利用单细胞转录组学绘制结直肠癌中肿瘤内细菌的细胞图谱

Tumor-associated microbes are increasingly investigated across cancer types. In CRC, intratumoral microbes are associated with detrimental treatment responses and clinical outcomes. However, critical gaps remain in understanding how individual cells within the TME respond to microbial presence. Defining these cellular responses could provide mechanistic insights into microbe-host interactions in cancer and reveal novel therapeutic targets. To map the cellular landscape of intratumoral bacteria in CRC, we developed a custom analytical pipeline integrating single-cell transcriptomics with metagenomic profiling. We performed scRNA-seq on 150 samples from 146 CRC patients across 85 primary tumors, 60 liver metastases, and 5 peritoneal metastases. Unsupervised clustering was performed using Seurat 5.1.0, with cells annotated using PanglaoDB and canonical markers. Single-cell reads were aggregated and mapped against NCBI and SILVA databases. Starting with 1.09x10 9 reads from 650,485 cells, we applied stringent quality filters including trimming, human genome filtering (removing 98.7% of reads), and a custom k-mer diversity filter to exclude likely contaminants. This final filter eliminated 96% of remaining reads and 92% of initially identified taxonomies, addressing the challenge of ambient contamination in scRNA-seq data. Our pipeline ultimately retained 18,115 high-confidence bacterial reads (<0.001% of total) from 4,888 cells (0.8%), representing 1,237 unique taxa. After contaminant exclusion, we identified 89 species across 48 genera, including Bacteroides fragilis, Parvimonas micra, Gemella morbillorum, and Fusobacterium nucleatum. Bacterial reads were mapped to cells using barcodes. Clustering identified 16 cell populations with bacterial signals across all major cell types and tumor sites. In this exploratory single-cell analysis, cells from primary colon tumors exhibited increased bacterial diversity compared to cells from liver metastases (p<0.01). Within primary colon tumors, cells from MSI-H tumors (n=15) demonstrated significantly higher bacterial reads than cells from MSS tumors (n=67, 95% CI [2.425, 12.691], p = 0.006). Furthermore, the presence of bacteria-positive cells in primary colon tumors was associated with significantly worse patient survival compared to tumors without detectable intracellular bacteria (HR 4.90, 95% CI [1.08, 22.2], p = 0.039). Our findings demonstrate that bacterial reads, although present at extremely low abundance, can be retrieved from scRNA-seq data when stringent contamination controls are applied. The taxonomic resolution achieved provides proof-of-concept for investigating microbe-host interactions at single-cell resolution. This study establishes a methodological foundation for leveraging single-cell technologies to probe the complex interplay between tumors and their microbiome.