Ethics statement

Ethical approval for the EPIC-Norfolk study was obtained from the Norfolk Research Ethics Committee³¹. The S3 study was approved by the local Medical Ethical Review Board and accepted by the competent authority, the Central Committee on Research on Human Subjects (NL73478.029.20). For both studies, written informed consent was obtained from all participants. For the MARS cohort, inclusion of patients was done using an opt-out method approved by the institutional review boards of the Academic Medical Center Amsterdam and the University Medical Center Utrecht (IRB No. 10-056 C). The Medical Ethics Committee of the Academic Medical Center, Amsterdam (the Netherlands), approved the CAST study (reference: NL52003.018.15) and all patients gave written informed consent. For all studies, this included permission for the analyses performed and consent to publish our findings. All research was conducted in accordance with the Declaration of Helsinki. Animal experiments were conducted in accordance with the Dutch Experiment on Animals Act and European Directives, and approved by the Central Authority for Animal Experiments and the Animal Welfare Body of the Amsterdam UMC.

EPIC-NORFOLK cohort

To assess the relationship between plasma tryptophan metabolites and incident severe (i.e. either hospitalization or mortality) lower respiratory tract infections in the general population, we used data from the population-based European Prospective Investigation into Cancer (EPIC)-Norfolk cohort. Details on the study design, in- and exclusion criteria, measurement of metabolites, and usage of national registries for clinical outcomes have been previously described^31,62. In brief, the EPIC-Norfolk cohort included middle-aged participants from the general population of Norfolk (United Kingdom), of whom 13,464 were selected for metabolomic profiling. Non-fasted plasma samples were collected at baseline in 1993−1997, and stored in liquid nitrogen. The Discovery HD4 platform (Metabolon) was used for untargeted metabolomic measurements, details of which have been previously described³¹. Metabolite levels were normalized and rescaled to a mean of zero and standard deviation of one³¹. For the purposes of this study, fifteen tryptophan metabolites were investigated; other metabolites were not analyzed. Our primary outcome was hospital admission or mortality due to a lower respiratory tract infection during follow-up after collection of a plasma sample at study inclusion. We opted for this outcome to identify the more severe cases of pneumonia, rather than those that could be treated on an outpatient basis. Participants were linked to national hospitalization and mortality registries as previously described³¹. Patients were identified as having reached the primary outcome if one of the following ICD-9 or ICD-10 codes was registered as reason for hospitalization or cause of death: 466, 480-487, 513, J10-J22, and J85. Mortality from other causes was treated as competing risk.

Severe CAP patient cohort

Patients with severe community-acquired pneumonia (CAP) were included as part of the Molecular Diagnosis and Risk Stratification of Sepsis (MARS) study (ClinicalTrials.gov identifier NCT01905033). Details on the study design, inclusion criteria and procedures have been previously described^32,33,34. In brief, MARS was a prospective observational in the ICUs of the Academic Medical Center (Amsterdam) and the University Medical Center Utrecht, both in the Netherlands, conducted between January 2011 and January 2014. Based on prospectively collected data during admission, the likelihood of CAP (and other infections) at ICU admission was post hoc labeled as ‘none’, ‘possible’, ‘probable’, or ‘definite’ by an adjudication committee, as previously detailed⁶³. In addition, patients were post hoc labeled as having sepsis (based on fulfilling the Sepsis-3 criteria⁶⁴). For the current study, patients admitted to the ICU with ‘probable’ or ‘definite’ CAP and sepsis were included. Patients transferred from other ICUs, with other concurrent infections, or no sepsis within 24 h of ICU admission were excluded. For patients with multiple admissions for sepsis, only the first was included. To enable the measurement of tryptophan metabolites and assess correlations with host immune response biomarkers, we included those patients of whom citrate plasma was collected within 24 h of ICU admission, and host response biomarkers were measured (i.e. enrolled within the first 2.5 years of the MARS project). Sex was determined based on self-report or as captured in the electronic health records. The causative pathogen of included patients was determined based on all available microbiology results. Our primary clinical outcome for severe CAP patients was mortality at 90 days after ICU admission. Sensitivity analyses included outcome assessment at an earlier (30 days) and later timepoint (1 year). To assess the relationship between IAA and respiratory failure, we used the highest respiratory component of the SOFA score, and calculated the number of ventilator-free days (i.e., alive and not mechanically ventilated), both adjudicated 90 days after ICU admission. Among severe CAP patients with a bacterial causative pathogen, bacteremia was considered present when the pathogen causing the pneumonia was (also) detected in a blood culture.

Healthcare workers employed at the Amsterdam University Medical Center and included in the prospective observational S3 study served as healthy controls (Netherlands Trial Register NL8645). Details have been previously described^35,36. Data on baseline BMI and comorbidities were collected after the initial study enrollment and thus not available for all participants. Citrate plasma, obtained prior to SARS-CoV-2 vaccination, of healthcare workers of similar age and sex (self-reported) as the patient cohort, was used for quantification of tryptophan metabolites.

Quantification of tryptophan metabolites

Blood citrate plasma, BALF and murine lung homogenate samples were shipped to the University of Tours (UMR 1253, iBrain, University of Tours, Inserm, France) to measure tryptophan metabolites using a previously described and validated method^18,37. In brief, liquid chromatography coupled with high resolution mass spectrometry was used to quantify absolute amounts of tryptophan and sixteen tryptophan metabolites. For each compound, a calibration curve was created by calculating the intensity ratio obtained between the metabolite and its internal standard. The total amount of indoles was calculated by summing the concentrations of IAA, indole-3-aldehyde, indole-3-lactic acid, and indole-3-sulfate. Similarly, the sums of metabolites from the serotonin and kynurenine pathway were calculated. For both severe CAP patients and healthy controls, the quantification and analysis of tryptophan metabolites were not part of a preregistered analysis plan.

Whole blood transcriptomics

Details on whole blood transcriptomics have been previously described^32,33,34,65. In brief, in a subset of severe CAP patients (n = 66; those enrolled during the first 1.5 years of the MARS study), whole blood was collected in PAXgene tubes (Becton-Dickinson, Breda, the Netherlands) within 24 h of ICU admission, and stored at −80 °C. The PAXgene blood mRNA kit (Qiagen, Venlo, the Netherlands) was used to isolate total RNA. RNA (with an integrity number ≥6) was processed and hybridized to the Affymetrix Human Genome U219 96-array, and scanned using the GeneTitan instruments at the Cologne Center for Genomics (Cologne, Germany).

Host immune response biomarkers measurements

We employed a Luminex multiplex assay (R&D Systems Inc, Minneapolis, MN) on a BioPlex 200 (BioRad, Hercules, CA) to measure 20 host immune response biomarkers reflective of inflammatory cytokines, inflammation and organ damage, and the endothelial and coagulation response in severe CAP patients, as earlier detailed^34,65,66,67. All host immune response biomarkers were measured in EDTA anti-coagulated plasma obtained within 16 h after ICU admission.

Human intervention trial

This proof-of-concept, single center, randomized, open-label, controlled intervention trial was part of the CAST (C1-inhibitor in allergic ASThma patients) study, in which the effect of C1-inhibitor or microbiota depletion on lung inflammation in asthma patients with allergy was investigated. The study protocol is registered at ClinicalTrials.gov (identifier: NCT03051698) on 14 February 2017. Details on the study design, screening procedures, in- and exclusion criteria and participant characteristics have been previously published^42,68. In short, patients (aged 18−45 years) with intermittent-to-mild asthma according to criteria of the Global Initiative for Asthma, sensitization to house dust mite and no clinically significant abnormalities during physical examination, hematological and biochemical screening were included. Following inclusion, subjects were randomly assigned to either broad-spectrum antibiotics or no antibiotic control. All participants collected a fecal sample at baseline, which was stored at −20 °C at home and transported to the study center for storage at −80 °C within 24 h. Participants allocated to the antibiotics group received 7 days of oral broad-spectrum antibiotics (ciprofloxacin, 500 mg every 12 h; vancomycin, 500 mg every 8 h; and metronidazole, 500 mg every 8 h). This antibiotic regimen is similar to regimens previously used by our group^69,70, and chosen based on the assumption that extensive gut microbiota alterations would make the effects of gut microbiota most evident. After a 36 h washout period (without antibiotic administration—to avoid direct interference of the antibiotics with any of the subsequent measurements), blood was drawn from all participants, and participants from the antibiotics group collected a second fecal sample (storage was performed as described above). In addition, as detailed earlier^42,68, a bronchoscopy was performed in all participants to instill one lung segment with house dust mite and lipopolysaccharide, and saline in the contralateral segment. During a second bronchoscopy 7 h later, both instilled lung segments were lavaged with 8 successive 20 mL aliquots of saline. Bronchoalveolar lavage fluid (BALF) from both instilled lung segments was centrifuged (400 g, 10 min, 4 °C) and supernatant was stored at −80 °C until further analysis. We quantified tryptophan metabolites in BALF from the saline-challenged lung segment. Although the modulation and analysis of gut microbiota were pre-specified in the study protocol, quantification of tryptophan levels in plasma and BALF were not part of the pre-specified analyses. All outcomes of pre-specified analyses have been previously published^42,68.

Gut microbiota analysis

Bacterial microbiota were analyzed as previously described by our group^10,42. In summary, DNA was extracted by a repeated bead beating protocol. Next, DNA was purified with the Maxwell RSC Blood DNA Kit (Promega, Madison, WI) and eluted in 50 μL DNAse free water. Twenty nanograms of DNA were used for the amplification of the 16S rRNA gene with the V3-V4 341 F forward primer (5’-CCTACGGGNGGCWGCAG-3’) and the 805 R reverse primer (5’-GACTACHVGGGTATCTAATCC-3’) for 25 cycles. Amplified product was purified using AMPure XP beads (Beckman Coulter, Indianapolis, IN; according to manufacturer’s guidelines). Purified products were equimolar mixed and 250 bp paired-end sequenced with 2 × 251 cycles on an Illumina MiSeq platform (GATCBiotech, Constance, Germany) using V3 chemistry, according to manufacturer’s instructions. Sequence reads were analyzed as follows. Read pairs with perfect matching forward and reverse barcodes were assigned to their corresponding samples. The forwards and reverse reads were length trimmed at 240 and 210, respectively, which were inferred and merged with ASVs using DADA2 (V1.5.2). The assignment of taxonomy was done using the DADA2 implementation of the RDP classifier and SILVA 16S reference database. Relative abundances were used throughout this study. Differences in overall microbiota composition (β-diversity) were assessed by permutational ANOVA using Bray-Curtis dissimilarities (adonis function, vegan package, 9999 permutations), and visualized using principal coordinates analysis. DESeq2 was used to test for differentially abundant taxa (restricted to genera present with at least 100 reads in at least 20% of the participants).

Mice experiments

Specific pathogen-free female C57BL/6 J mice were purchased from Charles River, and housed in individually ventilated cages in rooms with a controlled temperature (20–26 °C with 30–70% humidity) and 12 h light-dark cycle at the Animal Research Institute Amsterdam (ARIA) facility of the Academic Medical Center facility under standard care. Mice were acclimatized for 1 week prior to experiments. All experiments were conducted with mice between 9 and 11 weeks of age at time of infection.

IAA, AhR-inhibitor, FICZ and ILA pretreatment

Mice were pretreated by daily intraperitoneal injections with 50 mg/kg bodyweight indole-3-acetic acid (Sigma Aldrich) for 2 weeks. Control mice received daily intraperitoneal injections with saline. For experiments with AhR inhibition, CH-223191 (Sigma Aldrich) was given by daily intraperitoneal injections (10 mg/kg bodyweight) for 2 weeks, prior to IAA administration. ILA (50 mg/kg bodyweight; Sigma Aldrich) and FICZ (100 µg/kg bodyweight; Sigma Aldrich) were also administered by intraperitoneal injections for 2 weeks. Dose, duration and route of pretreatment are based on previous publications^22,48.

Infection models

A detailed protocol for our experimental K. pneumoniae-induced pneumonia model has been previously published⁴⁶. In summary, K. pneumoniae ATCC43816 (K2:O1) was grown from frozen aliquots in Tryptic Soy Broth in a shaking incubator (5% CO2, 37 °C). After overnight growth, one milliliter was transferred to fresh Tryptic Soy Broth, grown to midlogarithmic phase, washed with saline and diluted to a final concentration of 1 × 10⁴ colony-forming units (CFUs) per 50 µL. One day after completion of pretreatment, infection was induced as follows: for the induction of experimental pneumonia, inhalation anesthesia with isoflurane (2-3% in 100% oxygen) was applied and mice were subsequently intranasally inoculated with 50 µL bacterial suspension (10⁴ CFUs). At 12, 24, 36 or 42 h following infection, mice were sacrificed by intraperitoneal injection of ketamine/dexmedetomidine. Of note, for the assessment of ROS release, mice were not infected but sacrificed 24 h after completion of pretreatment.

Following infection and euthanasia, blood was collected via cardiopuncture. The left lung, spleen and liver were harvested and homogenized in four volumes of sterile PBS. Blood and organ homogenates were serially diluted in sterile PBS, plated onto blood agar plates, incubated overnight at 37 °C, and CFUs counted for determination of bacterial growth. Bronchoalveolar lavage fluid (BALF) was collected from right lung lobes using 1 mL phosphate buffered saline (PBS).

Lung pathology scores

Directly after sacrifice, a section of the left lung was fixed in 4% formalin and embedded in paraffin for routine histology. Hematoxylin and eosin-stained paraffin sections from murine lung tissue were scored as previously described^4,50. A specialized pathologist, blinded for pretreatment group, analyzed sections for bronchitis, edema, interstitial inflammation, intraalveolar inflammation, pleuritis, thrombi, and endothelialitis and graded on a scale of 0 (absent) to 3 (severe). The total lung inflammation score was calculated as the sum of the scores for each parameter, the maximum total score being 21.

Flow cytometry

Flow cytometry was performed as earlier detailed^71,72. In brief, BALF was centrifuged at 400 g for 10 min (4 °C). Supernatant was stored at −20 °C until further analysis. The cell pellet was resuspended in sterile PBS, washed and stained following manufacturer’s recommendations with eFluor 780 fixable viability dye (Thermo Fisher), PE-eFluor 610 rat anti-mouse CD45 (clone 30-F11), FITC rat anti-mouse Ly-6G (clone 1A8), and Alexa fluor 647 rat anti-mouse Siglec-F (clone E50-2440; all Biolegend). Following incubation, cells were washed and analyzed by flow cytometry (Cytoflex-S, Beckman Coulter). Precision counting beads (BD Bioscience) were added to quantify cell populations. FlowJo software (Becton Dickinson) was used to detect neutrophils based on surface markers: viable, CD45 + , Siglec F- and Ly-6G+ using FlowJo software (Becton Dickinson), as previously detailed^71,72 and shown in Supplementary Fig. 11.

Murine inflammation and organ damage markers

Inflammatory markers (TNF, IL-6, IL-10, monocyte chemoattractant protein-1) in murine BALF supernatant and plasma were quantified by cytometric bead array (mouse inflammation kit, BD Biosciences) according to the manufacturer’s instructions. IL-1β and chemokine CXC motif ligand (CXCL) 1 were measured in BALF supernatant using a commercial ELISA kit (R&D systems, Minneapolis, MN, US). To quantify oxidative stress in the lungs following pneumonia, nitrotyrosine was measured in lung homogenate by ELISA (Abcam, Cambridge, UK)²¹. The department of clinical chemistry at our hospital (Amsterdam UMC, the Netherlands) quantified organ damage markers (LDH, AST and ALT) in murine blood plasma using a c702 Roche Diagnostics machine.

ROS production assay

ROS production by neutrophils was measured as previously described^50,73. After completion of pretreatment, neutrophils were isolated from murine bone marrow using a single layer of Percoll (62.5%; Sigma). After washing with Hank’s balanced salt solution (HBSS), bone-marrow derived neutrophils were rested in HBSS^+/+ (containing 1.26 mM CaCl₂ and 0.49 mM MgCl₂) for 30 min at room temperature. Neutrophils in HBSS^+/+ were seeded in fetal calf serum precoated 96-well plates at a concentration of 1 × 10⁵ cells/well, together with isoluminol (50 µM; Sigma-Aldrich), and horse radish peroxidase (15 U/mL; Sigma-Aldrich). Neutrophils were stimulated with K. pneumoniae ATCC43816 (K2:O1) (grown as described above) at a concentration of 1 × 10⁶ CFUs/well to achieve a multiplicity of infection of 10. In parallel, phorbol myristate acetate (Sigma) was added as positive control and HBSS^+/+ as negative control. To load bacteria onto cells, plates were centrifuged at 500 × g for 3 min. Immediately after, chemiluminescence was measured every 3 min for 2 h using a Synergy HT plate reader (BioTek).

Statistical analysis

All statistical analyses were performed in R (version 4). Wilcoxon rank-sum tests were used to compare continuous variables between groups. For the EPIC-NORFOLK cohort, competing risk regression models were used to assess associations between the primary outcome (hospitalization or mortality due to lower respiratory tract infections) and tryptophan metabolites, with Benjamini-Hochberg correction for multiple comparisons. Mortality from other causes was treated as competing risk.

To calculate effect sizes for differences in tryptophan metabolites between severe CAP patients and controls, between surviving and non-surviving patients at 90 days, and between participants randomized to controls or antibiotics, we calculated Hedges’ g⁷⁴. For the differences in tryptophan metabolites between CAP patients and controls, linear regression models with log-transformed tryptophan metabolite concentrations were used to control for potential confounding by exposure to any antibiotic prior to sample collection. Since especially anti-anaerobic antibiotics have important effects on gut microbiota, we additionally controlled for exposure to anti-anaerobic antibiotics (piperacillin-tazobactam, meropenem, metronidazole, clindamycin, and amoxicillin with clavulanic acid) using separate linear regression models.

Cox proportional hazards models were used to assess the relationship between log2-transformed IAA concentrations and mortality, at 30 and 90 days, and 1 year. Time zero was defined as the day of ICU admission. Multivariable models were adjusted for age, sex, body mass index, disease severity, antibiotic exposure prior to sample collection, causative pathogen and comorbidities (diabetes, malignancy, immunocompromised state, cardiovascular, renal, and respiratory disease) at ICU admission. We quantified the severity of illness with the Sequential Organ Failure Assessment score. Pearson correlations were calculated between log2-transformed plasma IAA concentrations and the number of ventilator-free days, adjudicated 90 days after ICU admission. The association between IAA at ICU admission and the probability of bacteremia was assessed by univariable logistic regression. For the correlations between tryptophan metabolites and host immune response biomarkers (log2-transformed), Pearson correlations were calculated with Benjamini-Hochberg correction for multiple comparisons. Associations between IAA and gene expression were assessed by linear models, and Gene Set Enrichment Analysis (genes ranked by t-statistic) was applied using Reactome pathways with Benjamini-Hochberg correction^75,76. Reactome pathways related to ‘Immune System’ (R-HSA-168256) were assessed. Linear mixed models were used for dynamics of tryptophan metabolites during experimental pneumonia in mice. Two-tailed level of significance was set at (adjusted) p 

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.