Blood mNGS: an effective non-invasive diagnostic tool for Pneumocystis jirovecii pneumonia

Abstract

Background

Pneumocystis jirovecii pneumonia (PJP) is a life-threatening opportunistic infection. Colonization is prevalent but cannot be reliably distinguished from active infection by conventional methods. Metagenomic next-generation sequencing (mNGS) is a promising diagnostic tool, but the value of blood mNGS for diagnosis, microbial community comparison, and outcome-related associations in PJP remains unclear.

Methods

We analyzed 73 suspected PJP patients with paired BALF and blood mNGS. Using strict diagnostic criteria, patients were classified as: PJP (n = 50) and P. jirovecii colonization (PJC, n = 23). Bioinformatic analyses compared compartment-specific microbiota. BALF-blood concordance and associations between P. jirovecii load and outcomes were evaluated.

Results

BALF showed higher α-diversity than blood (both Shannon and Simpson, P < 0.001), whereas β-diversity showed no compartmental segregation. BALF identified 216 species versus 43 in blood; however, the top-10 species were concordantly ranked (90% concordance). Blood mNGS distinguished PJP from PJC with an AUC of 0.80 (specificity 95.7%, sensitivity 62.0% at RPM > 4.8), outperforming BALF mNGS (AUC 0.76), blood PCR (AUC 0.64) and BALF PCR (AUC 0.73). Gram-negative bacteria accounted for a large proportion of blood taxa (75% of top 20 taxa), while BALF showed additional fungal taxa including Aspergillus fumigatus. LEfSe identified matrix-specific taxa: oral commensals in PJC-BALF. Blood P. jirovecii load correlated positively with LDH (r = 0.34, P = 0.0035), CRP (r = 0.34, P = 0.0031), and BDG (r = 0.26, P = 0.025), and was higher in non-survivors (P < 0.05).

Conclusion

Blood mNGS may serve as a non-invasive, highly specific complementary tool for PJP diagnosis and broader microbiological assessment.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12866-026-05408-7.

Summary of the article’s main point

This study, using paired BALF and blood samples, supports blood mNGS as a non-invasive, specific tool for PJP diagnosis and broad microbiological assessment, while identifying descriptive associations between blood P. jirovecii load and clinical outcomes, and comparing its performance with conventional assays.

Introduction

Pneumocystis jirovecii pneumonia (PJP) is the most frequent AIDS-defining illness when CD4⁺ T-cell counts fall below 200 cells/µL [1], and mortality in this group ranges from 10 to 20% [2]. In the expanding non-HIV immunocompromised population, incidence is rising [3–5] and case-fatality has been reported to reach 30–60%, with cancer patients bearing the highest risk [6].

However, P. jirovecii colonization, detecting the organism or its DNA without typical pneumonia symptoms, can be prevalent in clinical practice and is closely associated with immune status, underlying diseases, and environmental exposure [7]. In immunocompromised patients, clinical manifestations are often non-specific, making it difficult to reliably distinguish colonization from active infection. This distinction can affect clinical decision-making regarding the initiation and duration of targeted anti-infective therapy. Microscopic examination of biopsy tissue, bronchoalveolar lavage fluid (BALF), or induced sputum using various staining techniques is considered the gold standard for PJP diagnosis [8, 9]. Conventional diagnostics including imaging and serum biomarkers (β-D-glucan [BDG] and lactate dehydrogenase [LDH]) exhibit low specificity for PJP diagnosis [10]. Polymerase chain reaction (PCR) detection of P. jirovecii nucleic acids in lower respiratory specimens significantly improves diagnostic accuracy, with sensitivity ranging from 94% to 100% and specificity from 79% to 96% [11]. Although quantitative PCR thresholds have been proposed to help differentiate colonization from active infection, this distinction remains clinically challenging due to frequent asymptomatic respiratory tract colonization [7, 12]. Meanwhile, invasive sampling barriers in critically ill patients further complicate diagnosis. Although blood-based PCR is emerging [13], its poor sensitivity severely limits clinical utility [14], and co-infections may require broader pathogen detection [15].

To address these limitations, metagenomic next-generation sequencing (mNGS) has emerged as an unbiased pathogen detection tool [16]. Circulating microbial cell-free DNA in peripheral blood may serve as a potential indicator of pulmonary barrier impairment, microbial translocation, and circulating microbial DNA signals, thereby providing a potential approach for diagnostic assessment and outcome-related analyses. However, blood mNGS for PJP diagnosis is underexplored. The potential of blood-based mNGS, which provides a less invasive alternative to BALF, distinguish between infection and colonization, and characterize the broader microbial community in PJP patients, may help characterize disease-related microbiological patterns and inform clinical interpretation.

This study aims to evaluate the diagnostic performance of mNGS in paired blood and BALF for PJP, comparing it with conventional PCR methods. We assessed the potential of blood mNGS to differentiate between P. jirovecii infection and colonization, characterize microbial communities in PJP patients, and investigate the consistency between BALF and blood microbiota profiles. Additionally, we explore the descriptive association between P. jirovecii detection by mNGS in both sample types and mortality.

Methods

Ethical approval

Ethical approval was granted by the Medical Ethics Committee of West China Hospital, Sichuan University (No. 2023 − 890). Written informed consent was obtained from all participants or legal guardians in accordance with the Declaration of Helsinki.

Study design

This retrospective cohort was derived from 203 consecutive adults in whom P. jirovecii was detected by mNGS in BALF at West China Hospital between December 2023 and December 2024. We included 73 patients who had paired BALF and blood samples collected within 7 days of each other. Using 2019 EORTC/MSGERC framework [8], two independent respiratory specialists adjudicated cases into P. jirovecii pneumonia (PJP, n = 50) or P. jirovecii colonization (PJC, n = 23) (Fig. 1 and Table S1). Detailed diagnostic assessments for each patient are provided in Table S2.

Fig. 1.

Flowchart of patient enrolment and group assignment. Flowchart illustrating the patient selection process and final group allocation. mNGS: metagenomic next-generation sequencing; BALF: bronchoalveolar lavage fluid. aDetailed “mycological evidence”, “host feature” and “clinical feature” are shown in Table S1

Specimen collection and processing

Paired BALF and blood were placed in DNA-stabilising Cryo.S™ tubes and stored at − 80 °C until January 2025, then all samples were processed in a single batch. Three blank negative controls (NC) were introduced during both BALF and blood workflows (total n = 6). Total nucleic acids (DNA and RNA) were extracted simultaneously from 0.3 mL of thoroughly mixed sample using the TIANamp Micro DNA Kit (TIANGEN). All steps from extraction to sequencing were performed by one technician on dedicated equipment to minimize batch effects.

Construction of library and sequencing

Separate DNA and RNA libraries were constructed from the extracted total nucleic acids. Library quality control was performed using the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA), confirming insert sizes of 200–300 bp. Library concentrations were quantified with the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific). High-throughput sequencing was conducted on the MGISEQ-2000 platform (Shenzhen, China) at Precision Medicine Center of West China Hospital (Chengdu, China). Sequencing generated 50-bp single-end reads with approximately 100 million reads per sample.

Bioinformatics analyses

DNA-seq and RNA-seq raw data were first filtered by fastp, and then the filtered reads were aligned to the human genome (GRCh38) using HISAT2 (2.2.1 release) to remove human sequences [17]. The remaining data were removed from low-complexity sequences and then compared to the BGI Microbial Reference Database PMDB [18] using Kraken2. The six blank NC samples used the same analysis process (Table S4). The read counts of all the species were then normalized to reads per million (RPM) before further analysis. Microorganisms were retained for analysis if they met all the following criteria for subsequent analyses: (1) no engineered microbial strains commonly found in NC samples; (2) no known contamination; and (3) only human viruses were selected. The relative abundance (the abundance of a particular species in each sample divided by the sum of the abundance of all species in that sample) was used for subsequent statistical comparisons. The raw sequencing data supporting this study have been deposited in the China National GeneBank DataBase (CNSA: https://db.cngb.org/cnsa/) under accession number CNP0007814.

P. jirovecii PCR detection

BALF (300 µL) was centrifuged (12 000 × g, 10 min); blood (500 µL) was lysed 1:1 with 3.7% NaOH then centrifuged (11 700 × g, 4 min). A 121-bp fragment of the P. jirovecii mitochondrial LSU rRNA was amplified using primers PjF1/PjR1 and TaqMan-MGB probe PjSL on a QuantStudio 6 Pro (Thermo Fisher) [19]. Cycling: 50 °C 2 min, 95 °C 4 min, then 40 cycles of 95 °C 15 s and 60 °C 45 s. Samples were run in duplicate alongside positive (ATCC PRA-111) and extraction controls. Positivity was defined as Ct ≤ 40.0 with RNase P Ct ≤ 32.0, verified against a 5 copies/µL limit of detection [20].

Statistical Analysis

Statistical analyses were conducted in R 4.4.1. Baseline characteristics were summarized using the tableone package. Continuous variables were reported as mean (SD) if normally distributed, or median [IQR] otherwise. Between-group comparisons were performed using Wilcoxon rank-sum tests. Categorical variables are presented as n (%) and compared via χ² or Fisher’s exact tests. Microbiome relative abundances were calculated as species reads/total microbial reads. Alpha diversity and between-group differences were evaluated with Kruskal–Wallis or Wilcoxon tests, and beta diversity was analyzed principal coordinate analysis (PCoA) on Bray-Curtis dissimilarity with ANOSIM [21]. All multiple comparisons were adjusted via Benjamini-Yekutieli false discovery rate (FDR) correction. Spearman correlations (r ≥ 0.30) examined associations between microbial and clinical variables; receiver operating characteristic (ROC) curves and diagnostic accuracy were assessed using the pROC package. LDA effect size (LEfSe) identified differentially abundant taxa [22]. Significance thresholds: P < 0.05, FDR q < 0.10. The statistical and plotting code is available on GitHub (https://github.com/huaxiyiyuan/mNGS-pjp).

Results

Patient demographics and clinical characteristics

In total, 73 patients with paired BALF and blood mNGS data were included in the analysis. Based on predefined criteria, patients were categorized into two groups: PJP (n = 50) and PJC (n = 23) (Table 1). Clinical symptoms differed significantly, with fever (P = 0.001) and hypoxemia (P = 0.002) being more prevalent in the PJP group. Immunosuppressive agent use was significantly higher in the PJP group (64.0%) compared with the PJC group (34.8%) (P = 0.038). Laboratory findings revealed significantly elevated LDH (IU/L) and BDG (ng/mL) levels in the PJP group compared with the PJC group (P < 0.001 for both). mNGS results showed significantly higher median P. jirovecii RPM in both blood (20.50 vs. 1.00) and BALF (229.00 vs. 20.21) of the PJP group (P < 0.001 for both). Although not statistically significant (P = 0.399), mortality rates trended higher in the PJP group (44.00%, 22/50) compared to the PJC (30.4%, 7/23).

Table 1.

Demographic and Clinical Characteristics of Patients

	PJP(N = 50)	PJC(N = 23)	P value
Age (years) (mean (SD))	55.18 (12.61)	56.83 (14.09)	0.619
Gender			0.229
male, n/N (%)	28 (56.0)	17 (73.9)
female, n/N (%)	22 (44.0)	6 (26.1)
Clinical symptoms
fever, n/N (%)	41 (82.0)	9 (39.1)	0.001
cough, n/N (%)	39 (78.0)	16 (69.6)	0.628
dyspnea, n/N (%)	47 (94.0)	18 (78.3)	0.11
hypoxemia, n/N (%)	36 (72.0)	7 (30.4)	0.002
Duration of hospitalization (days), median [IQR]	26.00 [17.25, 45.00]	20.50 [11.23, 32.12]	0.123
Immunosuppressive drug therapy, n/N (%)
corticosteroid	35 (70.0)	14 (60.9)	0.615
immunosuppressive agents	32 (64.0)	8 (34.8)	0.038
biological agents	6 (12.0)	3 (13.0)	1
Underlying_diseases, n/N (%)			0.026
AIDS	5 (10.0)	1 (4.3)
Chronic respiratory diseases	4 (8.0)	3 (13.0)
Hematology diseases or solid organ tumor	8 (16.0)	5 (21.7)
Immune-mediated inflammatory disease	22 (44.0)	6 (26.1)
Organ failure	1 (2.0)	0 (0.0)
Organ transplantation	10 (20.0)	3 (13.0)
Others	0 (0.0)	5 (21.7)
Positive imaging findings, n/N (%)	49 (98.0)	20 (87.0)	0.17
LDH (IU/L), median [IQR]	454.00 [353.75, 652.00]	221.00 [178.50, 261.00]	< 0.001
BDG (ng/L), median [IQR]	287.20 [96.61, 599.36]	37.00 [37.00, 44.21]	< 0.001
Pneumocystis nucleic acid testing, n/N (%)	48 (96.00%)	12 (100.00%)	< 0.001
Positive PCR(Blood), n/N (%)	14 (28.00%)	0 (0.00%)	0.016
mNGS (RPM), median [IQR]
Blood	20.50 [2.00;408.00]	1.00 [0.00, 2.50]	< 0.001
BALF	229.00 [46.00;2398.75]	20.21 [7.01, 300.81]	< 0.001
Outcome, n/N (%)			0.399
Improvement	28 (56.0)	16 (69.6)
Death	22 (44.0)	7 (30.4)

Bold values represent statistically significant intergroup differences, with statistical significance defined as P < 0.05

Abbreviations: PJP Pneumocystis jirovecii Pneumonia, PJC Pneumocystis jirovecii Colonization, AIDS Acquired Immunodeficiency Syndrome, LDH Lactate Dehydrogenase, BDG β-D-Glucan, PCR Polymerase Chain Reaction, mNGS Metagenomic Next-Generation Sequencing, RPM Reads Per Million, BALF Bronchoalveolar Lavage Fluid, SD Standard Deviation, IQR Interquartile Range

Concordance of microbial detection between BALF and blood samples

In the paired-sample analysis, metagenomic analysis demonstrated substantial agreement in pathogen detection between 73 paired blood and BALF samples. Although BALF exhibited greater species richness (216 vs. 43, Fig. 2A), commonly detected taxa were consistently identified across both compartments, with 30 species co-detected (Fig. 2A). The combined top 10 taxa (totaling 12 unique species) exhibited substantial overlap between blood and BALF (Fig. 2B–C). Acinetobacter baumannii dominated both sample types (blood: 36.7%; BALF: 25.7%), while key pathogens including P. jirovecii (blood: 4.8% vs. BALF: 11.6%), Escherichia coli (12.3% vs. 7.6%), Pseudomonas aeruginosa (9.5% vs. 6.2%), and Klebsiella pneumoniae (3.4% vs. 5.4%) were consistently detected in both matrices. In addition, the relative abundance of P. jirovecii, A. baumannii, E. coli, and P. aeruginosa differed between the two groups (Fig. 2D). We attempted to determine the differentially abundant species using LefSe to better understand this microbial dissimilarity between blood and BALF (Fig. 2E). P. jirovecii showed higher relative abundance in BALF (P < 0.001), whereas A. baumannii, E. coli, and P. aeruginosa demonstrated significant higher relative abundance in blood (all P < 0.001) (Fig. 2F). For the remaining species, the Wilcoxon rank-sum test was used to obtain the P values for between-matrix comparisons between the blood and BALF, which are presented in Table S3.

Fig. 2.

Microbial concordance between paired blood and BALF specimens. A Venn diagram showing total and shared species detected in blood and BALF. Average compositions of relative abundance of the top 10 species for (B) Blood and (C) BALF. D Alterations of all the top 10 species between the two groups. E LEfSe cladogram highlighting differentially abundant taxa in blood and BALF. F The relative abundance of P. jirovecii, A. baumannii, E. coli and P. aeruginosa between blood and BALF. LDA scores represent the degree of difference between the two groups. The p-values were confirmed by the Wilcoxon rank-sum test. LDA, linear discriminant analysis; P values < 0.05 were considered statistically significant

Diagnostic performance of mNGS and PCR

ROC curves were constructed to distinguish PJP from PJC (Fig. 3). Blood mNGS used an RPM cut-off of 4.8 and yielded an AUC of 0.80 (95% CI 0.70–0.89), corresponding to a sensitivity of 62.0% (31/50) with high specificity (95.7%, 22/23). BALF mNGS, set at 30.8 RPM, displayed a marginally higher sensitivity (72.0%, 36/50) but specificity was 82.6% (19/23). Blood PCR (CT < 39.96) showed a low sensitivity of 28.0% (14/50) despite 100.0% specificity, and BALF PCR (CT < 30.71) reached 40.0% sensitivity (20/50) with 95.7% specificity (22/23). Collectively, blood mNGS showed the highest AUC among the evaluated assays for differentiating active PJP from colonization in this cohort.

Fig. 3.

Comparative Diagnostic Performance for Differentiating PJP and PJC

Co-detected pathogen profiles between PJP and PJC

To investigate potential co-detected microorganisms associated with P. jirovecii infection in blood and BALF specimens, we analyzed the top 20 most abundant microorganisms in the PJP group and compared their distribution with the PJC group (Fig. 4A for blood; Fig. 4B for BALF). Gram-negative bacteria dominated blood (75%, 15/20) and BALF (70%, 14/20); BALF additionally harboured fungi (P. jirovecii 2nd, Aspergillus fumigatus 12th). Sankey diagrams further visualized the hierarchical relationships and flow of these top 20 microorganisms across the two subgroups (Fig. 4C). LEfSe was then applied to identify taxa that were differentially enriched between the PJP and PJC groups. In blood (Fig. 4D), P. jirovecii, Human betaherpesvirus 5 and Candida glabrata were enriched in PJP, while E. coli, P. aeruginosa, Staphylococcus haemolyticus, K. pneumoniae, and Stenotrophomonas maltophilia were characteristic of PJC. In BALF (Fig. 4E), P. jirovecii again was identified as an enriched taxon of the PJP group. The PJC group showed enrichment of oral commensal taxa, including Atopobium parvulum, Prevotella salivae, Corynebacterium propinquum, Streptococcus parasanguinis and Streptococcus oralis.

Fig. 4.

Differential abundance of co-detected microorganisms and LEfSe-defined microbial biomarkers across PJP and PJC subgroups in blood and BALF. Top 20 co-detected taxa in blood (A) and BALF (B) within the PJP and PJC groups. C Alterations of all the top 20 species among PJP and PJC. Subgroup-specific microbial signatures identified by LEfSe. D Blood-derived differential taxa among PJP and PJC (LDA ‌‌≥ 3.0). E BALF-derived differential taxa among PJP and PJC (LDA ‌‌≥ 3.0)

Compartmental and subgroup diversity patterns

To investigate differences in microbial richness and community structure, we compared alpha- and beta-diversity metrics across sample types and clinical subgroups. Blood microbiota exhibited significantly lower α-diversity than BALF (Simpson, P < 0.001, Figure S1A; Shannon, P < 0.001, Figure S1B). Notably, PCoA based on Bray-Curtis dissimilarity revealed no distinct clustering between blood and BALF compartments (Figure S1C). Within blood, PJP and PJC displayed comparable α-diversity (Simpson, all P > 0.5, Figure S1D; Shannon, all P > 0.5, Figure S1E), yet PJP samples showed separation in PCoA (Figure S1F). In BALF, microbial α-diversity differed significantly between PJP and PJC, with lower Simpson and Shannon indices observed in the PJP group (P = 0.0033 and P = 0.0028, respectively; Figure S1G–H), and Beta-diversity analysis by PCoA showed separation of community structures between the PJP and PJC groups (Figure S1I).

Association of P. jirovecii loads with clinical outcomes

Non-survivors had higher blood P. jirovecii loads than survivors (blood mNGS-RPM, P < 0.05, Fig. 5A), whereas P. jirovecii loads and relative abundance in either compartment did not differ in BALF (all P > 0.05). Blood P. jirovecii loads correlated positively with the value of serum BDG (RPM r = 0.26, P = 0.025; relative abundance r = 0.27, P = 0.02), while BALF load did not (Fig. 5B and Figure S2 B). Both the RPM and relative abundance of P. jirovecii correlated with LDH and CRP: blood RPM–LDH r = 0.34, P = 0.0035; BALF RPM–LDH r = 0.32, P = 0.0051; blood RPM–CRP r = 0.34, P = 0.0031; BALF RPM–CRP r = 0.35, P = 0.0025 (Fig. 5C–D and Figure S2C-D).

Fig. 5.

Association of P. jirovecii load with clinical outcomes. A Comparison of blood and BALF mNGS-RPM between survivors and non-survivors. B Correlation between the RPM of P. jirovecii and the value of serum BDG in blood and BALF. C–D Correlations of the RPM of P. jirovecii with the value of LDH and CRP in blood and BALF

Discussion

This study provides a comprehensive analysis of the diagnostic performance of mNGS in both BALF and blood for PJP, as well as insights into the microbial landscape associated with PJP. We observed consistent microbial communities between BALF and blood samples. Additionally, blood mNGS demonstrated better performance in differentiating active PJP from colonization relative to BALF mNGS and PCR. Notably, higher blood P. jirovecii loads determined by mNGS were detected in non-survivors than in survivors and showed a descriptive association with serum markers commonly evaluated in PJP, suggesting that blood P. jirovecii load may serve as a potential descriptive marker associated with clinical outcomes.

BALF exhibited significantly greater species richness (216 vs. 43 taxa) and Shannon/Simpson diversity than blood, reflecting the continuous exposure of the lower respiratory tract to environmental and upper-airway flora. Despite the differences in diversity, the lack of distinct clustering in the PCoA analysis between BALF and blood samples indicates a considerable degree of similarity in overall microbial composition [23]. Thirty species were simultaneously recovered in both compartments, and nine of the combined top-10 microorganisms were shared, with A. baumannii, P. jirovecii, E. coli, P. aeruginosa and K. pneumoniae among the most abundant taxa in both matrices. However, the relative abundance of these microorganisms differed significantly between compartments. LEfSe analysis revealed that P. jirovecii was significantly enriched in BALF, consistent with its pulmonary tropism and role in PJP. Conversely, A. baumannii, E. coli, and P. aeruginosa exhibited significant blood-tropic enrichment, suggesting circulating microbial DNA signals or compartment-specific detection patterns. These findings align with prior reports of gut-derived bacterial translocation in critical illness [24], and are consistent with potential compartment-specific microbial signals: pulmonary-to-systemic dissemination for P. jirovecii, and systemic seeding of gram-negative pathogens via translocation from the gut or other reservoirs [11]. Importantly, the 90% concordance in top 10 microorganisms between blood and BALF supports the potential utility of blood mNGS as a less invasive complementary approach when bronchoscopy is contraindicated, particularly for identifying circulating microbial DNA signals. However, the enrichment of P. jirovecii in BALF highlights the superior sensitivity of lower respiratory samples for fungal pathogens. Taken together, these compartment-specific abundance patterns suggest that blood and BALF mNGS may provide complementary microbiological information.

The high rate of P. jirovecii colonization complicates distinguishing colonization from active PJP [25]. Our evaluation of mNGS and PCR on blood and BALF samples revealed a higher AUC for blood mNGS (AUC: 0.80) compared to BALF mNGS (0.76), blood PCR (0.64), and BALF PCR (0.73). In this cohort, blood mNGS achieved high specificity (95.7%) with suboptimal sensitivity (62.0%). While guidelines recommend BALF-based diagnosis [26], its invasive collection poses risks (e.g., pneumothorax, infection spread), especially in critically ill patients. Blood sampling is minimally invasive and may offer a practical advantage. The high specificity of blood mNGS, which was higher than that of BALF mNGS (82.6%) and comparable to that of BALF PCR (95.7%), supports its potential value as a complementary rule-in approach for PJP in this cohort. Compared with blood PCR, blood mNGS showed higher sensitivity (62.0% vs. 28.0%) while maintaining high specificity (95.7% vs. 100.0%). However, the sensitivity of blood mNGS remained modest, indicating that a negative blood mNGS result cannot exclude PJP. The biological basis and clinical implications of blood P. jirovecii DNA detection require further investigation in prospective studies. This difference may be partly related to mNGS’s untargeted approach [27], enabling detection of cell-free DNA at lower loads without reliance on specific amplification [28], unlike PCR. In summary, blood mNGS may serve as a potential minimally invasive complementary diagnostic tool for PJP, with high specificity and a high positive predictive value (96.9%), enhancing feasibility particularly in critically ill patients. However, blood mNGS showed suboptimal sensitivity (62.0%), and a notable proportion of clinically adjudicated PJP cases yielded negative results by blood mNGS; therefore, a negative result cannot exclude PJP, and this limitation should be considered in clinical application.

Alpha diversity did not differ among PJP and PJC cohorts, yet PJP patients exhibited higher absolute abundances of bacteria, fungi and viruses, reflecting the cumulative immunosuppressive effects of prolonged corticosteroids, cytotoxic chemotherapy or transplantation [29–31]. Such patients are generally more susceptible to multiple concurrent infections or colonization by opportunistic pathogens [32]. The higher abundance of diverse microorganisms in PJP patients suggests a complex microbial landscape in immunocompromised hosts and highlights the potential for co-infections that may complicate diagnosis and treatment [33]. mNGS uncovered multiple concomitant microorganisms that would have escaped targeted assays, emphasizing its value for comprehensive pathogen mapping, provided that the results are interpreted cautiously in the appropriate clinical context to avoid unnecessary antimicrobial use.

Non-survivors displayed significantly higher blood RPM values than survivors (P < 0.05), whereas BALF levels were indistinguishable between groups. This dissociation suggests the possibility that blood mNGS may reflect disease severity or alveolar–capillary barrier disruption, rather than solely local pulmonary organism burden; however, this interpretation remains hypothesis-generating and requires further validation. Notably, heterogeneous immunosuppressive conditions, including corticosteroid use, chemotherapy, hematologic malignancy, and critical illness, may independently influence microbial translocation and circulating DNA levels, which should be considered in prognostic interpretation. Consistent with this interpretation, blood RPM correlated with LDH (r = 0.34, P = 0.004), a positive correlation with the extent of pulmonary damage in PJP cases [34], with BDG (r = 0.26, P = 0.03), a serum fungal cell-wall marker commonly used in the diagnostic evaluation of PJP [35], and with CRP (r = 0.34, P = 0.003), the acute-phase reactant that tracks systemic inflammation. Taken together, these correlations reveal preliminary links between blood P. jirovecii load and systemic inflammation and lung injury, which may offer exploratory clues for future studies on risk stratification before clinical application.

Our findings also have potential implications for the clinical interpretation and application of mNGS in suspected PJP. For patients with high clinical suspicion of PJP, particularly those who are critically ill or at high risk for bronchoscopy-related complications, blood mNGS may serve as a rapid, minimally invasive initial diagnostic test, while BALF mNGS remains valuable for PJP diagnosis by providing lower respiratory tract microbiological evidence. In clinical practice, simultaneous testing of both blood and BALF by mNGS may be considered to maximize diagnostic yield, with positive detection in both compartments supporting clinically significant infection rather than isolated airway colonization.

Notably, mNGS frequently detects multiple potential pathogens in immunocompromised patients. In scenarios where P. jirovecii is identified in both blood and BALF samples, even if another organism such as A. baumannii exhibits higher relative abundance, clinical decision-making should prioritize the overall clinical picture, including radiological features, inflammatory markers, and response to empirical therapy. Given the high specificity of blood mNGS for PJP in our cohort, concurrent detection of P. jirovecii in both blood and BALF may support clinically significant infection, and therapy targeting PJP may be considered after clinical assessment, while considering the presence of other potentially pathogenic bacteria. Combined interpretation of mNGS results with clinical context remains essential to reduce the risk of overlooking severe opportunistic infections such as PJP.

The retrospective single-center design and modest sample size may limit generalizability. Restricting analyses to 73 patients with paired BALF and blood samples from an initial cohort of 203 may also introduce additional selection bias. Despite rigorous adjudication, misclassification bias cannot be excluded. In addition, broad organism detection by mNGS may identify microorganisms that do not necessarily represent clinically significant infection; if overinterpreted, such co-detections could potentially contribute to unnecessary antimicrobial use. RPM thresholds require prospective validation across diverse immunocompromised populations. Finally, causality between blood RPM and mortality remains to be evaluated in longitudinal studies.

Conclusion

Blood mNGS may serve as a highly specific, non-invasive complementary diagnostic tool for PJP and may provide broader microbiological information. In this cohort, blood P. jirovecii load showed a descriptive association with mortality and systemic inflammatory markers, suggesting a potential descriptive outcome-related association. Although it does not universally replace BALF-based diagnostics, it may offer a useful alternative for high-risk patients for whom invasive procedures are contraindicated. Further prospective validation is needed to determine whether RPM thresholds can improve clinical interpretation when integrated with established diagnostic criteria.

Abbreviation

PJP

Pneumocystis jirovecii pneumonia

mNGS

metagenomic next-generation sequencing

BALF

Bronchoalveolar lavage fluid

PCR

Polymerase chain reaction

PJC

Pneumocystis jirovecii colonization

LDH

Lactate dehydrogenase

BDG

β-D-glucan

RPM

Reads per million

ROC

Receiver operating characteristic

AUC

Area under the curve

CI

Confidence interval

Ct

Cycle threshold

FDR

False discovery rate

IQR

Interquartile range

LEfSe

LDA effect size

PCoA

Principal coordinate analysis

CRP

C-reactive protein

NC

Negative controls

Authors’ contributions

The authors appreciate all the participants in this research. Y.Z and T.N were responsible for the idea and design of the article. Collection of data: Y.C, A.Z, M.L, S.L and X.G. Analysis of the data: Y.C, S.L, A.Z, Y.W, Y.Y, Q.W and M.H. Writing the manuscript: Y.C, S.L and A.Z. Revising the manuscript: Y.Z, S.L and T.N. The final draft was read and approved by all authors before submitting the manuscript.

Funding

This work was supported by National Key Research and Development Program of China (No. 2022YFC2406804), National Natural Science Foundation of China (No. 82370192, U24A20680, 82402588), 1.3.5 Project of High Altitude Medicine (No. GYYX24003), West China Hospital, Sichuan University, the Natural Science Foundation of Sichuan Province (No. 2024NSFSC1746), the Scientific Research Fund of Yunnan Provincial Department of Education (2024J0304), and the Kunming Medical Joint Special Project of Yunnan Provincial Science and Technology Plan Project (202401AY070001-295).

Data availability

The raw sequencing data supporting this study have been deposited in the China National GeneBank DataBase (CNSA: https://db.cngb.org/cnsa/) under accession number CNP0007814.

Declarations

Ethics approval and consent to participate

Ethical approval was granted by the Medical Ethics Committee of West China Hospital, Sichuan University (No. 2023 − 890). Written informed consent was obtained from all participants or legal guardians in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.