This retrospective study was conducted at Renji Hospital, Shanghai Jiao Tong University School of Medicine. A total of 61 patients were enrolled from 20 June 2021 to 20 October 2024. Among them, 23 were classified as CIP/RP group (of whom, 16 had CIP and 7 had RP), and 38 patients were diagnosed with LC-P. Inclusion criteria for CIP or RP included: (1) the cancer patients had received immunotherapy or radiotherapy; (2) imaging studies showed new pulmonary infiltrates (radiologic patterns of CIP include cryptogenic organizing pneumonia, ground glass opacities and interstitial pneumonia, while radiologic features of RP are ground glass opacities or consolidation conforming precisely to the shape of the radiation field.); (3) LRT infection or lung tumor progression was excluded; and (4) the patients were not treated with antibiotics or steroid within 2 weeks.

BALF samples were collected from all 61 patients according to the standard procedures. The lung was lavaged with 100mL of sterile saline solution and the BALF recovery rate was more than 35%. BALF samples for further mNGS were then transported to the hospital laboratory under cold-chain conditions. Clinical information and laboratory results were also collected when the patients were sampled.

The TIANamp Magnetic DNA Kit (Tiangen) was used to extract DNA. Quantity and quality of DNA were assessed using the Qubit and NanoDrop (Thermo Fisher Scientific), respectively. DNA libraries were prepared using the Hieff NGS C130P2 OnePot II DNA Library Prep Kit for MGI (Yeasen Biotechnology) according to the manufacturer’s protocols. Agilent 2,100 was used for quality control and DNA libraries were 50 bp single-end sequenced on MGISEQ-200. Raw sequencing data were split by bc12fastq2 (version 2.20), and high-quality sequencing data were generated using Trimmomatic (version 0.36) by removing low-quality reads, adapter contamination, duplicated and shot (length, 36bp) reads. Human host sequences were subtracted by mapping to human reference genome (hs37d5) using bowtie2 (version 2.2.6). Reads that could not be mapped to the human genome were retained and aligned with the microorganism genome database for microbial identification by Kraken (version 2.0.7), and species abundance estimating by Bracken (version 2.5.0). The microorganism genome database contained genomes or scaffolds of bacteria, fungi, viruses, and parasites (download from GenBank release 238, ftp://ftp.ncbi.nlm.nih.gov/genomes/genbank/).

We performed the microbial diversity analysis using R software (version 4.0.1). The alpha-diversity was assessed by taxonomic profiles, and the beta-diversity was estimated by Bray- Curtis distance. PERMANOVA (vegan) was used to analyze beta-diversity differences. Differences of the relative genus abundances were tested by the Kruskal-Wallis test (Kruskal.test package). Only genera with greater than 1% mean abundance and 40% prevalence were compared. Linear discriminant analysis (LDA) effect size (LEfSe) was performed to assess the statistical differences of the relative abundance of microorganisms between CIP/RP and LC-P patients. Spearman’s correlations between clinical indicators and relative genus abundances were determined by R package cor. test and adjusted by false discovery rate. A random forest binary classification model integrating key microbes and significantly clinical indicators was assessed by Receiver Operating Characteristic Curve.

Student’s t-test or Mann-Whitney U test was used to compare the continuous variables. For categorical variables, Chi-square test or Fisher’s exact test was used to explore the association. All significance tests were two-tailed and a P value < 0.05 was considered statistically significant.

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Renji Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China (KY2021-102-B). Informed consent was obtained from all patients.

The study included 61 patients. Among these patients, 23 were classified as grade1–2 CIP/RP (of whom, 16 had CIP and 7 had RP), while 38 were confirmed to have LC-P. The demographic and clinical characteristics of the patients are detailed in Table 1 . The patients in the CIP/RP group were younger than those in the LC-P group. Comorbidities comprised chronic obstructive pulmonary disease (COPD, 17.4%), hypertension (13%) and diabetes (13%) in the CIP/RP group. The CIP/RP group had lower levels of lymphocytes (P = 0.001), EOS (P = 0.008), and ALB (P = 0.006) in peripheral blood than those in the LC-P group. In BALF, the percentage of lymphocyte was significantly high in the CIP/RP group (P = 0.010). The CIP/RP group also had higher levels of C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), Krebs Von den Lungen-6 (KL-6), and D-dimer, but no significant differences were observed in these clinical indicators.

At species level, alpha-diversity was based on ACE, Chao1, Shannon, and Simpson indexes ( Supplementary Figure 1A ). However, there were no significant differences in ACE (P = 0.63), Chao1 (P = 0.63), Shannon (P = 0.8), and Simpson (P = 0.92) indexes. Additionally, PCoA and PC analysis based on the Bray-Curtis distances also showed that no difference was observed in beta-diversity (P = 0.481 and P = 0.477, respectively) ( Supplementary Figure 1B ). We then analyzed the lung microbial composition in CIP/RP and LC-P groups. In the CIP/RP group, the top five phyla were Pseudomonadota (27.0%), Bacteroidota (26.1%), Bacillota (19.4%), Actinomycetota(19.0%) and Peploviricota (3.0%) ( Figure 1A ). The top five genera included Prevotella (22.7%), Rothia (8.7%), Pseudomonas (7.8%), Streptococcus (7.7%), and Veillonella (6.7%) ( Figure 1B ). The top five species were Prevotella melaninogenica (11.6%), Rothia mucilaginosa (9.7%), Pseudomonas aeruginosa (6.0%), Prevotella jejuni (5.6%) and Haemophilus parainfluenzae (4.1%) ( Figure 1C ). In the LC-P group, the top five phyla included Bacteroidota (27.2%), Pseudomonadota (22.2%), Bacillota (21.6%), Actinomycetota (17.4%) and Ascomycota (4.2%) ( Figure 1A ). The top five genera were Prevotella (19.4%), Rothia (9.2%), Streptococcus (6.9%), Veillonella (6.6%) and Haemophilus (4.8%) ( Figure 1B ). The top five species included Rothia mucilaginosa (6.9%), Prevotella melaninogenica (5.4%), Prevotella jejuni (4.7%), Prevotella pallens (4.1%) and Haemophilus parainfluenzae (3.7%) ( Figure 1C ).

We further analyzed the differential relative abundance of top 10 phyla, top 20 genera, and top 20 species between CIP/RP and LC-P groups. At the phyla level, no significant differences were observed in the relative abundance of the top 10 phyla. At the genera level ( Figure 2A ), the relative abundance of Porphyromonas (P = 0.028) and Neisseria (P = 0.010) were significantly lower in the CIP/RP group than those in the LC-P group. No significant differences were observed in the relative abundance of the other genera between the two groups. At the species level ( Figure 2B ), the relative abundance of Prevotella melaninogenica (P = 0.018) and Cytomegalovirus humanbeta5 (P = 0.010) were significantly higher in the CIP/RP group than those in the LC-P group. However, the relative abundance of Neisseria subflava (P = 0.024) and Porphyromonas pasteri (P = 0.045) were significantly lower in the CIP/RP group than those in the LC-P group. No significant differences were observed in the relative abundance of the other species between the two groups. 22 discriminative features were identified by LEfSe. Among them, 13 taxa were discriminative for the CIP/RP group and 9 taxa were discriminative for the LC-P group ( Figure 2C ). At the genera level, the Cytomegalovirus (LDA scores >4, P = 0.010) was significantly higher in the CIP/RP group while Porphyromonas (LDA scores >4, P = 0.028) and Neisseria (LDA scores >4, P = 0.010) were abundant in the LC-P group. At the species level, Prevotella melaninogenica (LDA scores >4, P = 0.018) and Cytomegalovirus_humanbeta5 (LDA scores >4, P = 0.010) were significantly higher in the CIP/RP group while Neisseria_subflava (LDA scores >2, P = 0.024) and Porphyromonas pasteri (LDA scores >2, P = 0.045) were significantly more abundant in the LC-P group.

A Spearman correlation analysis was used to further explore the relationship between top 20 genera (or species) and clinical indicators. A two-dimensional heatmap showed the results ( Figures 3A, C ). We mainly focused on differential microbiota and clinical indicators with significant difference. At the genera level, the Porphyromonas had no correlation with all clinical indicators. The Neisseria showed a positive correlation with ALB. At the species level, the Prevotella melaninogenica had a negative correlation with B lymphocytes, NK, and lymphocytes in peripheral blood. The Cytomegalovirus humanbeta5 showed a negative correlation with B lymphocytes and ALB in peripheral blood. The Neisseria subflava and Porphyromonas pasteri had no correlation with significantly different clinical indicators. Furthermore, a CCA analysis showed that the levels of peripheral blood EOS, ALB, and BALF-L were found to have a strong relation with the top 20 genera or species ( Figures 3B, D ). In addition, a random forest binary classification model was constructed. The model integrated Prevotella melaninogenica and significantly clinical indicators including peripheral blood lymphocyte, EOS, ALB and BALF-L, yielding an AUC of 0.755 ( Figure 3E ).