Tumor-Asociated Microbiota in Esophageal Squamous Cell Carcinoma

Background: Human intestinal tract microbiome dysbiosis plays an emerging pivotal role in tumorigenesis of gastrointestinal tract cancers. For esophageal squamous cell carcinoma (ESCC), the esophageal microbiota plays a critical role during the pathogenesis. The microbiome of esophageal can impact its host decades before the onset of ESCC, and can interact with the host’s physiological situation, which are affected by lifestyle factors, including diet, obesity, alcohol and tobacco use, and oral hygiene.Our objective is to analyze the composition of the ESCC-associated microbiota and characterize its contribution to the development of ESCC. The esophageal microbiota was prospectively investigated in 18 patients with ESCC and 11 patients with physiological normal (PN) esophagus by 16S rRNA gene proling, using next-generation sequencing. Results: The microbiota composition in tumor tissues of ESCC patients is signicantly different from that of patients with PN tissues. The ESCC microbiota was characterized by reduced microbial diversity, by decreased abundance of Bacteroidetes, Fusobacteria and Spirochactes. Employing these taxa into a microbial dysbiosis index demonstrated that dysbiosis microbiota had good capacity to discriminate between ESCC and PN esophagus. Functional analysis of the microbiota characterized that ESCC microbiota had altered nitrate reductase and nitrite reductase functions when compared with PN group. The observations were conrmed in other validation cohorts. Conclusions: Detailed analysis of the microbiota of the ESCC patients revealed that tumor exhibit a dysbiotic microbial community when compared with PN groups. We characterized microbial compositional changes, and further identied signicant enrichments of Treponema amylovorum, Streptococcus infantis, Prevotella nigrescens, Porphyromonas endodontalis, Veillonella dispar, Aggregatibacter segnis, Prevotella melaninogenica, Prevotella intermedia, Prevotella tannerae, Prevotella nanceiensis and Streptococcus anginosus in

biomarkers/therapy strategy for esophageal carcinoma. Further studies are needed to clarify the pathogenesis of esophageal cancer and to explore new diagnostic and therapeutic possibilities. · The intestinal tract microbiota, containing at least 38 trillion bacteria, is critical for the maintenance of homeostasis and health [5]. Progress in metagenome-wide association studies of fecal samples has characterized some important microbial markers of colorectal cancer (CRC) [6,7], and the causal effect of bacteria on cancer has been recognized [8]. Also, the microbiome has been discovered to be involved in the initiation and progression of various types of cancer, such as liver cancer [6,9,10]. Experimental evidence indicates that the human intestinal microbiome can in uence tumor development and progression in the gastrointestinal tract by damaging DNA, activating oncogenic signaling pathways, producing tumor-promoting metabolites, and suppressing the antitumor immune response [6,7,11,12,13,14]. Esophagus is an important part of the upper digestive system and more and more attentions are paid to the relationship between microbiome and esophageal cancer. Esophageal cancer is one of the most aggressive malignant cancers. Treatment strategies provided by conventional therapies have limited improvements in clinical outcome. It is then critical to seek innovative clinical strategies for treating this type of cancer. As intestinal tract microbiota plays important roles during tumorigenesis, exploiting microbiota for cancer prevention or treatment may be feasible.
However, only a small number of studies characterized the human esophageal microbiota in health and disease [15]. The major ndings in esophageal adenocarcinoma were that lipopolysaccharides, a major structure of the outer membrane in gram-negative bacteria, can upregulate gene expression of proin ammatory cytokines via activation of the Toll-like receptor 4 and NF-κB pathway and promote the occurrence of Barrett esophagus and adenocarcinoma [16,17]. Host interactions with microbiota in esophagitis, Barrett's esophagus, esophageal adenocarcinoma and ESCC can be different. Here we focused on the microbiota of ESCC. As for ESCC, the microbiome was less well characterized. Only few studies suggested that the change of microbiota such as Fusobacterium were associated with the occurrence of ESCC [18,19]. The association between the change of esophageal microbiome and ESCC development has not been well elucidated [15,20]. Therefore, to investigate the relationship between the changes in esophageal mucosal microbiota and the occurrence of ESCC, we conducted a prospective study and performed high-throughput pro ling of the esophageal mucosal microbiota in ESCC cases and normal controls. The next-generation sequencing (NGS) of the 16S rRNA gene were used to determine microbiota communities potentially associated with ESCC. We have demonstrated microbial relative abundances at the phylum and genus level for ESCC. We illustrated the impact of ESCC-associated bacterial taxa during the pathogenesis of ESCC.

Patients
This prospective observational study was conducted according to a protocol approved by the respective Institutional Ethics Committees of the First A liated Hospital of Sun Yat-sen University and according to the Declaration of Helsinki. This GASTO1039 study was registered at http://www.chictr.org.cn/ (the Chinese Clinical Trial Registry: ChiCTR1800018897). Eighteen patients with esophageal squamous cell carcinoma (ESCC) and eleven volunteers (individuals with physiological normal (PN) esophagus and no evidence of esophageal diseases) from the First A liated Hospital of Sun Yat-sen University were included in the discovery cohort between October 2018 to March 2019 (Supplementary table S1 and S2).   A validation cohort of an additional 20 ESCC patients and 4 volunteers were enrolled between from April  2019 to December 2019 (Supplementary table S1 and S2). Tissues for analysis came from patients undergoing routine esophagogastroduodenoscopy for the investigation of upper gastrointestinal symptoms or surgical resection. All patients provided written informed consents.
16S rDNA gene sequencing DNA was extracted from healthy group (PN group) and tumor group (T group) according to the protocol of E.Z.N.A.® Bacterial DNA Kit (Omega Bio-tek, Norcross, GA, USA). The 16s rDNA V4 hypervariable region was ampli ed using primers 515F 5'-GTGCCAGCMGCCGCGGTAA-3' and 806R 5'-GGACTACHVGGGTWTCTAAT-3'. PCR products were puri ed with AmpureXP beads to keep the target fragment. At the end 2*250bp paired-end reads were generated by sequencing on the Illumina HiSeq2500 platform. The primers were assessed using PrimerProspector Software Package (see Supplementary gure S1).

Sequencing data analysis
High-quality data were acquired as the in-house procedure from BGI Co., Ltd, China (Shenzhen, China to eliminate low-quality reads, N reads and so on. Then the clean reads were overlapped to obtain tags using FLASH (v1.2.11) [21]. The clean tags which were dereplicated and ltered singletons to cluster into operational taxonomic units (OTUs) at 97% sequence similarity using USEARCH (v9.1.13) in order to obtain OTUs representative sequences and otu abundance in each sample [22]. Every OTU were assigned to the Greengene Database (v201305, http://greengenes.secondgenome.com/downloads) at the similarity threshold of 0.5 using RDP Classi er (v2.2) [23,24]. Alpha diversity analysis was performed by mothur (v1.31.2) [25]. The nonparametric tests were adopted in alpha diversity analysis. Wilcoxon rank sum test was used between two groups, and Kruskal test was used in the comparison of three groups or more than three groups in alpha diversity analysis. Beta diversity analysis were assessed by QIIME (v1.9) [26]. The results of beta diversity were showed by the principle coordinate analysis (PCoA) of weighted and unweighted UniFrac distance. Differences in beta diversity were evaluated by ANOSIM analysis of similarity and Mantel correlation analysis with 999 permutations [27].

Differential taxonomy analysis
The comparisons of genera relative abundance were performed by Metastats (http://metastats.cbcb.umd.edu/) and linear discriminant analysis (LDA) effect size (LEfSe) between PN and T groups [28,29]. The genus with greater than 3 at the base of P-value <0.05 were considered to choose.

Functional metagenome predictions
We predicted the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Clusters of Orthologous Groups (COG) function by Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt, v1.0.0) [30], after constructing the close reference of OTU representative sequences from the discovery cohort Qiime software. The accuracy of the predicted metagenome was evaluated by the Nearest Sequenced Taxon Index (NSTI) value. NSTI demonstrates the mean value of phylogenetic distance between all microbiota OTUs and closest sequenced genomes in one sample. The NSTI value is lower, the accuracy of the predicted metagenome is higher. The predicted KEGG and COG function analysis by STAMP software using two-sided test with Welch's t-test corrected with Benjaminin-Hochberg FDR [31]. The LEfSe analysis was based on the relative abundance of predicted KEGG Pathways. The correlation analysis was performed between the relative abundance of predicted KEGG Level 2 Pathways and microbial dysbiosis index (MDI).
Logistic regression and receiver operating characteristic (ROC) analyses ROC curves were constructed to re ect the discriminatory potential of the microbiota abundance to detect esophageal cancer. ROC curves and P-values were analyzed according to Wilson/Brown method recommended by GraphPad Prism v8.0.2. Good reproducibility is showed by area under the curve (AUC) results. AUC represents the area under the curve, and CI represents the con dence intervals. The AUC value is higher, and the accuracy of ROC curve is better.

16S rRNA microbiota pro ling in physiological normal esophagus and ESCC
We compared the 16S rRNA gene of the esophageal microbiota between patients with ESCC (T) and patients with physiological normal esophagus (PN) by NGS. After sequencing and quality ltering, more than 1171914 million clean Tags were obtained corresponding to a mean of 310 OTU and clean tags 40410 per sample. The number of clean Tags was not signi cantly different between PN and T groups (supplementary gure S2A). To determine the number of biologically signi cant OTUs, we classi ed the OTUs of PN and T groups according to the sequences of the Greengenes Named Isolated database. The results indicated that the frequency of bacteria OTUs was major fraction in OTU classi cation (supplementary gure S2C). Furthermore, Venn diagram displaying the number of common and speci c OTUs identi ed between PN and N groups (supplementary gure S2D).

The pro le of esophageal microbiota demonstrates difference in physiological normal esophagus and ESCC
We computed the alpha diversity of microbes of the T and PN using the OUTs, Shannon index, Simpson index and Good's coverage. On average, patients with ESCC had a signi cantly lower number of OUTs than PN esophagus ( Fig. 1A and supplementary gure S2F). However, the Shannon index and Simpson index were not signi cantly different between the two groups (Supplementary gure S2B and E). To estimate the bacterial diversity, we used Good's coverage estimator to determine the proportion of total bacterial species represented in samples of each group. Statistical analysis of Good's coverage showed that T groups had signi cantly different numbers of species when compared with PN groups (Fig. 1B) by measuring beta diversity using both unweighted and weighted UniFrac phylogenetic distance matrices. The microbiota composition of patients with T groups was signi cantly different from that of PN groups (ANOSIM R = 0.7879, P = 0.001; and ANOSIM R = 0.6346, P = 0.001, for unweighted and weighted distances, respectively; Fig. 1C and D).
Age is one of the risk factors for ESCC development. Since patients with ESCC were signi cantly older than patients with PN esophagus in our cohorts (supplementary table S1), we next asked whether the microbial pro le was different between the two groups. Overall, age factor was not in uenced the microbiota pro les of full sample set (supplementary gure S3A and B). However, compared with agematched microbiota in patient with ESCC and PN esophagus using unweighted and weighted UniFrac distance matrices, we found that microbiota composition was signi cant in the two clinical settings (supplementary gure S3C and D). Furthermore, in the age-matched comparisons, the microbial alpha diversity in ESCC patients was dramatic reduced (supplementary gure S3E, p = 0.001). Intriguingly, in the cancer progress comparisons, we found that patients with ESCC had signi cantly decreased microbial diversity than patients with PN and PreT (pre-cancer) (supplementary gure S4C). The microbiota composition of patients with T groups was signi cantly different from that of PN and PreT groups (ANOSIM R = 0.53029, P = 0.001; and ANOSIM R = 0.4792, P = 0.001, for unweighted and weighted distances, respectively; supplementary gure S4F and G). The relative abundance of Fusobacterium spp. gradually increased from PN esophagus to ESCC, while the abundance of Proteobacteria was decreased (Supplementary gure S4E). However, there are no statistically signi cant differences in the microbiota pro les of ESCC patients with gender and tumor stage (supplementary gure S4A, B, D, H and I).
Altogether, these results showed that there are signi cant decreased in microbial diversity and composition in ESCC.
The age-matched comparisons of the bacteria taxa in patients with ESCC and physiological normal esophagus was performed by LEfSe analysis (Supplementary gure S3G). To determine the relationships between disease-associated taxa and the abundance of Fusobacteria, we subtracted the Fusobacteria reads and re-analyzed the library from the dataset by LEfSe analysis. In support of the above, Streptococcus, and Prevotella in ESCC were enriched (supplementary gure S5). To con rm ESCCenriched and depleted taxa, we used 16 s rRNA-seq data from a discovery cohort of ESCC. In this dataset, we found signi cant decreases in the abundance of Geobacillus, Anoxybacillus, Thermus, Lactocccus, Pseudomonas, Klebsiella, Blastomonas, and Acinetobacter in ESCC compared to physiological normal esophagus (Fig. 3D). To illustrate that our results were not biased by microbiota pro ling pipeline, we used a second validation cohort to con rm. In agreement with the results obtained in discovery cohort, the enrichments of Selenomonas, Peptostreptococcus, Fusobacterium spp., and Acinetobacter were con rmed in 19 genera as identi ed by the LEfSe analysis (Fig. 3E).

Escc Demonstrates Microbial Dysbiosis
We combined the 19 relevant taxa which characterized in patients with ESCC and PN esophagus and estimated the microbial dysbiosis index (MDI). The esophageal microbiota of patients with ESCC had a higher MDI than that patients with PN esophagus both in discovery cohort and validation cohort (Fig. 4A). Moreover, similar results were also observed in age-matched subset of discovery cohort (Supplementary gure S3F). In Fig. 4B and C, we found that MDI had a signi cant inverse correlation with the alpha diversity (r=-0.971, P < 0.0001) and a positive correlation with the beta diversity (r = 0.3956; P < 0.001), indicating that esophageal microbiota has a high degree of dysbiosis, consistent with reduced bacterial diversity. We further assessed if MDI could be use employed to distinguish between ESCC and physiological normal esophagus. By ROC analysis, the MDI had a good performance in identifying ESCC in discovery cohort (AUC = 95.96%, Fig. 4D) and validation cohort (AUC = 93.75%, Fig. 4E). Additionally, The MDI displayed enhanced sensitivity and speci city to monitor ESCC by using single taxa (supplementary gure S6). In validation cohorts (as shown in Fig. 3E), we con rmed the differential abundance of Fusobacterium sp., Peptostreptococcus sp., Selenomonas sp. and Acinetobacter sp. We next re-calculated the MDI with these 4 genera. The data showed that microbiota was imbalanced in the discovery cohort, validation cohort and AUCs in the ROC analysis (supplementary gure S7).  Fig. 5A and B). Collectively, these data indicated that the change of nitrate and nitrite reductase functions of ESCC microbiota is present in ESCC.

Discussion
In this study, we pro led the microbial alterations in the tumor and PN tissues of ESCC patients using analyses based on 16S rRNA gene sequencing. We have shown that the microbiota composition in tumor tissues of ESCC patients is signi cantly different from that of patients with PN tissues. The microbial dysbiosis of ESCC tumor tissues was characterized by decreased microbial diversity, and increased Bacteroidetes, Fusobacteria and Spirochactes. In addition, we also found that the functional features of ESCC microbiota demonstrate reduced nitrate reductase and nitrite reductase functions. Together, our results explored the microbiota spectrum of ESCC patients and revealed the microbial signature associated with ESCC for diagnosis of patients.
The microbiota is less characterized in ESCC than in esophageal adenocarcinoma [20,32]. Only a few studies have the composition of microbiota on ESCC. Here we focused on the pathological microbiota characterization in ESCC.  [33] and that high burden of F. nucleatum, which inhabits the oral cavity and causes periodontal disease, in ESCC correlates with poor RFS [34]. Also, Porphyromonas gingivalis, a Gram-negative bacterial species involved in periodontal diseases, is a biomarker of ESCC [35]. In our studies Fusobacterium nucleatum, also called by a recently coined name oncobacterium because of its association with cancer [36], is again abundant in ESCC tissues of our cohorts. Here our data further characterize top 10 bacteria strains that may in uence or directly participate in carcinogenesis and progression of ESCC. Treponema amylovorum is a subgingival plaque bacteria species involved in chronic periodontitis [37]. Streptococcus infantis is a part of the salivary microbiome [38]and a prevalent bacteria in breast cancer [39]. Aggregatibacter segnis is an oral bacterial species in oral cancer [40]. Porphyromonas endodontalis, is a part of the salivary microbiome, and its lipopolysaccharide (LPS), which can trigger NFκB signaling [41], is enriched in infected root canals and apical periodontitis [42]. Veillonella dispar, a bacterium in oral mucosa, is involved in autoimmune hepatitis [43,44]. Streptococcus anginosus is enriched in gastric cancer [45] and dental implant-related osteomyelitis [46]. Prevotella intermedia is a biomarker for periodontal disease [47]. Prevotella melaninogenica is a causative agent of periodontitis [48]. Prevotella nigrescens, which elicits TLR2 signaling and p65-Mediated In ammation [49], is an oral bacterial species causing for respiratory tract infections. Further, Prevotella intermedia, Prevotella melaninogenica, and Prevotella nanceiensis are 3 periodontopathic bacteria species causing oral malodor and oral health issues [50]. They have cysteine and serine proteinases activity that may regulate tumorigenesis [50]. Lastly, Prevotella tannerae, a periodontopathogenic bacteria, is associated with an increased oral squamous cell carcinoma risk [51]. The resident microbial ecosystem in the esophagus is affected by both oral and gastric bacteria, but our data show that microbiota of ESCC seems to be dominated by oral bacteria. As mentioned above, top enrichments in microbial composition of ESCC are mainly from oral bacteria strains, we conclude that oral microbial dysbiosis leads to the occurrence and development of ESCC. These data suggest that poor oral health is linked to increased risk of developing ESCC. We speculate that these speci c bacterial strains in oral samples might be utilized as screening tools to assess the risk for ESCC to identify high risk individuals for more invasive screening procedures (e.g., endoscopy). It is possible that the altered abundances of these bacterial strains have a causative role in ESCC development. Their detailed mechanisms in promoting ESCC require further investigation.
We have depicted the diversity of the gut microbiota in ESCC, but the role of most bacterial species in ESCC remains largely unknown. The complexity of the ESCC microbiota, with a plethora of uncharacterized host-microbe, microbe-microbe, and environmental interactions, contributes to the challenge of advancing our knowledge of the ESCC microbiota-cancer interaction. We hypothesized that the microbes or microbiota contribute to the oncogenic evolution of ESCC. To address the microbiotacancer interaction, we have investigated KEGG pathways that were enriched in ESCC-associated bacteria.
One of the top-ranked KEGG pathway in ESCC microbiota is nitrate reductase function. The nitrate reductase might transform nitrate to nitrite, a precursor of nitrosamines, which are carcinogens associated with ESCC. Nitrate is critical inorganic nitrogen sources for microbes, and many bacteria express assimilatory nitrate reductase (NAS) to catalyze the rate-limiting reduction of nitrate to nitrite. For examples, it is interesting to note that fast-growing environmental mycobacteria carry nasN, while slowgrowing pathogenic mycobacteria are lacking [52]. Also, it has been shown that nitrite-oxidizing phylum Nitrospirae is reduced in gastric cancer [53], causing decreased nitrate/nitrite reductase functions. These data suggest that nitrate reduction plays role in pathogenic/neoplastic progression. And our observation that reduction of nitrate reductase functions in the microbiota of ESCC may impose pathogenic effects during ESCC progression and development. Given that ESCC-associated taxa can impact the nitrate regulation, it is possible that targeting the microbiota involved in nitrate regulation may be effective and bene cial to ESCC patients.
A recent comprehensive investigation of microbiomes across seven cancer types (not yet including ESCC) indicates that intracellular bacteria are widespread in tumors [39]. Particularly, 19 prevalent bacteria are characterized [39], including genera of streptococcus and fusobacterium, which are found in our ESCC microbiota. It is not clear whether the enriched bacterial species or genera identi ed in our study can reside in ESCC to facilitate the microenvironment to boost cancer growth. These species may impose immune in ammatory and metabolic burden, and further studies are warranted. ESCC is one of the most aggressive cancers and is therapeutic resistant including radioresistance [54]. The dysbiosis of esophageal microbiota could be the culprit of these impacts. For example, higher F. nucleatum burden leads to poor response to neoadjuvant chemotherapy [34], suggesting the possibility that a strategic intervention against this bacterium may signi cantly improve therapeutic response in patients with ESCC. Our detection of these 10 bacterial strains in ESCC may affect on the e cacy of radiotherapy, chemotherapy, or immune checkpoint inhibitor therapy. Indeed, microbiota composition can in uence the treatment e cacy of immune checkpoint inhibitor in melanoma [55,56]. As immune checkpoint inhibitors (e.g., nivolumab, or pembrolizumab) are being used to treat ESCC [57], unraveling ESCC microbiota-drug interactions and e cacy warrants further investigation.

Conclusion
In summary, our studies suggest that ESCC-speci c microbial taxa may serve as sensitive and speci c clinical diagnostic markers. It is possible that targeting these bacterial strains may be effective and bene cial to ESCC patients. Correct microbial assessment will aid in the detection and treatment of ESCC in the future.