Engineering breath biomarkers for respiratory disease

Abstract

Human breath contains many volatile metabolites. However, only a handful of breath tests are currently used in the clinic to monitor disease due to bottlenecks in biomarker identification. Here we show that we can engineer breath biomarkers for respiratory disease by local delivery of protease-responsive nanoparticles to the lungs. In this work, we built and optimized nanosensors that shed volatile reporters upon cleavage by neutrophil elastase, a protease with elevated activity in multiple lung pathologies including bacterial infection and alpha-1 antitrypsin deficiency (AATD). We demonstrate that, after intrapulmonary delivery into mouse models with acute lung inflammation, volatile reporters are liberated from nanosensors and expelled in breath at levels detectable by mass spectrometry. Furthermore, breath signals can identify diseased mice with high sensitivity as early as 10 min after nanosensor administration. Using our nanosensors, we were able to complete serial breath tests to monitor dynamic changes in neutrophil elastase activity during lung infection and assess the efficacy of a protease inhibitor therapy that directly targets neutrophil elastase for treatment of AATD.

Clinical efforts to monitor states of health and disease have historically centered on blood and urine assays. As an alternative, breath is a practical and potentially informative clinical analyte because it can be sampled non-invasively and contains hundreds of trace volatile organic compounds (VOCs) that are produced in the body as metabolites and from environmental exposure^1–5. However, harnessing endogenous breath volatiles for detecting and monitoring disease has been a challenging and slow process due to several factors. Low and variable VOC concentrations (ppt to low ppm) and confounding VOCs from diet and the environment have hindered the identification of robust breath biomarkers⁶. Furthermore, untargeted approaches historically using gas chromatography-mass spectrometry (GC-MS) to screen as many potential biomarkers as possible suffer from statistical challenges (i.e. false correlations due to overfitting models to high dimensional datasets) and technical challenges (i.e. trade-off between number of VOCs analyzed and individual VOC resolution).⁷ Therefore, few endogenous breath volatiles are currently used in clinical diagnostics.

An alternative to endogenous breath biomarkers is the in vivo administration of exogenous agents that are metabolized into volatile products by disease-specific molecular processes. Examples include the ¹³C-urea breath test for H. pylori detection and ¹³C-methacetin breath test for liver fibrosis, in which isotope-labeled small molecules are ingested and selectively metabolized by H. pylori-secreted urease and cytochrome P450 hepatic enzyme, respectively, into ¹³CO₂^8,9. In another example, small intestinal bacterial overgrowth (SIBO) is diagnosed by tracking peaks in breath hydrogen levels that correspond to bacterial fermentation after disaccharide ingestion¹⁰. These clinical tests leverage known enzyme biology to produce breath readouts, and in the case of ¹³C breath tests, produce a volatile that is not naturally found in the body, thereby maximizing signal-to-noise ratio (SNR). The application of exogenous agents for breath tests have, thus far, been limited to diagnosing gastrointestinal and liver diseases. However, as the direct source of breath, the lungs are optimally situated for interrogation using enzymatic activity-based breath tests. Current diagnostic tools for respiratory illness include chest x-rays, lung function tests (i.e. spirometry), sputum cytology, and/or microbiological tests which have poor specificity or, in the case of microbiological testing, can delay appropriate treatment due to testing time. Breath tests can potentially enable rapid disease detection and monitoring and, in combination with current tests, provide further actionable information at the point of care.

Here we report a class of nanoscale agents, volatile-releasing activity-based nanosensors (vABNs), that were engineered for intrapulmonary delivery and subsequent release of volatile reporters in response to protease activity. Proteases play an active role in the pathology of numerous diseases, including those localized to the lung (e.g. lung cancer, lung fibrosis, COPD, and acute respiratory infections)^11,12. To leverage protease activity for disease detection, we have previously engineered activity-based nanosensors (ABNs) consisting of mass- or ligand-encoded reporters tethered to long-circulating nanoparticle scaffolds via protease-cleavable peptide linkers. After intravenous administration, peptide linkers are cleaved by target proteases in diseased tissues, which releases reporters into circulation for subsequent renal clearance. Similarly, after intrapulmonary delivery, vABNs are exposed to proteases in the diseased lung tissue microenvironment that cleave surface-conjugated peptides and release bio-orthogonal volatile reporters into the alveolar space for exhalation (Fig. 1). Using our initial ABN diagnostic platform, we have previously demonstrated the feasibility of protease-driven urinary readouts for early disease detection, disease classification, and treatment monitoring in preclinical models of cancer, liver fibrosis, thrombosis, and infection^13–24. In this work, neutrophil elastase (NE), a serine protease found in neutrophil azurophil granules, was selected as the target protease. During inflammation, NE is released by neutrophils to kill microbial pathogens and regulate leukocyte recruitment through chemokine proteolysis²⁵. Due to its elastin-rich composition, the lung is also susceptible to damage by elastases. Therefore, NE plays a prolific role in numerous lung diseases²⁶, including pulmonary infection which is the most common risk factor for acute lung injury (ALI)²⁷, and alpha-1 antitrypsin deficiency (AATD) in which NE activity is elevated and damaging to the lungs due to the absence of its inhibitor²⁸. Here, we report the development of the vABN platform and demonstrate the ability to track NE activity in the lungs through breath analysis within ten minutes of vABN delivery. Using vABN breath tests, we show we can non-invasively monitor NE activity in diseases such as lung infection and AATD in which NE has been implicated as a pathology-contributing protease. Furthermore, we demonstrate the use of vABN breath tests to monitor NE inhibition during AATD treatment. .

Figure 1. Schematic of approach.

(a) VOC-modified peptide substrates are formulated into volatile-releasing activity-based nanosensors (vABNs) by conjugation onto an 8-arm PEG nanocarrier and delivered into the lungs via intratracheal injection, (b) Extracellular proteases produced during respiratory disease cleave surface- conjugated peptide substrates thereby releasing VOC reporters. Upon release from substrates, reporters rapidly transition into gas phase and are then exhaled. Upon release from vABNs, VOC reporters transition from an undetectable (grey) to detectable (orange) state, (c) Breath is collected into a receptacle (e.g. evacuated glass vial), and the VOC reporter is quantified using mass spectrometry.

Results

Development of protease-sensing ABNs with a volatile readout (vABNs).

To engineer vABNs, we first established a method to chemically modify peptide substrates to release a volatile reporter upon cleavage. We selected a fluorogenic tetrapeptide substrate optimized by Kasperkiewicz et al. with the sequence Ac-Nle(O-Bzl)-Met(O)₂-Oic-Abu-ACC (ACC = fluorophore reporter 7-amino-4-carbamoylmethylcoumarin) as our model peptide substrate due to its selectivity for human NE, catalytic efficiency (k_cat/K_m = 4.79 × 10⁷ M⁻¹∙s⁻¹), and cleavage site between the C-terminal residue (Abu) and the fluorophore reporter²⁹. Furthermore, this peptide was previously shown to bind specifically to NE in mouse lung lysates³⁰. This cross-species reactivity enables us to evaluate vABNs in mice with the potential for translation in humans. Using the fluorogenic tetrapeptide substrate as a starting point, we hypothesized that an amine-containing VOC could be attached via an amide bond to the peptide C-terminus in place of the fluorophore reporter while maintaining substrate susceptibility to NE. Furthermore, we predicted this attachment chemistry would enable the VOC to be released with its original chemical structure after substrate cleavage and recover its volatile properties to undergo phase transition for mass detection as a gas (Fig. 2a). Hydrofluoroamines (HFAs) with the chemical formula CF₃(CF₂)_xCH₂NH₂ were selected as bio-orthogonal VOC reporters due to their high volatility. In this work, HFA reporters will be referred to as HFAx, where x is the CF₂ chain length (full chemical names and CAS registry numbers provided in Supplementary Table 1). To test our hypothesis, we synthesized an HFA1-modified NE peptide substrate and used mass spectrometry to characterize substrate and reporter behavior in the presence of active NE (Fig. 2b). Before NE addition, no HFA1 was detected in the headspace of open vials containing the solubilized, intact substrate. After addition of purified human NE to the vial, the modified substrate was cleaved between the Abu residue and HFA reporter, and we observed rapid partitioning of freed reporters from solution into the vial headspace.

Figure 2. vABNs are activated by human neutrophil elastase and release a volatile reporter detectable by mass spectrometry.

(a) Chemical structure of the NE peptide substrate with an HFA reporter at the C-terminus. An N- terminal cysteine is included in the hydrophilic linker for conjugation to a nanoparticle carrier, (b) MALDI-MS spectrum confirming release of HFA1 from the peptide substrate after cleavage by NE (bottom) and real-time vapor analysis MS spectrum confirming partitioning of the freed reporter into the reaction headspace (top), (c) After nanoformulation of HFA-1 modified substrates into vABNs, TEM imaging was completed for particle sizing, (d) Measurement of HFA1 released from vABNs as a function of NE concentration, (e) Measurement of HFA1 released from vABNs in response to NE and other respiratory disease-associated proteases to assess sensor specificity. (d, e, mean ± sd., n = 3).

To enable breath-based monitoring of neutrophil elastase activity in the lungs, we formulated the HFA-releasing substrate into vABNs for intrapulmonary delivery. Low molecular weight macromolecules readily escape the lungs by diffusion across the thin alveolar walls into blood circulation³¹. Therefore, HFA1-modified NE substrates were conjugated to 40 kDa 8-arm polyethylene glycol (PEG) nanocarriers, which are well-distributed throughout the lungs after intratracheal instillation (Supplementary Fig. 1) and have a half-life of several hours, which we have shown prolongs substrate retention in the lungs^15,32. The completed vABNs were an average of 8.7±2.1 nm in diameter (Fig. 2c). Following substrate nanoformulation, we sought to confirm that surface-conjugated substrates remain susceptible to NE cleavage. Cleavage assays were completed by incubating vABNs with recombinant or purified proteases in sealed vials for a fixed time, after which signal from accumulated reporters in the vial headspace was measured. vABNs demonstrated a linear dose-dependent response to both human and mouse NE down to picomolar protease concentrations (Fig. 2d, Supplementary Fig. 2a). Furthermore, we confirmed that reporter release was specifically catalyzed by NE with minimal release triggered by proteases upregulated in other lung pathologies (Fig. 2e, Supplementary Fig. 3). We found this to be true both in buffer conditions optimized for each protease (Fig. 2e) as well as in modified Gamble’s simulated lung fluid (SLF), a solution that mimics the fluids and phospholipids released by type II alveolar cells in the deep lung³³. SLF was modified to pH 6.5 to represent pH depression during inflammation³⁴ and supplemented with DPPC, the most abundant phospholipid in lung surfactant³³ (Supplementary Fig. 3).

Development of a mathematical model for in silico prediction of breath signal after vABN delivery.

Having established the feasibility of vABN synthesis, we then sought to optimize vABNs for detection sensitivity, in order to monitor NE activity in mouse models with acute lung inflammation. To aid sensor optimization, we built a physiologically-based pharmacokinetic (PBPK) model to investigate parameters important for breath signal. PBPK models for the respiratory tract have been well-established in the toxicology field.^35–39 By incorporating cleavage kinetics, we were able to develop a PBPK model that is governed by a set of differential equations representing the sequence of events leading to a breath readout following vABN delivery (Fig. 3a, Supplementary text). These events include (i) accumulation of vABNs in diseased lung tissue, (ii) peptide cleavage by proteases/reporter release, (iii) partitioning of freed reporters from lung tissue to the respiratory lumen, (iv) exhalation of freed reporters, and (v) breath analysis. The model output (i.e. breath signal over time) resembles washout curves from existing breath tests in which we observe rapidly rising breath signal due to metabolic breakdown of the substrate, peak breath signal, and a gradual decline in breath signal as the administered dose is depleted ^9,40. We hypothesized that protease abundance, vABN dose, cleavage kinetics, and partitioning of freed reporters into air (over tissue or blood) were key parameters that would influence the generation of breath signal. These parameters were varied in the model to determine how they might affect detection sensitivity. At higher protease concentrations ([NE]), signal curves are narrowed with higher peak breath signal and earlier return of breath signal to baseline, which can be attributed to faster cleavage of the injected vABN dose (Fig. 3b). Of note, peak breath signal is predicted to occur within 15 min of vABN delivery with a return to baseline by 2h, suggesting the potential for a rapid breath readout and repeated dosing for longitudinal measurements. vABN dosing (N) in the micromolar substrate concentration range is predicted to generate breath signal at ppb levels, which is well above the ppt detection limit of mass spectrometry (Fig. 3c). With increasing vABN dose, we observe increased signal intensity and broadening of signal peaks until a vABN dose is achieved such that even at higher doses, the breath signal remains stable due to establishment of a substrate reservoir. While protease concentrations are inherent to the disease state, and sensor dosing can be easily adjusted, sensor cleavage rate and reporter partitioning are governed by vABN composition. vABNs can be synthesized with HFA reporters of varying size, which have different propensities to partition into the three compartments – lung tissue (t), blood (b), or respiratory lumen/air (a). Partitioning coefficients, H_b:a and H_t:a, define the ratio of reporters between two compartments at equilibrium. In the model, we observe decreased breath signal with increasing H_b:a due to loss of reporter molecules to the blood compartment (Fig. 3d).

Figure 3. In silico prediction of breath signal after vABN delivery and in vivo validation of model predictions.

(a) A PBPK model built to predict breath signal from HFA reporters released from vABNs in the lungs. Key parameters in the mathematical model such as (b) NE concentration ([NE]) (c) vABN dose (A/) and (d) blood-air partition coefficient of the reporter (H_ba) were varied to confirm model functionality. The completed model was then used to predict breath signal from vABNs containing reporters HFA1, HFA3, HFA5, and HFA7 to determine optimal reporter size. Parameters that change with reporter size were H_ba and the catalytic efficiency of vABN cleavage (k_cat/K_m). For each vABN, empirical values for both parameters were determined (Table 1) and used in the model to investigate the effect of each parameter in isolation (e, f) and their combined effects (g). A prediction for HFA7 is not included in (e) due to difficulty in solubilizing HFA7 for characterization of its partition coefficients, (h) Schematic of in vivo vABN testing in an acute pneumonia model, (i) Breath signal after intrapulmonary delivery of vABNs at two different doses: 10 or 100 |jM vABN by peptide concentration (mean ± s.d., n =4–5). (j) Breath signal after delivery of 10 pM vABNs with HFA1, HFA3, HFA5, or HFA7 reporters (mean ± s.d., n = 3–4). (k) Immunofluorescence staining for neutrophil elastase (green) in healthy lungs (top left) versus infected lungs (top right). Scalebar = 200 |jm. Fluorescent image showing co-localization of vABNs (magenta) with extracellular neutrophil elastase (green) (bottom left) and inside cells (bottom right). Scalebar = 10 pm. White arrows indicate vABNs of interest.

After establishing our PBPK model, we sought to use the model to predict optimal reporter size for our sensor. Partition coefficients were determined empirically by introducing a fixed quantity of vaporized HFA compound – HFA1, HFA3, or HFA5 – into vials containing blood or lung tissue and by quantifying HFAs in the headspace after equilibration. Partition coefficients for HFA7 could not be determined due to difficulty in solubilizing the pure compound for subsequent evidenced by decreasing H_b:a with increasing reporter size (Table 1). Therefore, if solely considering partitioning, vABNs with larger HFAs (HFA3 and HFA5) are expected to produce greater breath signal than vABNs with smaller HFAs (HFA1) (Fig. 3e). However, we hypothesized that the benefits of partitioning may be offset by changes to cleavage rates upon exchanging HFA1 for larger, bulkier HFA reporters which may adversely affect interaction of the substrate cleavage site with the NE catalytic pocket. To test this hypothesis, a panel of vABNs with HFA1, HFA3, HFA5, or HFA7 reporters were synthesized and catalytic rate constants (k_cat) and Michaelis-Menten constants (K_m) for substrate cleavage were determined empirically to be used in the PBPK model (Table 1). The optimized peptide substrate used in our vABNs was derived from the tetrapeptide substrate, Ac-Ala-Ala-Pro-Val-Acc²⁹. The original and optimized substrates have catalytic efficiencies spanning a difference of four orders of magnitude (10³-10⁷ M⁻¹∙s⁻¹). Our vABNs with HFA1 reporters were found to have a catalytic efficiency falling in the middle of that range (2.79 × 10⁵ M⁻¹∙s⁻¹). The reduced catalytic efficiency is likely due to modification of the P1’ position⁴¹ as well as contribution of nanoparticle sterics to protease-substrate interaction²². The catalytic efficiency is further reduced by up to two orders of magnitude when the HFA1 reporter is replaced by larger reporters. Thus, when catalytic efficiency was considered independent of all other parameters, vABNs with smaller HFAs (HFA1 and HFA3) were predicted to produce breath signal up to an order of magnitude greater than vABNs with larger HFAs (HFA7) (Fig. 3f). The combined effects of reporter partitioning and cleavage rates are less intuitive. The PBPK model predicts that vABNs with HFA1 and HFA3 reporters will produce the highest breath signal within the first 15 min after sensor administration (Fig. 3g). Using the model, we also explored the contribution of particle size to breath signal by varying $k_{Tissue}^{NP}$, the parameter used to describe the rate of particle transport from the lumen into the underlying lung tissue (i.e. the compartment containing the neutrophil elastase) (Supplementary Fig. 11c, Supplementary Text). While nanoparticle formulations of the peptide substrate are predicted to generate prolonged, elevated breath signal in infected mice relative to free peptide, microparticle formulations are predicted to generate sub-ppb breath signal due to poor tissue permeability. Therefore, we focused on nanoformulations of the peptide substrates.

Table 1 |.

Reporter partition coefficients and cleavage kinetic constants for vABNs.

Reporter	MW	Ht:a	Hb:a	Ht:b	Kcat,s−1	Km, μM	Kcat/Km, M−1-S−1 (X 105)
HFA1	149.1	34.49	51.26	0.67	3.10 (±1.15)	10.91 (±1.35)	2.79 (±0.74)
HFA3	249.1	36.64	31.50	12	0.89 (±0.09)	4 42 (±1 42)	2.13 (±0.67)
HFA5	349.1	30.82	18.73	1.65	2.37 (±2.31)	56.13 (±40.73)	0.50 (±0.27)
HFA7	449.1	n a	n a	n a	0.07 (±0.02)	22.34 (±1 84)	0.03 (±0.01)

Catalytic rate constants and Michaelis-Menten constants shown represent the mean ± s.d. (n = 3).

In vivo investigation of protease-driven breath signal.

Under the guidance of in silico predictions, we sought to confirm that vABNs could, in fact, produce a detectable breath signal in response to NE activity in the lungs. To do so, we employed a mouse model of acute pneumonia in which lung infection is established via intratracheal instillation of the gram-negative bacterial pathogen, P. aeruginosa²⁴ (Fig. 3h). In this model, NE protein levels in lung homogenates were found to increase by ~20-fold by 12h after inoculation (Supplementary Fig. 4), which enables rapid evaluation of breath readouts for NE activity. We first sought to generate HFA washout curves for two vABN doses equivalent to 10 and 100 μM NE substrate to verify the production of detectable, dose-dependent breath signal (Fig. 3i). After intrapulmonary delivery of vABNs containing HFA1 reporters, breath samples were collected periodically over 1–3 h using previously reported methods^42,43. Mice were placed in a breath collection apparatus consisting of a 100cc syringe connected to a stopcock valve with a 23G needle. For each timepoint, the syringe was sealed for 2 min to allow breath volatiles to accumulate in the syringe headspace. After 2 min, the valve was opened and 55-cc of headspace was displaced into 5 12-cc Exetainers® by puncturing Exetainer® septa with the needle and pushing the syringe plunger. Reporters in the collected syringe headspace were subsequently quantified using mass spectrometry. As predicted by the model, breath signal was produced in the detectable ppb range. Maximum breath signal for 10 and 100 μM doses from infected mice were 3.3 ± 1.1 ppb and 21.3 ± 8.9 ppb, respectively, and were observed within 10–20 min after vABN administration. Infection breath signal was elevated relative to breath signal from healthy controls and returned to control levels by 1 and 3 h for the low and high dose, respectively (Fig. 3i). Area-under-the-curve (AUC) analysis showed that total exhaled HFA scaled with dose (46.1 and 398 pmol HFA1 for 10 and 100 μM, respectively). As the whole mouse is placed inside the breath collection chamber, we sought to confirm that HFA reporters in the syringe headspace were predominantly derived from the breath without significant contribution from other sources such as the urine. Urine produced during the 30 min period following vABN dosing was collected by voiding the bladder, and the urine was immediately placed in sealed Exetainers®. As in prior studies, reporter signal in the headspace of the Exetainers® was measured using mass spectrometry (Supplementary Fig. 5a). Reporter signal from urine (representative of signal integration over 30 min) was on average <2.1% of the total reporter signal detected in the syringe headspace at the 10 min timepoint (representative of signal integration over 2 min) for the cohort of infected mice. In healthy controls, reporter signal from urine was on average <3.7% of the total reporter signal detected in the syringe headspace. Therefore, reporters in the syringe headspace are predominantly from the breath within our time frame of interest. Interestingly, reporter levels in the headspace of urine samples were still able to differentiate between healthy and infected mice with high sensitivity and specificity (AUROC = 1.0, p = 0.0039) (Supplementary Fig. 5b–c). Given that the HFA1 reporter is not endogenous to biological systems, control experiments were also completed to confirm the absence of signal in the HFA1 detection channel on the mass spectrometer for breath samples from mice not dosed with vABNs (Supplementary Fig. 6).

Having completed the first in vivo demonstration of vABN-generated breath signal, we next sought to validate predictions for optimal vABN composition. Reporter washout curves were established for vABNs with HFA1, HFA3, HFA5, and HFA7 reporters at the 10 μM dose (Fig. 3j). As predicted, vABNs with smaller HFA reporters (HFA1 and HFA3) resulted in the highest peak breath signal and most rapid signal kinetics and, thus, are the best reporters for detection sensitivity. Given that the vABN platform will be used to discriminate between diseased and healthy states, we next investigated diseased breath signal relative to healthy breath signal (signal-to-noise ratio, SNR). SNR values were determined for HFA1- and HFA3-releasing vABNs to identify the sensor with better sensitivity and specificity for elevated NE activity in the lungs. We found that the vABN with HFA1 reporter had higher average SNR at the 10 μM dose (Supplementary Fig. 7).. More importantly, we expected that larger HFA reporters would contribute to greater hydrophobicity of the overall vABN construct and, therefore, lead to nanoparticle aggregation and precipitation after delivery into the lungs. Therefore, vABN stability was assessed in modified Gamble’s simulated lung fluid (SLF). SLF is frequently used in the pharmaceutical sector to assess the dissolution of inhaled drug formulations to determine bioaccessibility³³. vABNs with HFA1 reporters demonstrated the greatest stability in SLF (Supplementary Fig. 8). While vABNs with HFA3 reporter precipitated immediately at 100 μM concentration (Supplementary Fig. 8a), greater than 99% of vABNs with HFA1 reporter remained in solution for up to 24h after sample preparation (Supplementary Fig. 8b). As predicted, HFA5- and HFA7-containing vABNs were also unstable, with 38% and 10% precipitating out of solution, respectively, within the first hour after sample preparation. Similar trends were observed at 10 concentration (Supplementary Fig. 8c). In toxicity studies, we observed no acute toxicity 24h after administration of HFA1-containing vABNs at up to 100 μM concentration (Supplementary Fig. 9). Given its SNR, stability in a simulated lung fluid, and limited toxicity, we moved forward with HFA1-containing vABNs for all subsequent experiments.

Due to the overall increased proteolytic activity in the lungs during infection, we sought to verify that breath signal was specifically driven by NE activity. To do so, PA01-inoculated mice were treated with sivelestat before vABN delivery. Sivelestat specifically inhibits NE and has been previously evaluated in clinical trials for use in preventing infection-associated acute lung injury⁴⁴. In in vitro studies, we verified that mouse NE can be inhibited using sivelestat to prevent cleavage of vABNs (Supplementary Fig. 2b). With the inhibitor, ~75% breath signal in infected mice was attenuated thereby demonstrating the role of NE activity in driving reporter release from vABNs in vivo (Supplementary Fig. 10). Using immunohistological staining and confocal imaging, we confirmed that vABNs co-localize with extracellular NE (Fig. 3k). Though cellular uptake of vABNs was also observed in lung tissue sections, we expect reporter release to be predominantly caused by extracellular NE activity due to the rapid breath signal kinetics. Furthermore, we found that peptide substrate delivery on a nanocarrier resulted in prolonged elevated breath signal in infected mice compared to delivery of unformulated, free peptide (Supplementary Fig. 11), demonstrating that the nanocarrier increases accessibility of peptides to NE. Taken together, this data demonstrates the feasibility of using vABNs to generate a rapid breath readout for protease activity in the lungs.

Breath signal can be used to monitor NE activity in respiratory diseases.

Disease progression and resolution are marked by changes in protease activity⁴⁵. Therefore, protease activity-sensing tools can be leveraged to monitor disease during treatment. Having observed dynamic changes to breath signal in response to NE inhibition, we were encouraged to employ serial vABN breath tests to monitor changes to NE activity during the progression and resolution of pneumonia. To model disease dynamics, mice were inoculated with P. aeruginosa to trigger acute inflammation and neutrophil recruitment to the lungs and then treated at 1d and 2d post-inoculation with an antibiotic, ciprofloxacin, to kill bacteria and allow neutrophil clearance (Fig. 4a). H&E-stained lung tissue sections from mouse models clearly showed increasing neutrophil infiltration at 8h and 1d, after which, we observed diminishing neutrophil presence at 4d and 7d (Fig. 4b). A corresponding rise and fall of NE protein was observed in lung homogenates from inoculated mice (Fig. 4c). Thus, both the onset and resolution of inflammation were represented at the sampled timepoints. Breath signal 10 min after vABN dosing was measured at each timepoint and exhibited dynamics similar to that of NE protein levels (elevated at 8h and 1d, reduced by 4d, and back to baseline signal by 7d) (Fig. 4d). Repeated vABN dosing did not elevate breath signal in healthy controls. Receiver operating characteristic (ROC) curve analysis indicated that inoculated mice can be identified with high sensitivity and specificity at 8h and 1d via breath analysis (AUROC = 1.0) (Fig. 4e). Furthermore, we were able to observe resolution of inflammation through convergence of diseased breath signal with healthy breath signal, as indicated by decreasing AUROC with time. Thus, we demonstrated practical application of the vABN platform to monitor neutrophilic inflammation during and after bacterial infection.

Figure 4. Breath signal after intrapulmonary delivery of vABNs can be used to monitor in vivo neutrophil elastase to assess respiratory disease progression.

(a) Experimental timeline for monitoring NE activity before and after antibiotic treatment in acute pneumonia mouse models, (b) H&E-stained lung tissue sections showing the influx and recession of neutrophils after infection and antibiotic treatment (scalebar = 100 |jm). (c) NE protein levels measured from lung homogenates (mean ± s.d., n = 3–6 mice) and (d) corresponding breath signal at each experimental timepoint before and after infection (mean ± s.d., n = 3–7 mice), (e) ROC curves showing the ability of breath signal to distinguish PA01-inoculated mice from healthy controls over the course of infection. Distinction between control and infected groups diminishes as inflammation is resolved. An AUROC of 1.0 indicates prefect distinction while the dashed red line represents an AUROC of 0.5 at which there is no distinction between the inoculated and control groups, (c, d, one-way ANOVA with Tukey’s multiple comparisons test, **p < 0.01, ****p < 0.0001.)

Since NE is active in numerous respiratory diseases, the clinical utility of NE breath tests lies in monitoring diseases in which NE is known to contribute to pathology as opposed to screening for the presence of disease. An example of such a disease is alpha-1 antitrypsin deficiency (AATD), a genetic condition caused by mutation of the SERine Proteinase Inhibitor family A member 1 (SERPINA1) gene. Alpha-1 antitrypsin (A1AT) is a protein inhibitor of NE and is expressed endogenously in the body. In the absence of its inhibitor, NE activity is dysregulated in the lungs after inflammatory insults (i.e. smoke inhalation, respiratory infections) and, therefore, leads to damage and enlargement of alveoli (i.e. emphysema). Prophylactic A1AT is administered weekly in AATD patients to prevent emphysema, and lung function and density are currently monitored using forced expiratory volumes (FEV) and CT imaging⁴⁶, respectively. Aside from these highly downstream readouts that assess NE activity-mediated lung damage, there are currently no pharmacodynamic (PD) biomarkers which can be used to optimize A1AT dosing regimen to minimize troughs in NE inhibition⁴⁷. Therefore, we sought to assess the utility of our engineered breath signal as a PD biomarker for A1AT treatment (Fig. 5). To test whether vABNs could be used for this purpose, mice were administered a 1-mg prophylactic A1AT dose, followed by a lipopolysaccharide (LPS) challenge to induce neutrophil recruitment, and a vABN breath test to assess NE inhibition. Three different A1AT dosing schedules – D-1, D-2, and D-3 – corresponding to A1AT dosing 1, 2, or 3 days before LPS challenge (Fig. 5b) were implemented to observe changes to breath signal with A1AT clearance from the lungs. On the day of the breath test (i.e. 2, 3, and 4 days after intrapulmonary A1AT delivery), approximately 11.1%, 3.2%, and 1.2% of the initial dose remained in the lungs, respectively (Fig. 5c). At these timepoints, NE protein levels were elevated due to delivery of immunostimulatory LPS into the lungs (Fig. 5d). Interestingly, we observed lower NE protein levels in mice with the greatest remaining A1AT relative to the untreated, LPS-challenged mice, which may be attributed to suppressed neutrophil recruitment, a well-documented anti-inflammatory effect of A1AT⁴⁸. As expected, untreated, LPS-challenged controls had elevated breath signal, and prophylactic A1AT dosing 1–2 days before the LPS challenge was able to significantly reduce breath signal (Fig. 5e). More importantly, breath signal was able to capture the diminishing capacity of a single prophylactic A1AT dose to inhibit NE activity over time. By day 3 after A1AT administration, the 1 mg dose no longer protected against NE activity as indicated by the absence of significant difference between breath signal of untreated controls and the D-3 cohort. This is further supported by the significant negative correlation between breath signal and A1AT in the lungs (Fig. 5f, Pearson’s r = −0.6018, p = 0.0293). Additionally, ROC curve analysis showed the ability of breath signal to robustly distinguish between untreated controls and mice treated 1 or 2 days before LPS challenge (AUROC = 0.97 and 0.91, respectively) (Fig. 5g). However, mice treated with A1AT 3 days before LPS challenge could not be identified from untreated controls (AUROC = 0.50). Thus, we have demonstrated that vABNs can be used to directly assess A1AT inhibition of NE in the lungs to characterize the duration of therapeutic efficacy, which can be used to inform dosing frequency. While these studies verified we could monitor A1AT activity in vivo, they were completed in wildtype mice with intact A1AT expression. Therefore, A1AT dosing in this context was supplementary to endogenous A1AT, and the therapeutic goal in these studies was suppression of NE activity to baseline levels observed in healthy controls rather than normal NE activity that is expected during acute inflammation. To better recapitulate the goal of A1AT therapies, vABN breath tests were completed in genetically-engineered mouse models of AATD with SERPINA1a-e gene knockout⁴⁹. In these studies, we sought to determine if vABN breath tests could be used to resolve the difference between normal NE activity during acute inflammation in wild-type (WT) C57BL/6 mice and pathologic, elevated NE activity during acute inflammation in C57BL/6 mice with SERPINA1a-e gene knockout (Fig. 6a). With this ability to resolve normal and elevated NE activity, we hypothesized that we would then be able to assess the ability of A1AT therapies to normalize NE activity in AATD. To test this hypothesis, vABN breath tests were conducted in both WT and SERPINA1a-e knockout mice before and after an LPS challenge (Fig. 6b). WT and SERPINA1a-e knockout mice had equivalent baseline breath signals before LPS challenge (Fig. 6c). After LPS administration, significant higher breath signal was observed in SERPINA1a-e knockout mice compared to WT mice (Fig. 6d), thus, demonstrating the ability of vABNs to identify pathologic NE activity (Fig. 6e, AUROC = 0.98, p = 0.003). vABN breath tests were also completed in knockout mice given 1 mg prophylactic A1AT, and breath signal in this treatment group was significantly lower than that in untreated knockout mice (p < 0.05) and indistinguishable from breath signal from WT mice with acute inflammation (Fig. 6e, AUROC = 0.60, p = 0.568). Thus, with these studies, we illustrate the capacity of our vABN platform to assess protease regulation during disease treatment.

Figure 5. vABN breath signal can be used to assess drug target engagement in the lungs.

(a) Schematic showing vABN reporter release in the presence of active NE (top) and attenuated reporter release due to NE inhibition by AAT (bottom), (b) AAT dosing schedules to assess duration of NE inhibition after a single prophylactic dose of AAT. (c) Percent of the initial dose of AAT (%ID) and (d) NE in the lungs on the day of the breath test (mean ± s.d., n = 8). (e) Breath signal 24 h after LPS is administered in the lungs of untreated mice (red) and AAT-pretreated mice (black) (mean ± s.d., n = 4–20). (f) Dot plot showing correlation between breath signal and remaining AAT in the lungs, (g) ROC curves showing the ability of breath signal to distinguish between untreated mice with LPS-triggered NE activity and mice given a 1 mg dose of AAT 1, 2, or 3 days before the LPS challenge. Distinction between untreated and treated mice diminishes with increasing time between AAT dosing and the LPS challenge. No protection is observed by day 3 after a single 1 mg AAT dose. (b, c, one-way ANOVA with Tukey’s multiple comparisons test, *p < 0.05, ****p < 0.0001.)

Figure 6. vABN breath tests can resolve normal NE activity in wild-type mice (WT) versus pathologic NE activity in SERPINA1a-e knockout (KO) mice.

(a) Schematic showing expected trends for vABN cleavage in WT C57BL/6J and C57BL/6J SERPINA1a-e KO mice, (b) Experimental timeline to determine breath signal before and after delivery of an inflammatory stimulus (LPS) to the lungs, (c) Baseline breath signal measured on Day −3 (mean ± s.d., n = 7–8). (d) Breath signal measured on Day 1 after LPS dosing ± prophylactic A1AT. Circle and square symbols indicate WT and KO mice, respectively (mean ± s.d., n = 7–8, one-way ANOVA with Tukey’s multiple comparisons test, *p < 0.05, ***p < 0.001, ****p < 0.0001). (e) The red ROC curve shows the ability of breath signal to robustly differentiate between normal NE activity in WT mice and uninhibited NE activity in KO mice during inflammation. In contrast, the blue ROC curve shows the loss of discernable difference in breath signal between WT and KO mice due to prophylactic A1AT treatment.

Discussion

In this work, we engineered a diagnostic platform for sensitive detection of aberrant protease activity in the lungs via breath analysis. This platform consists of nanoformulated peptide substrates, or vABNs, which are delivered into the lungs and leverage protease activity to release bio-orthogonal, volatile reporters for a breath readout (Fig. 1). To demonstrate the clinical utility of our platform, we designed a vABN breath test for neutrophil elastase. NE is released into the extracellular environment during the host response against bacterial pathogens and, when dysregulated, can lead to lung tissue damage. We successfully synthesized and optimized vABNs for sensitive detection of NE (Fig. 2–3) and demonstrated that, after intratracheal instillation into pneumonia mouse models, vABNs release reporters into breath at levels detectable by mass spectrometry (Fig. 3). We found that breath signal can robustly identify mice with elevated NE activity in the lungs as early as 10 min after vABN administration, and showed that vABN breath tests can be used to monitor dynamic changes in NE activity during respiratory infection (Fig. 4). Furthermore, we demonstrated that the engineered breath signal is a suitable PD biomarker for therapies such as A1AT, which directly target NE activity (Fig. 5–6). Cumulatively, our in vivo studies demonstrate that this vABN platform can be used to monitor lung protease activity despite differences in biological parameters (i.e. sex, age, mouse strain and genetic background), which are generally confounding factors for endogenous biomarkers (Supplementary Table 2).

Breath tests are practical diagnostic tools due to the ease and non-invasive nature of breath sampling. Many types of sampling devices (ReCIVA® breath sampler) and receptacles (vials, bags, sorbent tubes) are now commercially available to streamline breath collection. Moreover, major strides have been made to miniaturize mass- and ion mobility-based gas analysis tools with ppt-ppb detection limits (e.g. field asymmetric ion mobility spectrometry (FAIMS)⁵⁰) to bring breath analysis to the point-of-care. The rate-limiting step in translation of breath volatiles to clinical diagnostics is the identification of disease-specific breath biomarkers. The workflow for breath biomarker identification is time- and labor-intensive and involves breath collection from large multicenter patient cohorts, GC-MS analysis of breath samples, data analysis to identify biomarker candidates, and finally biomarker validation in independent patient cohorts. Biomarker identification is also hindered by highly variable breath VOC composition and volatiles in the diet and environment which can confound breath measurements^51,52. As a result, there are currently only four approved breath tests using endogenous VOCs, one of which is indicated for respiratory disease⁵³.

Here, we illustrate with our vABN platform that breath biomarkers can be engineered de novo by leveraging disease biology and stimulus-responsive nanomaterials. Using this bottom-up approach, we can maximize SNR by tuning key parameters through vABN composition that otherwise cannot be controlled in endogenous breath biomarkers (i.e. volatile production rate, proclivity of volatiles to partition into breath). Furthermore, by using VOC reporters that are absent in the body, diet, and environment, we can minimize noise. A final enabling feature of vABNs is their modular structure. We can build vABNs that sense other proteases simply by modifying the amino acid sequence in the peptide substrate component. Furthermore, we can explore different carriers for pulmonary delivery to create a substrate reservoir in the lungs. Here, we focused on polymeric PEG nanocarrier scaffolds, which we showed prolongs elevated breath signal for sensitive detection of neutrophil elastase activity (Supplementary Fig. 11a–b). The function of the nanocarrier is thus analogous to the test meal in the H. pylori breath test, which is used to delay gastric emptying to prolong urease substrate residence time in the stomach⁵⁴. In summary, this study provides a framework for engineering breath biomarkers for respiratory disease. We show that through simple integration of volatiles into nanomaterials, we can expand beyond the conventional optical readout for protease activity and apply this technology for non-invasive disease monitoring.

To successfully translate our platform for clinical use, next steps include working towards aerosolized delivery of vABNs into the lungs. In this work, we employed intratracheal instillation to deliver a controlled dosage of vABNs into the deep lung to establish feasibility. Moving forward, vABN formulations will be optimized for delivery via devices such as metered-dose inhalers, dry-powder inhalers or nebulizers, and breath signal will be studied after aerosolized vABN delivery. Using our PBPK model with parameter values modified for humans (i.e. scaled vABN dosing, respiratory rates and volumes, NE concentrations, and non-specific cleavage), we predict that the same order magnitude breath signal can be achieved in humans to monitor lung protease activity (Supplementary Fig 12, Supplementary text). To validate this, future work includes breath studies in large animal models. For breath analysis closer to the point-of-care, we will further explore pre-concentration of breath samples onto sorbent tubes, which entails adsorbing VOCs in large sampling volumes onto solid supports for off-line analysis⁵⁵. Pre-concentration enables sample analysis with less sensitive devices such as FAIMS⁵⁰. A caveat of the current vABNs includes the use of HFA reporters. While we did not observe adverse reactions in vivo after dosing up to 100 μM substrate, it is worth noting that these HFAs have not been previously used in humans and should be further characterized. Material safety data sheets indicate that the LC50 for the corresponding non-fluorinated n-alkylamine, n-propylamine, is >2000 ppm (4h), which is several orders of magnitude higher than our reporter concentrations. Work is currently underway to incorporate alternative classes of VOC reporters with more well-established toxicity profiles, which will also enable us to multiplex vABNs to created disease-specific volatile signatures in breath. While we have demonstrated that single protease detection is useful to monitor disease and treatment, single biomarkers suffer from poor disease specificity. Therefore, we predict that measurement of multiple proteases through multiplexed vABNs can provide the needed specificity for disease detection as we have demonstrated previously using urinary reporters^13,15,23.

Methods

vABN synthesis and characterization.

HFA-modified peptides were synthesized by CPC Scientific (>95% purity). Briefly, the peptide substrate, Ac-CKKK(Cy5)-PEG4-Nle(OBzl)-Met(O)₂-Oic-Abu-OH, was synthesized on Fmoc-Abu-CTC resin via standard Fmoc solid phase peptide synthesis. The peptide was cleaved from the resin using 30% HFIP in DCM for 30 min and subsequently coupled to the HFA reporter using DIC/HOBt coupling reagents in DCM under stirring conditions at room temperature for 2h. The finished product was deprotected in TFA for 2.5h, precipitated and washed in chilled ether 2 times, and dried under vacuum overnight. The crude peptide was then purified using RP-HPLC. For vABN synthesis, HFA-modified peptides were conjugated to 40 kDa maleimide-functionalized 8-arm PEG (Jenkem) overnight at room temperature in DI water (2:1 molar ratio of peptide to maleimide groups). Excess peptides were removed using spin filters (Millipore, 10 kDa MWCO), and completed vABNs were stored in DI water at 4°C. vABNs were visualized using TEM imaging. Samples were prepared by placing a 1 mg/mL vABN solution onto a TEM grid, wicking away the solution, and staining the sample with 2% uranyl acetate.

Real-time vapor analysis.

A triple quadrupole mass spectrometer with 1 L/min sampling rate and ppq to ppt detection limit⁵⁶ was used to quantify HFA reporter concentrations in the headspace of in vitro reactions and breath samples collected in glass vials during in vivo experiments. Multiple reaction monitoring (MRM), in which both [M+H]⁺ precursor ions and the [M-HF]⁺ product ions are monitored, was used for sensitive and selective detection of HFA reporters. Data was acquired using Analyst® software.

In vitro cleavage assays.

To investigate reporter release as a function of NE concentration, vABNs equivalent to 10 μM substrate were reacted with pM to nM neutrophil elastase in a 400-μL final reaction volume in capped 6-mL glass vials. After reacting for 10 min on a shaker, vapor analysis of the reaction headspace was completed by opening vials at the inlet of the triple quadrupole mass spectrometer. Protease specificity was subsequently assessed by reacting 10 μM vABN with 40 pM NE, PR3, GZMB, CTSB, CTSD, MMP9, or MMP13 and analyzing reporter signal in the headspace using the same method as before. For each protease, separate reaction vials were prepared for each timepoint separated by 5 min intervals to determine HFA reporter signal over time. To predict breath signal using the PBPK model, reporter release kinetics were further characterized for vABNs containing reporters HFA1, HFA3, HFA5, and HFA7. The catalytic rate constant (k_cat) and Michaelis-Menten constant (K_m), for vABNs were derived by first determining initial cleavage velocities for different micromolar substrate concentrations reacted with a fixed 20 pM neutrophil elastase concentration, plotting the values in a Lineweaver-Burke plot, and fitting a line to find the x-intercept (−1/K_m) and y-intercept (1/v_max). k_cat was calculated by dividing v_max by the NE concentration. All cleavage assays were completed in enzyme-specific buffers (NE, PR3, GZMB: 50 mM Tris, 300 mM NaCl, 0.05% (w/v) Brij-35, pH 7.5; CTSB: 25 mM MES, 2 mM DTT, 0.05% (w/v) Brij-35, pH 5.0; CTSD: 0.1M sodium acetate, 0.2M NaCl, 0.05% (w/v) Brij-35, pH 3.5; MMP9, MMP13: 50 mM Tris, 10 mM CaCl₂, 300 mM NaCl, 20 μM ZnCl₂, 1 mM MgCl₂, 0.05% (w/v) Brij-35, pH 7.5.

PBPK model.

For a complete description of the model, variables, and equations, refer to the SI Text.

Partition coefficient studies.

Partition coefficients for HFA reporters were derived empirically to be used in the PBPK model. To determine tissue:air partition coefficients (H_t:a) and blood:air partition coefficients (H_b:a) of HFA1, HFA3, and HFA5, blood in EDTA solution and lung tissue were collected from female CD-1 mice and aliquoted into 20-mL volatile organic analysis (VOA) glass vials with rubber septa caps (Thermo Scientific™). 5 μL of pure HFAs were vaporized inside separate VOA glass vials and 10 μL of vaporized HFA was transferred to empty reference vials and vials containing EDTA solution, blood, or lung tissue using gastight syringes. Vials were then allowed to equilibrate for 4h at 37°C, after which, 10 μL of headspace from each vial was sampled using a gastight syringe and introduced into the triple quadrupole mass spectrometer. Partition coefficients were calculated using equations established by Gargas et al.⁵⁷ For H_EDTA:a and H_t:a: $H_{\mathrm{i}:\mathrm{a}}=\frac{C_{ref}{V}_{vial}-C_i\left(V_{vial}-V_i\right)}{C_i{V}_i}$, where c represents HFA concentration in the headspace, v represents volume, and i represents the compartment of interest (i.e. EDTA or tissue). For the H_b:a: $H_{b:a}=\frac{C_{ref}{(V}_{vial}-V_{EDTA})-C_b\left(V_{vial}-{V_{EDTA}-V}_b\right)+(C_{ref}-C_b)H_{EDTA:a}{V}_{EDTA} }{C_b{V}_b}$.

Lung infection models.

All animal studies were approved by the Massachusetts Institute of Technology’s Committee on Animal Care and were completed in accordance with the National Institutes on Health Guide for the Care and Use of Laboratory Animals. For lung infection breath studies, 7–8 week old female CD-1 mice (Charles River) were administered an inoculum of 1.5 × 10⁶ cfu P. aeruginosa (strain PA01) in 50 μL PBS via intratracheal instillation. The inoculum was prepared by diluting an overnight culture 1:10–1:50 in LB media, allowing the secondary culture to grow to an OD~0.5, washing the secondary culture twice with PBS, and resuspending the bacteria in PBS. Total neutrophil elastase content in the lungs was quantified 4, 8, 12, 24, and 48h after inoculation by running ELISA assays on lung homogenates (R&D Systems, Mouse Neutrophil Elastase/ELA2 DuoSet ELISA). Lung homogenates were prepared by harvesting the lungs, placing lungs in gentleMACS C tubes containing 2 mL chilled cOmplete™ protease inhibitor cocktail (Roche) in PBS, and homogenizing using a gentleMACS dissociator. Bacterial burden was quantified by plating 1:10–1:10000 dilutions of lung homogenates on solid LB agar with overnight incubation and subsequent colony counting.

Breath collection and analysis.

For vABN breath tests, 10 μM vABNs in 50 uL PBS were administered into mice via intratracheal instillation. 10 min after vABN administration (unless otherwise specified), mice were placed inside a breath collection apparatus consisting of a 100cc syringe (Wilburn Medical USA) connected to a stopcock valve (Cole Parmer, UX-30600–05) with a 23G needle (BD, 305145)⁴². The syringe was sealed for 2 min to allow breath volatiles to accumulate in the syringe headspace. After 2 min, the valve was opened and 55-cc of headspace was displaced into 5 12-cc Exetainers® (Labco Limited UK) by puncturing the rubber septum with the needle and pushing the syringe plunger. To measure HFA reporter signal in breath samples, Exetainers® were uncapped at the inlet of a triple quadrupole mass spectrometer and analyzed for 0.25 min each. Signal peaks were integrated to determine total counts and converted to ppb units.

Infection-monitoring studies.

To model dynamic changes in NE activity during the onset and resolution of infection, CD-1 mice were treated with an antibiotic (10 mg/kg ciprofloxacin via IP injection) 24 h and 48 h after inoculation with PA01. vABN breath tests were completed at baseline before inoculation and 8h, 1 d, 2 d, 4 d, and 7 d after inoculation. In a separate cohort, mice that were infected at the same time as those receiving the vABN breath tests were euthanized at the experimental timepoints for histology and quantification of NE in lung homogenates. Lung tissues for histology were paraffin-embedded, and 5 μm-thick tissue sections were H&E stained and imaged using a slide scanner.

A1AT augmentation studies.

A 1-mg prophylactic dose of human A1AT (Sigma Aldrich, A6150) was administered via intratracheal instillation into 8–10-week old female C57BL/6N mice (Charles River). 1, 2, or 3 days following A1AT dosing, 50 μg pure E. coli LPS (Invivogen, tlrl-3pelps) in 50 μL PBS was administered via intratracheal instillation to induce acute lung inflammation. 24 h after the LPS challenge, vABN breath tests were completed. NE and human A1AT in lung homogenates was quantified using an ELISA (Bethyl Laboratories, Inc., E88–122).

Breath studies in genetically-engineered mouse models of AATD.

WT C57BL/6 mice and Serpina1a-e knockout mice of C57BL/6 background⁴⁹ were a kind gift from the laboratory of Christian Mueller (U Mass). Experiments were conducted on male mice at 27–30 weeks old. A single vABN breath test in this study consists of administering 20 μM vABNs via intratracheal instillation and collecting a 2-min breath sample 10 min after vABN administration. To measure reporter signal in breath samples, 15cc of breath (3cc drawn from each of the 5 Exetainers per mouse) were injected into the port of a PTR-MS (Ionicon PTR-TOF 1000Ultra). WT and Serpina1a-e knockout mice were divided into the following experimental groups: LPS- (healthy controls), LPS+, and A1AT+LPS+ (n = 7–8). At Day −3, vABN breath tests were completed on experimental groups at baseline (without any inflammatory stimuli in the lungs). On Day −1, A1AT+LPS+ mice were administered a 1-mg prophylactic dose of human A1AT in 50 μL PBS via intratracheal instillation. On Day 0, mice were given 40 μg S. enterica LPS (Sigma Aldrich L2137) in 40 μL PBS via intratracheal instillation. 24h after LPS-dosing, vABN breath tests were completed.

Statistical analysis.

ANOVA and ROC curve analyses were completed in GraphPad Prism 8.