Profiling Fusobacterium infection at high taxonomic resolution reveals lineage-specific correlations in colorectal cancer

The bacterial genus Fusobacterium promotes colorectal cancer (CRC) development, but an understanding of its precise composition at the species level in the human gut and the relevant association with CRC is lacking. Herein, we devise a Fusobacterium rpoB amplicon sequencing (FrpoB-seq) method that enables the differentiation of Fusobacterium species and certain subspecies in the microbiota. By applying this method to clinical tissue and faecal samples from CRC patients, we detect 62 Fusobacterium species, including 45 that were previously undescribed. We additionally reveal that Fusobacterium species may display different lineage-dependent functions in CRC. Specifically, a lineage (designated L1) including F. nucleatum, F. hwasookii, F. periodonticum and their relatives (rather than any particular species alone) is overabundant in tumour samples and faeces from CRC patients, whereas a non-enriched lineage (designated L5) represented by F. varium and F. ulcerans in tumours has a positive association with lymphovascular invasion.

C olorectal cancer (CRC) is a common malignant disease worldwide. The gut microbiota is now recognised as an important player in CRC 1 . Emerging evidence has demonstrated that the bacterial genus Fusobacterium in the human gut promotes CRC development [2][3][4][5] .
Among Fusobacterium members, Fusobacterium nucleatum is the most studied species that is associated with CRC 3 . It is enriched in the gut microbiota of CRC patients 5 and promotes carcinogenesis through multiple mechanisms, including increasing tumour cell proliferation by upregulating miR-21 6 , promoting carcinogenesis by modulating E-cadherin/β-catenin signalling and inducing Wnt/β-catenin 7,8 , inducing chemoresistance by modulating autophagy 9 , and protecting tumours from immune cell attack 10 . It can also be found in the liver metastases of Fusobacterium-associated primary tumours 11 . Although most CRC-related studies have chosen F. nucleatum (especially, subsp. nucleatum) as a model organism, the precise compositions of gut Fusobacterium communities at the species level remain largely unknown. Yet, a metagenomic study revealed that there were multiple fusobacterial taxa in the gut of southern Chinese populations 12 . A similar study in the same population showed that non-F. nucleatum species were also predominant and might have distinctive correlations with different host diseases 13 . Moreover, F. nucleatum itself has four subspecies, namely, nucleatum, animalis, vincentii (inclusive of fusiforme) and polymorphum, which are phylogenetically divergent and can arguably be classified as separate species 14,15 . It has been suggested that subsp. animalis-but not subsp. nucleatum-might be prevalent in CRC tissues 16 . Therefore, despite the extraordinary progress achieved in determining the oncogenic mechanisms of F. nucleatum with type strains, the uncertainties in the compositions of gut Fusobacterium communities have complicated efforts to establish aetiological relationships between Fusobacterium members and CRC. Understanding the species-level composition of Fusobacterium community is important to elucidate the roles played by Fusobacterium in the development of CRC and could provide precise targets that may aid early diagnosis and even treatment of CRC.
Major gut microbiota profiling approaches include 16S rRNA gene amplicon sequencing and metagenomic sequencing. But the 16S rRNA gene displays high similarity within Fusobacterium, and furthermore, it does not provide species-level resolution for this genus when used in amplicon sequencing 13,17 . In contrast, rpoB, which is also a widely used prokaryotic genotyping marker in the assessment of genetic relatedness, is sufficiently polymorphic to distinguish Fusobacterium species 17 . The metagenomic sequencing is capable of species identification, but it relies mostly on reference genomes of known species.
This study aims to resolve the precise compositions of gut Fusobacterium communities in CRC patients to probe the roles played by Fusobacterium members. By taking advantage of comparative genomic analysis, we devise a method that allows Fusobacterium species and certain subspecies to be differentiated in the microbiota and investigated the compositional features of Fusobacterium communities in the tumours and faeces of CRC patients. This study may promote the understanding of how Fusobacterium members promote CRC development and provide further directions to better model this malignant disease and study relevant host-gut microbe interactions.

Results
Species-level taxonomic analysis of Fusobacterium with sequenced genomes. To develop a new approach that can differentiate Fusobacterium species and some subspecies in the microbiota, we sought to search for taxonomic marker candidates in sequenced Fusobacterium genomes. A total of 150 genomes from the NCBI database and seven genomes sequenced by this study (Supplementary Data 1) were used. Taxonomic analysis of the Fusobacterium genus was firstly conducted with the genomes to serve as a reference for subsequent investigation. Wholegenome ANIb analysis was performed in a pairwise manner to ensure accurate taxonomic designation (Supplementary Data 2 and Fig. 1), as there could be inaccurately labelled species names in the database. The species names of 17 strains were reassigned, while 15 F. nucleatum and 10 Fusobacterium necrophorum strains were classified further into subspecies (Supplementary Data 1). We found that similar to most bacteria 18 , an ANIb cut-off of 94% could effectively delineate Fusobacterium species (Fig. S1), except F. nucleatum. The F. nucleatum strains commonly displayed >90% ANIbs with each other, but 90% ANIb could not be used as the species boundary since the strains also showed >90% ANIbs with F. hwasookii and Fusobacterium canifelinum. However, the 94% ANIb accurately defined the four F. nucleatum subspecies (nucleatum, animalis, polymorphum and vincentii), suggesting that these subspecies can be considered as four separate species, consistent with previous reports 14,15 . We also found that the F. necrophorum strains could be categorised into three clades, two of which corresponded to the subspecies necrophorum and funduliforme, while one might represent a different subspecies. However, unlike F. nucleatum, all members of F. necrophorum showed >95% intra-species ANIbs, and the subspecies boundaries were apparent at a cut-off of 97.5%. The 157 strains were ultimately classified into 19 species and four F. nucleatum subspecies (Fig. 1). The classification was in accordance with the wholegenome phylogeny of these genomes (Fig. S2). Of note, in the whole-genome phylogenetic tree, the F. nucleatum subspecies were also not in the same clade but were instead located on distinct major branches parallel to other species, which again supported the hypothesis that the four subspecies should rank as species. Thus, in the following analyses, they were considered as separate species (unless otherwise specified). As such, we clarified the taxonomy of Fusobacterium based on whole-genome analyses.
Development of a Fusobacterium rpoB amplicon-sequencing (FrpoB-seq) method for species-level designation of Fusobacterium. Based on the genomic data, we found that rpoB was an accurate high-resolution taxonomic marker for Fusobacterium. The rpoB was present in one copy in each genome. The phylogenetic trees based on rpoB and the 16S rRNA gene were both capable of distinguishing Fusobacterium species, showing consistency with the taxonomy described above (Fig. S3). Moreover, rpoB was more polymorphic than the 16S rRNA gene for specieslevel identification, as its intra-and inter-species identities had almost separable distributions ( Fig. 2A). In the rpoB tree, all the studied species formed distinct clades, outperforming the 16S rRNA gene tree (Fig. S3). In particular, rpoB could clearly distinguish Fusobacterium periodonticum and Fusobacterium pseudoperiodonticum and even the subspecies of F. necrophorum, with only one exception, namely, that an F. necrophorum subsp. necrophorum strain (DAB) was grouped in the clade of F. necrophorum subsp. funduliforme (Fig. S3). We also noted that in both the rpoB and 16S rRNA gene trees, the four F. nucleatum "subspecies" formed separate clades parallel to other species, in line with the whole genome-based findings (Fig. S3). Notably, all the studied Fusobacterium species could be robustly classified into certain lineages (Fig. S3). Thus, the rpoB was employed to develop a Fusobacterium species differentiation method. We next performed multiple sequence alignment of the Fusobacterium rpoB genes to search for a short representative region compatible with amplicon sequencing. We searched for   THCT5A4  THCT7A2  21_1A  HMSC065F01  CTI-3  4_8  3_1_33  CTI-5  F0419  THCT6B3   ATCC 25832  NCTC13121  593A  Marseille-P2749  ATCC 29250  clos_1_1  NCTC10723  AM25-1  OM06-15BH  ATCC 9817  AM25-18LB  ATCC 27725  NCTC10560  HMSC073F01  AM22-17  TM07-10  Fv113-g1  81A6  12_1B  OF01-15LB  NCTC12112  ATCC 49185  3_1_5R  CMW8396  ATCC 25563  KG35  DAB  BFTR- polymorphic regions of 300-500 bp bounded by >20 bp conserved sequences. Three candidate regions were identified (Fig. 2B), and only Region 1 was selected as it was found to be specific to Fusobacterium based on comparison against the NCBI nucleotide database. The selected region was~370 bp in length and located close to the 5′ end of the gene. Notably, the phylogeny of this region retained the capacity of distinguishing Fusobacterium species (Fig. 2C). The termini of the region werẽ 25 bp in length and highly conserved among Fusobacterium species, based on which a set of universal primers was designed (Fig. 2D). The primers yielded a single specific band from different Fusobacterium isolates (Table S1) and human faecal samples (Fig. S4A). We also assessed whether the primers were specific to Fusobacterium using in silico analysis against the NCBI     A Distributions of pairwise intra-and inter-species identities of the 16S rRNA (n = 2610 and n = 17,696, respectively) and rpoB genes (n = 3068 and n = 20,802, respectively) from the genomes. B Locations of three candidate regions compatible with amplicon sequencing found in rpoB. The reference length and coordinates (in parentheses) in the rpoB gene of F. nucleatum subsp. nucleatum ATCC 25586 are given. Region 1 was selected for further study. C Phylogenetic trees of full-length rpoB and their corresponding selected rpoB region. Strain names are given in parentheses for the species with only one sequenced genome available. Branches of the same species/ subspecies or those otherwise illustrated in the orange box (corresponding to the orange triangle) are compressed as applicable. There is an exception in the sub-tree of F. necrophorum subsp. funduliforme, which is illustrated in Fig. S3. Strain names are also provided for those that could not be compressed together. D Alignment of the non-redundant terminal conserved sequences of the selected rpoB region and universal primers designed based on these conserved regions. Asterisks denote consensus bases. Variations are denoted by nucleotide-specific colour shades. E Distribution of pairwise intra-and inter-species identities of the selected rpoB region (n = 3068 and n = 20,802, respectively). Complete gene sequences available in the sequenced genomes were used for analysis (144 16S rRNA gene and 155 rpoB sequences). F. naviforme genomes (n = 2) were not included. The analysis considered the four F. nucleatum subspecies as separate species. Source Data are provided as a source data file.
database. Certain putative non-specific amplification was detected with multiple primer mismatches, but most of the species detected were environmental or animal-inhabiting microorganisms (Table S2). Therefore, this region was adopted for Fusobacterium rpoB amplicon sequencing (FrpoB-seq). Based on ANIb analysis, we were unable to define a species boundary of sequence identity for this region, as it had 95.1-100% (median: 99.7%) intra-species and 74.7-99.7% (median: 80.4%) interspecies identities (Fig. 2E), we, therefore, adopted a phylogenybased annotation strategy in the following analysis. But in most cases, 98% could be considered as an efficient boundary (Fig. 2E) and was also used to aid annotation. Notably, Fusobacterium naviforme was not included in the analysis, as its rpoB gene did not show such conservation. Moreover, F. naviforme showed a distant relationship to other studied Fusobacterium species ( Fig. 1 and S3). Finally, we set up mock samples and tested the sensitivity of FrpoB-seq. We found the limit to achieve successful construction of a standard sequencing library was 1000-10000 overall Fusobacterium rpoB copies per reaction, while upon successful FrpoB-seq, detection of a particular species could be achieved at~100 rpoB copies per reaction. In addition, the detection result was generally in accord with the composition in mock samples (Fig. S4B). With the above efforts, we developed a FrpoB-seq method for species-level designation of Fusobacterium.
Implementation of FrpoB-seq in tissue and faecal samples of CRC patients. FrpoB-seq was then conducted with tissue (551 tumour-adjacent tissue pairs) and faecal samples from CRC patients (n = 94) and healthy individuals (n = 95). FrpoB-seq was accomplished in 556 (304 tumour and 252 normal) tissue and 99 (61 CRC and 38 control) faecal samples. Factors resulting in unaccomplished implementation included low or non-detectable Fusobacterium abundance, low amounts of sample DNA, and/or limited library quality. Notably, most of the sequencing data (91.42% and 99.73% in tissue and faecal samples, respectively) were specific to the Fusobacterium rpoB target (Fig. S4C). Nonspecific data mainly corresponded to the DNA of humans and some gut microbes (Table S3). The study used a 100% threshold to generate operational taxonomic units (OTUs), which were subsequently subjected to phylogeny-based species annotation (Fig. S5). Strikingly, 62 species were identified, and 45 (72.6%) of them were previously undescribed (Supplementary Data 3).
The rpoB sequences of the previously undescribed species all showed <98% (mostly <97%) identity to those of the known species, also sufficient to classify them as separate species. To further validate those species, the study performed deep metagenomic sequencing (average~10 Gbp) in a subset (n = 35; Supplementary Data 3) of faecal samples covering 25 putative undescribed species (Fig. S6). The existence of five putative undescribed species was directly confirmed in the matching samples (Fig. S6). However, in samples associated with the remaining putative undescribed species, the selected rpoB region or even the entire rpoB gene was not covered by metagenomic sequencing, likely due to the low abundance of Fusobacterium communities. We alternatively search for Fusobacterium-specific rpoB fragments of other region, 16S rRNA genes and other genomic sequences that belonged to undescribed species and indeed the corresponding evidence was obtained (Fig. S6). But in such cases, the FrpoB-seq data could not be exactly mapped. We additionally searched for the rpoB of undescribed Fusobacterium species in the public metagenomic datasets of two cohorts, a Southern Chinses population cohort (n = 556, average~7.5 Gbp) reported containing non-nucleatum Fusobacterium 12 , and a Korean population cohort (n = 106, average >30 Gbp) with ultra-high sequencing depth 19 . Eight putative undescribed Fusobacterium identified by this study were confirmed in the two cohorts (Fig. S6). The results highlighted that there were unknown Fusobacterium members in the human gut, but future validations are still needed.
Meanwhile, the phylogeny based on the rpoB region indicated that all Fusobacterium species could be categorised into nine lineages (designated L1-L9) (Figs. 3 and S5). This lineage classification was consistently observed in the ANIb matrixbased dendrogram, the phylogenetic trees of 16S rRNA genes, full-length rpoB and the selected rpoB regions of the studied genomes, as well as, the tree of the rpoB sequences obtained by FrpoB-seq (Figs. S3 and S7), implying that could be an important genetic feature of Fusobacterium.  Compositional features of the Fusobacterium communities in colorectal cancer tissues. To investigate the specific roles of various Fusobacterium members in CRC, the compositional features of Fusobacterium communities in the tissues of CRC patients were further analysed. Quantification with the designed rpoB-targeted primers showed that the relative Fusobacterium abundance among the total bacterial population was significantly higher in tumour tissue than in the normal mucosa (Fig. 4A)    lower Fusobacterium abundance in tumours than younger patients (Fig. 4C). In addition, higher Fusobacterium loads were observed in microsatellite instability-high (MSI-H) tumours than in microsatellite stable (MSS) tumours (Fig. 4D). The tumoural Fusobacterium abundance showed no association with other analysed pathological features, including the KRAS mutation status and EGFR or p53 immunohistochemical (IHC) staining results (Fig. S8).
Next, we study the patterns of Fusobacterium communities with the FrpoB-seq data (Supplementary Data 3). We found that the Fusobacterium communities in both tumour (n = 304) and normal (n = 252) tissues displayed surprisingly diverse compositional patterns at the species level (Fig. 4E). In total, 60 species were identified in those samples, with varied detection rates. Notably, in addition to the canonical taxon F. nucleatum, Fusobacterium mortiferum, Fusobacterium varium, and Fusobacterium sp. oral taxon 370 were also prevalent. Among the F. nucleatum "subspecies", animalis and polymorphum were highly prevalent, while nucleatum showed the lowest detection rate. At the lineage level, L1, L4, L5 and L9 were prevalent, and L2, L5, L6 and L7 were rare, while L3 was not detected. Between one and 20 species (median: seven) were detected in each sample, and the tumours contained more species than the normal mucosa (Fig. 4F). Strikingly, the Fusobacterium community in a sample was generally predominated by one of its species members; in over 96% of the samples, the most abundant species accounted for >50% of the total Fusobacterium community (Fig. 4G).
The Fusobacterium species compositional patterns in tumours did not show clear differences but rather were similar to those in the corresponding adjacent normal mucosa. Among the 201 pairs of samples for which FrpoB-seq was successfully applied, 170 (84.6%) had compositional patterns clustered within sister edges of the dendrogram or at least the same major clade (Figs. 4H and S9). Interestingly, in the remaining 31 pairs with distinct patterns, the predominant species exhibited an apparent shift from non-L1 lineages in the normal mucosa to the L1 lineage in tumours (Fig. 4I).
We next used the FrpoB-seq data for tumour and normal mucosa tissues to probe individual Fusobacterium members associated with CRC. Statistical analysis at the species level was not particularly productive due to the wide variation in species composition among samples, especially because the predominance of a particular species, even a frequently detected one, was confined to a relatively small proportion of the samples. We, therefore, conducted the analysis at the lineage level instead. Notably, the proportions of L1 and L4 in Fusobacterium communities showed differences between the tumour and normal mucosa samples (Fig. 4J, K). After incorporating qPCR data for total Fusobacterium quantification, we found that L1 was the only lineage that showed a difference in relative abundance among the total bacterial population, which was significantly increased in tumours (Fig. 4K).
We further assessed whether the Fusobacterium lineages in tumours were associated with particular pathological characteristics ( Fig. 5A and Supplementary Data 4). The abundance of L1, L4 or L9 in tumours did not display an association with the examined pathological features (Figs. 5B and S10). In contrast, the abundance of L5 in tumours, though not different from that in the paired normal mucosa, exhibited a strong positive association with lymphovascular invasion (Fig. 5B, C). The above results uncovered that Fusobacterium had lineage-specific correlations in CRC.
Compositional features of the Fusobacterium communities in faecal samples from colorectal cancer patients. Next, we studied the compositional features of Fusobacterium communities in faecal samples. Notably, the results were consistent with the findings in tissues. Higher Fusobacterium loads were found in the faeces of CRC patients than in that of controls (Fig. 6A). The Fusobacterium abundance among the total bacterial population in the faeces of CRC patients showed no association with the stage (Figs. 6B and S11) but displayed a weak but significant negative correlation with age (Fig. 6C, D). FrpoB-seq of 99 samples revealed 40 species with varied detection rates (Supplementary Data 3 and Supplementary Data 5), which were generally consistent with the distribution in tissue samples (Fig. 6E). More species were detected in the faeces of CRC patients than in that of the controls (Fig. 6F). The apparent feature of one-species predominance in the samples was maintained in the faeces (Fig. 6G). Compositional patterns at the species level between CRC and control samples could still not be differentiated. In addition, the compositions in faeces generally showed consistent patterns to those in the paired tissue samples, as observed in patients with available FrpoB-seq data for tissue and faecal samples (Figs. 6H and S12). In addition, consistent with the tissue-derived results, L1 was found at higher levels in terms of both percentage in the Fusobacterium community and relative abundance among the total bacterial population in CRC samples than in control samples (Fig. 6I, J). Importantly, L1 abundance exhibited predictive value for CRC (AUC = 0.862), performing better than the percentage or abundance of the canonical taxon F. nucleatum or the abundance of total Fusobacterium (Fig. 6K). These results further supported that L1 was associated the development of CRC.

Discussion
This study clarified the taxonomy of Fusobacterium, devised a Fusobacterium detection method that enabled accurate profiling of the species composition of gut Fusobacterium in CRC patients and revealed that Fusobacterium members may play distinct but lineage-specific roles in CRC.
Genotyping approaches have greatly facilitated the clarification of Fusobacterium taxonomy 3,14,15,17 . However, concerns have recently been raised regarding the taxonomic rank of the subspecies of F. nucleatum. Due to the sufficient divergence revealed by genomic analyses, several studies have proposed that the subspecies should be considered as separate species 14,15 . Our comprehensive taxonomic analyses conclusively indicate that the "subspecies" are indeed separate species. Although most studies, including our previous study 20 , regard F. nucleatum as one species, as all the subspecies have been isolated from and detected in CRC tumours and collectively display an association with CRC 3 , we show that the tumorigenic features do not justify grouping the subspecies as one species, since these features are actually exhibited by a wider spectrum of species belonging to the same lineage.
We further found that rpoB-based analysis could distinguish Fusobacterium species at fine resolution and applied this finding by developing the FrpoB-seq method, which allowed us to easily and stringently profile Fusobacterium species in microbiota-containing samples. Using FrpoB-seq, we were able to achieve high-depth profiling of this genus, which allowed us to uncover undescribed low abundance species. There were a striking number of previously undescribed species residing in the human gut. The diversity of Fusobacterium species in the gut microbiota is much higher than previously known. The compositions of gut Fusobacterium communities are seemingly stable within individuals but differed widely among individuals. This phenomenon implies that there might be multiple species involved in CRC, rather than one particular species. Indeed, we found that the L1 lineage which includes F. nucleatum, F. periodonticum and F. pseudoperiodonticum was overabundant in CRC, while L5 which includes F. varium and Fusobacterium ulcerans was associated with lymphovascular invasion. These results indicate that Fusobacterium members may have distinct but lineage-specific pathogenic behaviours in CRC, which might be attribute to the different distributions of putative invasion-associated virulence genes in theses lineages 15 . Thus, the L1 and L5 can be considered as different biomarkers or targets in CRC.
This research had some limitations. The suitability of FrpoB-seq for samples containing undetectable levels of or no Fusobacterium may be limited. Additionally, FrpoB-seq does not cover F. naviforme. However, the taxonomic position of F. naviforme remains in doubt 21 , and it may not be a member of the Fusobacterium genus and likely to be rare in the human gut 12   single-centre study focused on CRC. To expand the understanding of the distribution of Fusobacterium members in humans and their associations with CRC as well as other diseases, large-scale studies on different populations with different disease statuses are required.
In conclusion, with the FrpoB-seq approach, we provide a high-resolution view of the gut Fusobacterium communities in CRC patients, uncover a considerable number of undescribed Fusobacterium species and reveal that Fusobacterium members may play distinct lineage-specific roles in CRC. This study can help to clarify the association between gut Fusobacterium members and CRC and precisely probe the oncogenic culprit. We envision that FrpoB-seq will serve as a useful tool aiding future studies.   Bacterial strains. The bacterial strains used in this study are listed in Table S1. Nineteen Fusobacterium strains were isolated from tumour tissues as previously described 10 . Tumour tissues were placed in tryptic soy broth (#11104, Beijing SanYao) with 0.05% cysteine-hydrochloride (C7352, Sigma-Aldrich), disaggregated and plated on isolation media. Multiple media were used in parallel for isolation, including Crystal Violet Erythromycin Agar (CVE) 22 ; Fusobacterium egg yolk agar (FEA) 23 ; Fusobacterium Selective Agar (T1141, Rui Chu Biotech) supplemented with 5% defibrinated sheep blood (Shanghai HuaKang); josamycin/vancomycin/ norfloxacin (JVN) agar 24 ; brain-heart infusion agar (HB8478, Hope Biotech) supplemented with crystal violet, neomycin sulphate and vancomycin (8.75, 37.5 and 6.25 μg/ml, respectively, Sangon Biotech); and Columbia blood agar (Comagal Microbial). The 16S rRNA genes were sequenced with the Sanger method for species identification. All strains were cultivated on Columbia blood agar at 37°C in a jar containing an AnaeroPack System (C-1, Mitsubishi Gas Chemical).
Clinical samples. For Fusobacterium isolation, freshly resected tumour tissues were collected from ten CRC patients. For species-level Fusobacterium characterisation, fresh-frozen tumour and adjacent normal tissue samples from 551 primary CRC patients receiving surgical resection and faecal samples from 94 primary CRC patients receiving surgical resection (>18 years old, either sex) were retrieved from the Biobank of Shanghai Tenth People's Hospital. Among the patients, there were 29 who had both tissue and faecal samples collected and characterised. Faecal samples were also collected from 95 healthy volunteers. Patients who had received antibiotics or probiotics within one month prior to sample collection were excluded. Clinical and pathological information of patients was obtained from their medical records. Demographic information was collected from healthy volunteers.
Public bacterial genomic data. A total of 192 Fusobacterium genomes were retrieved from the NCBI database (ftp://ftp.ncbi.nih.gov/genomes/). Duplicated genomes or those not meeting the RefSeq criteria were excluded. Finally, 150 unique genomes, including 40 complete and 110 draft genomes, were included for analysis (Supplementary Data 1).
Average nucleotide identity analysis. The sequenced and downloaded genomes (n = 157) were subjected to pairwise whole-genome average nucleotide identity calculated with BLAST (ANIb) analysis with JSpecies v1.2.1 18 . Notably, each pair of genomes was examined twice, using one genome as the query and the other as the subject and vice versa, which might have generated two slightly different values due to the nature of the algorithm.
Multiple sequence alignment and phylogenetic analysis. Multiple sequence alignments were performed with MUSCLE 29 , and phylogenetic trees were subsequently constructed with MEGA5 30 using the maximum likelihood algorithm and 1000 bootstrap replicates. Whole-genome based phylogenetic analysis was conducted with kSNP3 31 .
Quantitative polymerase chain reaction. Total DNA from the faeces and tumour tissues was prepared with the cetyltrimethylammonium bromide (CTAB)-based method. The abundance of Fusobacterium was quantified by quantitative polymerase chain reaction (qPCR) with the primers fuso-rsub-F2 (5′-GCCTCATTTTH YTDGARTTYCAATT-3′) and fuso-rsub-R2 (5′-ACDACTCTTTCHGCCCATT KHHAT-3′), which were designed in this study to target the Fusobacterium rpoB gene. The reference was the bacterial 16S rRNA gene, which was quantified with a combination of five primers used together (5′-CNACGCGAAGAACCTTANC-3′, 5′-ATACGCGARGAACCTTACC-3′, 5′-CTAACCGANGAACCTYACC-3′, 5′-CA ACGCGMARAACCTTACC-3′, and 5′-CGACRRCCATGCANCACCT-3′) 32 . Primers were synthesized by Generay Biotech (Shanghai). qPCR was conducted with a TB Green Premix Ex Taq II kit (RR820, Takara) on a 7500 Real-Time PCR system (Applied Biosystems) according to the manufacturer's instructions. In each reaction, 90 ng of DNA was used. The thermal cycling conditions were 50°C for 2 min, 95°C for 30 s, and 35 cycles of 95°C for 5 s, 50°C for 30 s and 70°C for 30 s. If Fusobacterium was not detected, 45 cycles were then applied. A ΔCt value was calculated for each sample. Human blood DNA from healthy volunteers and total DNA of F. nucleatum subsp. nucleatum ATCC 25586 were used as negative and positive controls, respectively. A faecal sample tested to be negative for Fusobacterium was also used as a negative control. To assess the amplification performance of fuso-rsub-F2/R2, standard PCR was performed with a Premix Taq™ kit (RR903, TAKARA) using the conditions same as above. The primers Fn-F (5′-CA ACCATTACTTTAACTCTACCATGTTCA-3′ and Fn-R (5′-GTTGACTTTACAG AAGGAGATTATGTAAAAATC-3′) widely used for F. nucleatum detection were also used for comparison 5,33 . For qPCR, a repeat was made for each sample, while for standard PCR, experiment was performed in triplicate.
Fusobacterium rpoB amplicon sequencing. FrpoB-seq was conducted with the primers fuso-rsub-F2 and fuso-rsub-R2 at BGI, Shenzhen, following a standard amplicon sequencing procedure. For amplification, 30 ng of DNA extracted from a clinical sample was used in each reaction. The PCR conditions were 95°C for 3 min; 35 cycles of 95°C for 30 s, 50°C for 45 s and 72°C for 45 s; and 72°C for 10 min. Highthroughput sequencing was performed on the Illumina HiSeq 2500 platform. Pairedend reads with an overlap of ≥15 bp and a mismatch rate of <0.1 were assembled with FLASH v1.2.11 34 . Operational taxonomic units (OTUs) were generated with 100% identity via USEARCH (https://drive5.com/usearch/). Multiple sequence alignment of the OTUs based on the rpoB amplicon sequences and the corresponding rpoB regions from the 157 genomes was conducted, followed by phylogenetic tree construction; this phylogenetic tree was used to annotate the OTUs at the species level. The percentages of the derived species in the Fusobacterium communities were calculated based on the FrpoB-seq data. To further estimated the relative abundance of each species in the microbiota, the FrpoB-seq-derived percentages were normalised to the relative abundance of total Fusobacterium determined by qPCR. Serially diluted (10×) mock samples containing nine Fusobacterium species and subspecies (Table S1) with available strains were firstly used to assess the minimum overall Fusobacterium load allowing successful construction of a standard sequencing library. Then, mock samples containing serially Fig. 6 FrpoB-seq identified similar Fusobacterium patterns in faeces and that lineage 1 abundance was predictive of CRC. A Relative abundance of total Fusobacterium among bacteria in faecal samples from colorectal cancer (CRC) patients and healthy controls. Mann-Whitney test. B Relative abundance of Fusobacterium in faecal samples from CRC patients at different stages. Kruskal-Wallis test was followed by Dunn's multiple comparison test. The p value of Kruskal-Wallis test is shown. C Relative abundance of Fusobacterium in faecal samples from CRC patients in different age groups separated by the median age. Mann-Whitney test. D Correlation analysis between the relative abundance of Fusobacterium in faecal samples and patient age. Spearman correlation test. E Fusobacterium species compositions were detected in the faecal samples from CRC patents and healthy controls. The black or blue histogram on the right indicates the detection rate of each species or defined lineage, respectively. F The number of detected species in faecal samples. Mann-Whitney test. G Distribution of the percentage of the most abundant species in each sample. H Fusobacterium species compositions in faecal samples and matching normal and tumour tissues. Species with a <5% proportion are not shown. L, lineage. I Overall Fusobacterium lineage compositions in the CRC and control faecal samples. J Percentages in the Fusobacterium community (upper heatmap) and estimated relative abundance among bacteria (lower heatmap) of the defined lineages compared between CRC and control. The scale in E is applicable to the percentage heatmap. Lineages with low detection rates were not included. The Mann-Whitney test followed by the Benjamini-Hochberg correction was used, and adjusted p values are shown. K Receiver operating characteristic (ROC) curve for predicting CRC. AUC, the area under the ROC curve. In brackets are the 95% confidence intervals. The data were obtained from the participants in (E). Pairwise comparisons of ROC curves were conducted with MedCalc (the DeLong method) followed by the Benjamini-Hochberg correction, and adjusted p values are shown. For (A)-(C) and (F), medians and interquartile ranges are shown. All statistical analyses are two-sided where applicable. Source Data are provided as a source data file and in Supplementary data 3.
diluted F. nucleatum subsp. nucleatum or F. varium supplemented by F. mortiferum were used to assess the sensitivity of FrpoB-seq. A mock sample containing defined percentages of the nine species and subspecies (Fig. S4B) was also used to validate the amplification performance.
Metagenomic sequencing and public metagenomic data. Metagenomic sequencing of faecal samples were conducted at BGI, Shenzhen, following a standard procedure. For library construction, 1 μg DNA was used. Highthroughput sequencing was performed on the BGI DNBSEQ platform at an average~10 Gbp depth. Contigs were assembled with MEGAHIT 35 , and then used to search for undescribed Fusobacterium species. A detailed pipeline is shown in Fig. S6. Publicly available metagenomic data were retrieved from the NCBI SRA database and the accessions were PRJNA557323 for the Southern Chinses population cohort and PRJNA678426 for the ultra-deep sequencing (Korean population) cohort 12,19 .
Statistics. Data were analysed with GraphPad Prism (version 5), SPSS (version 19) or R (version 4.0.2). The Mann-Whitney test was used for comparisons between two groups. The Wilcoxon signed-rank test was used to analyse paired data. For comparisons among multiple groups, the Kruskal-Wallis test followed by Dunn's multiple comparison test was used. To identify taxa that were differentially abundant between two groups from the FrpoB-seq data, the Mann-Whitney test followed by the Benjamini-Hochberg correction was used. The Spearman rank correlation test was used for correlation analysis. Pairwise comparisons of ROC curves were conducted with MedCalc (the DeLong method). A two-sided p < 0.05 was considered statistically significant.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data generated or analysed during this study are included within the paper, its Supplementary information/data files, and public repositories. The bacterial genome data generated in this study have been deposited in the NCBI GenBank database under the accession CP071099 for strain THCT5A4, CP071098 for THCT6B3, CP071097 for THCT7A2, CP071096 for THCT7E2, CP071093 for THCT14A3, CP071092 for THCT14E2 and CP071094-CP071095 for THCT15E1. Other raw sequencing data generated in this study have been deposited in the NCBI SRA database under the accession number PRJNA715828. Detailed information of 192 publicly available Fusobacterium genomes retrieved from the NCBI nucleotide database is listed in Supplementary data 1 with accession numbers included. The publicly available metagenomic datasets used were retrieved from the NCBI SRA database under the accessions PRJNA557323 and PRJNA678426. Source data are provided with this paper.