioTek) upon excitation at 450 nm in increasing concentrations of DMSO (Sigma-Aldrich).

In vitro bacterial labeling and confocal assessment

Eight-well Lab-Tek II Confocal Chambers (VWR) were coated in fibronectin (for mammalian cell experiments) or poly-d-lysine (for bacteria) at 37°C for 20 min and then washed in phosphate-buffered saline (PBS). Bacteria were counterstained and coincubated with NBD-PMX at the required concentrations in a single eight-well chamber. For coculture assays, neutrophils, mononuclear cells, or alveolar macrophages were seeded in each well (1 × 105 per well) for 20 min with Syto 60 (5 μM; Invitrogen), and nonadherent cells were aspirated before bacterial and NBD-PMX inoculation. For imaging in human lung tissue, lung tissue was dissected into thin sections and incubated in a 48-well plate with bacteria. To each well, 5 μM Syto 82 (Invitrogen) and NBD-PMX at required concentration were added, incubated for 15 min, and, without a wash step, were placed on a glass slide, and a 13-mm coverslip was placed over the sample and edges were sealed. A laser scanning confocal imaging system (LSM510, Carl Zeiss), incorporating an upright Axioskop FS2 microscope (63× objective), was used. “Green” fluorescence (for NBD) was excited with a dedicated 488-nm line (detected with META detector at 500 to 530 nm), Syto 82 nuclear acid dyes were excited with a dedicated 543-nm line (detected with META detector at 570 to 610 nm), and Syto 60 nuclear acid dyes were excited with a dedicated 633-nm line (detected with a META detector at 660 to 700 nm). Analysis was performed with ImageJ (version 1.46r, National Institutes of Health); the Syto channel was thresholded, and the region of interest generated was quantified on the NBD channel. All experiments were performed at least three times unless otherwise stated. For analyses where direct comparison of fluorescence was made, the confocal settings for the NBD channel were kept identical between experiments.

Flow cytometry

NBD-PMX (50 μl) or PBS was added to 50 μl of counterstained bacteria and incubated for 5 min at 37°C. Samples were diluted to 500 μl immediately before analysis using the BD FACSCalibur (Becton Dickinson, San Jose) flow cytometer capturing 50,000 of counterstained gated events. Voltages remained constant throughout the experiment, and data were collected on a logarithmic scale. Analysis was performed using FlowJo version 7.6.5 (TreeStar Inc.), where the counterstain was gated to eliminate debris and artifact, followed by histogram analysis of the FL-1 (probe) channel. For quantification, the mean of the FL-1 channel was recorded, and data were presented as the mean fluorescence from independent experiments.

GLP toxicity study

A single-dose, intratracheal rat toxicity study was undertaken by Covance Laboratories (Harrogate, UK) under GLP conditions and conducted in accordance with the requirements of the UK Home Office Animals Scientific Procedures Act 1986. Crl:WI(Han) strain rats (Charles River Laboratories) were acclimatized for 2 weeks before a single intratracheal dose of 100 μg of NBD-PMX or PBS control. Animals were monitored daily, provided with ad libitum water and maintenance diet. Animals were weighed daily, and subgroups were sacrificed at 48 hours or day 14 after dosing (dosing day as day 1). Assessments were undertaken by the CRO. Blood sampling from the abdominal aorta of animals at necroscopy was obtained for hematology (by flow cytometry methods), coagulation (by turbidometry methods), and clinical chemistry (by optimized ultraviolet, colorimetric, bromocresol green, or ion-selective electrode methods) and analyzed by the CRO under GLP conditions.

Ovine model

Ewes, which were destined for cull, were terminally euthanized with an overdose of anesthetic. The right pulmonary artery was identified, cannulated, and perfused with 1000 ml of 0.9% NaCl with free drainage from the left ventricle. The trachea was intubated with an 8.0 endotracheal tube, and lungs were placed in a neonatal incubator with an ambient temperature of 37°C and a humidity of 65% and ventilated using a pressure-controlled ventilator (Vivo PV 403, Breas Medical). For positive control data, GFP S. aureus was instilled via a flexible catheter (1.5 mm diameter APC catheter, Erbe) and bronchoscopic guidance into a naive segment and subsequently imaged with an OEM imaging system (Cellvizio Lab, Mauna Kea Technologies), capturing images at 12 frames/s with a field of view of ~305 μm by 430 μm. Imaging was undertaken with a 1.5 mm diameter S-1500 fiber (Mauna Kea Technologies), using a 488-nm laser excitation source with constant laser power between experiments. For in situ bacterial labeling after 1 hour of optimal ventilation, bronchoscopy (Pentax EB-1530 T3 Video Bronchoscope) was undertaken with proximal wedging of the bronchoscope at individual segments, followed by instillation of 2 ml of bacteria or PBS control via a flexible catheter (1.5 mm diameter APC catheter, Erbe). One hour later, a separate catheter was introduced into the working channel, and NBD-PMX was instilled into the distal bronchopulmonary segments. Then, the S-1500 fiber was passed down the working channel, and the same segment was imaged with up to five transbronchial passes and recorded (recorded at 12 frames/s for a minimum of 2 min). For BAL, the bronchoscope was wedged distally, and 20 ml of 0.9% NaCl was instilled and carefully withdrawn with lavage yields of 40 to 50%. Control segments were anatomically distinct and/or in the contralateral lung and imaged before the introduction of bacterial cultures to the lung to prevent contamination. The bronchoscope was decontaminated between each bacterial segment imaged.

Image analysis

Bespoke image processing algorithms were developed to detect bacteria (data files S1 and S2), which appear as bright speckles in OEM imaging, using two commonly used image processing algorithms, namely, Laplacian of Gaussian (LoG) and Difference of Gaussian (DoG) filtering (5153). The algorithm was developed in the MATLAB software environment (MathWorks Inc.), and all digital videos were transferred offline in the manufacturer’s proprietary file format (.mkt) from Cellvizio (Mauna Kea Technologies) to a PC workstation. Each video file was deconstructed into individual frames, with each frame being processed independently. For each frame, irrelevant background content was suppressed by subdividing the image into multiple parts, calculating the average regional pixel signal intensity, and then suppressing all pixels within the region, which were below an experimentally derived cutoff value of between four to six times the SD. Each frame then underwent LoG and DoG filtering to enhance bright spots (combined filter applied); the center of each bright spot was localized, and the size and coordinates of its location were recorded. A further criterion defined the maximum spot size (range from 1 to 7) so as to discard clusters of unresolvable bright/enhanced regions (size of 4 or more) that could not be attributed to the detection of bacteria. Frames were considered positive if there were >80 dots per frame for ex vivo data. This arbitrary cutoff value was selected on the basis of ex vivo negative (PBS segments) and positive control segments (GFP bacteria). Data are presented as the proportion of frames over a video sequence containing >80 dots per frame. For in vivo human imaging, the field of view was doubled and, consequently, a dot per frame of >160 was assigned. Furthermore, because there was a higher proportion of background noise due to increased elastin autofluorescence in aged humans, a second size criterion was applied excluding small single pixel dots (excluding size 1) to diminish noise. Because not all frames contained relevant information, due to positioning of the OEM fiber or movement, redundant frames from the in vivo study were not included in the analysis. For inclusion, a target must have been present in a sequence of greater than five continuous frames. Frames were excluded if any of the following were present: (i) the absence of background fluorescence, (ii) the presence of motion blur, (iii) >50% of image obscured because of fluid bubbles, or (iv) large airway or bronchus imaging.

Statistical analysis

All experiments were performed at least three times, and results were expressed as means ± SEM, unless otherwise stated. Data were analyzed by Student’s t test, two-sample t test, or ANOVA, and analyses are described in each figure legend. Binary classifications are demonstrated as a receiver operator characteristics curve with the AUC and 95% CI. Significance was determined as P < 0.05, and all analyses were undertaken on GraphPad Prism (version 5.01 for Windows, GraphPad Software).



Fig. S1. Progressive lengthening of the linker leads to loss of Gram selectivity.

Fig. S2. NBD-PMX has improved signal-to-noise ratio at lower concentrations in the presence of pulmonary surfactant.

Fig. S3. NBD-PMX binding with PMX-resistant strains of B. cenocepacia is reduced compared to P. aeruginosa.

Fig. S4. NBD-PMX demonstrates no cell membrane toxicity and no in vivo toxicity after intratracheal instillation in mice.

Fig. S5. OEM coupled with NBD-PMX enables bioburden estimation in the distal ovine lung.

Fig. S6. NBD-PMX labels Gram-negative bacteria endobronchially.

Fig. S7. Ex vivo confocal labeling of species from BAL in patient D11.

Fig. S8. Synthetic route of NBD-PMX probe.

Fig. S9. Analytical high-performance liquid chromatography of NBD-PMX.

Fig. S10. Analytical high-resolution mass spectrometry of NBD-PMX.

Table S1. NBD-PMX demonstrates no changes in weight, hematological, coagulation, or clinical chemistry parameters in GLP toxicity study.

Table S2. NBD-PMX is stable in aqueous formulation over 24 months.

Table S3. Bronchiectasis patient demographics, blood results, adverse events, and microbiological growth.

Table S4. ICU patient demographics for six patients included in the analysis, blood results, adverse events, and microbiological growth.

Data file S1. MATLAB file containing image analysis algorithm for preclinical images.

Data file S2. MATLAB file containing image analysis algorithm for clinical images.

Movie S1. GFP-expressing S. aureus instilled in ovine bronchopulmonary segment and imaged with OEM.

Movie S2. Patient D1 imaged after administration of NBD-PMX (no bacterial signal seen).

Movie S3. Patient D2 imaged after administration of NBD-PMX (bacterial signal seen).

Movie S4. Patient D3 imaged after administration of NBD-PMX (no bacterial signal seen).

Movie S5. Patient D4 imaged after administration of NBD-PMX (no bacterial signal seen).

Movie S6. Patient D5 imaged after administration of NBD-PMX (bacterial signal seen).

Movie S7. Patient D6 imaged after administration of NBD-PMX (bacterial signal seen).

Movie S8. ICU patient D7 imaged after administration of NBD-PMX (bacterial signal seen).

Movie S9. ICU patient D8 imaged after administration of NBD-PMX (no bacterial signal seen).

Movie S10. ICU patient D9 imaged after administration of NBD-PMX (no bacterial signal seen).

Movie S11. ICU patient D11 imaged after administration of NBD-PMX (bacterial signal seen).

Movie S12. ICU patient D12 imaged after administration of NBD-PMX (no bacterial signal seen).

Movie S13. ICU patient D13 imaged after administration of NBD-PMX (no bacterial signal seen).

Reference (54)


Acknowledgments: We thank S. Johnston for technical assistance of flow cytometry and M. Vendrell for support for chemical characterization. Funding: This work was supported by Wellcome Trust, the Department of Health Healthcare Innovation Challenge Fund (HICF-0510-069), and the Engineering and Physical Sciences Research Council Interdisciplinary Research Collaboration “Proteus” (EP/K03197X/1). The GMP activities were supported by the National Institute for Health Research (NIHR) BRC GMP Unit at Guy’s and St. Thomas’ NHS Foundation Trust and NIHR Biomedical Research Centre based at Guy’s and St. Thomas’ NHS Foundation Trust and King’s College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health. Ethics statement: All experiments using human samples in vitro were performed after the approval of the appropriate regional ethics committee and with informed consent of the patients [BAL for alveolar macrophages (REC no.: 07/S1102/20), blood for isolation of neutrophils and mononuclear cells (REC no.: 08/S1103/38), and human lung tissue (REC no.: 13/ES/0126)]. Animal experiments were performed under UK Home Office Animals Scientific Procedures Act 1986 (project license number 60/4434). Ovine lungs were from ewes destined for cull and were euthanized under Schedule 1 of Animals (Scientific Procedures) Act 1986. Author contributions: S.V.C., A.M.-F., T.A., N.A., and M.B. performed chemical design and synthesis. A.R.A., N.M., E.S., B.M., C.H., M.B., and K.D. designed and performed the in vitro biological assays. E.S. and K.D. designed and performed the murine work. A.R.A., N.M., T.H.C., D.S.C., E.S., and K.D. designed, set up, and performed the ex vivo ovine lung experiments. A.R.A., T.H.C., A.M., A.T.H., A.M.S., T.W., C.H., and K.D. designed and undertook the in vivo study. A.R.A., N.M., and K.D. undertook data analysis. N.H. and J.R.G. provided cells from BALF and clinical isolates of bacterial strains and microbiology guidance, respectively. C.G. designed the image analysis algorithm with input from A.R.A., N.M., and K.D. C.H., M.B., T.W., and K.D. conceived and supervised the project. A.R.A. and K.D. wrote the manuscript. All authors approved the manuscript. Competing interests: C.H., M.B., and K.D. have a shareholding in Edinburgh Molecular Imaging Ltd. K.D. has received travel and attendance fees from Mauna Kea Technologies as a consultant for an advisory board. A.R.A., T.H.C., N.M., and K.D. have received travel fees for attendance at educational conferences supported by an unrestricted educational grant from Mauna Kea Technologies. T.H.C. has received travel fees for attendance at an educational conference from Edinburgh Molecular Imaging Ltd. K.D., M.B., C.H., T.W., N.M., and N.A. are inventors on a patent (WO2016075483A1) held by the University Court of the University of Edinburgh that covers the probe and method of use. The other authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Data and materials used in the study are available subject to a material transfer agreement and subject to costs for manufacture if required.

Stay Connected to Science Translational Medicine

Navigate This Article