WO2016112488A1

WO2016112488A1 - Biomarkers for colorectal cancer related diseases

Info

Publication number: WO2016112488A1
Application number: PCT/CN2015/070584
Authority: WO
Inventors: Qiang FENG; Dongya ZHANG; Jun Wang
Original assignee: BGI Shenzhen Co Ltd
Current assignee: BGI Shenzhen Co Ltd
Priority date: 2015-01-13
Filing date: 2015-01-13
Publication date: 2016-07-21
Anticipated expiration: 2017-07-13
Also published as: EP3245298A4; EP3245298B1; EP3245298A1; CN107208141B; CN107208141A

Abstract

The present invention provides biomarkers and methods for predicting the risk of a disease related to microbiota, in particular colorectal cancer (CRC).

Description

BIOMARKERS FOR COLORECTAL CANCER RELATED DISEASES

CROSS-REFERENCE TO RELATED APPLICATION

None

FIELD

The present invention relates to biomarkers and methods for predicting the risk of a disease related to microbiota, in particular colorectal cancer (CRC) related diseases.

BACKGROUND

Colorectal cancer (CRC) is the third most common form of cancer and the second leading cause of cancer-related death in the Western world (Schetter AJ, Harris CRC (2011) Alterations of microRNAs contribute to colon carcinogenesis. Semin Oncol 38: 734-742, incorporated herein by reference) . A lot of people are diagnosed with CRC and many patients die of this disease each year worldwide. Although current strategies, including surgery, radiotherapy, and chemotherapy, have a significant clinical value for CRC, the relapses and metastases of cancers after surgery have hampered the success of those treatment modalities. Early diagnosis of CRC will help to not only prevent mortality, but also reduce the costs for surgical intervention.

Current tests of CRC, such as flexible sigmoidoscopy and colonoscopy, are invasive and patients may find the procedures and bowel preparation to be uncomfortable or unpleasant.

The development of CRC is a multifactorial process influenced by genetic, physiological, and environmental factors. Regarding environmental factors, the lifestyle, particularly dietary intake, may affect the risk of CRC developing. Western diet, rich in animal fat and poor in fiber, is generally associated with an increased risk of CRC. Thus, it has been hypothesized that the connection between the diet and CRC, may be the influence that the diet has on the colon microbiota and bacterial metabolism, making both relevant factors in the etiology of the disease (McGarr SE, Ridlon JM, Hylemon PB (2005) . Diet, anaerobic bacterial metabolism, and colon cancer. J Clin Gastroenterol. 39:98-109； Hatakka K, Holma R, El-Nezami H, Suomalainen T, Kuisma M, Saxelin M, Poussa T,

H, Korpela R (2008) . The influence of Lactobacillus rhamnosus LC705 together with Propionibacterium freudenreichii ssp. shermanii JS on potentially carcinogenic bacterial activity in human colon. Int J Food Microbiol. 128: 406-410, incorporated herein by reference) .

SUMMARY

Embodiments of the present disclosure seek to solve at least one of the problems existing in the prior art to at least some extent.

The present invention is based on the following findings by the inventors:

Intestinal microbiota analysis of feces DNA has the potential to be used as a noninvasive test for finding specific biomarkers that may be used as a screening tool for early diagnosis of patients having CRC, thus leading to a longer survival and a better quality of life. To carry out analysis on gut microbial content in CRC patients, the inventors carried out a protocol for a Metagenome-Wide Association Study (MGWAS) (Qin, J. et al. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature 490, 55-60 (2012) , incorporated herein by reference) based on deep shotgun sequencing of the gut microbial DNA from 128 Chinese individuals (cohort I) . The inventors identified and validated 140, 455 CRC-associated gene markers. To exploit the potential ability of CRC classification by gut microbiota, the inventors developed a disease classifier system based on the 20 gene markers that are defined as an optimal gene set by a minimum redundancy -maximum relevance (mRMR) feature selection method. For intuitive evaluation of the risk of CRC disease based on these 20 gut microbial gene markers, the inventors calculated a healthy index (CRC index) . The inventors'data provide insight into the characteristics of the gut metagenome related to CRC risk, a paradigm for future studies of the pathophysiological role of the gut metagenome in other relevant disorders, and the potential usefulness for a gut-microbiota-based approach for assessment of individuals at risk of such disorders.

It is believed that gene markers of intestinal microbiota are valuable for increasing cancer detection at earlier stages due to the following. First, the markers of the present invention are more specific and sensitive as compared with conventional cancer markers. second， analysis of stool promises accuracy, safety, affordability, and patient compliance. And samples of stool are transportable. As compared with colonoscopy requiring bowel preparation, polymerase chain reaction (PCR) -based assays are comfortable and noninvasive, so people will participate in a given screening program more easily. Third, the markers of the present invention may also serve as tools for therapy monitoring in cancer patients to detect the response to therapy.

BRIEF DISCRIPTION OF DRAWINGS

These and other aspects and advantages of the present disclosure will become apparent and more readily appreciated from the following descriptions taken in conjunction with the drawings, in which:

Fig. 1 shows distribution of P-value association statistics of all microbial genes in the study. The association analysis of CRC p-value distribution identified a disproportionate over-representation of strongly associated markers at lower P-values, with the majority of genes following the expected P-value distribution under the null hypothesis. This suggests that the significant markers likely represent true rather than spurious associations.

Fig. 2 shows species involved in gut microbial dysbiosis during colorectal cancer. Differential relative abundance of two CRC-associated and one control-associated microbial species consistently identified using three different methods: MLG, mOTU and IMG database.

Fig. 3 shows enrichment of Solobacterium moorei and Peptostreptococcus stomatis in CRC patient microbiomes.

Fig. 4 shows the Receive-Operator-Curve of CRC specific species marker selection using random forest method and three different species annotation methods. A, IMG species annotation using cleanreads to IMG version 400. B, mOTU species using published methods (E. M.

E. M.

M. C. C. Gomes-Marcondes, Leucine modulates the effect of Walker factor, a proteolysis-inducing factor-like protein from Walker tumours, on gene expression and cellular activity in C2C12 myotubes. Cytokine 64, 343 (10//, 2013) , incorporated herein by reference) , C, All significant genes clustered using MLG methods (M. R. Rubinstein et al. , Fusobacterium nucleatum promotes colorectal carcinogenesis by modulating E-cadherin/beta-catenin signaling via its FadA adhesin. Cell Host Microbe 14, 195 (Aug 14, 2013) , incorporated herein by reference) and the species annotation using IMG version 400.

Fig. 5 shows stage specific abundance of three species that are enriched in stage II and later, using three species annotation methods: MLG, IMG and mOTU.

Fig. 6 shows species involved in gut microbial dysbiosis during colorectal cancer. Relative abundances of three enriched in CRC-associated microbiomes, during different stages of CRC (three different species annotation methods were used) .

Fig. 7 shows minimum redundancy maximum relevance (mRMR) method to identify 20 gene markers that differentiate colorectal cancer cases from controls. Incremental search was performed using the mRMR method which generated a sequential number of subsets. For each subset, the error rate was estimated by a leave-one-out cross-validation (LOOCV) of a linear discrimination classifier. The optimum subset with the lowest error rate contained 20 gene markers.

Fig. 8 shows principal component analysis (PCA) based on profiles of 20 gene markers separates CRC cases and control individuals. First and second principal components associate with CRC status (PC1 and PC2 explain 31.9％and 13.3％of variance, respectively) . Compare this with the analysis based on 2.1 million genes, where no separation can be observed.

Fig. 9 shows discovering gut microbial gene markers associated with CRC. CRC index calculated for CRC patients (black) and control individuals (gray) from this study, shown along patients and control individuals (gray) from earlier studies on type 2 diabetes and inflammatory bowel disease. The box depicts the interquartile ranges between the first and third quartiles, and the line inside denotes the median. CRC indices for CRC patient microbiomes are significantly different from the rest.

Fig. 10 shows ROC analysis of CRC index from 20 gene markers in Chinese cohort I, which shows excellent classification potential with an area under the curve of 0.99.

Fig. 11 shows CRC index using 20 gene markers in 128 samples.

Fig. 12 shows CRC index , which classifies with an area under the receiver operating characteristic (ROC) curve of 0.97.

Fig. 13 shows correlation between quantification by the metagenomic approach versus quantitative polymerase chain reaction (qPCR) for four gene markers.

Fig. 14-1 shows that ROC analysis reveals moderate potential for classification using CRC index, with an area under the curve of 0.71.

Fig. 14-2 shows CRC index, which classifies with an area under the receiver operating characteristic (ROC) curve of 0.85.

Fig. 15 shows validating robust gene markers associated with CRC. Quantitative PCR abundance (in log10 scale, zero abundance plotted as -8) of two gene markers (m1704941: butyryl-CoA dehydrogenase from F. nucleatum, m1696299: RNA polymerase subunit beta, rpoB, from P. micra) were measured in cohort II consisting 47 cases and 109 healthy controls. (a) CRC index based on the two genes clearly separates CRC microbiomes from controls. (b) CRC index classifies with an area under the receiver operating characteristic (ROC) curve of 0.84. (c, d) The two marker genes show relatively higher incidence and abundance starting in CRC stage II and III compared to control and stage I microbiomes.

Fig. 16 shows CRC index (only using 1696299) , which classifies with an area under the receiver operating characteristic (ROC) curve of 0.80.

Fig. 17 shows CRC index (only using 1704941) , which classifies with an area under the receiver operating characteristic (ROC) curve of 0.69.

DETAILED DESCRIPTION

Terms used herein have meanings as commonly understood by a person of ordinary skill in the areas relevant to the present invention. Terms such as “a” ， “an” and “the” are ont intended to refer to only a singular entity, but include the general class of which a specific example may be used for illustration. The terminology herein is used to describe specific embodiments of the invention, but their usage does not delimit the invention, except as outlined in the claims.

In one aspect, the present invention relates to a gene marker set for predicting the risk of colorectal cancer (CRC) in a subject comprising one or more of the genes as set forth in SEQ ID NOs: 1 to 20.

In another aspect, the present invention relates to use of the gene marker set of the present invention for predicting the risk of colorectal cancer (CRC) in a subject, via the steps of:

1) collecting a sample j from the subject and extracting DNA from the sample；

2) determining the abundance information of each of gene marker in the gene marker set； and

3) calculating the index of sample j by the formula below:

A_ij is the relative abundance of marker i in sample j, wherein i refers to each of the gene markers in said gene marker set；

N is a subset of all CRC-enriched markers in the gene marker set；

M is a subset of all control-enriched markers in the gene marker set；

and |N| and |M| are the sizes (number) of the biomarker respectively in these two subsets；

wherein an index greater than a cutoff indicates that the subject has or is at the risk of developing colorectal cancer (CRC) .

In yet another aspect, the present invention relates to use of the gene marker set of the present invention for preparation of a kit for predicting the risk of colorectal cancer (CRC) in a subject, via the steps of:

1) collecting a sample j from the subject and extracting DNA from the sample；

3) calculating the index of sample j by the formula below:

N is a subset of all CRC-enriched markers in the gene marker set；

M is a subset of all control-enriched markers in the gene marker set；

In another aspect, the present invention relates to a method for diagnosing whether a subject has colorectal cancer or is at the risk of developing colorectal cancer, comprising:

1) collecting a feces sample j from the subject and extracting DNA from the sample；

2) determining the abundance information of each of the marker in a gene marker set comprising one or more of the genes as set forth in SEQ ID NOs: 1 to 20； and

3) calculating the index of sample j by the formula below:

N is a subset of all CRC-enriched markers in the gene marker set；

M is a subset of all control-enriched markers in the gene marker set；

In one specific embodiment, the abundance information is gene relative abundance of each of gene marker in the gene marker set which is determined by means of sequencing method.

In another specific embodiment, the abundance information is gene relative abundance of each of gene marker in the gene marker set which is determined by a qPCR method.

In yet another specific embodiment, the cutoff value is obtained by a Receiver Operator Characteristic (ROC) method, wherein the cutoff corresponds to when AUC (Area Under the Curve) reached at its maximum.

In a preferred embodiment, the gene marker set of the present invention consists of SEQ ID NOs: 1 to 20, more preferably, the gene marker set of the present invention consists of SEQ ID NOs: 1, 9, 13 and 16, most preferably, the gene marker set of the present invention consists of SEQ ID NOs: 13 and 16. In another preferred embodiment, the gene marker set consists of SEQ ID NO: 13.

In yet another aspect, the present invention relates to use of a marker as set forth in SEQ ID NO: 13 or rpoB gene encoding RNA polymerase subunit β as a gene marker for predicting the risk of colorectal cancer (CRC) in a subject, wherein the enrichment of said gene marker in a sample of the subject relative to a control sample is indicative of the risk of colorectal cancer in the subject

The present invention is further exemplified in the following non-limiting Examples. Unless otherwise stated, parts and percentages are by weight and degrees are Celsius. As apparent to one of ordinary skill in the art, these Examples, while indicating preferred embodiments of the invention, are given by way of illustration only, and the agents were all commercially available.

Example 1. Identifying 20 biomarker and use gut healthy index to evaluate their colorectal cancer risk

1.1 Sample collection

1.1.1 Sample collection in China

Cohorts I (Table 1, used in Example 1, consisting of 74 colorectal cancer patients and 54 control subjects) and cohort II (Table 13, used in Example 3, consisting of 47 colorectal cancer patients and 109 control subjects) : Stool samples were collected between 2002 and 2012 in the Prince of Wales Hospital, Hong Kong. The inclusion criteria of all the samples were: 1) not taking antibiotics or other medications, with no particular diets (diabetic, vegetarian, etc) and with normal lifestyle (without extra stress) for a minimum 3 months； 2) a minimum of 3 months after any medical intervention； 3) no history of colorectal surgery, any kind of cancer, or inflammatory or infectious diseases of the intestine. Subjects were asked to collect stool samples in standardized containers at home and store in their home freezer immediately. Frozen samples were then delivered to the hospital in insulating polystyrene foam containers and stored at -80℃ immediately until further analysis.

1.1.2 Sample collection in Denmark

Cohort III (Table 15, used in Example 3, consisting of 16 colorectal cancer patients and 24 control subjects) : Stool samples were collected from individuals referred to colonoscopy due to symptoms associated with colorectal cancer or from patients who had been diagnosed with colorectal cancer and referred to large bowel resection for their primary cancer disease. All individuals were included at their visit to the out-patient clinic either before colonoscopy or before the operation and always before bowel evacuation. The individuals received a stool collection set including a tube without stabilizing buffer and were instructed to collect a stool sample at home one or two days before initiation of large bowel evacuation. Every included individual kept the sample refrigerated at -18℃ and contacted a research nurse who collected the sample. At the laboratory stool samples were immediately snap frozen in liquid nitrogen and subsequently stored at -80℃ under 24/7 electronic surveillance until analysis.

All included individuals thus underwent complete colonoscopy either as the primary examination of after the subsequent operation. Exclusion criteria were previous adenoma, previous colorectal cancer and previous or present other malignant diseases.

The collection of stool samples and the recording of data from the included individuals were performed according to the Helsinki II declaration. The protocol was approved by the Ethics Committee of the Capital Region of Denmark (H-3-2009-110) and the Danish Data Protection Agency (2008-41-2252) .

Table 1 Baseline characteristics of colorectal cancer (CRC) cases and controls in cohort I. FBG: fasting blood glucose； ALT/GPT: alanine transaminase/glutamate pyruvated transaminase； BMI: body mass index； DM: diabetes mellitus type 2； HDL: high density lipoprotein； TG: triglyceride； eGFR: epidermal growth factor receptor； TCHO: total cholesterol； Cr: creatinine； LDL； low density lipoprotein； TNM: tumor node metastasis staging system.

1.2 DNA extraction

Chinese samples: Stool samples were thawed on ice and DNA extraction was performed using the Qiagen QIAamp DNA Stool Mini Kit (Qiagen) according to manufacturer’s instructions. Extracts were treated with DNase-free RNase to eliminate RNA contamination. DNA quantity was determined using NanoDrop spectrophotometer, Qubit Fluorometer (with the Quant-iTTMdsDNA BR Assay Kit) and gel electrophoresis.

Danish samples: A frozen aliquot (200 mg) of each fecal sample was suspended in 250 μl of 4 M guanidine thiocyanate- 0.1 M Tris (pH 7.5) and 40 μl of 10％N-lauroyl sarcosine. Then, DNA extraction was conducted using bead beating method as previously described (J. J. Godon, E. Zumstein, P. Dabert, F. Habouzit, R. Moletta, Molecular microbial diversity of an anaerobic digestor as determined by small-subunit rDNA sequence analysis. Applied and environmental microbiology 63, 2802 (Jul, 1997) , incorporated herein by reference) . The DNA concentration and its molecular size were estimated by nanodrop (Thermo Scientific) and agarose gel electrophoresis.

1.3 DNA library construction and sequencing

DNA library construction was performed following the manufacturer’s instruction (Illumina HiSeq 2000 platform) . The inventors used the same workflow as described previously to perform cluster generation, template hybridization, isothermal amplification, linearization, blocking and denaturation, and hybridization of the sequencing primers (Qin, J. et al. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature 490, 55-60 (2012) , incorporated herein by reference) .

The inventors constructed one paired-end (PE) library with insert size of 350bp for each sample, followed by a high-throughput sequencing to obtain around 30 million PE reads of length 2x100bp. High-quality reads were obtained by filtering low-quality reads with ambiguous ‘N’ bases， adapter contamination and human DNA contamination from the Illumina raw reads, and by trimming low-quality terminal bases of reads simultaneously. 751 million metagenomic reads (high quality reads) were generated (5.86 million reads per individual on average)

1.4 Species annotation of IMG genomes

For each IMG genome, using the NCBI taxonomy identifier provided by IMG, the inventors identified the corresponding NCBI taxonomic classification at species and genus levels using NCBI taxonomy dump files. The genomes without corresponding NCBI species names were left with its original IMG names, most of which were unclassified.

1.5 Data profile construction

1.5.1 Gene, KEGG Ortholog (KO) and genus profiles

The inventors mapped the high-quality reads to the gene catalogue to a published reference gut gene catalogue established from European and Chinese adults ( (Qin et al. 2012, supra) (identity >＝ 90％) , based on which the inventors derived the gene, KO, and genus profiles using the same method of the published T2D paper (Qin et al. 2012, supra) .

1.5.2 mOTU profile

Clean reads were aligned to mOTU reference (total 79268 sequences) with default parameters (S. Sunagawa et al. , Metagenomic species profiling using universal phylogenetic marker genes. Nature methods 10, 1196 (Dec, 2013) , incorporated herein by reference) . 549 species level mOTUs were identified, including 307 annotated species and 242 mOTU linkage groups without representative genomes, which were putatively Firmicutes or Bacteroidetes.

1.5.3 IMG-species and IMG-genus profiles.

Bacterial, archaeal and fungal sequences were extracted from IMG v400 reference database (V. M.Markowitz et al. , IMG: the Integrated Microbial Genomes database and comparative analysis system. Nucleic acids research 40, D115 (Jan, 2012) , incorporated herein by reference) downloaded from http: //ftp. jgi-psf. org. 522, 093 sequences were obtained in total, and SOAP reference index was constructed based on 7 equal size chunks of the original file. Clean reads were aligned to reference using SOAP aligner (R. Li et al. , SOAP2: an improved ultrafast tool for short read alignment. Bioinformatics 25, 1966 (Aug 1, 2009) , incorporated herein by reference) version 2.22, with parameters “-m 4 -s 32 -r 2 -n 100 -x 600 -v 8 -c 0.9 -p 3”. Then, SOAP coverage software was used to calculate read coverage of each genome, normalized with genome length, and further normalized to relative abundance for each individual sample. The profile was generated based on uniquely mapped reads only.

1.6 Analysis of factors influencing gut microbiota gene profile

From the reference gene catalogue (Qin et al. 2012, supra) , the inventors derived a subset of 2.1M (2, 110, 489) genes that appeared in at least 6 samples in all 128 Hong Kong samples, and generated 128 gene abundance profiles using these 2.1 million genes. The inventors used the permutational multivariate analysis of variance (PERMANOVA) test to assess the effect of different characteristics, including age, BMI, eGFR, TCHO, LDL, HDL, TG, gender, DM, CRC status and location, on gene profiles of 2.1M genes. The inventors performed the analysis using the method implemented in package “vegan” in R， and the permuted p-value was obtained by 10, 000 times permutations. The inventors also corrected for multiple testing using “p.adjust” in R with Benjamini-Hochberg method to get the q-value for each gene.

When the inventors performed permutational multivariate analysis of variance (PERMANOVA) on 19 different covariates, only CRC status and CRC stage were significantly associated with these gene profiles (q < 0.05, Table 2) . Thus the data suggest an altered gene composition in CRC patient microbiomes that cannot be explained by other recorded factors.

Table 2 PERMANOVA analysis of microbial gene profiles in cohort I. The analysis was conducted to test whether clinical parameters and CRC status have significant impact on the gut microbiota with q<0.05. BMI: body mass index； DM: diabetes mellitus type 2； FBG: fasting blood glucose； HDL: high density lipoprotein； TG: triglyceride； eGFR: epidermal growth factor receptor； TNM: tumor node metastasis staging system； TCHO: total cholesterol； Cr: creatinine； LDL； low density lipoprotein. ALT/GPT: alanine transaminase/glutamate pyruvated transaminase.

1.7 CRC-associated genes identified by MGWAS

1.7.1 Identification of colorectal cancer associated genes

The inventors performed a metagenome wide association study (MGWAS) to identify the genes contributing to the altered gene composition in CRC. To identify the association between the metagenomic profile and colorectal cancer, a two-tailed Wilcoxon rank-sum test was used in the 2.1M gene profiles. The inventors got 140, 455 gene markers, which were enriched in either case or control with P<0.01 (Fig. 1) .

1.7.2 Estimating the false discovery rate (FDR)

Instead of a sequential p-value rejection method, the inventors applied the “qvalue” method proposed in a previous study (J. D. Storey, R. Tibshirani, Statistical significance for genomewide studies. Proceedings of the National Academy of Sciences of the United States of America 100, 9440 (Aug 5, 2003) , incorporated herein by reference) to estimate the FDR. In this analysis, the statistical hypothesis tests were performed on a large number of features of the 140, 455 genes. The false discovery rate (FDR) was 11.03％.

1.8 Taxonomic alterations in CRC microbiomes

The inventors examined the taxonomic differences between control and CRC-associated microbiomes to identify microbial taxa contributing to the dysbiosis. For this, the inventors used taxonomic profiles derived from three different methods, as supporting evidence from multiple methods would strengthen an association. First, the inventors mapped metagenomic reads to 4650 microbial genomes in the IMG database (V. M. Markowitz et al. , IMG: the Integrated Microbial Genomes database and comparative analysis system. Nucleic acids research 40, D115 (Jan, 2012) , incorporated herein by reference) (version 400) and estimated the abundance of microbial species included in that database (denoted IMG species) . Second, the inventors estimated the abundance of species-level molecular operational taxonomic units (mOTUs) using universal phylogenetic marker genes (S. Sunagawa et al. , Metagenomic species profiling using universal phylogenetic marker genes. Nature methods 10, 1196 (Dec, 2013) , incorporated herein by reference) . Third, the inventors organized the 140, 455 genes identified by MGWAS into metagenomic linkage groups (MLGs) that represent clusters of genes originating from the same genome (Qin et al. 2012, supra) , annotated the MLGs at species level using IMG database whenever possible, grouped MLGs based on these species annotations, and then estimated the abundance of these species (denoted MLG species) .

1.8.1 Identification of colorectal cancer associated MLG species

Based on the identified 140, 455 colorectal cancer associated marker genes profile, the inventors constructed the colorectal cancer associated MLGs using the method described in the previous type 2 diabetes study (Qin et al. 2012, supra) . All genes were aligned to the reference genomes of IMG database v400 to get genome level annotation. An MLG was assigned to a genome if >50％constitutive genes were annotated to that genome, otherwise it was termed as unclassified. Total 87 MLGs with gene number over than 100 were selected as colorectal cancer associated MLGs. These MLGs were grouped based on the species annotation of these genomes to construct MLG species.

To estimate the relative abundance of an MLG species, the inventors estimated the average abundance of the genes of the MLG species, after removing the 5％lowest and 5％highest abundant genes. Relative abundance of IMG species was estimated by summing the abundance of IMG genomes belonging to that species. Genus abundances were estimated by analogously summing species abundances.

1.8.2 CRC-associated species

Above analysis identified 28 IMG species, 21 mOTUs and 85 MLG species that were significantly associated with CRC status after stratifying by colonoscopy as a confounding factor (Wilcoxon rank-sum test, q<0.05； see Table 3) . Eubacterium ventriosum was consistently enriched in the control microbiomes across all three methods (Wilcoxon rank-sum tests - IMG: q ＝ 0.002； mOTU: q ＝ 0.0049； MLG: q ＝ 3.33x10^-4) . On the other hand, Parvimonas micra (q<7.73x10^-6) , Solobacterium moorei (q<0.011) and Fusobacterium nucleatum (q<0.00279) were consistently enriched in CRC patient microbiomes across all three methods (Fig. 2, Fig. 3) , while Peptostreptococcus stomatis (q <7.73x10^-6) was enriched according to two out of three methods. PERMANOVA analysis showed that only CRC status (P ≤ 0.013 from all three methods) and colonoscopy(P ＝ 0.079 from two methods) explained the quantitative variation in the three CRC-enriched species. All other non-CRC specific factors could not explain the variation with statistical significance (P > 0.18； Table 4) . P. stomatis has recently been shown to significantly associated with CRC, and S. moorei has previously been associated with bacteremia. The results confirmed this association in a new cohort with different genetic and cultural origins. However, a highly significant enrichment of P. micra - an obligate anaerobic bacterium that can cause oral infections like F. nucleatum - in CRC-associated microbiomes is a novel finding. P. micra is involved in the etiology of periodontis, and it produces a wide range of proteolytic enzymes and uses peptones and amino acids as energy source. It is known to produce hydrogen sulphide, which promotes tumor growth and proliferation of colon cancer cells. P. micra may represent opportunities for non-invasive diagnostic biomarkers for CRC.

1.9 Species level analysis

In order to evaluate the predictive power of these taxonomic associations, the inventors used the random forest ensemble learning method (D. Knights, E. K. Costello, R. Knight, Supervised classification of human microbiota. FEMS microbiology reviews 35, 343 (Mar, 2011) , incorporated herein by reference) to identify key marker species in the species profiles from the three different methods. This analysis revealed that 17 IMG species, 7 species-level mOTUs and 27 MLG species were highly predictive of CRC status (Table 5) , with predictive power of 0.86, 0.89 and 0.96 in ROC analysis, respectively (Fig. 4) . P. micra was identified as a key species from all three methods, while F. nucleatum, P. stomatis and S. moorei were identified from two out of three methods, providing further statistical support for their association with CRC status.

1.9.1 MLG species marker identification

Based on the constructed 87 MLGs with gene numbers over than 100, the inventors performed the Wilcoxon rank-sum test to each MLG with Benjamini-Hochberg adjustment, and 85 MLGs were selected out as colorectal associated MLGs with q<0.05. To identify MLG species markers, the inventors used “randomF orest 4.5-36” package in R vision 2.10 based on the 85 colorectal cancer associated MLG species. Firstly, the inventors sorted all the 85 MLG species by the importance given by the “randomForest” method. MLG marker sets were constructed by creating incremental subsets of the top ranked MLG species, starting from 1 MLG species and ending at all 85 MLG species. For each MLG markers set, the inventors calculated the false predication ratio in the 128 Chinese cohorts (cohort I) . Finally, the MLG species sets with lowest false prediction ratio were selected out as MLG species markers. Furthermore, the inventors drew the ROC curve using the probability of illness based on the selected MLG species markers.

1.9.2 IMG species and mOTU species markers identification

Based on the IMG species and mOTU species profiles, the inventors identified the colorectal cancer associated IMG species and mOTU species with q<0.05 (Wilcoxon rank-sum test with Benjamini-Hochberg adjustment) . Subsequently, IMG species markers and mOTU species markers were selecting using the random forest approach as in MLG species markers selection.

1.9.3 MLG, IMG and mOTU species Stage enrichment analysis:

Encouraged by the consistent species associations with CRC status and to take advantage of the records of disease stages of the CRC patients (Table 1) , the inventors explored the species profiles for specific signatures identifying early stages of CRC. The inventors hypothesized that such an effort might even reveal stage-specific associations that are difficult to identify in a global analysis. To identified which species were enriched in the four colorectal cancer progress or health control, the inventors did Kruskal test for the MLG species with gene number over 100, and all IMG species and mOTU species with q<0.05 (Wilcoxon rank-sum test with Benjamini-Hochberg adjustment) to get the species enrichment by the highest rank mean among four CRC stages and control. And the inventors also compared the significance between each two groups by pair-wise Wilcoxon Rank sum test.

In Chinese cohort I, several species showed significantly different abundances in different stages. Among these, the inventors did not identify any species enriched in stage I compared to all other stages and control samples. Peptostreptococcus stomatis, Prevotella nigrescens and Clostridium symbiosum were enriched in stage II or later compared to control samples, suggesting that they colonize the colon/rectum after the onset of CRC (Fig. 5) . However, Fusobacterium nucleatum, Parvimonas micra, and Solobacterium moorei were enriched in all four stages compared to controls and were most abundant in stage II (Fig. 6) , suggesting that they may play a role in both CRC etiology and pathogenesis, and implying them as potential biomarkers for early CRC.

1.10 CRC biomarker discovery

The inventors proceeded to identify potential biomarkers for CRC from the 140, 455 genes identified by the MGWAS approach, using the minimum redundancy maximum relevance (mRMR) feature selection method (H. Peng, F. Long, C. Ding, Feature selection based on mutual information: criteria of max-dependency, max-relevance, and min-redundancy. IEEE transactions on pattern analysis and machine intelligence 27, 1226 (Aug, 2005) , incorporated herein by reference) . To eliminate the confounding effects of colonoscopy, the inventors selected genes that were significant even after stratifying for colonoscopy, which resulted in 102, 514 genes. However, since the computational complexity of mRMR method did not allow us to use all 102, 514 genes, the inventors had to reduce the number of candidate genes. First, the inventors selected a stricter set of 24, 960 genes with higher statistical significance (P < 0.001； FDR ≤ 5.23%). Then the inventors identified groups of genes that were highly correlated with each other (Kendall’s τ ＞ 0.9) and chose the longest gene in each group, to generate a statistically non-redundant set of 11, 128 significant genes. Finally, the inventors used the mRMR method and identified an optimal set of 20 genes that were strongly associated with CRC status (Fig. 7, Table 6 and Table 7) . PCA (principal component analysis) using these 20 genes showed good separation of CRC patients from controls (Fig. 8) . PERMANOVA analysis showed that only CRC status, stage and fasting blood glucose explained the variation in the 20 marker gene abundances with statistical significance (P ≤ 0.01； see Table 8) . Although the inventors cannot rule out other confounding factors, the results suggest that the 20 marker genes characterize differences between CRC and control microbiomes. The inventors calculated a simple CRC index based on un-weighted log relative abundance of these 20 markers, which clearly separated the CRC patient microbiomes from the control microbiomes, as well as from 490 fecal microbiomes from two previous studies on type 2 diabetes in Chinese individuals (Qin et al. 2012, supra) and inflammatory bowel disease in European individuals (J. Qin et al. , A human gut microbial gene catalogue established by metagenomic sequencing. Nature 464, 59 (Mar 4, 2010) , incorporated herein by reference) (Fig. 9, median CRC-index for patients and controls in this study were 7.31 and -5.56, respectively； Wilcoxon rank-sum test, q < 6x10^-11 for all five comparisons, see Table 9) . Classification of the 74 CRC patient microbiomes against the 54 control microbiomes using the CRC index exhibited an area under the receiver operating characteristic (ROC) curve of 0.99 (Fig. 10) , while the areas under the curve (AUC) for classifying type 2 diabetes and IBD patients against the corresponding controls were 0.658 and 0.738, respectively, suggesting that the patterns captured by the index are predominantly CRC-specific. At the cutoff 0.7383 in Fig. 10, true positive rate (TPR) was 0.99, and false positive rate (FPR) was 0.07, indicating that the 20 gene markers could be used to accurately classify CRC individuals.

1.10.1 Minimum Redundancy Maximum Relevance (mRMR) feature selection framework

To establish a colorectal cancer classification only by gut metagenomic markers, the inventors adopted an mRMR method to perform a feature selection. The inventors used the “sideChannelAttack“ package from R to perform an incremental search and found 128 sequential markers sets. For each sequential set, the inventors estimated the error rate by leave-one-out cross-validation (LOOCV) of a linear discrimination classifier. The optimal selection of marker sets was the one corresponding to the lowest error rate. In the present study, the inventors made the feature selection on a set of 102, 514 colorectal cancer associated gene markers. Since this was computationally prohibitive to perform mRMR using all genes, the inventors derived a statistically non-redundant gene set. Firstly, the inventors pre-grouped the 102, 514 colorectal cancer associated genes that are highly correlated with each other (Kendall correlation > 0.9) . Then the inventors chose the longest gene as representative gene for the group, since longer genes have a higher chance of being functionally annotated, and will attract more reads during the mapping procedure. This generated a non-redundant set of 11, 128 significant genes. Subsequently, the inventors applied the mRMR feature selection method to the 11, 128 significant genes and identified an optimal set of 20 gene biomarkers that are strongly associated with colorectal cancer for colorectal cancer classification, which were shown on Table 6 and Table 7. The gene id is from the published reference gene catalogue as Qin et al. 2012, supra

1.10.2 Definition of CRC index

To exploit the potential ability of disease classification by gut microbiota, the inventors developed a disease classifier system based on the gene markers that the inventors defined. For intuitive evaluation of the risk of disease based on these gut microbial gene markers, the inventors calculated a gut healthy index (CRC index) .

To evaluate the effect of the gut metagenome on colorectal cancer, the inventors defined and calculated a CRC index for each individual on the basis of the selected 20 gut metagenomic markers by mRMR method. For each individual sample, the CRC index of sample j that denoted by I_j was calculated by the formula below:

A_ij is the relative abundance of marker i in sample j, wherein i refers to each of the gene markers in said gene marker set； .

N is a subset of all CRC-enriched markers in these selected gut metagenomic markers (namely, a subset of all patient-enriched markers in selected biomarkers related to the abnormal condition) .

M is a subset of all control-enriched markers in these selected gut metagenomic markers (namely, a subset of all control-enriched markers in selected biomarkers related to the abnormal condition) .

wherein the subset of CRC-enriched markers and the subset of control-enriched markers are shown in Table 7.

And |N| and |M| are the sizes (number) of these two sets, wherein |N| is 8 and |M| is 12.

Larger the CRC index, higher the risk of disease. Smaller the CRC index, more healthy the people. The inventors can build an optimal CRC index cutoff based on a large cohort. If the test sample CRC index is larger than the cutoff, then the person is in higher disease risk. And if the test sample CRC index is smaller than the cutoff then he is more healthy at low risk of disease. The optimal CRC index cutoff can be determined by a ROC method when AUC (Area Under the Curve) reached at its maximum.

1.10.3 Receiver Operator Characteristic (ROC) analysis

The inventors applied the ROC analysis to assess the performance of the colorectal cancer classification based on metagenomic markers. Based on the 20 gut metagenomic markers selected above, the inventors calculated the CRC index for each sample. The inventors then used the “Dalm” package in R to draw the ROC curve.

1.10.4 CRC index validation

After establishing CRC-index, the inventors calculated the CRC-index in Chinese cohort I consisting 128 individuals (Fig. 11, Table 10) , and 490 individuals from two previous studies on type 2 diabetes in Chinese individuals (Qin et al. 2012, supra) and inflammatory bowel disease in European individuals (J. Qin et al. , A human gut microbial gene catalogue established by metagenomic sequencing. Nature 464, 59 (Mar 4, 2010) , incorporated herein by reference) . The ability of the CRC index to distinguish CRC patients from the rest was compared using Wilcoxon rank-sum test with Benjamini-Hochberg adjustment for Chinese CRC cohorts, T2D cohorts and IBD cohorts.

Table 9 CRC index estimated in CRC， T2D and IBD patient and healthy cohorts.

Table 10.128 samples’ calculated CRC index (CRC patients and non-CRC controls)

Example 2. Validating the 20 biomarkers

The inventors validated the discriminatory power of the CRC classifier using another new independent study group, including 15 CRC patients and 15 non-CRC controls that were also collected in the Prince of Wales Hospital .

For each sample, DNA was extracted and a DNA library was constructed followed by high throughput sequencing as described in Example 1. The inventors calculated the gene abundance profile for these samples using the same method as described in Qin et al. 2012, supra. Then the gene relative abundance of each of the markers as set forth in SEQ ID NOs: 1-20 was determined. Then the index of each sample was calculated by the formula below:

A_ij is the relative abundance of marker i in sample j, wherein i refers to each of the gene markers as set forth in SEQ ID NOs 1-20；

N is a subset of all CRC-enriched (case) markers in these 20 selected gut metagenomic markers (namely, a subset of all patient-enriched markers in selected biomarkers related to the abnormal condition) ；

M is a subset of all control-enriched markers in these 20 selected gut metagenomic markers (namely, a subset of all control-enriched markers in selected biomarkers related to the abnormal condition) ；

wherein the subset of CRC-enriched markers and the subset of control-enriched markers are shown in Table 7；

Table 11 shows the calculated index of each sample and Table 12 shows the relevant gene relative abundance of a representative sample V1 and V30. In this assessment analysis, the areas under the curve (AUC) for classifying was 0.9733 (Fig. 12) . At the cutoff 0.9945, true positive rate (TPR) was 1, and false positive rate (FPR) was 0.2, validating that the 20 gene markers could be used to accurately classify CRC individuals.

Table 11.30 samples’ calculated CRC index

Table 12. Gene relative abundance of Sample V1 and V30

Thus the inventors have identified and validated 20 markers set by a minimum redundancy - maximum relevance (mRMR) feature selection method based on 140, 455 CRC-associated markers. And the inventors have built a gut healthy index to evaluate the risk of CRC disease based on these 20 gut microbial gene markers.

Example 3 Validation of gene markers by qPCR

Abundances of four randomly selected gene markers, including two enriched in control (m181682 (SEQ ID NO: 4) and m370640 (SEQ ID NO: 6) ) and two enriched in patient (m482585 (SEQ ID NO: 11) and m1704941 (SEQ ID NO: 16) ) , were further evaluated in 96 stool samples of the sequenced cohort (51 cases and 45 controls, a subset of cohort I) and cohort II of 156 samples (47 cases and 109 controls) using TaqMan probe-based qPCR. Primers and probes were designed using Primer Express v3.0 (Applied Biosystems, Foster City, CA, USA) . The qPCR was performed on an ABI7500 Real-Time PCR System using the

Universal PCR Master Mixreagent (Applied Biosystems) . Universal 16S rDNA was used as internal control and abundance of gene markers were expressed as relative levels to 16S rDNA.

3.1 Evaluating CRC biomarkers using targeted quantitative PCR

The biomarkers were derived using the admittedly expensive deep metagenome sequencing approach. Translating them into diagnostic biomarkers would require reliable measurement by simple, affordable and targeted methods such as quantitative PCR (qPCR) . To verify this, the inventors randomly selected two case-enriched and two control-enriched gene markers and measured their abundances by qPCR in a subset of 96 samples selected from cohort I (51 cases and 45 controls) . Quantification of each of the four genes by the two platforms (metagenomic sequencing and qPCR) showed strong correlations (Spearman r＝0.81-0.95, Fig. 13) , suggesting that the gene markers could also be reliably measured using qPCR. Next, in order to validate the markers in previously unseen samples, the inventors measured the abundance of these four gene markers using qPCR in 156 fecal samples (47 cases and 109 controls) from an independent Chinese cohort (cohort II； see Table 13) . The two control-enriched genes did not show significant associations (P > 0.31； Table 14) . On the other hand, the CRC-enriched gene markers (m1704941, butyryl-CoA dehydrogenase from F. nucleatum； m482585, RNA-directed DNA polymerase from an unknown microbe) significantly associated with CRC status after stratifying by colonoscopy (P ＝ 0.0015 and P ＝ 0.045, respectively, see Table 14) . However, only the gene from F. nucleatum remained significant after a Mantel-Haenszel test adjusted for colonoscopy (odds ratio＝18.5, P ＝ 0.0051) . The CRC index based on the abundances of the four genes only moderately classified CRC microbiomes from control microbiomes (AUC＝0.73) , perhaps suggesting that choosing randomly from the list of 20 biomarkers was not an effective strategy. Nevertheless, the gene from F. nucleatum was present only in 4 out of 109 control microbiomes, suggesting a potential for developing specific diagnostic tests for CRC using fecal samples.

3.2 Accurate qPCR biomarkers identified by validation in an independent metagenomic cohort

To identify robust biomarkers that can have a more general applicability, the inventors evaluated all 20 gene markers using fecal metagenomes from a cohort with different genetic background and lifestyle: 16 CRC patients and 24 control individuals from Denmark (cohort III) . These were symptomatic individuals referred to colonoscopy and all samples were blinded before DNA extraction and analyses (see Table 15) . When mapped to the 4.3 million gut microbial genes, the 40 Danish microbiomes exhibited significantly higher gene richness and gene alpha diversity, both in cases (Wilcoxon rank-sum tests, gene count: P ＝ 1.94x10^-5 ； Shannon’s index： P ＝ 5.85x10^-5) and controls (gene count: P ＝ 0.0017； Shannon’s index： P ＝ 9.34x10^-4； Table 16) , agreeing with a recent study and suggesting differences in gut microbial community structure between the Chinese and Danish populations (J. Li et al. , An integrated catalog of reference genes in the human gut microbiome. Nature biotechnology 32, 834 (Aug, 2014) , incorporated herein by reference) . Among the 102, 514 genes associated with CRC status in Chinese cohort I, only 1, 498 genes could be validated in the Danish microbiomes. However, CRC-enriched genes were shared significantly more between the two populations than control-enriched genes (1, 452 out of 35, 735 CRC-enriched vs. 46 out of 66, 779 in control-enriched； two-tailed chi-squared test, chi-squared＝2576.57, P < 0.0001) . Over half (53.6％) of the 1, 452 CRC-enriched genes were from just three species: Parvimonas micra (389 genes) , Solobacterium moorei (204 genes) and Clostridium symbiosum (177 genes) (see Table 17) . At the species level, P. micra was enriched in CRC microbiomes using all three methods, while P. stomatis, G. morbillorum, and S. moorei were enriched according to two methods (Wilcoxon rank-sum test, q <0.05； Table 18) . Notably, all the species that were validated by at least one method were CRC-enriched. These results suggest that changes in the colorectal environment during CRC development and progression may facilitate the growth of similar species across the two populations, potentially leading to the reduced microbial diversity observed in the CRC patients, in line with earlier observations by others (J. Ahn et al. , Human gut microbiome and risk for colorectal cancer. Journal of the National Cancer Institute 105, 1907 (Dec 18, 2013) , incorporated herein by reference) . The CRC index using 20 gene markers discovered in Chinese cohort I marginally differentiated the Danish patient microbiomes from the control ones (Wilcoxon rank-sum test, P ＝ 0.029) and exhibited moderate classification potential (area under ROC curve 0.71, Fig. 14-1) . Only four out of the 20 genes (two from P. anaerobius and one each from P. micra and F. nucleatum) were associated with CRC status in the Danish cohort III (Wilcoxon rank-sum test， q≤0.06； all CRC-enriched； see Table 19) . Among the factors the inventors had recorded, only CRC status could explain the variation in these four genes (PERMANOVA P ≤ 0.0001； see Table 20) , suggesting that these signatures are CRC-specific. CRC index using these four genes could classify CRC patients accurately with area under ROC curve of 0.85 (Fig. 14-2, Table 21) . At the cutoff -16.68, true positive rate (TPR) was 0.75, and false positive rate (FPR) was 0.08333. This higher AUC validated that the 4 gene markers could be used to classify CRC individuals. Two of the four genes were transposases from Peptostreptococcus anaerobius. The third gene (m1704941, butyryl-CoA dehydrogenase from F. nucleatum) was incidentally among the two genes successfully validated using qPCR in Chinese cohort II. The fourth gene from P. micra was the highly conserved rpoB gene (namely m1696299 (SEQ ID NO: 13, with identity of 99.78％) encoding RNA polymerase subunit β， often used as a phylogenetic marker (F. D. Ciccarelli et al. , Toward automatic reconstruction of a highly resolved tree of life. Science 311, 1283 (Mar 3, 2006) , incorporated herein by reference) .

For each sample, DNA was extracted and a DNA library was constructed followed by high throughput sequencing as described in Example 1. The inventors calculated the gene abundance profile for these samples using the same method as described in Qin et al. 2012, supra. Then the gene relative abundance of each of the markers as set forth in SEQ ID NOs: 1, SEQ ID NO: 9, SEQ ID NO: 13 and SEQ ID NO: 16 was determined. Then the index of each sample was calculated by the formula below:

A_ij is the relative abundance of marker i in sample j, wherein i refers to each of the gene markers as set forth in said gene marker set； .

N is a subset of all CRC-enriched (case) markers in these 4 selected gut metagenomic markers (namely, a subset of all patient-enriched markers in selected biomarkers related to the abnormal condition) .

wherein the subset of CRC-enriched markers arethe marker as set forth in SEQ ID NOs: 1, SEQ ID NO: 9, SEQ ID NO: 13 and SEQ ID NO: 16；

|N| is the sizes (number) of the biomarkers in the subset, wherein |N| is 4.

wherein an index greater than a cutoff indicates that the subject has or is at the risk of developing colorectal cancer.

Table 21. 40 samples’ gene relative abundance and calculated CRC index

The inventors decided to evaluate the diagnostic potential of the two non-transposase genes in the independent Chinese cohort II using qPCR. As these were originally discovered in Chinese cohort I and validated in Danish cohort III, cohort II serves as a suitable independent validation cohort of these genes, notably in a different platform. The inventors performed additional qPCR measurements of rpoB from P. micra, which showed a significant enrichment in CRC patient microbiomes in cohort II (Wilcoxon rank-sum test, stratified by colonoscopy, P ＝ 8.97x10^-8) . Mantel-Haenszel odds ratio adjusted for colonoscopy was 20.17 (95％ confidence interval 4.59-88.6, P ＝ 3.36x10^-7) . Combined qPCR measurements (primers in Table 22) of the two genes (1696299 (SEQ ID NO: 13) and 1704941 (SEQ ID NO: 16) ) clearly separated case from control samples in Chinese cohort II (Wilcoxon rank-sum test stratified by colonoscopy, P ＝1.404x10^-8, Fig. 15a) . Their combined abundance accurately classified CRC samples in Chinese cohort II with an improved area under the ROC curve of 0.84 (cutoff -13.38, true-positive rate＝0.723, false-positive rate＝0.073； Fig. 15b, Table 23), validating that the 2 gene markers could be used to classify CRC individuals. The accuracy was slightly better than that in a recent study (AUC＝0.836, true-positive rate＝0.58, false-positive rate＝0.08) , even though they used a combination of abundances of 22 species using metagenomic sequencing (G. Zeller et al. , Potential of fecal microbiota for early-stage detection of colorectal cancer. Molecular systems biology 10, 766 (2014) , incorporated herein by reference) . The Mantel-Haenszel odds ratio (adjusted for colonoscopy) for detecting at least one of the two markers by qPCR in CRC patients was 22.99 (P ＝5.79x10^-8, 95％confidence interval 5.83-90.8) . When stratifying the cohort into early stage (stages I-II) and late stage (stages III-IV) cancer patients, the classification potential and the odds ratio were still significant (see Table 24) . Abundance of these two genes was significantly higher compared to control samples starting from stage II of CRC (Fig. 15c-d) , agreeing with the results from species abundances, and providing proof-of-principle that fecal metagenomes may harbor non-invasive biomarkers for the identification of early stage CRC.

For each sample, DNA was extracted as described in Example 1. The inventors performed qPCR as described above. Then the gene relative abundance of each of the markers as set forth in SEQ ID NO:13 and SEQ ID NO: 16 was determined. Then the index of each sample was calculated by the formula below:

N is a subset of all CRC-enriched (case) markers in these 2 selected gut metagenomic markers (namely, a subset of all patient-enriched markers in selected biomarkers related to the abnormal condition) .

wherein the subset of CRC-enriched markers arethe marker as set forth in SEQ ID NO: 13 and SEQ ID NO: 16；

|N| is the sizes (number) of the biomarkers in the subset, wherein |N| is 2.

The inventors also used one of the two gene markers to calculate CRC index respectively (Table 23). The area under the ROC curve was 0.80 only using 1696299 (cutoff -6.762, true-positive rate＝0.6383, false-positive rate＝0.05505, Fig. 16) and the area under the ROC curve was 0.69 only using 1704941 (Fig. 17) . The results showed that gene marker 1696299 (SEQ ID NO: 13) was the robust biomarkers, which also could be used to classify CRC individuals uniquely.

The inventors have demonstrated, for the first time, the potential for CRC diagnosis through affordable targeted detection methods for microbial biomarkers in fecal samples. Two recent studies reported on potential CRC diagnosis using metagenomic sequencing of the fecal microbiome, with the same accuracy as ours (in terms of area under the receiver-operating curve) . While the 16S ribosomal RNA gene based study used 5 operational taxonomic units to classify CRC from healthy samples in a cohort notably without any cross-validation (J. P. Zackular, M. A. Rogers, M. T. t. Ruffin, P. D. Schloss, The human gut microbiome as a screening tool for colorectal cancer. Cancer prevention research 7, 1112 (Nov, 2014) , incorporated herein by reference) , the metagenomic shotgun study used 22 species-level taxonomic units to accurately classify CRC patients notably in an independent cohort (G.Zeller et al. , Potential of fecal microbiota for early-stage detection of colorectal cancer. Molecular systems biology 10, 766 (2014) , incorporated herein by reference) . The inventors have shown that using just two gene markers, discovered in 128 Chinese individuals and validated in 40 Danish individuals, the inventors could accurately classify CRC patients from control individuals in an independent qPCR validation cohort of 156 Chinese individuals. The significant improvement in the classification potential (from AUC＝0.73 to AUC＝0.84) by using a gene (rpoB gene from P. micra) validated in the Danish cohort reiterates the importance of validating newly discovered biomarkers in independent cohorts with different genetic and environmental background.

Table 22. Sequence Information for the primers and probes for the selected 2 gene markers

Table 23 156 samples’ qPCR gene relative abundance and calculated CRC index

Table 3 IMG, mOTU and MLG species associated with CRC with q-value < 0.05.85 MLG species were formed after grouping 106 MLGs with more than 100 genes using species annotation when available.

Claims

A gene marker set for predicting the risk of colorectal cancer (CRC) in a subject comprising one or more of the genes as set forth in SEQ ID NOs: 1 to 20.
Use of the gene marker set of claim 1 for predicting the risk of colorectal cancer (CRC) in a subject, via the steps of:

1) collecting a sample j from the subject and extracting DNA from the sample；

2) determining the abundance information of each of gene marker in the gene marker set； and

3) calculating the index of sample j by the formula below:

A_ij is the relative abundance of marker i in sample j, wherein i refers to each of the gene markers in said gene marker set；

N is a subset of all CRC-enriched markers in the gene marker set；

M is a subset of all control-enriched markers in the gene marker set；

and |N| and |M| are the sizes (number) of the biomarker respectively in these two subsets；

wherein an index greater than a cutoff indicates that the subject has or is at the risk of developing colorectal cancer (CRC) .
Use of the gene marker set of claim 1 for preparation of a kit for predicting the risk of colorectal cancer (CRC) in a subject, via the steps of:

1) collecting a sample j from the subject and extracting DNA from the sample；

2) determining the abundance information of each of gene marker in the gene marker set； and

3) calculating the index of sample j by the formula below:

A_ij is the relative abundance of marker i in sample j, wherein i refers to each of the gene markers in said gene marker set；

N is a subset of all CRC-enriched markers in the gene marker set；

M is a subset of all control-enriched markers in the gene marker set；

and |N| and |M| are the sizes (number) of the biomarker respectively in these two subsets；

wherein an index greater than a cutoff indicates that the subject has or is at the risk of developing colorectal cancer (CRC) .
The use of claim 2 or claim 3, wherein the abundance information is gene relative abundance of each of gene marker in the gene marker set which is determined by means of sequencing method.
The use of claim 2 or claim 3, wherein the abundance information is gene relative abundance of each of gene marker in the gene marker set which is determined by a qPCR method.
The use of any one of claims 2-5, wherein the cutoff value is obtained by a Receiver Operator Characteristic (ROC) method, wherein the cutoff corresponds to when AUC (Area Under the Curve) reached at its maximum.
The use of any one of claims 2-6, wherein the gene marker set consists of SEQ ID NOs: 1 to 20.
The use of any one of claims 2-6, wherein the gene marker set consists of SEQ ID NOs: 1, 9, 13 and 16.
The use of any one of claims 2-6, wherein the gene marker set consists of SEQ ID NOs: 13 and 16.
The use of any one of claims 2-6, wherein the gene marker set consists of SEQ ID NO: 13.
A method for diagnosing whether a subject has colorectal cancer or is at the risk of developing colorectal cancer, comprising:

1) collecting a feces sample j from the subject and extracting DNA from the sample；

2) determining the abundance information of each of the marker in a gene marker set comprising one or more of the genes as set forth in SEQ ID NOs: 1 to 20； and

3) calculating the index of sample j by the formula below:

A_ij is the relative abundance of marker i in sample j, wherein i refers to each of the gene markers in said gene marker set；

N is a subset of all CRC-enriched markers in the gene marker set；

M is a subset of all control-enriched markers in the gene marker set；

and |N| and |M| are the sizes (number) of the biomarker respectively in these two subsets；

wherein an index greater than a cutoff indicates that the subject has or is at the risk of developing colorectal cancer (CRC) .
The method of claim 11, wherein the abundance information is gene relative abundance of each of gene marker in the gene marker set which is determined by means of sequencing method.
The method of claim 11, wherein the abundance information is gene relative abundance of each of gene marker in the gene marker set which is determined by a qPCR method.
The method of claim 11, wherein the cutoff value is obtained by a Receiver Operator Characteristic (ROC) method, wherein the cutoff corresponds to when AUC (Area Under the Curve) reached at its maximum.
The method of any one of claims 11-14, wherein the gene marker set consists of SEQ ID NOs: 1 to 20.
The method of any one of claims 11-14, wherein the gene marker set consists of SEQ ID NOs: 1, 9, 13 and 16.
The method of any one of claims 11-14, wherein the gene marker set consists of SEQ ID NOs: 13 and 16.
The method of any one of claims 11-14, wherein the gene marker set consists of SEQ ID NO: 13.
Use of a marker as set forth in SEQ ID NO: 13 or rpoB gene encoding RNA polymerase VXEXQLW β as a gene marker for predicting the risk of colorectal cancer (CRC) in a subject, wherein the enrichment of said gene marker in a sample of the subject relative to a control sample is indicative of the risk of colorectal cancer in the subject.