The cancer-associated microbiota are known to influence cancer development and progression, most notably for colorectal cancer (
1–
5). Unbiased genomic analyses have revealed an enrichment of
Fusobacterium nucleatum in human colon cancers and adenomas relative to noncancerous colon tissues (
6,
7). These observations have been confirmed in studies of multiple colon cancer patient cohorts from around the world (
8–
12). Increased tumor levels of
F. nucleatum have been correlated with lower T cell infiltration (
13); with advanced disease stage and poorer patient survival (
10,
11,
14); and with clinical and molecular characteristics such as right-sided anatomic location,
BRAF mutation, and hypermutation with microsatellite instability (
9,
12,
15).
Studies in diverse experimental models have suggested a pro-tumorigenic role for
Fusobacterium. Feeding mice with
Fusobacterium (
16–
18), infection of colorectal cancer cell lines with
Fusobacterium (
19–
21), and generation of xenografts derived from
Fusobacterium-infected colorectal cancer cell lines (
17) were all observed to potentiate tumor cell growth. Suggested mechanisms have ranged from enhanced tumor cell adhesion and invasion (
17,
19,
22) to modulation of the host immune response (
16,
23) to activation of the Toll-like receptor 4 pathway (
17,
20,
21). However, not all animal or cellular studies of
Fusobacterium have demonstrated a cancer-promoting effect (
24). A recent editorial has highlighted the importance of studying
Fusobacterium infection in colon cancer as a component of the diverse microbiota within the native tumor microenvironment (
25).
To investigate the role of
Fusobacterium and its associated microbiota in native human colorectal cancers, we analyzed five independent cohorts of patient-derived colorectal cancers for
Fusobacterium and microbiome RNA and/or DNA. Where technically possible, we performed
Fusobacterium culture and tested the effect of antibiotic treatment upon the growth of propagated patient-derived colon cancer xenografts. These cohorts (table S1) include: (i) 11 fresh-frozen primary colorectal cancers and paired liver metastases (frozen paired cohort); (ii) 77 fresh-frozen primary colorectal cancers with detailed recurrence information (frozen primary cohort); (iii) published data from 430 resected fresh-frozen colon carcinomas from The Cancer Genome Atlas (
26) (TCGA cohort), together with data from 201 resected fresh-frozen hepatocellular carcinomas from TCGA (
27); (iv) 101 formalin-fixed paraffin-embedded colorectal carcinomas and paired liver metastases (FFPE paired cohort); and (v) 13 fresh primary colorectal cancers used for patient-derived xenograft studies (xenograft cohort).
Using the frozen paired cohort, we tested whether we could culture viable
Fusobacterium species from primary colorectal carcinomas and corresponding liver metastases. Quantitative polymerase chain reaction (qPCR) analysis showed that 9 of 11 (82%) snap-frozen primary tumors (table S2) were positive for
Fusobacterium in the primary tumor [patients one through nine (P1 through P9)]; we could isolate
Fusobacterium species from 73% of these tumors (
n = 8 of 11 tumors; P1 through P8) (
Fig. 1A). In addition, we cultured
Fusobacterium species from two liver metastases (P1 and P2) from
Fusobacterium-positive primary tumors. Five metastatic specimens had inadequate amounts of tissue for culture but were positive for
Fusobacterium by qPCR (P3 through P7), for a total of seven primary-metastatic tumor pairs (64%) testing positive for
Fusobacterium by qPCR (
Fig. 1A). This finding extends previous results showing the presence of
Fusobacterium nucleic acids in hepatic and lymph node metastases of colon cancer (
7,
22,
28) to now demonstrate that viable
Fusobacterium species are present in distant metastases.
To address whether the same
Fusobacterium is present in primary cancers and metastases, we performed whole-genome sequencing of pure
Fusobacterium isolates from primary and metastatic tumors from two patients (P1 and P2). For both patients, the primary-metastatic tumor pairs harbored highly similar strains of
Fusobacterium, with >99.9% average nucleotide identity, despite the tissue being collected months (P2) or even years (P1) apart (
Fig. 1B and fig. S1). We cultured
Fusobacterium necrophorum subsp.
funduliforme from the primary colorectal tumor and liver metastasis of P1 and
F. nucleatum subsp.
animalis from the primary tumor and metastasis of P2. We also cultured other anaerobes, including
Bacteroides species, from the primary-metastasis pairs (table S3). Our finding of nearly identical, viable
Fusobacterium strains in matched primary and metastatic colorectal cancers confirms the persistence of viable
Fusobacterium species through the metastatic process and suggests that
Fusobacterium species may migrate with the colorectal cancer cells to the metastatic site.
To quantitate the relative abundance of
Fusobacterium and to evaluate the overall microbiome in the paired primary and metastatic tumors, we performed RNA sequencing of 10 primary cancers and their matched liver metastases from the frozen paired cohort (P1 to P6 and P8 to P11). PathSeq analysis (
29) of the RNA sequencing data showed that the same
Fusobacterium species were present, at a similar relative abundance, in the paired primary-metastatic tumors (
Fig. 1C, samples P1 to P6) and that the overall dominant microbiome was also qualitatively similar. In addition to
F. nucleatum and
F. necrophorum, primary cancer microbes that persisted in the liver metastases included
Bacteroides fragilis,
Bacteroides thetaiotaomicron, and several typically oral anaerobes such as
Prevotella intermedia and
Selenomonas sputigena (
Fig. 1C). In contrast, there was little similarity between bacterial sequences in the primary colorectal cancer and liver metastasis in the lone sample where
Fusobacterium was present in the primary cancer but not detected in the metastasis (
Fig. 1C, sample P8) or in the three samples with low or undetectable levels of
Fusobacterium in the primary cancer (
Fig. 1C, samples P9 to P11). Jaccard index analysis revealed a high correlation between the dominant bacterial genera in the primary tumor and metastasis for
Fusobacterium-positive pairs, but a low correlation between bacterial genera in the primary tumor and metastasis for
Fusobacterium-negative pairs (
Fig. 1D and fig. S2).
Targeted bacterial 16S ribosomal RNA (rRNA) gene sequencing on DNA from the 11 frozen paired samples confirmed that (i) Fusobacterium species are present in paired primary-metastatic tumors, (ii) the relative abundance of Fusobacterium is correlated between primary tumors and metastases, and (iii) the dominant microbial genera in the liver metastases correspond to those in the primary tumors, demonstrating microbiome stability between paired Fusobacterium-positive primary-metastatic tumors (P = 0.01) (fig. S3).
To investigate the relationship between Fusobacterium and cancer recurrence, we performed microbial culture and bacterial 16S rRNA gene sequencing in a blinded fashion on the frozen primary cohort of 77 snap-frozen colorectal cancers lacking paired metastases (n = 21 with recurrence, n = 56 without recurrence) (table S4). We discovered that 44 of 77 tumors (57%) had cultivable Fusobacterium species and 45 of 77 had >1% Fusobacterium relative abundance. We found no correlation between Fusobacterium load or culture with either recurrence or stable disease in this cohort (fig. S4).
To assess Fusobacterium persistence and its correlation with clinical parameters, we analyzed the 101 primary-metastasis pairs from the FFPE paired cohort (table S5). We found that 43% (n = 44 of 101) of primary colorectal cancers tested positive for Fusobacterium by qPCR and 45% (n = 20 of 44) of liver metastases arising from these primary tumors were Fusobacterium-positive (fig. S5A).
To determine the spatial distribution of
Fusobacterium in these tumors,
Fusobacterium RNA in situ hybridization (ISH) analysis was performed on five qPCR-positive primary-metastasis pairs from this cohort (table S6,
Fig. 2, and fig. S6). Both biofilm and invasive
F. nucleatum were observed in primary colorectal cancer (
Fig. 2, A to D). Invasive
F. nucleatum distribution was highly heterogeneous and focal, found in isolated or small groups of cells with morphology consistent with that of malignant cells and located close to the lumen and ulcerated regions.
F. nucleatum was also observed in glandular structures present in the tumor center and invasive margins, but to a lesser extent. In adjacent normal mucosa (when present),
F. nucleatum was exclusively located in the biofilm. In liver metastasis,
F. nucleatum was predominantly localized in isolated cells whose histomorphology is consistent with colon cancer cells (
Fig. 2, E to H), although occasional stromal
F. nucleatum could be observed as well. No
F. nucleatum was detected in the adjacent residual liver parenchyma.
Notably, none of the 57
Fusobacterium-negative primary colorectal tumors were associated with a
Fusobacterium-positive liver metastasis (
n = 0 of 57;
P = 0) (fig. S5A). Consistent with previous reports (
15), the presence of
Fusobacterium in paired primary tumors and corresponding metastases was enriched in metastatic cancers of the cecum and ascending colon (
n = 10 of 20
Fusobacterium-positive primary-metastasis pairs,
P = 0.002), (fig. S5B), whereas cancers that were
Fusobacterium-negative in both primary and metastatic lesions were more likely to be rectal cancers (
n = 29 of 57 of the
Fusobacterium-negative primary-metastasis pairs,
P = 0.016) (fig. S5B).
To assess the relationship between patient survival and
Fusobacterium presence in the primary cecum and ascending colon, we carried out PathSeq (
29) analysis on RNA sequencing data from the 430 primary colon adenocarcinomas in the TCGA cohort. Patients with cancer of the cecum and ascending colon exhibited worse overall survival than patients with non-cecum ascending colon cancer (
P = 0.01) (fig. S5C). Among patients with cecum and ascending colon tumors, we observed poorer overall survival in correlation with tumor
Fusobacterium load (fig. S5D) (
P = 0.004).
To determine whether
Fusobacterium is associated with primary liver hepatocellular carcinoma, we performed PathSeq analysis (
29) of RNA sequencing data from 201 primary liver tumors from the TCGA cohort. This analysis demonstrated that
Fusobacterium is rare in primary liver carcinomas and that the relative abundance of
Fusobacterium is significantly enriched in liver metastases arising from colorectal cancers compared with primary liver cancers (
P = 0.008) (
Fig. 1E).
PathSeq analysis of data from the TCGA cohort also confirmed that the microbes present in liver metastases of
Fusobacterium-positive colorectal carcinomas are similar to those associated with
Fusobacterium in primary colorectal carcinoma.
Selenomonas,
Bacteroides, and
Prevotella genera were shared between primary and metastatic colorectal cancers and also correlated with
Fusobacterium abundance in primary colon adenocarcinoma (
Fig. 1F, fig. S7, and table S7).
Given that metastatic colorectal carcinomas harbored cultivable Fusobacterium species, we wondered whether viable Fusobacterium could persist in xenografts from human colorectal cancers, which would provide a valuable model system for evaluating the effects of microbiota modulation on cancer growth. In a double-blinded approach, 13 fresh human primary colorectal tumors from the xenograft cohort were evaluated, by culture or qPCR, for the presence of Fusobacterium. In parallel, these tumors were implanted subcutaneously, by an independent investigator, into Nu/Nu mice to establish patient-derived xenografts (PDXs) (table S8). All five Fusobacterium–culture positive tumors resulted in successful xenografts (fig. S8), one of four qPCR-positive but culture-negative tumors gave rise to a successful xenograft, and none of the four Fusobacterium-negative tumors generated successful xenografts (P = 0.003). Tumor grade did not appear to significantly influence successful xenograft formation (P = 0.1) (fig. S9A), although we noted a modest association between Fusobacterium cultivability and high-grade tumors in this cohort (n = 4 of 5 tumors, P = 0.03) (fig. S9B).
Next, we sought to determine whether
Fusobacterium species would remain viable and stably associated with a xenograft. A PDX derived from an
F. nucleatum culture–positive colon cancer (COCA36) was passaged to xenograft generation F8 over 29 weeks and tested for
F. nucleatum. We cultured
F. nucleatum from this PDX for up to four generations and 124 days in vivo. All xenograft generations, from F1 through F8, were positive for
Fusobacterium by qPCR (
Fig. 3A). Additionally, we cultured other anaerobic bacteria, including
B. fragilis and
B. thetaiotaomicron, from both the primary tumor and PDXs. We further cultured
Fusobacterium species from PDXs generated from two additional patient tumors (table S9). qPCR and microbiome analysis of fecal pellets and oral swabs from the PDX-bearing animals were negative for
Fusobacterium species (fig. S10), arguing against the possibility of
Fusobacterium species arising from the endogenous murine microbiota.
To evaluate the overall microbiome stability and to identify bacteria that are persistently associated with the primary colorectal tumor and derived xenografts, we carried out unbiased total RNA sequencing followed by PathSeq analysis, which revealed that
F. nucleatum and other Gram-negative anaerobes, including
B. fragilis and
S. sputigena, persist in these PDX models for multiple generations (
Fig. 3B). The bacteria that persist within the PDX include the genera that we report to persist in distant-site metastases to the liver (
Fig. 1C) and that are enriched in
Fusobacterium-associated colorectal cancer from analysis of TCGA data (
Fig. 1F). Bacterial 16
S rRNA gene sequencing further confirmed the persistence of
Fusobacterium and co-occurring anaerobes in these primary colorectal tumors and derived xenografts (fig. S11).
Transmission electron microscopy showed that F. nucleatum isolates from both the primary colon carcinoma and PDX were invasive when incubated with human colon cancer cell lines HT-29 and HCT-116. Upon infection with F. nucleatum, we saw evidence of bacterial cells within vesicle-like structures in the cancer cell (fig. S12, A to C). We also observed evidence of bacterial adhesion and invasion in the respective patient xenograft tissue (fig. S12D).
Finally, we asked whether treatment of Fusobacterium-positive colon cancer xenografts with either (i) an antibiotic to which Fusobacterium is resistant or (ii) an antibiotic to which Fusobacterium is sensitive would affect tumor growth. We chose erythromycin as a resistant antibiotic because the F. nucleatum clinical isolates were resistant to high concentrations of erythromycin (minimum inhibitory concentration >25 μg/ml) (fig. S13A). After oral gavage of the Fusobacterium-harboring PDX COCA36, with erythromycin, we observed a slight decrease in tumor volume compared with mice treated with the vehicle control. However, erythromycin did not significantly affect the trajectory of tumor growth (P = 0.073) (fig. S13B), Fusobacterium tumor load (P = 0.98) (fig. S13C), or tumor cell proliferation (P = 0.3) (fig. S13D).
For a
Fusobacterium-killing antibiotic, we chose metronidazole because fusobacteria are known to be highly sensitive to this drug (
30). We then confirmed sensitivity of the
F. nucleatum isolate from PDX COCA36 (minimum inhibitory concentration < 0.01 μg/ml) (fig. S14). Because PDXs could not be generated from
Fusobacterium-negative primary tumors, we treated
Fusobacterium-free xenografts derived from HT-29 colon adenocarcinoma cells with metronidazole to assess whether metronidazole inhibits the growth of
Fusobacterium-negative colorectal carcinomas. This experiment revealed no significant change in tumor growth (
P = 0.88) (
Fig. 4A).
Finally, oral administration of metronidazole to mice bearing
Fusobacterium-positive PDXs resulted in a statistically significant decrease in the trajectory of tumor growth, compared with PDXs in mice treated with vehicle (
P = 0.0005) (
Fig. 4A). Treatment with metronidazole was associated with a significant decrease in
Fusobacterium load in the tumor tissue (
P = 0.002) (
Fig. 4B), as well as a significant reduction in tumor cell proliferation (
P = 0.002) (
Fig. 4C and fig. S15).
We have shown that (i) Fusobacterium is persistently associated with distant metastases from primary human colorectal cancers; (ii) invasive Fusobacterium can be detected in liver metastases by ISH; (iii) Fusobacterium co-occurs with other Gram-negative anaerobes in primary and matched metastatic tumors; (iv) Fusobacterium survives in colorectal cancer PDXs through multiple generations; and (v) treatment of a Fusobacterium-harboring PDX model with the antibiotic metronidazole decreases Fusobacterium load, cancer cell proliferation, and tumor growth.
The persistence of Fusobacterium and its associated microbiome in both metastasis and PDXs, as well as the ability of antibiotic treatment to reduce PDX growth, point to the potential of Fusobacterium, and its associated microbiota, to contribute to colorectal cancer growth and metastasis. On the basis of our observation that the dominant microbiome is highly similar in primary-metastatic pairs and the concordance of Fusobacterium strains found in primary tumors and paired metastases, we hypothesize that Fusobacterium travels with the primary tumor cells to distant sites, as part of metastatic tissue colonization. This suggests that the tumor microbiota are intrinsic and essential components of the cancer microenvironment.
Our results highlight the need for further studies on microbiota modulation as a potential treatment for
Fusobacterium-associated colorectal carcinomas. One concern is the negative effect of broad spectrum antibiotics on the healthy intestinal microbiota. Given that metronidazole targets a range of anaerobic bacteria, including co-occurring anaerobes that persist with
Fusobacterium, one would ideally want to develop a
Fusobacterium-specific antimicrobial agent to assess the effect of selective targeting of
Fusobacterium on tumor growth. Important questions raised by our findings are whether conventional chemotherapeutic regimens for colorectal cancer will affect the colon cancer microbiota and whether the microbiota will modulate the response to such therapies. A recent study, reporting that colorectal tumors with a high
Fusobacterium load are more likely to develop recurrence (
21), supports the concept that
Fusobacterium-positive tumors may benefit from anti-fusobacterial therapy. Our results provide a strong foundation for pursuing targeted approaches for colorectal cancer treatment directed against
Fusobacterium and other key constituents of the cancer microbiota.
Acknowledgments
This work was supported by NIH grants (R35 CA197568 to M.M., R35 CA197735 to S.O., K07 CA148894 to K.N., and R01 CA118553 and R01 CA169141 to C.S.F.) and Dana-Farber/Harvard Cancer Center GI SPORE P50 grant CA 127003 to C.S.F. (this grant also supports S.B., A.J.A., W.C.H., K.N., and S.O.). In addition, M.M. is supported by an American Cancer Society Research Professorship; W.C.H. is supported by the Hale Family Center for Pancreatic Cancer; C.S.F. is supported by the Project P Fund for Colorectal Cancer Research, Stand-up-to-Cancer (Colorectal Cancer Dream Team), the Chambers Family Fund for Colorectal Cancer Research, the Team Perry Fund, and the Clark Family Fund for GI Cancer Research; S.B. is supported by a Prevent Cancer Foundation Figdor Family Fellowship; G.S., N.M., E.E., P.N., and J.T. are supported by the Cellex Private Foundation; and P.N. is supported by the Banco Bilbao Vizcaya Argentaria Foundation. M.M. is on the Scientific Advisory Board of and holds stock in OrigiMed, a biotechnology company that provides sequencing information for cancer diagnostics. C.S.F. is on the Board of CytomX Therapeutics, a biotechnology company that is developing therapeutic antibodies for cancer, and is a paid consultant for Eli Lilly, Genentech, Merck, Sanofi, Five Prime Therapeutics, Merrimack Pharmaceuticals, Bayer, Agios Pharmaceuticals, Taiho Oncology, and KEW Group. M.M. and S.B. are inventors on U.S. Provisional Patent Application no. 62/534,672, submitted by the Broad Institute and Dana-Farber Cancer Institute, that covers targeting of Fusobacterium for treatment of colorectal cancer. All raw sequencing data from this study can be accessed at the National Center for Biotechnology Information (NCBI) under the bioproject PRJNA362951. Bacterial whole-genome sequences have been deposited at DNA Data Bank of Japan/European Nucleotide Archive/GenBank, with the following NCBI accession, GenBank assembly accession, and BioSample numbers, respectively: F. necrophorum subsp. funduliforme P1_CP patient P1 primary colorectal tumor (NPNF00000000, GCA_002761995.1, and SAMN07448029), F. necrophorum subsp. funduliforme P1_LM patient P1 liver metastasis (NPNE00000000, GCA_002762025.1, and SAMN07448030), F. nucleatum subsp. animalis P2_CP patient P2 primary colorectal tumor (NPND00000000, GCA_002762005.1, and SAMN07448031), and F. nucleatum subsp. animalis P2_LM patient P2 liver metastasis (NPNC00000000, GCA_002762015.1, and SAMN07448032).