Amplicon-based sequencing as a diagnostic tool for severe pneumonia in the ICU

Abstract

The microbiological diagnosis of severe pneumonia in intensive care unit (ICU) patients remains challenging. We investigated whether amplicon-based sequencing (ABS) could provide additional value in this setting. Bacterial 16S rRNA coupled with fungal internal transcribed spacer (ITS) ABS was compared with standard of care (SOC) comprising cultures, PCR and galactomannan on bronchoalveolar lavages of immunosuppressed and immunocompetent ICU patients presenting severe pneumonia (n = 24) and noninfected lung-transplanted patients (n = 12). The ABS-matched SOC results were complete for 20/36 (55.5%) samples. Regarding 16S rRNA, ten samples showed partial agreement, whereas five others were fully discordant. Discordances were observed mainly for gram-negative organisms, either for quantification (Enterobacteria) or identification (Enterobacter sp., Haemophilus sp.). Only 1 of the 36 ITS results were discordant with the SOC results. The ABS has proven to be more valuable in immunosuppressed patients. Despite limitations in bacterial identification, pathogens not reported by SOC were detected in 8 samples (coinfections, bacterial superinfections from viral pneumonia). ITS identification was reliable, and semiquantification helped distinguish P. jirovecii infection and colonization. The correlation between galactomannan and ITS was good. While 16S rRNA sequencing is interesting for severe pneumonia of undefined aetiology in IS patients, ITS could provide additional diagnostic value for severe fungal pneumonia.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-25323-7.

Introduction

Pneumonia is a common infection caused by a wide variety of pathogens. It can progress to acute respiratory distress syndrome (ARDS), requiring admission to an intensive care unit (ICU) and intubation. This severe progression can occur in both community-acquired pneumonia (CAP) and hospital-acquired pneumonia (HAP). Additionally, 10 to 25% of patients receiving mechanical ventilation (MV) develop ventilator-associated pneumonia (VAP)¹. VAP has a mortality rate above 30% and is responsible for the majority of antibiotic prescriptions in the ICU². CAP and HAP are caused by a wide range of bacteria, fungi and viruses that can be exogenous (virus and atypical bacteria) or endogenous and can be translocated from the oro-pharyngal flora¹. In contrast, VAP is an endogenous process that is caused primarily by bacterial agents that emerge from dysbiosis of the patients’ lung flora. This flora coevolves with the gut microbiota through the gut‒lung axis^3,4 and is influenced by various risk factors, including immunosuppression, neutropenia, increased age and thoracoabdominal surgeries¹. The microbial agents implicated vary depending on timing: early VAP (within 4 days after admission) is typically monobacterial and caused by community-acquired, antibiotic-susceptible agents. Late VAP, however, is usually caused by multiple drug-resistant (MDR) agents and/or is polymicrobial, involving endogenous bacteria such as gram-negative bacilli and S. aureus that can translocate through the gut‒lung axis⁵, as described above.

With continuous advances in surgical and medical care, the number of immunosuppressed (IS) patients is increasing. Both the duration and severity of immune suppression are also increasing⁶. In these patients, both CAP and HAP can be caused by common, well-described pathogens or rare, opportunistic fungi, environmental bacteria or even parasites^6–8. The laboratory diagnosis of these infections remains challenging. This often leads to delayed optimal treatment, resulting in increased morbidity and mortality rates^7–9.

Studies of the human microbiota have revealed how disruptions in microbial community balance play a key role in infections and immunological disorders. Contrary to previous beliefs, lungs are not sterile; they possess a rich microbiota, albeit with low biomass, that is directly influenced by the oral and gut microbiota^5,10–12. Over the last decade, many studies have investigated the diversity and modification of the lung microbiota in severe diseases^3,7,11,13,14.

The metagenomic technology used for microbiota description can theoretically identify and quantify an entire sample’s microbiological population in a single culture-independent run. Given the current challenges in accurately diagnosing pneumonia-causing agents in ICU patients¹⁵, we investigated the potential value of metagenomics in severe pneumonia cases. A previous study conducted at our institution reported that respiratory infections, mostly in IS patients, would be a perfect target for potential improvement of our microbiological diagnostic arsenal¹⁶. Compared with standard-of-care (SOC) methods, this approach enables faster documentation of both fastidious and nonculturable agents, enabling prompt treatment adjustments⁹. Additionally, metagenomic semiquantification of different microbiota components helps distinguish between colonization and infection, which can be difficult when conventional methods are used.

Metagenomics can be performed via two different methods: shotgun or amplicon-based sequencing (ABS). Shotgun sequencing can be used to identify any DNA pathogen, whereas ABS can be used to identify only bacteria and fungi through 16S rRNA and ITS genes. These sequences are mapped to 16S rRNA and ITS gene reference sets ^, and amplicon sequence variants (ASVs) are subsequently identified for taxonomic classification at the order, family, genus, or species level^3,14.

We designed this study to test ABS in three groups of patients based on their immune status and the presence or absence of severe pneumonia. Our primary objective was to evaluate the benefits of targeted ABS for both IS and IC ICU patients with severe pneumonia. We also sought to determine whether this technique could distinguish between colonization and invasive infections by opportunistic pathogens such as Pneumocystis jirovecii or Aspergillus fumigatus. Additionally, we studied these patients’ overall lung microbiota.

Methods

Study design

The study was prospectively carried out between July 2022 and April 2024 at the university hospital Hôpital Universitaire de Bruxelles (HUB)-Erasme, Brussels, Belgium, and included three groups of 12 patients: IS patients with a first episode of pneumonia lying in the ICU (group IIS), IC patients with a first episode of pneumonia lying in the ICU (group IIC) and noninfected IS patients who underwent lung transplantation (group NIIS) (the inclusion criteria are reported in Table 1; Methodology for inclusion in the Supplementary Material and Methods). Patients from each group were numbered as follows: group IIS, IIS.1 to IIS.12; group IIC, IIC.1 to IIC.12; and group NIIS, NIIS.1 to NIIS.12. Bronchoalveolar lavages (BAL) were sampled in the frame of the SOC procedure for groups IIS and IIC, as was the systematic BAL performed at month 6 after lung transplant for group NIIS. The sampling was performed according to the local standardized protocol fitting French guidelines¹⁹. The data collected included age, sex, clinical antecedents, ongoing antimicrobial treatment, clinical pulmonary infection score (CPIS), temperature, X-ray results, CRP, white blood cell (WBC) count, and PaO₂/FiO₂. Patient evolution on day 28 post sampling was ranked as follows: cured (C) according to the normalization of pulmonary, radiological and biological symptoms and the cessation of antibiotic therapy; still infected (I) according to the absence of normalization of pulmonary, radiological and/or biological symptoms despite the administration of antimicrobial therapy; or deceased (D).

Table 1.

Inclusion criteria.

Groups	Patient criteria	Clinical criteria	Biological and Radiological criteria (minimum 2 at the time of BALsampling)
IIS	·Minimum 18-year-old ·Secondary immunosuppression*	Signs of pneumonia leading to intubation  or Suspected first episode of VAP*	·CRP > 30 mg/L ·WBC counts < 4 or ≥ 12.103/mm3 ·PaO2/FiO2< 200 mmHg ·New or progressive radiographic infiltrate, consolidation, cavitation, or pleural effusion
IIC	·Minimum 18-year-old ·No immunosuppression	Signs of pneumonia leading to intubation or Suspected first episode of VAP*	·CRP > 30 30 mg/L ·WBC counts < 4 or ≥ 12.103/mm3 ·PaO2/FiO2< 200 mmHg ·New or progressive radiographic infiltrate, consolidation, cavitation, or pleural effusion
NIIS	·Minimum 18-year-old ·Lung transplantation < 1 year and > 5 month ·Excluding cystic fibrosis ·Secondary immunosuppression* ·Not undergoing a therapeutic antibiotic treatment	Absence of both signs of pneumonia and general signs of infection	·CRP < 30 mg/L ·WBC counts between 4 and 12.103/mm3 ·Negative pulmonary radiography

* See supplementary material and methods.

Standard-of-care Microbiological diagnosis

In the context of the SOC diagnostic procedure, all BAL fluids were cultured for bacterial and fungal documentation via the following methods: both 5% sheep blood agar Columbia (Benton Dickinson^®) and HAEM-2 polyvitex-enriched blood agar (Biomérieux^®) were inoculated with 10 µL of BAL fluid and incubated for 48 h at 35 °C -/+ 2 °C in a CO₂ atmosphere. Sabouraud agar plates (Biomérieux^®) were incubated at 35 °C -/+ 2 °C in a normal atmosphere for ten days. Galactomannan (GM) assay was performed with the Platelia kit (Bio-Rad^®) according to the manufacturer’s instructions. Additionally, an in-house multiplex PCR panel designed on a Taq Man Array card for the detection of 35 pathogens (8 bacteria, 4 fungi, and 23 viruses; see Supplementary Material and Methods) was also performed on 11/12 group IIS BAL²⁰. Leftover volumes of BAL fluid were stored at −80 °C before performing ABS without prior centrifugation.

The significance threshold for bacterial culture was set at 10³ colony-forming units (CFU)/mL of pure culture or 10⁴ CFU/mL of polymicrobial culture. Both fungal culture and PCR are reported to be positive without a quantification threshold. A GM index ≥ 1 is considered indicative of active pulmonary aspergillosis.

Microbiota sequencing

Total DNA was extracted from 300 µl of BAL fluid via the Agowa Mag DNA Extraction Kit (LGC Genomics^®, Berlin, Germany)^13,21. After DNA extraction, 16S qPCR was performed to assess the DNA concentration of each sample (Supplementary Data-1)²⁰.

Preamplification of the V4 hypervariable region of the 16S rRNA coding gene was performed by using the primers 515 F and 806R, which target the 16S coding region, following the protocol of Odendaal et al.. for low-biomass samples described in detail in its published version²⁰. The ITS1 region was preamplified with the universal primers ITS1F/ITS2¹⁵ via the Illumina protocol “Fungal sequencing and classification with the ITS Metagenomics”²² modified as described by Ruiz-Rodriguez et al.¹⁴ and barcoded with the same barcodes as those used for the 16S amplification product¹⁴.

Post-PCR quantification of DNA was performed for both 16S and ITS amplicons with the Quant-iT™ PicoGreen^® dsDNA Assay Kit (Thermo Fisher Scientific Inc.™, MA, USA)^15,16,22.

The products were gathered in a pooled library of 150 nM per sample, of which 10 nM was pipetted into the MiSeq cassette to carry out pair-ended sequencing via the MiSeq reagent V2 kit (2 × 250 bp) (Illumina^®, San Diego, CA, USA) on an Illumina MiSeq instrument (Illumina^®, San Diego, CA, USA).

Quality controls were used for each extraction and metagenomic amplification run. An additional negative control (blank) was added at the PCR step. The positive control was the commercialized Microbial Community Standard ZymoBIOMICS (Zymo Research^®, CA, USA) containing a panel of known diverse bacterial and fungal species, allowing the control of extraction, species level identification and semiquantification for both the 16S and ITS pipelines (Listeria monocytogenes, Pseudomonas aeruginosa, Bacillus subtilis, Escherichia coli, Salmonella enterica, Lactobacillus fermentum, Enterococcus faecalis, Staphylococcus aureus, Saccharomyces cerevisiae, Cryptococcus neoformans).

Bioinformatic analysis

The multiplexed raw data were processed through a bioinformatical pipeline for 16S and ITS analysis, and a minimum merge length of 150 bp was applied to both datasets to consider the varying ITS1 sequence lengths¹⁵ and enhance the quality of the taxonomic assignment. The assignment of ASVs was performed via the QIIME2 database (version QIIME 2 2024.5)²². In the absence of identification to the species level by QIIME2, the 16S rRNA or ITS consensus sequence built during the analysis by DADA2 (version 1.30.0) (see Supplementary Material and Methods) was manually tested on the Basic Local Alignment Search Tool (BLAST) NCBI core nucleotide database (core_nt) to test for possible further identification. These BLAST results were accepted if 100% identification similarity was obtained and for one species only (reported between brackets in Table 3). Quality metrics, alpha diversity parameters, statistical significance, and relative and absolute abundances were extracted from QIIME2-provided data.

Table 3.

Summary of standard of care (SOC) results and their ABS concordances for all patients with additional data on their antimicrobial treatments and evolution at D-28. Results of bacterial, fungal and viral aetiologies reported here are based on SOC (BCULT= bacterial culture, FCULT = fungal culture). In bold are reported the potential pathogenic species found by ABS with a relative abundance > 10% (excluding anaerobic and banal flora). PCR results come from the in-house respiratory multiplex panel. * Unsignificant level of growth for bacteria is < 1.000 CFU/ml of BAL on bacterial culture or presence of only banal flora from the upper respiratory tract; Presence of Candida species on fungal culture; PCR Pneumocystis at a Ct > 32 and absence of clinical and radiological signs specific for an infection by Pneumocystis jirovecii. ** Results of relative abundance given on Qiime2 online plugin (no decontamination, nor normalization). ° when needed, for the reporting of the ITS method, the relative abundance was first reported by genera and, when confronted to the BLAST database, by species (into brackets). # Lung transplant group received long term antibiotic prophylaxis. GM = galactomannan. Interpretation: T: total; P: partial; D: Discordant. Evolution: C = cured; I = still infected; D = deceased. CMV = Cytomegalovirus; HSV-1 = Herpes simplex virus 1; HHV6 = Human Herpes virus 6; SARS-Cov2 = SARS Coronavirus 2. AK = Amikacin; AMC = Amoxicillin-clavulanate; Anidula = Anidulafungin; AZT = azithromycin; Ceftri = Ceftriaxone; Fep = Cefepime; Isavuco = Isavuconazole; Line = linezolid; Mero = Meropenem; PIP-TZ = piperacillin/tazobactam; SXT = trimethoprim-sulfamethoxazole; Vanco = Vancomycin.

We first performed a decontamination of the raw results of the absolute abundance of each ASV provided by QIIME2 by removing the average number of reads of species amplified in the negative controls (Supplementary Data-2). The number of decontaminated reads was then normalized to the bacterial or fungal biomass of the sample for 16S and ITS, respectively¹⁵. The Shannon index values were subsequently calculated based on these modified values. As the results revealed no difference in the interpretation of relative abundance compared with the raw data, we decided to use the raw results from QIIME2 for clinical interpretation without passing through this decontamination normalization pipeline. The rationale behind this decision was to avoid an unnecessary time-consuming step.

Statistical analysis

Excel-Analyse-it software (Microsoft^®; version 2503) and the bioRender Graph online application (Biorender^®, Toronto, Ontario, Canada) (app.biorender.com) were used for other statistical analyses and visualization of the graphs. Means and standard deviations were used for normally distributed data, and medians with variance were used for nonnormally distributed data. The normality of the distribution of the parameters was tested with the Shapiro‒Wilk test (10% significance level). A confident statistical significance was set at a confidence interval of 95% (p < 0.05).

Comparisons between the SOC and ABS results

For IIS and IIC patients, pneumonia episodes were classified as bacterial, fungal, viral or undefined in aetiology based on the SOC results (Tables 1 and 3 & Supplementary Material and Methods).

To assess the total concordance between the SOC and ABS results, we decided to set a threshold of positivity for the ABS results. On the basis of our observations, we set it at 10% relative abundance of the total reads, as it allowed for total concordance with the SOC for a majority of the samples. Pathogens also had to be identified to the species level to be completely concordant. Therefore, concordance between ABS and SOC results was considered ‘complete’ if all the bacterial and fungal pathogen(s) reported by SOC were also identified by ABS with a relative abundance greater than or equal to 10% of the total reads. Candida species were not considered pathogenic. Partial concordance was reported when only some of the species identified as responsible for bacterial or fungal pneumonia were concordant, the others were either misidentified (discordant bacterial or fungal species) or did not pass the relative abundance threshold (< 10%) by ABS , or when ABS identified pathogens (≥ 10%) that were not reported by the SOC. Concordance was only assessed for pathogens that were detectable by both methods.

Results

Population description

Detailed demographic and clinical information on the 36 patients included is presented in Table 2. Seven IIS patients suffered from severe CAP leading to ICU admission, one developed severe HAP leading to ICU admission, and the remaining 4 developed VAP after ICU admission. Among IIC patients, 11 had VAP, and 1 had severe HAP. Antibiotic therapy was initiated just after BAL sampling for 8 IIS patients, while 4 patients received antibiotic therapy within 14 days before sampling. Ten IIC patients received antibiotic therapy after BAL sampling, and 5 received antibiotics during the 14 days before sampling. All NIIS patients were under a prophylactic regimen of azithromycin and trimethoprim-sulfamethoxazole at the time of sampling (Table 2). Evolution on day 28 revealed that 9 IIS patients died and 3 were cured of infection. No IIC patients died, 3 were still infected, and 9 were cured of infection (Tables 2 and 3).

Table 2.

Demographics and clinical characteristics of the included patients.

	Immunocompromised infected (IIS)	Immunocompetent infected (IIC)	Immunocompromised non-infected (NIIS)	P value
	n = 12	n = 12	n = 12
Age (years) median (SD)	60.5 (17)	40 (15.2)	57.5 (6.8)	0.0193
Gender ratio (F/M)	0.33	0.17	0.33	0.574
Chronic lung disease*	4	1	12	-
• Emphysema/COPD	2	1	7	-
• Pulmonary fibrosis	1	-	3	-
• Genetic disease and other	1	-	2	-
Short Corticosteroids treatment	1	-	-	-
Chronic immunosuppression	11	-	12	-
• Immunosuppressive regimen	5	-	12	-
• AIDS	2	-	-	-
• Hematological malignancy	2	-	-	-
• Solid tumor under chemotherapy	1	-	-	-
• Lung transplantation	3	-	12	-
• Other solid organ transplantation	1	-	-	-
CPIS in favour of VAP (7–12)	7	11	-	0.059
Pneumonia			-
• CAP	7	-	-
• HAP	1	1	-
• VAP	4	11	-	0.003
CRP median (mg/L), range; ND	85 [19–220]; 2	90 [35–330]; 2	1.45 [0.5–26]; 0	0.003
WBC count (10 3 /mm 3 )
• < 4, range	4	1	0
• ≥ 12, range	5	6	0
PaO2/FiO2 < 200 mmHg	10	6	-
Radiological signs: infiltrate, consolidation, cavitation, or pleural effusion	11	11	0
Antibiotic therapy before BAL (< 15 days)	4	5	0
Antibiotic prophylaxis at the time of BAL	-	-	12
Evolution D-28
• Cured from infection	3	9	-
• Still infected	0	3
• Deceased	9	0	-

* Before lung transplantation for the NIIS group.

ND: not done.

CRP: C-reactive protein.

CPIS: Clinical Pulmonary Infection Scale.

BAL: Bronchoalveolar lavage fluid.

CAP: community-acquired pneumonia.

HAP: hospital-acquired pneumonia.

VAP: ventilator-associated pneumonia.

Quality metrics of ABS

The bacterial and fungal biomasses of the negative controls (blanks) were significantly different from those of the patient samples (p = 0.03 and = 0.006, respectively) (Supplementary Data-3). All bacterial species of the mock community were amplified via 16S analysis, which validated the expected identification at the species level and semiquantification of the method^23,24. For ITS analysis, Cryptococcus neoformans was well identified, but Saccharomyces cerevisiae failed to appear in the mock community read profiles. However, the sample results revealed the presence of other diverse Saccharomycetales, such as Candida species (Table 3). Moreover, these Candida species were identified as diverse species and matched the SOC results. As S. cerevisiae is neither a commensal nor a pathogenic species of the lung, the ITS results were considered acceptable for further interpretation.

Lung microbiota quantification

The bacterial biomass was significantly greater in patients with bacterial infection (p < 0.001) than in those with NIIS, which was not the case for fungal biomass in fungal infections, although the p value was close to the significance threshold (p = 0.054) (Fig. 1). The bacterial biomass was significantly greater in the VAP samples than in the non-VAP samples (HAPs and CAPs) (p = 0.018), whereas the fungal biomass was significantly lower (p = 0.04) (Fig. 1). The GM index in BAL was positively correlated with the number of reads assigned to Aspergillus species by ITS (IS patients only), with an R² parameter = 0.607 and a p value < 0.0001 (Fig. 2).

Fig. 1.

Biomasses A. Bacterial biomass was significantly greater in patients with bacterial pneumonia (blue) than in non-pneumonic patients (green). Patients with undefined infections or fungal infections were excluded. B. Fungal biomass was greater in patients with fungal pneumonia (blue) than in non-pneumonic patients (green), but the difference was not statistically significant. Patients with undefined, viral or bacterial infections were excluded. C. Bacterial and fungal biomasses are significantly greater and lower, respectively, in VAP samples (red) (n = 15). The comparisons were performed with a Mann‒Whitney U test (p = 0.018 and 0.04, respectively).

Fig. 2.

Absolute number of Aspergillus reads compared with the GM index. A: Aspergillus absolute number of reads of patients reported with a positive (≥ 1) (green) or negative (< 1) (blue) GM compared by Mann-Whitney U test (p = 0.003). B: Linear regression between the GM index and the number of reads assigned to Aspergillus species (after decontamination and normalization to fungal biomass) showed a p value = < 0.0001 and an R² = 0.607.

Lung microbiota diversity and composition

The median bacterial raw reads of the clinical samples were 112,891 [37,811–156,343], and the median fungal reads were 31,098 [309–59,158]. A total of 135 bacterial genera were identified across the samples, and 90 were identified via ITS. At the species level, 382 different bacterial ASVs and 322 fungal ASVs were identified.

The bacterial community diversity (Shannon index - alpha diversity) was significantly different between the infected groups (IIS-IIC) and NIIS but not between IIS and IIC (p = 0.004) (Fig. 3). This index was not significantly different among the 3 groups for ITS (p = 0.02). Additionally, the bacterial composition differed between the groups. Samples from patients with bacterial pneumonia mainly contained the pathogens responsible for the infection, with or without commensals such as anaerobes. In contrast, NIIS group patients either presented with normal commensals, including anaerobes, or had negative results (Table 3).

Fig. 3.

Shannon diversity index for 16S and ITS analysis. Alpha diversity was represented by the Shannon index for samples grouped by the inclusion criteria. 16S indices are represented in blue, and ITSs are represented in green. Kruskal‒Wallis test p values are reported on the graph.

The absolute abundance of Candida species was significantly greater in infected patients than in noninfected patients (p = 0.01), whereas the absolute abundances of Penicillium, Malassezia and Teratosphaeriaceae species were significantly lower in VAP patients than in CAP, HAP and noninfected patients (p = 0.008, 0.005 and 0.02, respectively). No difference in bacterial or fungal species abundance was associated with immune status, previous antibiotic therapy ( within 14 days before sampling) or lung transplantation.

Microbiological diagnosis

The detailed results of SOC microbiological testing are displayed in Table 3. In total, 17 patients were diagnosed with bacterial pneumonia (6 in group IIS, 11 in group IIC); 3 patients were reported to have fungal infection, all of whom were in group IIS. Five patients were reported to have positive viral results, 4 of whom were associated with bacteria or fungi and were therefore considered to have bacterial or fungal pneumonia. The last patient had severe acute respiratory syndrome-coronavirus type 2 (SARS-CoV-2) monoinfection and was the only patient classified as having viral pneumonia. One patient in the NIIS group (NIIS.1) was reported to be positive by PCR for rhinovirus without biological signs of infection, and another patient (NIIS.3) was positive for “normal bacterial flora” associated with Staphylococcus aureus without any clinical signs of infection. All the other patients in the NIIS group had negative or “normal flora” positive cultures.

Diagnostic capacities of ABS

Altogether, 20 out of36 (55.5%) ABS results showed total concordance with the SOC and 6/36 (16.7%) discordances. Only one was due to ITS results. 16S ABS and SOC were indeed partially concordant for 10 out of 36 (27.8%) and discordant for 5 out of 36 (14%) samples (Fig. 4).

Fig. 4.

Schematic representation of concordant and discordant results between standard-of-care and amplicon-based sequencing results (A) and types of discordances observed (B). The absolute numbers and percentages of total samples and discordances are reported in the graph.

Among the 17 bacterial pneumonia patients, only 6 had completely concordant results. One discordance (IIC.5) occurred because ABS did not identify the major pathogen, Enterobacter cloacae complex, in the bacterial culture (75,000 CFU/mL). Instead, 10% of Streptococcus pneumoniae and 4% of Klebsiella sp. were amplified among a normal flora. Partial concordances due to misidentified H. influenzae also occurred twice (IIS.11 and IIC.11), as both, diagnosed in a mixed culture, were misidentified by ABS as H. haemolyticus.

Four of the ten remaining reported partial concordances and two discordances were due to additional potential pathogens found by 16S. The bacterial culture of Patient IIS.2 revealed 75,000 CFU/ml Pseudomonas aeruginosa, which was well identified by ABS but was associated with 33.7% Corynebacterium propinquum, which is described as a potential respiratory pathogen. Patient IIS.9 was reported to have a monoculture of Pseudomonas aeruginosa and ABS, with 83.5% of P. aeruginosa associated with 15.9% of Stenotrophomonas maltophilia, a frequently multidrug-resistant bacterium involved in pneumonia in IS patients. Patient IIC.9 was reported to have a monoculture of P. aeruginosa well retrieved by ABS at 58% but associated with 14.9% Citrobacter koseri. Patient IIS.12 had a concordance for Enterococcus faecium (50%); however, ABS identified an additional 21% of S. aureus that probably grew within what was identified as normal flora in culture. Similarly, among the discordant results, patient IIS.8 was diagnosed with viral SARS-CoV-2 pneumonia by SOC, and bacterial culture revealed normal flora only, but 16S revealed 30.5% S. aureus within this normal flora. Finally, patient IIS.1, who was diagnosed with undefined pneumonia, presented with 1,000 CFU/ml of normal flora not further identified on agar, which was, according to 16S results, composed of 86% Moraxella catarrhalis, a commensal species known to provoke pneumonia. In addition, patient IIS.3 is reported as having total concordant results due to negative SOC testing. However, 16S analysis revealed that 13.5% of Ureaplasma urealyticum, a pathogen not included in SOC testing but known to cause pneumonia in early-stage lung transplant patients such as IIS.3.

Quantification discordances also occurred in two discordant (IIS.10 and IIC.6) and four partially concordant samples (IIC.7, IIC.8, IIS.5, and NIIS.3). Indeed, IIC.10 was reported with 50,000 CFU/ml normal flora with 40% Escherichia coli, but only 6.5% were reported by 16S ABS; IIC.6 was reported with 5,000 CFU/ml S. aureus mixed with Hafnia alvei, but only 0.13 and 0.17% of the reads were assigned to these two species by ABS among a majority of mixed bacteria anaerobe taxa, without any notable predominance. More Enterobacteria were under quantified in other samples: Proteus vulgaris in Patient IIC.7 and E. coli in Patient IIC.8. In IIS.5 and NIIS.3, S. aureus was significantly affected by ABS but was considered a “normal flora” by SOC.

Three fungal infections were diagnosed by SOC, all in the IIS group, all of which were completely concordant with the ITS results, and all were caused by Aspergillus fumigatus, one of which was coinfected with Pneumocystis jirovecii (IIS.4). Patient IIS.5 was also considered possibly coinfected with P. jirovecii, as the specific PCR was weakly positive, with a “late” cycle threshold (Ct) of 34 (6 cycles later than the PCR result of IIS.4 (Ct 28)). Interestingly, the ITS amplification of P. jirovecii amplified only 0.9%, corroborating this “late” Ct.

The only discordance found in the NIIS group was due to the identification of one potential additional pathogen. Indeed, NIIS.2 had 75% of the reads assigned to Aspergillus sp. with a GM of 0.39 (which is below the threshold but significantly higher than that of other negative patients) and a negative culture (PCR was not performed). Notably, non-pathogenic Aspergillus reads were also identified in other patients, up to 11.6% in some cases. They were all associated with negative GM and PCR results with no other argument for fungal infection. Notably, IIS1, which was positive only for A. fumigatus by PCR in SOC, was well correlated with the results of the ITS analysis, with 2.4% of the reads assigned to A. fumigatus.

Correlation between evolution on day 28 and microbiological diagnosis.

A significant association between discordance, partial or total concordance of bacterial microbiological results and the clinical outcome on day 28 was demonstrated by a Pearson chi-square test (p = 0.04). Fungal pneumonias were excluded because they were all concordant and associated with fatal outcomes (Fig. 5). Patients whose bacterial infection may have been misdiagnosed by standard methods—indicated by a mismatch with ABS results—were more likely to experience poor clinical outcomes and be either dead or still infected by day 28 than those with fully matching diagnoses.

Fig. 5.

Evolution on Day 28 postinfection in 21 patients with pneumonia. Evolution on Day 28 was classified as deceased, still infected, or cured according to the normalization or not of pulmonary, radiological or biological symptoms and the cessation of antimicrobial therapy. For all patients, the status at D28 was compared with the presence of discordance (D) or partial (P) or total concordance (T) between the 16S ABS and SOC results. Patients with fungal pneumonia (n = 3) were excluded. The Pearson chi-square test p value = 0.04.

Discussion

Pneumonia remains one of the most common infections in both community and hospital settings. The range of potential pathogens is broad and spans all microorganism phyla, making microbiological diagnosis challenging, particularly in IS patients^7,15,16. To adapt our arsenal of diagnostic tests, we set up a preimplementation study of the usefulness of 16S and ITS metagenomics in IS and IC pneumonia patients.

Although multiple studies describing metagenomics (shotgun or ABS) as a clinically relevant diagnostic tool have been published, no “turn-key” solution exists for implementing these methods in a routine fashion in a clinical microbiology laboratory²⁵. Since these techniques remain both costly and time-consuming, clearly defining which specific clinical indications would benefit from an improved diagnosis is important. In addition, clinical microbiology laboratories should each develop tools designed to complete their existing arsenal of diagnostic tests. With this in mind, we designed a pilot study based on the needs of our own center for improving the etiological diagnosis of severe pneumonia. Despite being theoretically more informative, shotgun NGS was not our choice, as our main purpose was to use a tool that could be performed in a routine setting, with acceptable cost and turn-around time²⁶. We also intentionally chose a technique lacking viral investigation, as our center is equipped with a wide multiplex panel encompassing 23 species and subtypes of viruses²⁷. Even though this method is neither commercialized nor widely used, the development of fast multiplex panels targeting common viruses allows for accurate virus diagnosis in respiratory samples in a short turnaround time (45 min)^25,28,29. Eventually, 16S and ITS amplification could improve the ability to diagnose bacterial and fungal pathogens and help rare cases of parasitic infections or nontuberculosis mycobacteriosis undetected by our SOC PCR test³⁰.

Despite the small cohort size, our study included both infected and noninfected IS and IC patients, enabling a reasonable assessment of the tool’s clinical utility in a tertiary care hospital such as ours, which hosts a large ICU supporting haematology, heart–lung, and kidney transplant units. Alpha diversity was significantly lower for 16S analyses in the infected groups (IIS and IIC) than in the noninfected group (NIIS), which is in accordance with previous results^{3,13,14,25,31,32}. Additionally, high bacterial biomass was significantly associated with bacterial pneumonia in general and with VAP more specifically, as already described^3,13.

The 16S analysis resulted in 5 discordances and 10 partial concordances with SOC. These discrepancies were partly due to the limitations of 16S in identifying Enterobacter sp. (1 patient) at the genus level and Haemophilus influenzae (2 patients) at the species level. It is known that nontypeable H. influenzae isolates are genetically closely related to H. haemolyticus³³. As per Enterobacter cloacae complex, the same phenomenon of misidentification occurred with Klebsiella species, as they are also genetically closely related³⁴. Therefore, 16S may reveal a lack of specificity for species discrimination in these two genera, with an important impact on the diagnosis of bacterial infections. To overcome this problem, Charalampous et al.. used an added pathobiont-specific PCR for H. influenzae (and Streptococcus pneumoniae) to reinforce their identification via a 16S targeted ABS workflow³⁵. Another study that used a shotgun ABS on VAP also revealed that K. pneumoniae (along with P. aeruginosa and Acinetobacter baumannii) was not always well identified by ABS. The authors suggested that this was due to a “drowning” of information in the amplification of the oral flora³⁶. In the mock community we used as a control, Escherichia coli and Salmonella enterica were well distinguished and semiquantified, but Haemophilus, Enterobacter and Klebsiella were not tested. Positive controls for quality assessments in clinical settings should be specifically tailored to mock the communities expected in clinical samples³⁷.

Similarly, discrepancies in quantification were mostly observed for Enterobacteria, as illustrated by patients IIC.6, 7, 8 and 10. However, the fact that culture media are designed to highlight pathogenic bacteria among normal flora and anaerobes, which are impaired by their fastidious growth, obviously leads to an overestimation of Enterobacterales compared with their actual quantity in the sample. Moreover, Proteus sp., for example, which has the capacity to “swarm” and extend rapidly throughout agar, can easily be overquantified. We arbitrarily chose a threshold of 10% of reads for ABS to consider that the identified pathogen was present in significant abundance. However, S. aureus was found in more than 10% of several samples, for which it was not reported on bacterial culture or as a commensal flora, in patients with pneumonia but also in a patient clear of infection, as underlined in another study on VAP³⁸. S. aureus can cause bacterial pneumonia and viral pneumonia superinfection but is also part of the normal mucosal flora of the lower respiratory tract in chronic lung disease. Staphylococcus enrichment in the lower respiratory tract, however, was demonstrated to increase airway inflammation, which is linked to worse clinical outcomes, or to have adverse or triggering effects on CAP due to S. pneumoniae, H. influenzae or P. aeruginosa^25,39. The interpretation of such results would thus need to be performed in a larger cohort to understand the implications of such findings in the ABS report. Moreover, the significance threshold we chose might need to be adapted.

Finally, ABS allows the identification of potential additional pathogens for patients with a monobacterial SOC diagnosis or pneumonia of undefined aetiology either due to a lack of testing or to a positive bacterial culture but is below the threshold set for positivity in our laboratory (10³ CFU/mL). Among these, two bacteria—Moraxella catarrhalis and Corynebacterium propinquum—were possibly classified as normal flora on agar plates or may not have grown at all. C. propinquum is a discussed respiratory pathogen; however, it has been described in infections either in monoculture or in coinfection with P. aeruginosa or S. maltophilia^9,40,41.

As its presence is linked to misquantification and not evaluated by SOC, the involvement of anaerobic flora in pneumonia or the immune response must also be considered. VAPs in the ICU are linked to the microaspiration of the intestinal and oropharyngeal flora, which are widely composed of anaerobic genera. All VAP samples and almost all samples in our study were identified by ABS with anaerobic flora (mostly Prevotella, Fusobacterium, Veillonella, Porphyromonas, Alloprevotella and Actinomyces), which are related mostly to oropharyngeal flora. Studies that have evaluated the use of antibiotics to target anaerobes in pneumonia patients have shown no benefit^42,43. However, other reports associated phyla such as Prevotellaceae and Veillonellaceae with increased inflammation in the lower respiratory tract of HIV patients but also in non-ARDS patients or patients with healthy lungs^13,14,25,32. In our study, no genera were associated with immune status, antibiotic therapy or lung transplantation.

The ITS results revealed only one discordance with SOC, which was the identification by the ABS of A. fumigatus for NIIS.2. This highlights the difficulty of cultivating Aspergillus species routinely and demonstrates the added value of culture independent methods. However, the patient fit the definition for chronic airway colonization by Aspergillus, whose incidence after lung transplantation is common (up to 23%) ⁴⁵. The patient was already known to be colonized with A. fumigatus two years prior to lung transplantation but was never clinically infected. The identification and quantification of Aspergillus by SOC are performed via specific PCR, culture and GM dosing methods. The main limitations of SOC are twofold: the low sensitivity of culture and the limited disease prediction capability of specific PCR when used alone, although PCR can be combined with testing for genotypic azole resistance^45,46. The associations of GM and PCR were evaluated for the diagnosis of invasive pulmonary aspergillosis, with 98% specificity and 99% sensitivity⁴⁷. However, SOCs are currently not capable of diagnosing all cases of invasive aspergillosis, and the current definition of proven invasive aspergillosis is based on anatomopathological markers, which involve more invasive tests, such as biopsies^44,48. The positive correlation between the GM index and the number of reads of Aspergillus in our study suggests good concordance between the ITS region and SOC. In a previous study, the GM index was also reported to be capable of differentiating infection and colonization by Aspergillus⁴⁹. Taken together, the ITS was proven to be an interesting tool to improve the diagnostic capacity in the study of pulmonary fungal infections. The lack of Mucorales, Cryptococcus sp. or Scedosporium sp. infections in our cohort prevented us from assessing the diagnostic capacities for these fastidious pathogens and should be further tested.

The quantification of P. jirovecii in the two positive samples of the cohort yielded interesting results, as one was considered clinically relevant and infected, but the other was not considered infected based on clinicians’ appreciation. On the basis of the differences between relative and absolute abundances, we suspect a better semiquantification value of ITS than of SOC PCR, which has already been reported by Xing et al.. as statistically significant for 53 patients⁵⁰. Additionally, no other patients were reported with reads identified as P. jirovecii with neither ICs nor ISs, although colonization in the general population is reported to be 20–27%^51,52. Nevertheless, it must also be noted that all lung transplant patients were receiving trimethoprim‒sulfamethoxazole prophylaxis.

The assessment of the correlation between SOC and ITS was challenging, as GM and multiplex panels are only routinely performed for IS patients. However, for several IC patients, ABS results revealed up to 11.6% of the reads assigned to Aspergillus in their ITS samples, although the identified species were never reported as pathogenic. However, it has been reported that aspergillosis and the GM index could be linked to mechanical ventilation even in the absence of immunosuppression⁵³, and the patient with the highest percentage of Aspergillus in this group (IIC.12) had chronic hepatitis (no cirrhosis) and was an intravenous drug user, which is known to compromise immunity. This underlines the importance of reporting species and quantifying Aspergillus spp. Once again, an adaptation of the mock communities used for microbiological diagnostics must be considered to improve our ability to distinguish the absence and misidentification of several pathogens.

Overall, ITS analysis revealed greater added value for putative fungal infections than 16S did for bacterial infections. In the end, 16S, if tested alone, would have led to a default of diagnosis for several known respiratory pathogens. However, patients with discrepancies between ABS results and bacterial SOC diagnosis tended to have worse clinical outcomes, highlighting the potential added value of ABS. This finding may offer insights into other factors involved in airway infections⁷, helping to distinguish between colonization, infection, or polymicrobial infections³⁹. This potential added value of 16S was more obvious for IIS patients than for IIC patients because of the detection of multiresistant pathogens or fastidious pathogens such as Ureaplasma in IIS.3 lung transplant patients. This patient developed later a hyperammonaemia syndrome, which would have been diagnosed earlier and prevented owing to 16S testing. However, these infections are very rare, occur in a small niche of predisposed patients, and can eventually be monitored by screening donors before transplantation or by using a specific PCR^9,54. The method would first need to be improved and allow better isolate identification at the species level before being used in daily practice, and bacterial microbiota analysis by 16S rRNA amplification will certainly not replace the importance of obtaining a positive culture to perform resistance testing, although the resistome of the microbiota might be routinely assessed in the near future^15,55. In addition, to maximize its clinical impact, the duration of the methodology should be further improved. To this end, and to have an idea of its real clinical impact, it would be interesting to study its prospective time to results as well as the tangible changes that the results enable in terms of patient care.

To conclude, while 16S rRNA testing of BAL fluids might have potential applications in severe pneumonia, the actual incorporation of this test in our routine seems to be unattainable. This approach might allow the identification of nonculturable or fastidious agents for IS patients and patients who recently underwent lung transplantation but should display better identification performance for gram-negative bacilli and resistome-relevant information to be of potential broader interest. In contrast, ITS amplification in IS patients has added value in the diagnosis of severe fungal pulmonary infections and should be further explored in a larger set of patients. Finally, the interpretation of bacterial species identification or quantification with respect to SOC by skilled clinical microbiologists will be necessary, so they may, together with infectious disease specialists, pneumologists and intensivists, be able to reach actionable results in terms of patient management.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

CM, OVB and MHa: design, study protocol, writingCM, LN, NY, IE, MHi: data collection, sample processing and patient inclusionMLC: sample processing and metagenomic testingHI: bioinformatic analysisCM: interpretation results and statisticsTD, MLC, DB, LN, IE, MHi, HI: critical revision of the manuscriptAll the authors approved the final version.

Funding

Part of the sample processing was performed owing to a grant attributed to the LHUB-ULB annual funding request. This study was partially funded by the Laboratory of Clinical Biology of Universitair Ziekenhuis Brussel.

Data availability

The raw data were published at NCBI under Bioprojects PRJNA1193294.

Declarations

Competing interests

The authors declare no competing interests.

Ethics

Patients or their legal representatives signed an informed consent form before enrolment. The study was reviewed and approved by the local ethical committee of HUB-Erasme (reference P2022-099). This research was performed in accordance with the Declaration of Helsinki and European regulations for human research.