Systematic review: Gut microbiota in fecal samples and detection of colorectal neoplasms

ABSTRACT

Colorectal cancer (CRC) is a leading cause of cancer morbidity and mortality. Dysbiosis in the gut microbiota may be associated with CRC. This systematic review focuses on differences in gut microbial community between people diagnosed with CRC or adenoma and healthy individuals using fecal samples, emphasizing non-invasive fecal microbiome models for CRC early diagnosis.

Nineteen studies were identified in a systematic literature search of Pubmed, Web of Science and ScienceDirect. Several bacteria were reported to differ in abundance between CRC and adenoma cases and healthy controls, with Fusobacterium the most common. Fecal multi-bacterial predictive models used to distinguish CRC patients from healthy controls had reported areas under the receiver operating curve (AUCs) in external validation populations of 0.68–0.77.

Though advanced sequencing techniques could in the future complement current non-invasive methods for CRC early detection, more studies with high statistical power, comparable and reproducible methods and external validation of predictive models are needed.

Introduction

Colorectal cancer (CRC), the third most common cancer worldwide, is a leading cause of cancer mortality.¹ With 5-year survival rates of almost 90% for CRC diagnosed at stage I² and slow development over many years,³ perspectives for secondary prevention of CRC and reduction in mortality by early detection and population screening are much better than for most other cancers.^4-6 Feces-based test kits, the guaiac-based fecal occult blood testing (gFOBT) and fecal immunological tests (FITs) are used in many countries⁷ for non-invasive early detection of CRC and advanced adenomas, a leading precursor of CRC.

At least 15% of cancer burden worldwide is attributable to known intestinal infectious agents such as Helicobacter pylori.⁸ In the case of CRC, studies demonstrated an association between a disruption in the gut microbial-community balance, or “dysbiosis”, and cancer development.^9,10 The increased relative abundance of harmful bacteria on the one hand, and the depletion of other health promoting beneficial bacteria on the other hand may create a microenvironment more favorable to the development of CRC.^11,12

Concentrating on the possible use of fecal gut microbiome as a non-invasive diagnostic tool in population screening for CRC, comparable to FIT and gFOBT, we aimed to systematically review the literature on differences in the composition of the gut microbial community between people diagnosed with CRC or adenomas and healthy individuals, and the possible use of these differences in models for non-invasive early diagnosis of CRC and advanced adenomas. Due to significant differences in the bacterial population extracted from tumor tissue and feces from the same individual,^13-15 this systematic review specifically focuses only on gut microbiome in fecal samples.

Literature search strategy

A systematic literature search was conducted independently by two of the authors (EA, AK) in order to identify original studies examining the differences in gut microbial community in fecal samples between people diagnosed with CRC or adenomas and people without colorectal malignancies. The online databases Medline, Web of Science and ScienceDirect were searched on April 17^th 2017 using the following search terms: (Fecal AND microbiota AND CRC) OR (Microbiome AND colorectal AND cancer) OR (Fecal AND microbiome AND colorectal AND cancer) OR (Gut AND microbiome AND colorectal AND cancer). On Web of Science the search was refined by “article” category. On ScienceDirect the search was limited to “Journals”, years 2007–2017 and topics: gut microbiota, colorectal cancer, gut, human, cancer, colon cancer and tumor.

Duplicate studies were excluded as were irrelevant publication types such as symposium agendas. Titles and abstracts were screened independently by EA and AK and full texts were reviewed by each with reference lists searched for additional relevant studies. This systematic literature review was conducted according to the PRISMA guidelines.¹⁶

Inclusion criteria

Studies were included if they met all inclusion criteria: examining the gut microbial community in fecal samples from CRC or adenoma cases compared to healthy individuals; using human feces; original research with kind of bacteria found as outcome. Where same cases and controls were included in more than one publication, only the publication offering the most detailed information useful for this review was included. Studies published in letter or abstract forms were excluded as they did not include enough data for our review. The reviewers independently selected studies based on the inclusion criteria and disagreements were discussed until agreement was reached. Where studies included data from several sources (feces, tissue, mice) only fecal data analysis was included.

Data extraction and quality assessment

Included studies were evaluated for quality of information independently by EA and AK based on the Newcastle-Ottawa Scale,¹⁷ designed to assess the quality of case control studies, with some necessary amendments [Table 1]: a criterion for collecting fecal samples before preparation for colonoscopy was added in the place of the original criterion “ascertainment of exposure” since some previous studies demonstrated that bowel preparation for colonoscopy changes the gut microbial community.^18,19 We also added a criterion for statistical adjustment for multiple testing or correction for over-optimism. Thus, the number of quality criteria amounts to eight and the maximum point score to nine since the “comparability of cases and controls” criterion can receive up to two points.

Table 1.

Quality of included studies based on Newcastle-Ottawa Scale.

	Selection				Comparability of cases and controls		Exposure
First author/Pub. Year/ country (ref)	Cases validated by colonoscopy	Representative Cases	Community controls	Control status validated by colonoscopy	Age	Gender	Same exposure analysis for cases and controls	All fecal samples collected before bowel preparation	Adjustment for multiple testing / Correction for over optimism	Total (out of 9)
					−	−				5
Yu 2017, China41	+	+	+	+	−	+	+	−	+	7
Flemer 2017, Ireland15	+	[no info]	−	−	+	+	+	[no info]	+	5
Hale 2016, USA40	+	+	+	+	+	+	+	+	+	9
Baxter 2016 USA / Canada39	+	+	[no info]	+	−	−	+	−	+	5
Kasai 2016, Japan38	+	+	+	+	+	+	+	+	−	8
Mira-Pascual 2016 Spain36	+	+	+	+	−	−	+	+	+	7
Fukugaiti 2015, Brazil37	+	[no info]	[no info]	+	−	+	+	+	−	5
Goedert 2015, China35	+	+	+	+	+	−	+	+	+	8
Feng 2015, Austria34	+	+	+	+	+	+	+	[no info]	+	8
Zeller 2014, France20	+	+	+	+	+	+	+	+	+	9
Zackular 2014 USA / Canada33	+	+	+	+	+	−	+	−	−	6
Wu 2013, China (Beijing)32	+	+	[no info]	+	+	+	+	[no info]	[no info]	6
Weir 2013, USA31	+	+	[no info]	+	−	−	+	+	−	5
Ohigashi 2013, Japan30	+	+	−	+	+	+	+	+	+	8
Kostic 2013 USA / Scotland29	+	+	+	+	[no info]	[no info]	+	+	[not relevant]	6
Ahn 2013, USA (DC)23	+	+	−	+	+	+	+	+	+	8
Chen 2013, China28	+	+	+	+	+	+	+	[no info]	−	7
Wang 2012 China (Shang)27	+	+	−	+	−	+	+	+	−	6

Literature search results

The literature search process, shown in a PRISMA flow diagram in Fig. 1, yielded a total of 1111 records. After exclusion of 198 duplicates, titles and abstracts of 913 studies were screened for relevance. Studies deemed not relevant for the review topic (n = 755), published only as a short communication or in abstract form (n = 61) and non-English publications (n = 26) were excluded. The full texts of 71 studies were read independently by EA and AK and further 52 studies were excluded for not meeting all the inclusion criteria.

Figure 1.

PRISMA flow diagram.

Three studies included participants from the same French cohort and therefore only the report including the most detailed information needed for this review was included²⁰ while the other two were excluded.^21,22 Two other studies shared the same American cohort, therefore only one was included²³ and the other was excluded.²⁴ A study by Vogtmann et al²⁵ recently performed whole-genome shotgun sequencing on fecal samples from an American population that took part in the study by Ahn et al,²³ validating their results on a French population included in a study by Zeller et al,²⁰ and thus was not included in this review. One randomized clinical trial²⁶ was excluded since the study design involved giving the participants synbiotics which made it incomparable to the other selected studies. Although no design criterion was applied, all included studies were case-control studies. In total, nineteen studies were included in the qualitative analysis of the systematic review.

Study populations and study quality

The nineteen included studies^{15,20,23,27-42} were published between 2012 and 2017 [Table 2]. Eight studies^{20,23,28,30,34,38,40,42} reported when data and samples were collected, ranging between 1985 and 2015 with a gap of 2 to 28 years between collection of fecal samples and publication year. Fecal samples used for microbiome analysis ranged between 15 (8 CRC cases) in the smallest study by Weir et al³¹ to 490 (120 CRC cases) in the largest study by Baxter et al³⁹ A study by Hale et al⁴⁰ included 780 participants, 223 of whom were diagnosed with adenoma, but this study did not include any CRC cases. Although sample size is not a criterion in the Newcastle-Ottawa scale, it is a highly relevant factor for achieving statistical power and precision with larger studies potentially representing better the microbial community variety in the population. Eleven studies included a separate group of participants diagnosed with adenomas, joined in the analyses to either the CRC group,^33,36 the control group^20,30 or analyzed separately.^{20,28,29,33-35,39,40,42}

Table 2.

Characteristics of studies included in the systematic review.

		Study participants (males) Age mean ± sd / median (range)
First author/Pub. Year/Country	Data collection years	CRC	Adenoma	Healthy	Reason for colonoscopy	Stool collection	Temp. stool storage	Antibiotic use prior to stool sample	Database used for taxonomy assignment	Microbiome Analysis Method	Groups compared for bacterial differences	Statistical analysis for bacterial differences	Relevant design issues
		203 (117)	97 (50)	236 (87)
		Mean 66.5	60.5 ± 4.7	Mean 58
Yu 2017 China-Hong Kong 41	—	74 (48)Median 67 (34–89)	—	54 (33)Median 62 (50–73)	Symptoms + screening	Standard containers at home	−20°C	Not in 3 months	IMG	16S rRNAqPCR	CRC / healthy	Non parametric tests	More samples in cases taken after colonoscopy.
Flemer 2016 Ireland 15	—	43 (30)Median 63 (41–90)	—	37 (21)Median 68 (64–81)	Scheduled for colonic resection	Self-sampled	−80°C	Not 1 month	RDP, V.14	V3 16S rRNAqPCR	CRC / healthy & adenoma / healthy	Non parametric tests	Possible effects of bowel preparation or neoadjuvant therapy.
Hale 2016 USA 40	2001 – 2005	—	233 (133)66.5 ± 6.9	547 (310)66 ± 6.9	Screening	Self-collected: no buffer	−80°C within 48 hours	Not excluded	RDP	16S rRNA	adenoma / healthy	Machine learning
Baxter 2016, USA / Canada 39	—	120 (68)63 ± 12.9	198 (118)63.3 ± 11.5	172 (61)54 ± 9.9	Asymptomatic	Whole evacuated stool	−80°C	[no info]	RDP	V4	CRC / healthy & adenoma / healthy	Machine learning	Some samples collected 1–2 weeks after colonoscopy.
										16S rRNA
Kasai 2016 Japan 38	2012 – 2013	6(4)Mean 54	—	6(4)Mean 50	various reasons	[no info]	+4°C	Current antibiotics excluded	Techno Suruga Lab Microbial Identification	16S rRNA	CRC / adenoma / healthy	Non parametric tests
									DB-BA9
Mira-Pascual 2016 Spain 36	—	7771.14 (56–82)	8 (7)61.75 (41–85)	9 (5)53.78 (18–69)	Any indication + familial screening	[no info]	−80°C	[no info]	RDP	16S rRNAqPCR	CRC + adenoma / healthy	Non parametric tests
Fukugaiti 2015 Brazil 37	—	7(5)65 ± 1.1	—	10(8)55 ± 1.3	Various reasons	[no info]	−80°C	excluded	[not relevant]	qRT-PCR	CRC / healthy	Parametric tests
Goedert 2015, China 35	—	—	20 (12)	24 (7)	Screening after positive FIT	Self-collected at home	−20°C;	included	RDP	V3 – V4	adenoma / healthy	Non parametric tests	Antibiotics in last 6 m in adenoma as confounder
							Moved weekly			16S rRNA
							−80°C
Feng 2015 Austria 34	2010–2012	46 (28)	47 (23)	63 (37)	symptoms + screening	[no info]	−20°C for 48 h;	Not in 3 months	IMG	16S rRNA	CRC / adenoma / healthy	Non parametric tests
		67.1 (43–86)	66.5 (48–84)	67.1 (43–78)			−80°C
Zeller 2014 France / Denmark / Spain / Germany 20	2004–2006	53 (29)	42 (30)	61 (28)	F: Referral	[no info]	F: −20°C in 4 h	[no info]	KEGG	16S rRNA (for 129 partic.)	CRC / adenoma / health	Non parametric tests	CRC patients were significantly older on average
		67 (58.5–75.5)	64.1 (53–71)	63.0 (56–67)	G: recruited after diagnosis		G: −80°C				Or CRC / Adenoma + healthy
							H: −20°C then
							−80°C
Zackular 2014, USA / Canada 33	—	30 (21)59 ± 11	30 (18)61 ± 11.1	30 (11)55 ± 9.2	[no info]	Self-collected at home	−80°C	[no info]	RDP, V.9	V416S rRNA	healthy / CRC or adenoma, or CRC + adenoma	Non parametric tests	Stool samples collected 1–2 weeks after colonoscopy
Wu 2013 China 32	—	19 (10)58 ± 8.7	—	20 (11)53 ± 5.4	Cases: CRC patients	[no info]	Directly−80°C	Not in 3 months	RDP, V2.3	16S rRNAqPCR	CRC / healthy	Non parametric tests
Weir 2013 USA 31	—	8 (8)	—	7 (3)	[no info]	Self-collected	−20°C	Not in 2 months	Green Genes	V4	CRC / healthy	Parametric tests
		66.4 (47–84)		35.86 (24–67)						16S rRNA
Ohigashi 2013 Japan 30	2009–2010	938 ± 12.1	22 (11)67 ± 9.2	27 (16)65 ± 13.5	Recruited at hospital	Collected at hospital	+4°C	Current users excluded	[not relevant]	RT-qPCR	CRC / healthy + adenoma	Non parametric tests
Kostic 2013 USA / Scotland 29	—	27	28	31	[no info]	[no info]	[no info]	[no info]	[not relevant]	qPCR	CRC / healthAdenoma / healthy	Non parametric tests	Only Fusobacterium. Samples from EDRN of NCI.
Ahn 2013 USA (DC) 23	1985–1989	47 (28)	—	94 (56)	Recruited at hospital	Self-collected at home	−40°C	Not in 1 year	IMG / Green Genes	V3 – V4	CRC / healthy	Non parametric tests
		62.9		58.5						16S rRNA
Chen 2013 China 28	2010–2011	—	47 (24)58 ± 11	47 (24)57 ± 11	Various reasons	Self-collected at home	−80°C	Not in 6 months	Silva database	16S rRNAqPCR	CRC + adenoma / healthy	Parametric and Non parametric tests	Adenoma and healthy groups matched for age and gender.
Wang 2012 China 27	—	46 (24)Median 60 (42–77)	—	56 (27)Median 49 (40–54)	[no info]	[no info]	−80°C	Cases: no current. Controls: not in3 m	RDP	16S rRNAqPCR	CRC / healthy	Parametric and Non parametric tests

^*Abbreviations: CRC – colorectal cancer; qPCR – quantitative polymerase chain reaction; IMG – Integrated Microbial Genome; RDP – Ribosomal Database Project; KEGG – Kyoto Encyclopedia of Genes and Genomes; EDRN – Early Detection Research Network; NCI – National Cancer Institute.

All studies used frozen stool samples except for Ohigashi et al³⁰ and Kasai et al³⁸ who used stool samples stored in a temperature of +4°C before microbiome analysis. In population-wide CRC screening programs using stool-based tests, analysis is done on fresh stool samples within days of collection. These conditions are rarely possible in case-control studies, where stool may be frozen and stored for a few years before analysis and where collection methods, storage temperatures and duration before analysis of fecal samples may vary widely and have a differentiating effect on the results of microbiome analysis.^43-45

Fecal bacteria may also be affected by bowel preparation for colonoscopy, though evidence is somewhat limited.^13,19,20 Eleven studies indicated that fecal samples were taken before colonoscopy preparation and in one study the samples were taken 1–2 weeks after colonoscopy, arguing it is enough time for the gut microbial community to return to its pre-preparation state. Yu et al adjusted for timing of fecal sampling as a confounding factor in analyses since in a significantly higher number of cases fecal samples were collected after surgery.⁴¹

Antibiotics taken by study participants may also affect the microbiome. Five studies did not address antibiotics taken by the participants.^{20,29,33,36,39} In the 14 studies that addressed it, one mentions it as a limitation of the study that no data was available on antibiotics consumption,⁴⁰ three studies excluded participants that were taking antibiotics at the time of recruitment and the others excluded participants that used antibiotics before recruitment for a period ranging between one month to one year. In establishing the case and control status of participants, six studies did not state ascertainment that the controls had no history of CRC or adenoma diagnosis in the past.^{15,23,27,34-36} Thus, vast variation in most aspects of the studies limits the ability to synthesize the results together or compare the individual study results.

Bacteria detection

Bacterial detection method in three studies was qPCR alone for Quantitative Real-Time amplification of bacterial DNA,^29,37,42 ten studies used the 16S rRNA sequencing and 6 studies utilized both 16S rRNA and qPCR in analyzing the samples [Table 2]. Three studies^20,23,34 reported statistically significant differences in the gut microbial composition also after adjustment for multiple testing.

The studies used different reference databases for the assignment of taxonomy and different statistical methods to analyze possible differences between participants with CRC, adenoma or healthy controls with regards to their gut microbial community properties with most using non-parametric statistical tests and two^39,40 using machine learning.

Table 3 includes the bacteria found in at least two studies to significantly differ in abundance between cases and control groups. Two studies^20,34 have also looked at enterotypes, a classification of three distinct bacterial clusters used as a general descriptive measure of the gut microbiota.⁴⁶ There were also some contradicting findings, such as the genus Streptococcus that was found by Chen et al²⁸ and by Wang et al²⁷ to be more abundant in people diagnosed with adenomas and CRC respectively, while Feng et al³⁴ found this genus to be significantly more abundant in the healthy control group. Based on included studies, it seems there is no consensus as to which specific fecal bacteria differentiate the gut microbiome of people diagnosed with CRC or adenoma compared to healthy controls. Nonetheless, members of the Fusobacteria phylum, mainly the genus Fusobacterium, were the most common bacteria, reported by 10 studies, to be significantly more abundant in CRC cases compared to healthy controls.

Table 3.

Bacteria found in significantly different abundance in CRC, adenoma and controls in at least two studies.

Bacteria taxonomic level				Reported to be more abundant in:
Phylum	Order/Family	Genus	Species	CRC	Adenoma	Controls
Enterotype: Bacteroides				20,34	34
Firmicutes						20,23,40
	Clostridiales					23,33
	Lachnospiraceae			33,34		23
		Roseburia				27,28
			Roseburia intestinalis			20,42
	[Eubacteriaceae]	Eubacterium				20,27,28,41
			Eubacterium ventriosum			20,41
	[Peptoniphilaceae]	Parvimonas		20,34,41
			Parvimonas micra	39,41
	[Peptostreptococcaceae]	Peptostreptococcus		20,27,41
			Peptostreptococcus stomatis	39		41
	[Ruminococcaceae]	Ruminococcus				20,34
	[Veillonellaceae]	Selenomonas		20,23
	[Streptococcaceae]	Streptococcus		27	28	34
Fusobacteria				20,32,38,41
	[Fusobacteriaceae]	Fusobacterium		20,23,29,32-34,38,41	29
			Fusobacterium nucleatum	20,37,39,41,42	42
	[Leptotrichiaceae]	Leptotrichia		20,41
[Bacteroidetes]	Porphyromonadaceae			32	33
		Porphyromonas		20,23,27,33
	[Bacteroidaceae]	Bacteroides		20,34		27,28,33
[Actinobacteria]	[Coriobacteriaceae]	Atopobium		30,38
	[Bifidobacteriaceae]	Bifidobacterium				20,30,34
[Proteobacteria]	[Desulfovibrionaceae]	Bilophila		20,34

^*In Zackular et al³³ the findings are OTUs associated with the family/genus. Bacteria in square brackets were not reported on this level and are there for reference.

Multi-bacteria models for early detection of CRC

Eight studies included multi-bacteria predictive models that may potentially complement established tests such as gFOBT or FIT as a non-invasive early detection tool for CRC and its precursors [Table 4]. All but one study, by Liang et al,⁴² reported only point estimates of the area under the receiver operator characteristic curve (AUC) values found and did not report 95% confidence intervals (CIs) for these values. Furthermore, the reported AUCs are for the most part derived from the training (derivation) populations, and are therefore prone to overoptimism due to overfitting. The few AUCs derived from validation sets are marked in Table 4 with asterisks.

Table 4.

Multi–bacteria models for early detection of colorectal neoplasms.

		Population examined	Comparisons and Outcome AUC
Authors Year Country	Model includes	Cohort = C / A / H	C vs H	C vs A	C+A vs H	A vs H	C vs A+H	External validation
Liang 2017 China42	Fusobacterium nucleatum	I = 170/ NA / 200	0.87					Validated in independent Chinese cohort II
		II = 33 / NA / 36	0.68			0.61
	4 bacteria	I = 170/ NA / 200	0.89
		II = 33 / NA / 36	0.76
		I = NA / 97 / 200
Yu 2017 China / Denmark / Austria / France42	4 gene markers	C2 = 47 / NA / 109	0.73					Test: C2-Chinese
		A = 41 / NA / 55	0.77					Validation: D-Danish
		F = 53 / NA / 61	0.72					A-Austrian F-French
	20 gene markers	D = 16 / NA / 24	0.71					—
	2 gene markers	C2	0.84					—
Hale 2016 USA40	4 genera	NA / 233 / 547				0.66		—
Baxter 2016 USA / Canada39,47	23 OTUs + FIT result	I = 120 / 198/ 172	0.95		0.83	0.76		II – sequencing of buffered stool from FIT tubes
	28 OTUs	II = 101 / 162 / 141	0.83		0.69
	41 OTUs
Goedert 2015 China35	Rank relative abundance of 18 phylum-level taxa with and without sex	NA/20/24				0.77		—
	Rank relative abundance of 64 order-level taxa	NA / 20 / 24				0.77		—
Feng 2015 Austria34	15 MLGs	5 / 47 / 8					0.96	Internal Validation set also includes patients from test set
		46 /5/ 8			0.59
	10 MLGs + age	46 / 5 / 8			0.59
Zeller 2014 France / Denmark / Spain / Germany20	22 species	F = 53 / 42 / 61	0.84					Test: F-French
		H + F						Validation: Danish, Spanish & H-Healthy German
		H = NA / NA / 297	0.85					G-CRC German
		G + F	0.90
		G = 53 / NA / NA
Zackular 2014 USA / Canada33	6 OTUs + age + race + BMI	30 / NA / 30	0.92					—
	4 OTUs + gFOBT + BMI	30 / 30 / NA		0.97				—
	5 OTUs +age + race + BMI	NA / 30 / 30				0.90		—
	6 OTUs + age + gender + race	30 / 30 / 30			0.94			—

C- colorectal cancer; A – adenoma; H – healthy; AUC – area under the receiver operating characteristic curve; (g)FOBT – (guaiac) fecal occult blood test; BMI – body mass index; MLGs – metagenomics linkage groups; OTUs – operational taxonomic units.

^*AUC in independent validation populations.

Liang et al⁴² found that Fusobacterium nucleatum alone had in their training population an AUC of 0.87 (95%CI 0.83, 0.90). A model of four bacterial markers including Fusobacterium nucleatum offered better diagnostic performance with an AUC of 0.89 (95%CI 0.85, 0.92).

Baxter et al³⁹ used 23 different operational taxonomic units (OTUs) with FIT and received an AUC of 0.95 for the detection of CRC, albeit the result was not significantly higher than the result obtained using FIT alone (p-value = 0.90) in the same population. In addition, Baxter et al, using DNA from buffered stool from participants' FIT cartridges, constructed a model utilizing 28 OTUs with an AUC of 0.83 for differentiating healthy controls from those with CRC.⁴⁷

For distinguishing between adenoma and CRC groups, Zackular et al,³³ found that the gFOBT alone had an AUC = 0.62, while using a microbiome-based model with four OTUs had a significantly higher performance with AUC of 0.95 in the training population. However, independent validation remains yet to be performed.

Multi-bacteria models for early detection of adenoma

Lower performance was generally reported for classification models distinguishing adenomas from healthy controls. Liang et al⁴² used Fusobacterium nucleatum only model and had AUC = 0.61 (95%CI 0.55–0.67). A combination of FIT and 23 OTUs revealed in one study AUC of 0.76³⁹ and in another study the AUC of a model combining 10 metagenomic linkage groups (MLGs) and age was 0.59.³⁴ The model by Hale et al,⁴⁰ including 4 genera: 2 enriched in adenoma and 2 enriched in controls, had an AUC of 0.66 for predicting the existence of adenomas in their study population. In another study, classification models using 18 phylum-level taxa or 64 order-level taxa showed an AUC of 0.77 for both models.³⁵ Remarkably higher performance of the classification model including 5 OTUs plus age, race and BMI for detection of adenomas was demonstrated by Zackular et al (AUC = 0.90).³³

Both for detection of CRC and of adenomas the researchers used different bacteria, including some as of yet unnamed, at different taxonomy levels to comprise their models and there is not one common bacterial genus or species that is included in all the models. In addition, apart from in the study by Liang⁴² et al, no additional information such as sensitivity, specificity or other diagnostic/prognostic accuracy parameters of the models were included in the publications.

External validation of the multi-bacteria models for early detection

External validation of the multi-bacterial models was done in three studies. The model by Yu et al,⁴¹ with 2 case-enriched and 2 control-enriched microbiomic gene markers, gave an AUC of 0.73 for detecting CRC compared to healthy controls in the Chinese test population. The model was tested also in independent validation Austrian and French cohorts, with similar AUC values of 0.77 and 0.72 in the latter two respectively.

Zeller and colleagues²⁰ created a CRC classifier consisting of 22 species using data from a French cohort and received an AUC of 0.84. The same model revealed AUCs of 0.85 and 0.90 in the French population (156 participants) combined with either one German population consisting of 297 healthy individuals or another consisting of 38 CRC cases, respectively. Liang et al⁴² verified their two models, one containing Fusobacterium nucleatum only and the other a combination of four bacterial markers, in a separate smaller Chinese validation cohort and received lower AUCs of 0.68 (95%CI 0.55, 0.80) and 0.76 (95%CI 0.64, 0.87), respectively.

Discussion

Nineteen studies examining the difference in gut microbial community composition in human fecal samples between people diagnosed with CRC or adenoma and healthy individuals were included in this systematic review. The studies used different analysis protocols at different taxonomic levels to identify significant bacterial differences therefore enabling only limited comparisons. The most common bacteria found in eight of the studies to be in relatively higher abundance in CRC cases compared to healthy controls was the genus Fusobacterium.

The relatively high incidence of CRC and the high potential for survival when detected early have prompted a growing number of countries to adopt national screening programs. The most widely used non-invasive screening tool is the gFOBT but due to its relatively low sensitivity it is increasingly being replaced by more sensitive FITs.^48-50 Using colonoscopy for CRC screening, with detection and removal of adenomas, could prevent most CRC cases,⁵¹ but both adherence of the target population and available colonoscopy capacity remain limiting factors.^52-54

Fecal gut-microbiome for early detection of CRC

Advanced sequencing techniques to identify gut bacteria in feces may have in the future the potential to complement current non-invasive methods for CRC early detection and risk stratification. Eight studies included in this analysis developed bacteria based models. Some of these studies reported very high AUCs of up to 0.95 for discriminating CRC patients from healthy participants in the study population in which the models were trained. Most likely, however, these results are far too overoptimistic due to overfitting, as suggested by much lower AUCs in the order 0.68–0.77 that were found in the two studies in which the derived algorithms were validated in independent validation samples. Such models may nevertheless potentially be useful for stratification of the population for their cancer risk level based on the composition of their gut microbiome.

Fecal gut-microbiome for detection of colorectal adenomas

Several studies in this systematic review have also identified statistically significant differences in the abundance of certain bacteria between people with adenomas compared to healthy controls. Five studies reported AUCs of 0.61–0.90 for their bacteria based model for discerning participants with adenomas from healthy participants. Regrettably, none of these models was validated in an external validation set. The reported bacterial differences between carriers of adenomas and healthy controls are more likely to reflect a consequence rather than a bacterial cause of tumorigenesis given the much more limited structural changes in the large bowel resulting from adenomas. Flanagan et al⁵⁵ have found an over-abundance of F. nucleatum in cancerous compared to matched normal tissue (p < 0.0001) but Fusobacterium levels in colorectal adenomas were not significantly higher compared to healthy controls (p  =  0.06). Ito et al⁵⁶ found that F. nucleatum positivity was significantly higher in CRCs (56%) than in premalignant lesions of any histological type (p < 0.0001) and its positivity increased according to histological grade. Amitay et al⁵⁷ found that no statistically significant difference was detected in fecal prevalence of F. nucleatum between carriers of advanced and non-advanced adenomas and healthy controls while F. nucleatum was much more commonly found in the CRC group compared to all above groups. Moreover, within the CRC cases in this study, abundance of Fusobacterium was positively associated with more advanced cancer stages. In addition, in the current review, most included studies reported more pronounced differences in bacterial relative abundance between participants with CRC and healthy participants than between participants with adenomas and healthy participants [Table 3]. Moreover, the AUCs for the bacterial models used to distinguish CRC cases from healthy participants are significantly higher than those used to distinguish carriers of adenomas from healthy participants [Table 4].

Fusobacterium and CRC

Fusobacteria, the bacteria reported most commonly to differ between CRC patients and healthy controls in the included studies, are common residents in the human oral microbiome and are only found infrequently in the gut. Studies have shown that CRC tumors and their metastases possess specific glycans that underlie fusobacterial tumor enrichment⁵⁸ and a recent study reported that Fusobacterium was found in both primary and matched metastatic tumors and that treatment with the antibiotic metronidazole decreased Fusobacterium load, cancer cell proliferation and tumor growth.⁵⁹ There are several hypotheses as to how Fusobacterium and other bacteria may be contributing to CRC tumorigenesis. Rubinstein et al⁶⁰ found that F. nucleatum adheres to and invades colonic cells and induces oncogenic and inflammatory responses to stimulate growth of CRC cells through its unique FadA adhesion, a virulence factor facilitating the bacterium's adherence to cell surface. Kostic et al,²⁹ using a mouse model, found that through recruitment of tumor-infiltrating immune cells Fusobacteria generate a pro-inflammatory microenvironment that is conducive for colorectal neoplasia progression. Gur et al⁶¹ found in a mouse model that F. nucleatum-containing tumors inhibited NK cell cytotoxicity and tumor infiltrating lymphocyte cell activity via the interaction of the F. nucleatum protein Fap2 with human TIGIT (T cell immunoglobulin and ITIM domain). Mima et al⁶² found that a greater amount of F. nucleatum in human colorectal carcinoma tissue was associated with a lower density of CD3⁺ T cells in tumor tissue. These interactions of Fusobacterium with the tumor environment and with the immune system, though unable yet to provide concrete evidence of causality, shed light on the association between gut bacterial dysbiosis and the development of colorectal tumors and may further explain the association of the amount of F. nucleatum DNA in colorectal cancer tissue with shorter survival.⁶²

Synthesis of results and sources of bias and heterogeneity

The ability to draw definite conclusions from the reviewed studies is limited by the fact that although the study of the microbiome using high-throughput DNA sequencing is a rapidly evolving field, the absence of a gold-standard unified protocol leads to great heterogeneity in methodology and makes direct comparisons or potential pooling of results across studies not easily possible. Varying regions sequenced, depths of sequencing, methods and protocols for the extraction of the bacterial DNA, sample handling (collection, transportation, storage, time to analysis), varying PCR primers used for 16S microbial analysis, varying methods for assignment of taxonomy and different statistical approaches for data analysis (varying from established epidemiological methods to innovative machine-learning modelling) are all significant potential sources of bias and heterogeneity.^63-65

In addition, several recent reviews indicate that there is no “universal human gut microbiome” and that there are significant variations in the gut microbiota due to differences in ethnicity, geographic location, lifestyle, westernization and nutrition^66-69 though others could identify a small “core microbiota” found in the majority of their study population though here too the microbial community composition in the stool was significantly influenced by various exposures such as stool consistency and medication use.⁷⁰ This variety in the healthy microbiota may make it harder to find a unified list of bacteria to be included in a microbiome-based stool test for CRC and adenoma detection. In light of these findings, it is possible that emphasis should be put not only on the presence of specific bacteria but also on the absence of other bacteria, bacterial functioning in the gut environment and the relationship between the microbiota and the host immune system.^67,71

Nonetheless, as seen in the study by Yu et al,⁴¹ the same multi-bacteria model for discerning CRC from healthy controls and containing 4 gene markers which was tested independently in Chinese, Austrian and French populations gave similar results (AUCs 0.72–0.77). This indicates that although different populations from different geographical areas may have different gut microbial communities at baseline^72,73 it may be possible to develop feces based multi-bacterial models for the detection of colorectal neoplasms that may be used in a variety of settings and populations.

Tissue vs fecal-derived bacteria

As noted before, there are significant differences in the bacterial population extracted from tumor tissue and from feces.^13-15 This may explain why several prominent bacteria such as Streptococcus gallolyticus,^74,75 found in tissue analysis to be associated with CRC, are missing from the studies included in this review. While tissue microbial community analysis may contribute to a better understanding of the mechanisms by which the gut bacteria play a role in CRC carcinogenesis and progression and may be relevant from a therapeutic biomarker perspective, it is not relevant for using microbiome-based tests as an inexpensive non-invasive early diagnostic or risk-stratification tool to complement the currently-used FIT or gFOBT in population based screening programs.

To our knowledge this is the first review to systematically and comprehensively review the results and methodology of epidemiological studies looking into the differences in the gut microbial community as manifested in fecal samples between people diagnosed with CRC or advanced adenoma and people with normal colonoscopy results and their potential use for non-invasive early diagnosis of CRC and adenomas. Given that this is a young emerging field of research, with more than half of identified studies published from 2015 onward, there may be additional recent studies whose results have not been published as full articles at the time of the search.

In conclusion, based on the currently published data, there is limited but encouraging evidence that differences in fecal gut microbiome between people diagnosed with CRC or adenoma and persons without colorectal malignancy may be used to develop new, non-invasive and inexpensive fecal tests that could complement the repertoire of current non-invasive CRC screening tools on their own or in combination with established fecal occult blood tests. While it remains unclear to what extent these differences in the gut microbiome precede and contribute the occurrence of CRC, distinct patterns in the gut microbial community of CRC patients may still contribute to CRC early detection even if they should not be causally related to CRC development. Future research should focus on developing unified documented and reproducible protocols for studying the human gut microbiome from fecal samples so that results can be more comparable and conclusions can be drawn on a larger basis. These models should be externally and independently validated in ethnically and geographically diverse populations. A better knowledge of the changes in the composition of the gut microbial community during CRC development may not only assist in better early diagnosis and identification of those at high risk for developing CRC but may also help identifying novel CRC preventive approaches by modulating the gut microbiome through nutrition, life-style changes or more targeted approaches such as pro- or synbiotic therapy.

Funding Statement

No funding.

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.