Dear editor,
Patients with clear-cell renal cell carcinoma (ccRCC) potentially have a high recurrence rate of more than 40% [
1]. Patients receive radical or partial nephrectomy, sometimes assisted by targeted therapies or/and immune checkpoint inhibitors (ICIs), to improve overall survival (OS) [
2]. Large-cohort studies and large-scale human tissue sequencing have provided a series of molecular subtyping methods for precision medicine-based ccRCC risk stratification and therapeutic regimens, such as ClearCode34, the prognostic risk predictor [
3], and seven subsets for ICI and angiogenesis blockade outcomes [
4]. However, these subtyping results are all based on human genomes without including the host microbiota, which may be non-negligible genome components.
Trillions of microorganisms form a specific ecosystem in the human body, and their imbalance may influence human health and lead to diseases, including cancer [
5,
6]. Most previous microbiome studies utilized fecal samples but potentially missed information about microbes within organs [
7]. Data from next-generation sequencing for human tissues consist of microbiome read count [
8]. The Cancer Genome Atlas (TCGA) has explored the presence of bacteria in various tumors [
7] and supports diagnosis for some cancers [
8]. Thus, the analysis of microbial genetic data from TCGA deserves investigation into the crosstalk between clinical and molecular levels. Pan-cancer study evidence in seven types of malignancies revealed through 16S rRNA sequencing that the human tumor microbiome contains cancer-specific intracellular bacteria [
9]. However, this study did not include RCC samples, and thus, no hypothesis or conclusion on microbiome subtyping for ccRCC or morphological microbiota diversity has been reported.
Treatment-naïve patients were pathologically diagnosed with ccRCC, whose tumor and paired para-tumor tissues were aseptically obtained for 16S rRNA sequencing (Table S1, n = 31) and untargeted metabolome detection (n = 24 pairs; see Supplementary Material). After contaminations in bacterial genome profiling of low microbial biomass samples were removed, 16S rRNA sequencing data analysis showed that alpha diversity was increased in ccRCC tumor tissues compared with the matched para-tumor tissues (Fig.1; feature, P < 0.05; abundance-based coverage estimators (ACE), P < 0.01; Chao1, P < 0.01; Simpson, P < 0.05; Shannon, P < 0.05), indicating the emergence and elimination of certain microorganisms. Interestingly, the abundance of aerobic bacteria in the tumor tissues was decreased (Fig.1, P < 0.05), whereas no significant differences was observed at the anaerobic or facultatively anaerobic level, suggesting an alteration in oxygen demands in the tumor microenvironment (TME), i.e., hypoxia, a commonly seen phenomenon in ccRCC. Meanwhile, the genus of microbe with distinct characteristics between the tumor and normal tissues was detected (Fig.1). Bacterial RNA in ccRCC was detected using RNA fluorescence in situ hybridization (FISH) and generic probes targeting bacterial 16S rRNA (rRNA). Similar spatial shapes were revealed in Gram staining and 16S rRNA (Fig.1).
We then further explored ccRCC subtyping based on bacterial genome. We extracted bacterial sequencing reads from TCGA KIRC and removed the contamination interference as reported by Poore
et al. [
8] (Supplementary Material). We applied multiple clustering algorithms (IntNMF, ConsensusClustering and SNF; see Supplementary Material) to classify 428 ccRCC patients with complete clinical information and effective bacteria counts and obtained comprehensive subtyping from consensus sets with high consistency (Fig. S1A–S1C). Based on bacteria relative abundance, three subtypes were regarded as optimal clusters for ccRCC (Fig. S1D and S1E). The upregulation of bacteria genus in each subgroup is shown in Fig.1 and Supplementary Material. Survival analysis showed that the three subtypes were statistically different in terms of OS, with Group 1 or Group 3 having worse outcomes (
P = 0.0107) (Fig.1). Combined with bulk-seq data in these 428 patients, three subgroups displayed distinct mRNA expressions (Fig. S1E). The Gene Set Enrichment Analysis (GSEA) result suggested that Group 1 presented enriched metabolism reprogramming-related pathways (Fig.1, S1F, 2A, and 2B) and downregulated immune-activated pathways compared with Group 2, while Group 3 presented downregulated anti-tumor immune pathways and unclassical ccRCC-metabolism pathways (Fig.1, S2C, and S2D).
Considering that the tumor bacteria genome may influence human genetic characterization, we found that the top 20 mutated genes in the three groups had different features combined with the whole-exome sequencing data in the TCGA KIRC cohort. VHL, PBRM1, and TTN were the top three mutated genes. Of the three groups, Group 2 had the lowest VHL (47%) and PBRM1 (38%) mutation rate (Fig. S3), which were associated with metabolic function change and immunosuppression [
1]. SETD2 and BAP1 mutations were both more than 10% in each group (Fig. S3), ranking as more frequent mutations, and they were genes acting as the driver of invasiveness [
10]. Finally, we mapped the bacterial cohort profiles of the three subgroups by using the k-nearest neighbor classification algorithm. Combining the mapped profiles with the metabolomics data, we enriched distinct pathways (
P < 0.05) in our three groups by differential metabolite analysis (Fig.1 and S4). Glycerophospholipid metabolism was enriched in both Groups 2 and 1, while the linoleic acid metabolism was enriched in both Groups 1 and 3 but not arginine biosynthesis. These metabolome features may suggest that abnormal linoleic acid metabolism pathways could lead to worse prognosis, although further analysis is required due to the relatively few studies on ccRCC.
Our work has several limitations. The cohort in this multi-center genome study included only patients from the TCGA (n = 428) and our institution (n = 31). In addition, the matched metabolomics data were not totally consistent with the 16S rRNA sequencing data due to the underused paired sample tissues biomass.
In this study, we first identified three subtypes of the ccRCC bacteria microbial genome and found that they had distinct molecular features that were associated with different survival outcomes, thereby suggesting that the microbiota genome may have crosstalk with or even affect the human genome.
Electronic Supplementary Material
Supplementary material is available in the online version of this article at https://doi.org/10.1007/s11684-023-1029-3 and is accessible for authorized users.
PDF
(3052KB)
