Could gut microbiota serve as prognostic biomarker associated with colorectal cancer patients' survival? A pilot study on relevant mechanism

Evidences have shown that dysbiosis could promote the progression of colorectal cancer (CRC). However, the association of dysbiosis and prognosis of CRC is barely investigated. Therefore, we used 16S rRNA gene sequencing approach to determine differences in microbiota among tumor tissues of different prognosis and found that Fusobacterium nucleatum and Bacteroides fragilis were more abundant in worse prognosis groups, while Faecalibacterium prausnitzii displayed higher abundance in survival group. To further explore the prognostic value of the found bacteria, Kaplan–Meier and Cox proportional regression analyses were used and the results exhibited that high abundance of F. nucleatum and B. fragilis were independent indicators of poor patient's survival. Besides, the expression of major inflammatory mediator were analyzed using PCR and western blot methods, and it turned out that high abundance of F. nucleatum was associated with increased expression of TNF-α, β-catenin and NF-κB, while COX-2, MMP-9 and NF-κB were positively related with high B. fragilis level, and high level of F. prausnitzii showed lower expression of β-catenin, MMP-9 and NF-κB. Moreover, immunohistochemical analysis indicated that KRAS and BRAF expression were prominent in F. nucleatum and B. fragilis high abundance group, while MLH1 showed lower expression. In conclusion, F. nucleatum, B. fragilis and F. prausnitzii can be identified as useful prognostic biomarkers for CRC, and dysbiosis might worsen the patients' prognosis by up-regulating gut inflammation level.


INTRODUCTION
Colorectal cancer (CRC) is a common life threatening disease worldwide [1], with 700,000 annual mortalities making it the fourth most deadly cancer in both men and women [2]. In recent years, the 16S rRNA gene sequencing approach has been widely used as an effective tool to globally analyze the microbial community [3,4], and multiple studies have demonstrated that breakdown of the intestinal microbiota structure can promote carcinogenesis and development of CRC [5][6][7][8][9]. Comparative data about microflora in relation to survival of patients with colorectal cancer are scanty, but may be of clinical significance. Flanagan et al. demonstrated a significant association between Fusobacterium nucleatum level and patient outcome and suggested that F. nucleatum may have value as a prognostic indicator [5]. Boleij et al. found that the detection of Bacteroides fragilis toxin (BFT), which was produced by Enterotoxigenic Bacteroides fragilis (ETBF), increased in the mucosa of later staged CRC [10]. These studies show there is a possibility that some type of microbe will affect the prognosis of patients with CRC. Given that infection has gradually been accepted as a major driver of inflammation,

Research Paper
and various inflammatory mediators substantially contribute to metastasis [11,12]. We hypothesize that inflammation might be the key point between microbiota and prognosis of CRC.
In this study, we examined the microbial structure in CRC clinical tumor samples and assessed the correlation of microbiota with clinicopathologic features and with patient survival. The status of MLH1, BRAF and KRAS expression, as well as inflammation related TNF-α, COX-2, MMP-9, β-catenin and NF-κB of cancer tissues were assessed to calculate their correlation with different microbial phylotypes. This approach may reveal the pathological process of how microbiota could affect the prognosis of CRC patients.

Diversity and structural changes of the tumor microbiota in CRC patients with different prognosis outcome
Libraries of 16S rRNA V4 region amplicon sequences from 180 CRC tumor samples were sequenced. A total of 16,854,578 high-quality and classifiable reads were obtained from this study, with an average of 93,636 reads per sample. At 3% dissimilarity level, a total of 41,628 OTUs in all samples and an average of 231 OTUs per sample were identified.
The value of Good's coverage for each group was over 99%. We examined the estimators of community richness (observed species and Chao indexes) and diversity and evenness (Shannon and Simpson indexes) among groups ( Figure 1A). The only significant difference was detected between the survival group and recurrent group in Chao diversity index (Chao, 257 ± 88 vs. 397 ± 89, P= 0.03), demonstrating the significantly lower diversity found in survival group.
For beta diversity analysis, the microflora and compositions were analyzed and compared through the relative abundance of OTUs by using Bray-Curtis distance matrix and weighted Unifrac distance matrix for each group. Subsequent results of principal coordinates analysis (PCoA) exhibited the difference in bacterial community composition among groups. The first three principal component scores of Bray-Curtis distance matrix ( Figure 1B) and weighted Unifrac distance matrix ( Figure 1C) were 22%, 10%, 8% and 42%, 14%, 4%. Significant difference was detected in Bray-Curtis distance (P=0.011), suggesting that the community membership of each group was different.

Correlation of microbiota in CRC patients with clinicopathologic features
The significant difference of microbiota between survival group and non-survival group showed that B. fragilis and F. prausnitzii might be correlated with patient's survival in CRC. Besides, F. nucleatum, a well-studied detrimental bacteria which could promote CRC development and progression, also showed higher abundance in non-survival group. Therefore, we evaluated the relationship between the level of the three bacteria and clinicopathologic characteristics of CRC patients. Based on the relative abundance of each microbiota in tumor sample, patients were divided into high and low bacteria subgroups with the median relative abundance as the cutoff, that was B. fragilis high vs B. fragilis low (Cut-off: 2.04%), F. prausnitzii high vs F. prausnitzii low (Cutoff: 0.55%), and F. nucleatum high vs .F. nucleatum low (Cut-off: 0.52%). As shown in Table 1, high abundance of F. nucleatum was significantly correlated with positive lymph node metastasis (P =0.011). Furthermore, high abundance of F. prausnitzii and F. nucleatum was significantly correlated with worse depth of invasion (P =0.015 and 0.015).

Prognostic value of B. fragilis, F. prausnitzii and F. nucleatum
To assess the clinical significance of the three bacteria in CRC, Kaplan-Meier analysis and the logrank test were used to analyze the relationship between bacteria relative abundance in cancer tissue and patient' survival. We found that the 3-year OS was significantly lower in patients with high B. fragilis and F. nucleatum than in those with low abundance of these two microbiota (P= 0.001, P= 0.003). And patients with low abundance of F. prausnitzii showed worse 3-year OS, although the difference was not significant (P= 0.06). Similarly, patients with high B. fragilis and F. nucleatum were  significantly associated with poorer disease-free survival (DFS) rates than those with low abundance (P< 0.001, P= 0.001) ( Figure 2

Expression of inflammation related molecules in CRC tissues with different bacteria abundance
Quantitative RT-PCR was used to determine the expression of TNF, COX2, MMP9 and CTNNB (catenin beta) in tumor samples. As shown in Figure 3A, TNF was overexpressed in CRC tissues of the F. nuleatum high abundance group (P=0.0024) and F. prausnitzii low abundance group (P=0.0117) compared to their opposite groups. Higher expression of COX2 was found in B. fragilis high abundance group (P=0.0245) than in the low abundance one. And the expressions of MMP9 and CTNNB were positively correlated with high abundance of F. nuleatum (P=0.0005 and P=0.0189) and B. fragilis (P=0.0432 and P=0.0300), but negatively correlated with high abundance of F. prausnitzii (P=0.0147 and P=0.0197). Consistently, western blotting analysis further confirmed the results obtained from qRT-PCR ( Figure 3B). Moreover, the expression of NF-κB, which was detected by western blot, increased in F. nuleatum high abundance and B. fragilis high abundance groups, while decreased in F. prausnitzii high abundance group.

Bacteria levels related to other molecular features of cancer tissue
Immunohistochemistry was used to detect the differential expression of KRAS, BRAF and MLH1 in cancer tissues ( Figure 4). By comparing immunohistochemical scores of tumor samples with high or low bacteria abundance, we observed that KRAS was highly expressed in F.nucleatm high abundance group, B. fragilis high abundance group and F. prausnitzii low abundance group. The F. nucleatm high abundance group and B. fragilis high abundance group also exhibited a higher expression of BRAF and lower expression of MLH1 (Table 4).

DISCUSSION
Nowadays, cancer stage is the most important indicator for the prognosis of CRC patients, and multiple strategies were well designed based on different TNM stage. However, new tool to indicate more accurate clinical characteristics and effective therapeutic targets are still needed, so it is meaningful to find new predictor of CRC patients' prognosis which might lead to novel treatment method to improve patients' survival. In our study, we are the first to compare the microbial population among groups of cancer tissues divided by different postoperation prognosis status. Differences of F. nucleatum, B. fragilis and F. prausnitzii, found between non-survival and survival group, drew our special attention, because these three bacteria have relative high abundant, all of which counted for more than 1% of microbiota in species, and they are intimately related with colorectal cancer [5-8, 13, 14]. In addition, we found a correlation between high abundance of F. nucleatum and increased lymph nodes metastasis rates, which was consistent with previous study [15]. Tumor's depth of invasion was shown to be correlated with high F. nucleatum and low F. prausnitzii abundance. Further survival analysis confirmed the prognostic value of F. nucleatum, Fragilis and F. prausnitzii.
Another study has shown that patients with high levels of F. nucleatum had a significantly shorter survival time, which was similarly with our result [5]. But the number of patients enrolled in the cohort of that study was relatively small (32 patients). A more recent study reported a similar result using a larger databases of CRC cases in USA, which observed a correlation between high amount of tissue F. nucleatum DNA and higher CRC-specific mortality [16]. This result was promising, but the positive rate of bacteria DNA was relatively small which made the positive group less representative. ETBF, a major subtype of B. fragilis, was also associated with CRC through producing BFT [17]. A recent study suggested that BFT gene positivity was more prominent in later stage CRC, which showed a possible link between increased B. fragilis and worse prognosis of CRC [10]. In our study, we report, for the first time, that the high abundant of B. fragilis is correlated with poor patient's clinical outcome. Furthermore, increased F.prausnitzii, a well-known human intestinal probiotic bacteria, was shown to be related with better survival status, a result that has not been reported before.
The specific mechanisms by which gut microbiota affects the development of CRC are still not well understood. One of the most promising theories is that it is thought to be through microbe-driven intestinal inflammation [18]. Interestingly, F. nucleatum, B. fragilis and F. prausnitzii are all key players in modifying intestinal inflammation levels [6,7,19]. In a study about colorectal adenomas, the abundance of F. nucleatum was found to positively correlate with inflammatory cytokine genes expression including that of TNF, which was consistent with our result [20]. TNF-α is produced during the inflammatory response and can promote survival, attachment, and proliferation of metastatic colon cancer cells in a mouse model of lung metastasis depending on the activation of NF-κB by inflammation and cancer cells [21]. Moreover, through activation of NF-κB and STAT3, TNF-α can enhance epithelial-mesenchymal transition which are critical steps that allow polarized epithelial tumor cells to become mesenchymal like, enhancing cell migration and invasion [22,23]. Meanwhile, MMP-9 was also elevated in F. nucleatum high abundant group. MMP-9, known as an independent stimulus for increased cell migration [24], can be activated by inflammatory signals, such as NF-κB [25]. And, as previously mentioned, NF-κB can be activated by TNF-α in certain tumor microenvironment [21]. Furthermore, it has been demonstrated that F. nucleatum can bind to E-cadherin on epithelial cells via FadA and activates β-catenin which is a transcription factor in the Wnt signal transduction pathway [6], Wnt signaling is fundamental in CRC progression [26]. In addition, studies suggest that F. nucleatum can also generate a proinflammatory microenvironment by recruit tumor-infiltrating immune cell [27] and downregulate antitumor T cell-mediated adaptive immunity [28] to promote CRC progression.
B. fragilis, was also related with higher CTNNB expression in our study. BFT can alter epithelia structure  and function including cleavage of the tumor suppressor protein, E-cadherin, resulting in enhanced nuclear Wnt/βcatenin signaling that yields increased colonic carcinoma cell proliferation and metastasis [29][30][31]. PGE2, a major downstream mediator of COX-2, can activates β-catenindependent signaling, which promotes survival and proliferation. Besides increased COX-2 can also stimulate tumor angiogenesis by inducing production of VEGF and basic fibroblast growth factor, and it can increase tumor dissemination by altering the adhesive properties of cells and increasing matrix metalloproteinase activity [32,33]. Available evidence demonstrated that BFT, through activation of NF-κB, could stimulate intestinal epithelial cells to induce the expression of COX-2 and increased release of PGE2 [13]. In the meantime, MMP-9 is also elevated in the B. fragilis high abundance group which might also be explained by activation of NF-κB that was induced by BFT.
Several studies have shown that culture supernatant of F. prausnitzii exerts an anti-inflammatory effect both in vitro and in vivo [34]. A recent experiment demonstrated that a 15 kDa protein produced by F. prausnitzii possessed anti-inflammation properties through inhibition of NF-κB pathway in intestinal epithelial cells in an animal model [35], which is consistent with our finding. Notably, the down regulated NF-κB could decrease the expression of various inflammation related factors including βcatenin and MMP-9 [25,36], both of which showed lower expression level in F. prausnitzii high group in our experiment and could promote CRC metastasis. Furthermore, TNF was also overexpressed in F. prausnitzii low group, which further supports the idea that F. prausnitzii possess an anti-inflammation effect, although the potential mechanism is not thoroughly understood.
Evidences have shown that overexpression of KRAS and BRAF were markers of poor prognosis [37,38] which were fortunately consistent with our findings of prognostic values of F. nucleatum, B. fragilis and F.prausnitzii. In addition, MSI, the primary causes of which is hypermethylation of the MLH1 promoter, was also associated with clinical outcomes of CRC [39,40], and in this study, decreased expression of Mlh-1 had also been detected in F. nucleatum high and B. fragilis high abundant samples. These findings might imply that cancers which are accompanied with high abundance of F. nucleatum and B. fragilis or low abundance of F. prausnitzii are more invasive and inclined to metastasis. Furthermore, serrated adenocarcinoma, a subtype of CRC which is developed though serrated pathway, is characterized by high frequency of KRAS and BRAF mutation and MLH1 deficiency [41,42]. Evidence showed that this kind of CRC is likely to have less favorable 5-year survival [43]. Therefore, it came a question that whether there's an association between bacteria F. nucleatum or B. fragilis and serrated adenocarcinoma, which might be worth exploring in future studies.
Moreover, as obvious benefits of anti-epidermal growth factor receptor therapy were shown in patients with KRAS mutations [44,45] and sensitivity to irinotecan was found in the MLH1 deficiency cell model [46,47], our finding poses the question as to whether patients that suffered from dysbiosis should receive more aggressive chemotherapy? This might be a promising direction of future microbiotic study. The study has some limitations. First, our findings based on patients in a single center, and the prognostic value of these bacteria might require a multi-center study with larger data to validate this result. Second, the specific mechanism by which bacteria promote invasion and metastasis of CRC and unknown mutual effect that exist among different bacteria will surly need to be explored by well-designed animal and cell line experiments in the future.
In conclusion, our study is the first report demonstrating the prognosis value of B. fragilis and F.prausnitzii, and further validates the connection of F. nucleatum and cancer-specific mortality. Furthermore, the correlation between bacteria's relative abundance and inflammatory factors suggests that microbiota might impact patient's prognosis via inducing gut inflammation. Moreover, given that the tumor samples could be easily collected both in surgery and colonoscopy, our results will be useful in developing novel bacteria-related prognostic indicator for CRC, and encourage further investigation of the role played by microbiota in CRC pathology.

CRC patients
A total of 180 CRC patients were enrolled in our study. Patients with stages I-III cancer subjected to standard curative surgery, while stage IV CRC tissues were collected from patients who received palliative surgery to relieve serious cancer related contradiction. Surgeries were performed at the general surgery department of Affiliated Hospital of Qingdao University between 2010 and 2012, and all patients who received postoperative treatment were guided by the National Comprehensive Cancer Network Guidelines. Of all the patients, there were 108 males and 72 females with a mean age of 62.2 (age range 30-88 years). The median follow-up period was 47 months with a range from 36 to 59 months. The criteria for study enrollment were histopathological diagnosis of primary CRC, newly diagnosed and untreated, no history of other tumors, and the potential to follow up. Patients who used antibiotics within 2 months before operation, or were regularly using Non-steroidal anti-inflammatory drugs, statins or probiotics were excluded from the study. Other exclusions included chronic bowel disease, other signs of infections, food allergies and dietary restrictions.

Sample preparation
CRC tumor samples and adjacent normal tissue samples (at least 5cm from the tumor site) of these 180 patients were obtained from the gastrointestinal cancer specimen bank of Affiliated Hospital of Qingdao University, Qingdao, China. To be specific, surgically resected specimens were collected immediately after tumor removal and stored at -80°C until use. The TNM staging were determined according to the American Joint Committee on Cancer system and all specimens were graded histologically according to the World Health Organization classification criteria. Written informed consents of joining the specimen bank were obtained from all the patients before surgery, and the protocols used in the study were approved by the Ethics Committee of Affiliated Hospital of Qingdao University. Clinical and pathologic data were reviewed from gastrointestinal cancer database of Affiliated Hospital of Qingdao University, Qingdao, China.
So as to preliminarily detect the species of bacteria with prognostic value, subjects included in the study were subdivided based on their different survival conditions. From those, 92 patients corresponded to the survival group (people who lived more than 3 years without any sign of recurrence or metastasis), 28 to the non-survival group (people who died within 3years after surgery for CRC related causes), 31 to the recurrence group (people who experienced recurrence or metastasis of primary tumor within 3 years but survived), and 29 to the unclear group (people who lived more than 3 years with unclear history of recurrence or metastasis).

DNA extraction
DNA was extracted from all tumor samples using CTAB method with minimal modification. Concentration of DNA was measured by fluorometer or microplate reader, and sample integrity was tested by agarose gel electrophoresis (1% concentration of agarose Gel: 1 %; 150 V; 40 min electrophoresis time). All DNA samples were stored at -20°C until used.

PCR and sequencing analysis
Amplification of the V4 region of the bacterial 16S rRNA gene was performed by polymerase chain reaction (PCR) using universal primers 319F and 806R. The reaction mix consisted of Phusion High-Fidelity PCR Master Mix (NEB, Ipswich, MA, USA) and appropriate primer/probe pairs. The PCR program was as follows: 3 min denaturation at 98°C followed by 30 cycles of 45 s at 95°C (denaturation), 45 s for annealing at 55°C and 45 s at 72°C (extension), with a final extension at 72°C for 7 min. The PCR products were purified with AMPure XP beads (Agencourt Bioscience) to remove the unspecific products prior to library construction. The library was quantitated in two ways: the average molecule length was determined using the Agilent 2100 bioanalyzer instrument (Agilent DNA 1000 Reagents), and then quantified by real-time quantitative PCR (qPCR; EvaGreen TM). Sequencing of qualified libraries was performed by the BGI-Huada Genomices institute in Shenzhen using MiSeq System, with the sequencing strategy PE250 (PE251+8+8+251) or PE300 (PE301+8+8+301) (MiSeq Reagent Kit).

Western blot analysis
Western blotting was performed to detect the differences of TNF-α, COX-2, MMP-9, β-catenin and NF-κB in above-mentioned paired tissues. Total proteins were prepared from frozen tissue by Cellytic M cell lysis Reagent (Sigma-Aldrich Inc., St. Louis, MO, US). After being centrifuged at 12,000g for 20 min, the supernatants were loaded onto 10% SDS-PAGE gels, electrophoresed, and transferred to PVDF membrane (Millipore). Then, Membranes were incubated overnight at 4°C with TNF-α, COX-2, MMP-9, β-catenin and NF-κB antibodies (Abcam) respectively, β-actin was served as an internal loading control. After washes, membranes were incubated with appropriate secondary antibodies for 1 hr at room temperature, and bands were scanned using a ChemiDoc™ Touch Imaging System (Bio-Rad Laboratories, UK).

Statistical analysis
Metastats (http://metastats.cbcb.umd.edu/) and R (v3.0.3) are used to determine which taxonomic groups were significantly different between groups of samples. We adjusted the obtained P-value by a Benjamini-Hochberg false discovery rate (FDR) correction (function 'P.adjust' in the stats package of R (v3.0.3)) [52]. Continuous data are presented as mean±standard deviation, unless otherwise stated. The P-values for Bray-Curtis distance and Weighted-Unifrac distance were calculated by ANOSIM analysis. Differences between groups were analyzed using Student t test (two-tailed). The association between clinicopathological variables and differences in microbiota were examined by χ 2 tests. The categorical data were analyzed by a Fisher's exact test. Overall survival (OS) curves were analyzed using the Kaplan-Meier method, and differences were examined using log-