INTRODUCTION
Colorectal cancer (CRC) is the third most common cancer diagnosed worldwide. In 2012, CRC accounted for 1,361,000 new cases and 694,000 deaths globally (
1), while in the United States alone, 95,520 new cases and 50,260 deaths were recorded in 2017 (
2). The colonic microbiota may contribute to the development of colorectal cancer (
3,
4). Early studies, based on culture methods, indicated an association between a few bacterial species and CRC or healthy tissues (
5,
6). An increased risk for colon polyps was associated with increased abundances of
Bacteroides and
Bifidobacteria organisms, whereas
Lactobacillus and
Eubacterium aerofaciens were associated with the absence of tumors.
Over the past decades, the widely used 16S rRNA gene-based metataxonomics, the development of metagenomic methods based on next-generation sequencing (NGS) technology, and improved bioinformatic tools for big data analysis have afforded in-depth descriptions of the microbial compositions and functions of the gut microbiota. This has allowed for detailed comparison between feces specimens from CRC patients and those of healthy controls (
7–14) and for comparison between tumor-associated microbes and microbes associated with tissues not adjacent to tumors (
15–20). Among the diverse set of bacterial taxa identified in these studies,
Fusobacterium,
Bacteroides fragilis, and
Parvimonas micra were found to be consistently associated with the tumor tissues (
21,
22), while other taxa, like
Providencia (
23),
Roseburia,
Ruminococcus and
Oscillibacter (
24), and
Streptococcus gallolyticus (
25,
26) were found associated with tumor tissues in some, but not all, studies. Of note, one study described an association between a decreased relative abundance of
Roseburia in fecal samples of CRC patients compared to that in healthy volunteers (
27), in contrast to findings in tissue, further pointing to inconsistencies between fecal and tumor-associated microbiotas in CRC patients (
28). It was also shown that microbial compositions differ between right (ascending, proximal) and left (descending, distal) cancerous colonic mucosas, with higher relative abundances of
Prevotella,
Selenomonas, and
Peptostreptococcus in right colonic tumors and with higher abundances of
Fusobacterium,
Escherichia-
Shigella, and
Leptotrichia in left colonic tumors (
29). In addition, right-side tumors were marked by the presence of a bacterial biofilm, unlike left-side tumors (
21,
30,
31). Interestingly,
Fusobacterium was associated with right colonic tumors in one study (
32) and with both right and left colonic tumors in another (
21). Moreover, it was shown that the tumor-associated microbiota could vary with the stage of the tumor (
33).
Utilizing silver nitrate staining and a combination of laser-capture microdissection (LCM) and amplification of the 16S rRNA gene followed by deep sequencing, we previously showed that murine proximal colon crypts harbor a resident microbiota that we call crypt-specific core microbiota (CSCM). Regardless of the mouse line and breeding origin, this bacterial population is unexpectedly homogeneous and dominated by a restricted diversity of strictly aerobic genera, such as
Acinetobacter,
Delftia, and
Stenotrophomonas (
34). The aim of the present study was to investigate, using LCM technology and 16S rRNA gene sequencing, if human colonic crypts also harbor a consistent core microbiota and if CRC is associated with a dysbiotic core microbiota. Consequently, we characterized both the crypt-associated microbiota (CAM) and the mucosa-associated microbiota (MAM) in tumors and in their paired adjacent normal tissues in samples collected from the right and left colons of CRC patients. We compared these microbiotas with those associated with colonic biopsy specimens of healthy volunteers. Our results showed that, regardless of health status, human colonic crypts are colonized mainly with
Firmicutes but are also colonized with
Acinetobacter,
Delftia, and
Stenotrophomonas; however, they are in lower relative abundances than in murine proximal colon crypts. Nonfermenting
Proteobacteria were also detected in these samples. The proportions of bacteria previously shown to be associated with CRC were differentially represented in tumoral crypts from right and left colonic samples. For instance,
Fusobacterium and
Bacteroides fragilis were abundant in tumors from the right colon, whereas
Parvimonas micra was associated with tumors from the left colon. In healthy samples,
Faecalibacterium was more abundant in right than in left colonic crypts. Taken together, our results demonstrate the existence of a human CSCM and point to a specific localization of bacteria previously associated with CRC. The presence of an abnormal microbiota in colonic crypts is hypothesized to be linked to CRC oncogenesis, but further studies are needed to explore this association.
DISCUSSION
Most studies addressing the relationship between the microbiota and CRC have been based on an analysis of the bacterial compositions of either fecal samples or, to a lesser extent, tumoral tissues and their paired adjacent normal tissues. In the present study, we performed an in-depth analysis of the bacterial compositions in colonic crypts and mucosa-associated compartments in both CRC patients and individuals with normal colonoscopy results. First, we confirmed that colonic crypts in healthy volunteers are colonized by bacteria. While this result supports earlier published findings (
45), another group reported that no bacteria were found in colonic crypts of healthy individuals (
46). The combination of LCM and 16S rRNA gene sequencing applied in this study allowed for the identification of bacterial species colonizing the crypts as well as those associated with the mucosa. Interestingly, the three genera (
Acinetobacter,
Delftia, and
Stenotrophomonas) which we previously described as the crypt-specific core microbiota in murine proximal colon crypts (
34) were also retrieved in human normal colonic crypts, and OTUs associated with these genera belong to the common core microbiota. These bacteria were also found in crypts and mucosa-associated regions from tumor sites and their paired adjacent normal tissue, and we did not observe a significant decrease in the abundance of
Acinetobacter in tumoral samples compared to that in adjacent normal tissue, unlike results reported by other studies (
29,
47). Instead, we identified additional Gram-negative, aerobic, nonfermentative, environmental
Proteobacteria (such as
Ralstonia and
Acidovorax) that inhabit human colonic crypts. It is interesting to note that
Proteobacteria, such as
Comamonadaceae,
Moraxellaceae,
Pseudomonadaceae, and
Xanthomonadaceae, represent 31.9% and 35.2% of CAM and MAM, respectively, whereas the abundances of these phyla represent only 1% of the fecal microbiota (
48). Taken together, the present study was able to extend the concept of a human crypt-specific core microbiota (CSCM), which has so far been established only in rodents. Like its murine counterpart, human CSCM is marked by the strong representation of strictly aerobic and facultative anaerobic taxa that are likely to be part of a coevolutionary symbiosis driven by the necessity of maintaining a stable ecosystem in the vicinity of the crypt, a critical epithelial regenerative apparatus (
49).
In addition, this study shows that this CSCM is broadly conserved in tumoral and normal tissues adjacent to the tumor as wells as in control individuals but is marked by the added presence of pathobionts, i.e., commensal bacteria with pathogenic potential (
50), which are highly abundant at the expense of the bona fide CSCM in both tumoral crypts and the associated epithelium. We also found, despite the low number of patients and low abundances, a statistically significantly higher proportion of
S. gallolyticus (formerly
Streptococcus bovis) in tumoral samples than in the adjacent nontumoral samples, in accordance with the results of previous publications (
51,
52). We also found an increase in the abundance of
B. fragilis in tumoral crypts of right colons, even though this bacterium was also present in normal paired samples.
B. fragilis involvement in the oncogenic process leading to CRC (
53) was recently confirmed (
21,
54).
B. fragilis enterotoxin was found to induce reactive oxygen species (ROS)-dependent DNA damage, degradation of the tumor suppressor E-cadherin, and increased expression of Wnt, leading to cell proliferation. Similarly,
B. fragilis and
S. gallolyticus were described as “alpha-bugs,” which by themselves or via microbiota modification can drive the oncogenic process (
55). While
E. coli strains expressing the polyketide synthase (pks) island, encoding DNA-damaging colibactin (
56–58), have also been associated with CRC, this was not confirmed in the present study, as the presence of
E. coli was found in only a very low number of patients, without significant differences between tumoral and nontumoral samples. Similarly, we did not find
Providencia or
Shigella, two
Proteobacteria genera previously shown to be associated with the CRC tumor environment (
4,
23). In agreement with our results, other studies did not find an increase in
E. coli or
Shigella in cancer groups versus normal patients (
3) or a decrease in the abundance of
Escherichia-
Shigella (
29) or
Shigella (
15) in cancerous versus noncancerous tissues. As a matter of fact, fecal microbiotas of mucosa-associated bacteria significantly differ; hence, fecal samples cannot be considered an adequate reflection of the microbiota attached to mucosal surfaces (
24,
40,
41,
59). This needs to be considered before establishing bacterial biomarkers of early stages of CRC.
Interestingly, right-side and left-side tumors appear to differ in their dominant pathobionts, with
Fusobacterium isolates prevailing in the former and
P. micra in the latter, both belonging to a decompartmentalized oral microbiota that appears as a hallmark of cancer and inflammatory conditions of the gut (
14,
21,
60–64). The association of oral bacteria with tumors is not restricted to colorectal cancer; indeed
P. micra and
P. stomatis were found to be enriched in patients with gastric cancer (
65). Moreover, some oral pathogens have been associated with a higher risk of pancreatic cancer (
66). A “driver-passenger” model was proposed to explain the bacterial interactions occurring in CRC (
67). In this model, driver bacteria are involved in the initiation phase of CRC and are then replaced by passenger bacteria that promote tumorigenesis. In other terms, a first change in the gut microbiota allows for oral bacteria to colonize the gut mucosa, and by the disruption of the epithelial barrier, these bacteria promote oncogenesis (
60). Many studies mention an association between
F. nucleatum and CRC (
56). Interestingly, our study identified with the HITdb
F. periodonticum as associated with CRC. However, by further characterization using
rpoB sequencing, we confirmed that
F. nucleatum subsp.
polymorphum is actually the species associated with CRC. This result is in line with those of previous phylogenetic studies based on 16S rRNA gene sequence analysis reporting either that
F. periodonticum is indistinguishable from
F. nucleatum (
68) or that these two species are very close (
69,
70). Nonetheless,
F. periodonticum has also been previously identified in colon tumors (
21,
71).
F. nucleatum was proposed as a driver bacterium through its ability to adhere to and invade epithelial cells through its FadA adhesin, followed by an increase in the production of ROS, transcription factors, Wnt, and inflammatory proteins that stimulate the growth of CRC cells (
72). Moreover, it was shown that microsatellite instability was present in the right colon but that chromosomal instability was more frequent in the left colon, possibly due to differences in microbiota composition (
56,
73).
One limitation of the present and similar studies analyzing low-biomass samples, such as colonic crypt microdissected structures, is the risk of potential environmental contaminations leading to the misinterpretation of a result if one does not apply alternative techniques to 16S rRNA analysis, such as FISH or cultivation, both being impossible to apply to the entire array of 16S rRNA-identified taxa in a complex structure where bacteria are possibly weakly metabolically active. For instance, recent studies on the microbial compositions of placenta and amniotic fluid, which are expected to be poorly, if at all, colonized, indicate that 16S rRNA gene sequencing does not reveal differences in bacterial composition between samples and technical controls (
74,
75). As a matter of fact, the main sources of contamination stem from reagents, such as extraction buffers and PCR reagents, or even from cross-contamination between samples. One approach may be to remove the sequences present in negative-control samples. However, low levels of real sequences may be present in the negative controls due to cross-contamination, and the removal of such sequences may result in loss of relevant biological signals (
76,
77). Alternatively, one could remove the potential contaminants following 16S rRNA gene sequencing by deleting the sequences previously identified as contaminants in the literature. This might, however, arbitrarily eliminate relevant signals because the sequences were identified as contaminants on the basis of different extraction methods and PCR reagents. Some sequences assigned to phylotypes described in the literature as environmental contaminants, such as
Agrobacterium tumefaciens,
Pseudomonas monteilii, and
Chryseobacterium, were detected in our study but not necessarily together in the same samples (
Table S1B). Another known environmental contaminant is
Acinetobacter. However, using a probe specifically targeting the 16Sr RNA of this genus, we could visualize the presence of
Acinetobacter in our colonic human samples, thereby providing the necessary alternative method required to confirm the actual presence of the bacterium. As a reminder, in a previous study bearing on murine colonic crypts, we had been able to confirm the presence of
Acinetobacter not only by FISH but also by culture (
78). This demonstrates that
Acinetobacter is likely to be a universal commensal of colonic crypts following harnessing from environmental sources. In summary, in these studies of low-biomass samples, one is faced with the difficult exercise of extracting relevant signals from among contaminating noise that cannot be rationally eliminated. Confirmation by alternative approaches, such as FISH and culture, brings indisputable validation. The issue is more difficult for taxa that do not benefit from these alternative techniques. One can consider that environmental taxa with high signal levels that share common metabolic properties with
Acinetobacter (i.e., strictly aerobic, nonfermentative) may also be considered relevant. One therefore needs to perform a critical analysis on the basis of the above-described criteria but cannot at once delete all signals without reflection.
We are also conscious that a second limitation of the study was the low number of biopsy specimens from healthy volunteers; however, this does not alter the conclusions based on the comparison between tumoral and adjacent nontumoral samples from cancerous patients. In conclusion, based on this decompartmentalization and the differential microbial patterns present in the distinctive colon segments, the role of these pathobionts, such as Fusobacterium and Bacteroides fragilis, awaits further evaluation, as they may reveal key elements on the mechanisms involved in colon oncogenesis.
ACKNOWLEDGMENTS
We thank the study participants for their collaboration, Nathalie Jolly from the Pôle Intégré de Recherche Clinique at the Institut Pasteur for her help with the ethical statement of this study, Laurence Motreff for her help with Illumina library preparation and the MiSeq runs, Katja Brunner for editing the manuscript, and Armand Sobhani and Jeremy Gaudez for their contribution during their internship at the PMM unit during the course of their studies.
This work was supported by the European Research Council (PJS advanced grant 339579-DECRYPT), by the Inserm cross-cutting program Microbiota, and by a Danone research grant. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
We have no conflict of interest to declare.
T.P. designed the research, performed the research, analyzed the data, and wrote most of the manuscript; C.M. and B.R. performed the research and analyzed the data; A.S. performed bioinformatic analyses of the data and helped write the manuscript; A.A., J.T.-V.-N., and J.R. analyzed the data; and I.S. and P.J.S. designed the research, analyzed the data, and helped write the manuscript.