Microbial composition of tumorous and adjacent gastric tissue is associated with prognosis of gastric cancer

After sequencing, filtering, denoising, merging, chimeras removal, and normalizing to the minimum sequencing depth of 4153 sequences, the cohort was formed by 128 samples from 64 patients (each n = 64 tumorous GC tissue (T-GC) and non-tumorous GC (NT-GC)), from which roughly half a million reads were retrieved and merged into 5450 different phylotypes belonging to 19 phyla and 296 genera.

Variation of the bacterial communities in the stomach of GC patients

First, we evaluated the differences in microbial composition of mucosa from NT-GC and T-GC specimens from GC patients. As shown in the Fig. 2A, we observed comparable bacterial community in 28% (18 out of 64) of paired T-GC and NT-GC specimens, while 72% differed between T-GC and NT-GC tissues. The Bray–Curtis similarity for the samples clustering together was 56 ± 14%, while for the remaining samples it was 22 ± 14%, suggesting a very high intra-individual variation of the bacterial communities within the stomach between NT-GC and T-GC. Due to this intra-individual variation, both PERMANOVA and ANOSIM analyses showed no statistical differences between the bacterial communities in T-GC and NT-GC at all taxonomic ranks (at phylotype level: p = 0.7, F = 0.9 and p = 0.5, R = 0, respectively). The data on phylotype variation and abundance are shown in Supplementary Fig. S1. Intra-individual variation was found in 46 patients, where the bacterial community from NT-GC and T-GC of the same patient differed considerably. Despite this variability, a dominant phylotype with at least 10% of abundance was detectable in all samples. It is worth mentioning that phylotypes which are typically found only in the lower GI^8,9 were detected in the stomach tissues to a greater or lesser extent, such as Bacteroides, Clostridium sensu stricto, or Escherichia/Shigella, suggesting a possible shift in the colonization of microorganisms from the lower to the upper GI in patients with GC (Supplementary Tables S5 and S6).

Overall bacterial communities based on the Bray–Curtis similarities between the samples of the complete cohort (A). Only the samples of the same patient that clustered together are shown in green lines. If the abundance of H. pylori was > 15% this is denoted with black filled triangle and if < 15% with black triangle. The most abundant genus found in the sample is displayed with different colours. Species richness (B) and Shannon diversity (C) of every adjacent and tumour tissue sample. Global bacterial community in adjacent tissue and in tumour tissue at phylum (D) and genus (E) level. Relative abundances of the phylum Proteobacteria (F) and the genus Helicobacter (G) in adjacent and tumour tissue. Significantly different taxa are indicated by * if q < 0.05 and ** if q < 0.01. Percentage of abundance of the phylotypes belonging to H. pylori in each patient (H). The patients are sorted as a function of the total number of different phylotypes detected (right).

Characterization of bacterial community between tumorous and non-tumorous specimens

The species richness was overall comparable between NT-GC and T-GC samples (Fig. 2B). Analysis of the taxa distribution between the groups showed statistical difference only in the abundance of Proteobacteria (specifically Helicobacter pylori) between both tissues (Fig. 2D,E), with a greater abundance in NT-GC in comparison to T-GC (Proteobacteria q = 0.009 and Helicobacter q = 0.01). In correlation, the Shannon diversity was higher in T-GC than in NT-GC specimens (Fig. 2C), which is likely to reflect the H. pylori-linked changes. As expected, the variation of H. pylori abundance in both NT-GC and T-GC ranged from undetectable to highly abundant levels (Fig. 2F,G). In total, 8 different H. pylori phylotypes were recognized in H. pylori positive samples, while no phylotype was observed in 6 patients. Among subjects harbouring H. pylori sequences, 32 patients were infected with only one phylotype, 23 patients carried two phylotypes, and 3 patients were infected by three different phylotypes; however, more detailed characterization was not possible (Fig. 2H).

Mucosal microbial composition and OS

The survival data of 53 GC patients (83% of the cohort) were available for up to seven years (Table 1). To evaluate the effect of various factors, including age, gender, tumour region, and tumour staging (UICC and Lauren’s classification), we first performed OS analysis regarding those factors. The OS analysis revealed no significant difference related to age, gender, and tumour localization (Supplementary Fig. S2A–C). In comparison, patients with advanced tumour stages (III–IV) and Lauren´s diffuse tumour type showed worse OS compared to those with earlier tumour stages (I–II) and intestinal type (p = 0.03 and p = 0.049), respectively (Supplementary Fig. S2D,E). No difference was observed in frequently discovered genera between major histological tumour types based on Lauren´s or UICC classification (Supplementary Fig. S3A,B). Lower values were observed for the Shannon index and for species richness in diffuse tumours compared to intestinal type GC tumours (q = 0.029 and q = 0.036, respectively) (Supplementary Fig. S3C–F).

To evaluate the effect of bacterial diversity on the OS, we applied median abundance values as a cut-off for subgrouping. Neither species richness nor Shannon diversity in T-GC or NT-GC samples had any association with differences in OS in GC patients (Supplementary Fig. S3G–J). However, systematic analysis of bacterial taxa revealed that three out of eight families were associated with OS differences (Supplementary Fig. S4). Fusobacteriaceae, due to Fusobacterium, and Chitinophagaceae, due to Asinibacterium, were associated with statistically significant although discordant difference depending on the tissue type T-GC vs. NT-GC (Fig. 3). In the family Prevotellaceae, the genus Prevotella showed a residual association with worse OS. Higher abundance of Fusobacterium in T-GC was associated with a significantly worse OS (Fig. 3). Prevotella was associated with a trend similar to that of Fusobacterium but to a lesser extent (p = 0.045, Fig. 3C).

Kaplan–Meier-curve of the abundances of Fusobacterium in tumour tissue (A). Correlation of OS data (in days) with the abundance of Fusobacterium in tumour tissue, with linear regression and Spearman rank coefficient (B). Kaplan–Meier-Curves of the abundances of Prevotella in tumour tissue (C) and the abundances of Asinibacterium in adjacent tissue (D). Significant differences in OS are indicated by * if q < 0.05 and ** if q < 0.01. (E) Kaplan–Meier-Curve of the abundance of Fusobacterium based on publicly available TCMA data.

Neither Fusobacterium nor Prevotella were associated with differences in OS when the abundances were considered in NT-GC samples (Supplementary Fig. S5). However, patients with higher abundance of Asinibacterium in NT-GC had better OS (p = 0.0044), while no difference was found in T-GC (Fig. 3D and Supplementary Fig. S5). Since the abundance of Fusobacterium and Prevotella in adjacent and tumour tissue showed a strong positive correlation (Supplementary Fig. S5), we believe that an observed association with worse overall survival may be linked to functional role in tumorous tissues. Abundance of H. pylori was not associated with OS and the abundances in tumour and adjacent tissue are less correlated (Supplementary Figs. S4 and S5).

To confirm those results, we applied Cox regression analysis. In univariable Cox regression analyses, UICC III–IV (p = 0.04) and abundance higher than the median of Prevotella (p = 0.049) and Fusobacterium (p = 0.01) were significantly associated with the overall survival of GC patients. Multivariable analysis performed using Cox proportional hazards model revealed that diffuse type according to Lauren’s classification (p = 0.002) and high Fusobacterium abundance (p = 0.009) were independently predictive for the worse overall survival of patients with GC (Table 1).

To validate our findings, we accessed publicly available data from the TCMA²⁴ cohort of 107 patients with a survival data up to 7 years and including 16% of diffuse type tumours (our cohort: 53 patients with 7 years survival follow-up, 50% diffuse type). The Kaplan–Meier-curve analysis based on the abundance of Fusobacterium revealed a trend for a worse overall survival (p = 0.08) partially supporting our hypothesis (Fig. 3E).

Phylogenetic biodiversity of Fusobacterium species in the stomach

Next, we performed a detailed characterization of Fusobacterium species in the stomach following a phylotype annotation. The DNA sequences of the phylotypes taxonomically annotated as Fusobacterium detected in the present study (n = 64, 108 sequences) were compared with publicly available data from the previous study with phylotypes annotated also as Fusobacterium detected in healthy individuals (n = 12, 127 sequences)⁸ and linked to 53 types strains of the genus Fusobacterium available in NCBI database²⁵. The Supplementary Fig. S6 displays the mismatches of example 16S sequences annotated to different Fusobacterium species indicating the species resolution specially of the V1–V2 region (primers used in this work). After sequences alignment and phylogenetic tree construction, the majority of the phylotypes detected in GC patients clustered around the species F. nucleatum or F. necrophorum, whereas in healthy individuals, the vast majority of the phylotypes clustered together around the species F. periodonticum (Fig. 4A). Interestingly, when the abundances of the phylotypes belonging to Fusobacterium in both cohorts were compared, GC patients harboured higher abundances of F. necrophorum or F. nucleatum compared to healthy individuals in the stomach tissue (Fig. 4B). Furthermore, the majority of GC patients with F. periodonticum also had detectable F. nucleatum in much higher abundance. No significant difference could be found between the abundance of F. nucleatum in tumour and adjacent tissue of GC patients (Supplementary Fig. S7). In healthy individuals, however, F. periodonticum was more abundant than F. nucleatum. Similar analyses were performed with the phylotypes annotated as Prevotella and Asinibacterium, and no significant species differences between tumour and healthy tissues were found for those two genera (data not shown).

Phylogenetic tree built with the DNA sequences of the V1–V2 region of the 16S rRNA gene from 108 phylotypes annotated to Fusobacterium from the tumour tissue of the 64 GC patients of this study (red), 127 phylotypes annotated to Fusobacterium from corpus tissue of 12 healthy individuals previously published⁸ (green) and 53 type strains of Fusobacterium (black). Cluster in red contains sequences belonging to F. necrophorum, cluster in green contains sequences belonging to F. periodonticum, and cluster in violet contains sequences annotated as F. nucleatum (A). Percentage of abundances of F. necrophorum, F. periodonticum, and F. nucleatum in tumour tissues and healthy tissues (B).

Bacterial community network in GC patients

To understand the bacterial interaction in GC, we performed a bacterial network analysis. The extraordinary complexity of the bacterial community in the stomach tissue became evident when the co-occurrences between phylotypes were investigated (Fig. 5 and Supplementary Table S7). A total of 1191 positive and negative co-occurrences were detected between phylotypes (Fig. 5A). H. pylori (Phy1) only showed three negative correlations with three phylotypes belonging to Fusobacterium nucleatum (Phy8), Prevotella sp. (Phy27) and Selenomonas sp. (Phy184). However, Phy8, Phy27, and Phy184 showed 37, 35, and 24 neighbours in the bacterial network, respectively, suggesting that H. pylori does not co-occur with any bacteria and even is antagonistic to F. nucleatum and/or Prevotella sp. In contrast to H. pylori, all the interactions detected for both F. nucleatum and Prevotella sp. were positive, with the exceptions of Phy1 and Phy10 (Sphingobium limneticum), which negatively co-occurred with F. nucleatum, as well as Phy1, Phy2 (Asinibacterium), and Phy25 (Ralstonia detusculanense or picketii), which negatively co-occurred with Prevotella. These findings suggest that F. nucleatum and Prevotella have a close interplay with each other and with stomach bacterial communities in GC patients. In fact, both Phy8 and Phy27 had the maximum number of neighbours (together with Phy980 belonging to Ruminococcaceae) in the bacterial network (Fig. 5B and Supplementary Table S7).

Bacterial community network with the interactions between the phylotypes in GC samples (A). The annotation of the phylotypes displays only if the abundance was > 10% in at least one sample. Interactions were considered statistically significant if − 0.3 < ρ > 0.3 and if q-value was < 0.05. Betweenness centrality as a function of the number of neighbours from all nodes of the bacterial community network (B).