Application and development of noninvasive biomarkers for colorectal cancer screening: a systematic review

Background:

Colorectal cancer (CRC) is the second most common cause of cancer-related death (9.4% of the 9.9 million cancer deaths). However, CRC develops slowly, and early detection and intervention can effectively improve the survival rate and quality of life. Although colonoscopy can detect and diagnose CRC, it is unsuitable for CRC screening in average-risk populations. Some commercial kits based on DNA mutation or methylation are approved for screening, but the low sensitivity for advanced adenoma or early-stage CRC would limit the applications.

Main results:

Recently, researchers have focused on developing noninvasive or minimally invasive, easily accessible biomarkers with higher sensitivity and accuracy for CRC screening. Numerous reports describe advances in biomarkers, including DNA mutations and methylation, mRNA and miRNA, gut microbes, and metabolites, as well as low-throughput multiomics panels. In small cohorts, the specificity and sensitivity improved when fecal immunochemical testing combined with other biomarkers; further verification in large cohorts is expected. In addition, the continuous improvement of laboratory technology has also improved the sensitivity of detection technology, such as PCR, and the application of CRISPR/Cas technology. Besides, artificial intelligence has extensively promoted the mining of biomarkers. Machine learning was performed to construct a diagnosis model for CRC screening based on the cfDNA fragment features from whole-genome sequencing data. In another study, multiomics markers, including cfDNA, epigenetic, and protein signals, were also discovered by machine learning. Finally, advancements in sensor technology promote the applicability of volatile organic compounds in CRC early detection.

Conclusion:

Here, the authors review advances in early detection and screening of CRC based on different biomarker types. Most studies reported optimistic findings based on preliminary research, and prospective clinical studies are ongoing. These promising biomarkers are expected to more accurately identify early-stage patients with CRC and be applied in the future.

Introduction

Highlights

• Commercial DNA mutations or methylation kits showed low sensitivity in adenoma and early-stage colorectal cancer (CRC) screening.

• Combined with the progress made in wet and dry lab, cfDNA fragment features, or multiomics are developed for CRC screening.

• Microorganisms in fecal and volatile organic compounds in breath are promising biomarkers.

• A combination of multiomics biomarkers showed both high specificity and sensitivity.

• CRC detection performance improved in a fecal immunochemical test when combined with other biomarkers.

According to the Global Cancer Statistics 2020, colorectal cancer (CRC) ranks as the third most commonly diagnosed cancer and the second most common cause of cancer-related death1. The 5-year survival rates range from 90% for stage I disease to 12% for stage IV disease2. Therefore, the detection of early-stage CRC will lead to a higher survival rate. In addition, patients with early-stage disease may avoid adjuvant therapy after surgery and enjoy a higher quality of life2.

Most CRCs arise from adenomatous polyps or sessile serrated lesions, and since these alterations may slowly develop over ~5–10 years3, CRC is preventable if polyps are excised and proper treatment is administered. Moreover, CRC screening would decrease the risk of developing CRC by 44%, as reported by a prospective study4. Although the benefits of screening are obvious, ~33% of adults aged 50–75 are not screened as recommended5.

CRC screening guidelines differ by country, including the starting ages and screening strategies. However, colonoscopy and fecal immunochemical tests (FITs) are the most commonly mentioned for CRC screening6. The FIT test is relatively cheap and easy to complete at home, but the sensitivity (75% at 20 µg/g and 91% at 10 µg/g) and specificity (95% at 20 µg/g and 90% at 10 µg/g) for CRC diagnosis were affected by the threshold of FIT positivity7. Besides, FIT tests may miss treatable precursor lesions with their lower sensitivity (40% at 10 µg/g and 25% at 20 µg/g) in advanced adenomas (AAs) detecting7. On the other hand, although the colonoscopy was used as the gold standard for CRC diagnosis, its use in CRC screening has some limitations. Since colonoscopy requires full bowel cleansing and must be performed by professional technicians at a hospital, this costly, complex, and invasive procedure may limit patient access to the test and result in poor compliance rates.

The ideal screening method with high sensitivity and specificity would be low cost, easy to perform, and noninvasive or minimally invasive. Biological samples, such as stool or blood, that can be collected in a noninvasive and user-friendly manner will facilitate screening tests. In this review, we summarize the promising biomarkers under investigation.

DNA methylation

Stool-based test

Since colorectal adenomas and early cancers both occur in the intestinal tract and are in direct contact with feces, extracted tumor cells are mixed in the stool, which has strong tissue specificity and contains a significantly higher amount of tumor DNA.

Syndecan-2 (SDC2) is an oncogene in CRC and promotes the epithelial-mesenchymal transition. The expression of SDC2 was upregulated with advanced stage, vascular invasion, lymph node metastasis, and distant metastasis7. The aberrant SDC2 methylation frequently occurs in all stages of CRC8 and can be sensitively and specifically detected in stool and blood samples from CRC patients9–11.

SDC2 methylation is a major target in stool-based tests. One case–control clinical trial (NCT03146520) enrolled 585 participants, including 245 with CRC, three with AAs, and 245 healthy controls. The SDC2 methylation in fecal samples was analyzed using quantitative methylation-specific real-time PCR following linear target enrichment (LTE-qMSP) (EarlyTectColon; Table 1). Compared with colonoscopy as the gold standard, SDC2 methylation can detect stage 0–IV and stage I–II CRC with sensitivities of 90.2 and 89.1%, respectively, with a specificity of 90.2%10. Clinical studies (NCT04304131) to expand the sample size are ongoing15. The other two trials investigating SDC2 methylation in stool also reported considerable clinical results. In one trial (Chi-CTR-TRC-1900026409), 1110 Chinese participants (359 with CRC, 38 with AAs, and 512 healthy controls) were enrolled, and SDC2 methylation in stool could detect CRC, stage I–II CRC, and AAs with sensitivities of 83.3, 87.0, and 42.1%, respectively, with a specificity of 98.0%18. In another trial, 339 participants (102 with CRC, 50 with AAs, and 130 normal individuals) were enrolled, and two fragments of SDC2 CpG islands (SDC2-A and SDC2-B) were detected by qPCR (quantitative real-time PCR). The sensitivity of this test for CRC and AAs improved to 87.25 and 52%, respectively, while the specificity was 94.62%16,19. Large prospective screening cohorts should be studied to analyze factors, such as geography, age, and sex, that may affect test performance.

Table 1.

The indications and performance of commercial CRC screening kits.

Producer	Application	Sample	Population	Biomarkers	Sensitivity	Specificity	References
Exact Sciences	Screen	Stool	Average-risk adults 50–80 years old	Promoter methylation: NDRG4 BMP3; 7 gene mutations in KRAS gene; FIT	92.3% for CRC; 42.4–46.2% for AA	89.1–89.8%	12–14
Genomictree	Screen	Stool	Average-risk adults	SDC2 methylation	68.6–90.2% for CRC	90.2-91.7%	10,15–17
Creative Biosciences	Auxiliary diagnosis	Stool	Average-risk adults	SDC2 promoter methylation	83.8% for CRC; 42.1% for AA	98.10%	18
AmoyDx	Auxiliary diagnosis	stool	Average-risk adults	SDC2 methylation	1) CRC: SDC2-A:85. 29%; SDC2-B: 83.33% combine: 87.25% 2) AA: SDC2-A: 48%; SDC2-B: 52% combine: 52% 3) CRC (0–IV):87.52%	Normal control: SDC2-A: 96.15%; SDC2-B: 97.69%; combine: 94.62%; 2. CRC (0–IV): 94.62%	19
Epigenomics AG	Screen	Blood	Average-risk adults >50 years old	Methylation of Septin 9	75–79.3% for CRC; 27% for AA	96.7–98.9%	20–25
BioChain	Auxiliary diagnosis	Blood	–	Methylation of Septin 9	73–79.3% for CRC; 9.8–17.1% for AA	94.5–95.9%	26–28

CRC indicates, colorectal cancer.

Blood-based tests

SEPTIN9, a tumor suppressor gene, functions in cell division and cell cycle regulation29. The methylation in the promoter of SEPTIN9 would inhibit its expression and promote the progression of CRC30. The plasma-circulating mSEPT9 is derived from tumor-derived apoptotic cells31, and the abundance of SEPT9 methylation in the peripheral blood of CRC patients increases with the advanced pathological stage20.

The Epi proColon 2.0 assay (Epigenomics AG) (Table 1) is a qualitative assay for PCR-based detection of methylated SEPTIN9 from blood and is the FDA-approved (2016) blood test for colon cancer screening in average-risk adults older than 50 years. In early-stage CRC, the sensitivity of the Epi proColon 1.0 assay ranged from 48.2 to 73%, while the specificity ranged from 78.8to 91.5%20–22. The updated protocol used three separate PCR replicates and changed the positive threshold, which highly improved the sensitivity from 48.2–73% to 75–79.3% with a specificity of 96.7–98.9% for CRC detection. However, the sensitivity of detecting AA was 27%23,24. When verified in a Chinese cohort, the sensitivity of the new SEPTIN9 assay ranged from 73 to 79.3% and the specificity ranged from 94.5 to 95.9%, but the sensitivity was still lower for AA (9.8–17.1%)26,30 [NCT02540850]. These results indicated that the SEPTIN9 assay detected more late-stage CRC cases than early-stage CRC cases. However, the combination of the SEPTIN9 assay and FIT could improve sensitivity to 94.4–98% with a specificity of 69%26,27.

Another approved multilocus blood-based assay, ColonAiQ (Singlera Genomics), contains six methylation markers (SEPTIN9, SEPTIN9 region2, BCAT1, IKZF1, BCAN, VAV3). The sensitivity of ColonAiQ for the detection of CRC and AA was 86.7 and 42%, respectively, while the specificity was 92%, the sensitivity increased with advancing CRC stage and increasing tumor size. ColonAiQ has a higher sensitivity compared with FIT (88.3 vs. 59.7%) for all stages, especially for stage I (85.7 vs. 28.6%)32. The results of a prospective real world study in an average-risk population were reported at the 2022 AACR Annual Meeting. This trial enrolled 21 735 healthy volunteers aged from 40 to 80 years. ColonAiQ identified 1630 (7.5%) positive individuals, and of those, 103 were asymptomatic (four with CRC and 99 with AA) and were diagnosed by coloscopy. However, the positive predictive value is primarily estimated to be 2.2% for CRC and 15.2% for AA33.

cfDNA fragment features

The cfDNA fragment profile, including fragment sizes, end motifs, preferred ends, single-stranded jagged ends, and nucleosomal footprints, is referred to as ‘fragmentomics’34. The fragmentomic signatures in cancer patients are different from those in healthy individuals35; for example, fragments are shorter in length (<145 bp) in tumor-derived cfDNA from cancer patients36.

In a prospective study, Peng and colleagues identified five cfDNA fragment features by five machine learning models and developed a model to identify CRC or AA in healthy participants37 (Table 2). At a 94.8% specificity, this model discriminated early-stage CRC with a high sensitivity of 98.0% (95% CI: 94.2–99.6%) and differentiated AA with a sensitivity of 95.7% (95% CI: 85.2–99.5%). Even at 1X coverage, the sensitivity reached 94.4% (95% CI: 90.1–97.2%) with a specificity of 94.8%37. This cfDNA fragmentomics model is thus a promising approach for early CRC detection and screening.

Table 2.

The promising biomarkers for CRC early detection and screening in development.

Rank	Biomarkers	Biospecimen	Platform	Cohorts (train/validation)	Model	AUC	Specificity	Sensitivity	Location	References
1	Six methylation markers (SEPT9, SEPT9 region2, BCAT1, IKZF1, BCAN, VAV3)	Blood cfDNA	qPCR	CRC: 138/173; AA: 62/107; Benign and non-AA: 58/91; Healthy: 90/136	CRC	0.93	91.9%	86.1%	China	32
					AA	–	–	42.1%
				PreC study: 21,735 apparently healthy volunteers aging from 40 to 80 (median 61.8)	PPV is estimated as 2.2% for CRC, 15.2% for AA, 43.5% for adenoma	China				33
2	19 mutations of 4 genes (APC, KRAS, CTNNB1, BRAF)	Blood cfDNA	Xenonucleic acid (XNA)-mediated qPCR clamping technology	CRC: 137/–; precancerous lesions: 58/–	CRC	–	100%	92.2%	Chinese patients	38
					Precancerous lesions	–	95.2%	62.5%
3	5 cfDNA fragmentomic features: Fragment Size Ratio (FSR); Fragment Size Distribution (FSD); End Motif (EDM); BreakPoint Motif (BPM); Copy Number Variation (CNV)	Plasma cfDNA	WGS	Early-stage CRC: 149/149; AA: 46/46; Healthy: 115/116	CRC and AA vs. Healthy	0.988	94.8% (89.1–98.1%)	97.4% (94.1–99.2%)	Shanghai, China	37
					CRC vs. Healthy	0.99	94.8% (89.1–98.1%)	98.0% (94.2–99.6%)
					AA vs. Healthy	0.982	94.8% (89.1–98.1%)	95.7% (85.2–99.5%)
4	mRNA panel: LGALS4, CEACAM6, TSPAN8, and COL1A2	Whole blood	qRT-PCT	High-risk CRC (HR/CRC): 92; low risk (LR, polyps <10 mm): 36; Healthy (N): 67; Negative colonoscopy FIT positive (NFIT): 36	LR vs. Healthy	0.91	97%	79%	Italy	39
					HR/CRC vs. Healthy	0.88	87%	75%
					NFIT vs. Healthy	0.93	94%	82%
				FIT negatives (FITN): 174; False-positive FIT (NFIT): 36; Low risk lesions (LR): 36; High-risk lesions or CRC (HR/CRC): 92;	FITN vs. (HR/CRC, LR or NFIT)	0.87	76%	83%	Italy	40
5	29-gene panel	Peripheral blood mononuclear cells (PBMC)	qRT-PCR	Multicenters: CRC: 45/52; large adenomatous polyps (LAPs) 61/42; healthy: 75/75; colorectal diseases other than cancer: –/245	CRC vs. control	–	92.2% (85–97%)	74.1% (61–85%)	Switzerland, South Korea	41
					LAP vs. control	–	92.2% (85–97%)	52.3% (40–65%)
6	23-gene panel	Peripheral blood mononuclear cells (PBMC)	qRT-PCR	Multicenter	CRC vs. Healthy	0.95 (0.85–1.00)	92.3% (76.9–100.0%)	92.6% (81.5–100.0%)	Belgium, Germany, Italy	42
7	MiRFec model miR-421 and miR-27a-3p	Fecal	qRT-PCR	767 FIT positive: CRC: 67; AAs: 347; Nonadvanced adenoma: 136; Normal examination: 217	CRC vs. nonrelevant findings	0.899	50%	97.0% (NPV=98.9%)	Spain	43
					CRC and AAs vs. nonrelevant findings	0.705		79.0% (NPV=67.0%)
8	4-miRNA panel (hsa-miR-513a-5p, hsa-miR-628-3p, hsa-miR-193a-5p, and hsa-miR-210)	Plasma	qRT-PCR	Early-onset CRC (EOCRC): 72/77 Nondisease controls: 45/65	Stage I/II vs. nondisease	0.92 (0.84–0.96)	80%	92%	USA, Spain, Japan	44
					stage III/IV vs. nondisease	0.87 (0.79–0.92)	86%	79%
9	4 microbial gene markers	Fecal	qPCR	4 markers validated in Metagome data from Austrian and France Cohort: CRC: 41+53; Control: 55+61 qPCR validation cohort: CRC:47; Controls:109	CRC vs. Healthy	0.84	–	–	China, Denmark, France, Austria	45
10	Fusobacterium nucleatum (Fn), Bacteroides clarus (Bc), Roseburia intestinalis (Ri), Clostridium hathewayi (Ch), one undefined species (labeled as m7).	Fecal	qPCR	CRC: 170/33; AA: 97/–; Healthy: 200/36	CRC vs. Healthy	0.886	83.20%	83.80%	China	46
11	F. nucleatum vincentii and F. nucleatum animalis, Porphyromonas, asaccharolytica, and Peptostreptococcus stomatis	Fecal	mNGS	CRC: 53/38; AA: 42/-; Healthy: 61/297	CRC vs. Healthy	0.84	–	–	France, Germany, Demark, Spain	47
12	24 biomarkers of adenoma vs. cancer	Fecal	16s RNA	CRC: 217/0/52; AA: 306/102/57; Healthy: 252/70/0	Healthy vs. AA	0.8	0.62	0.82	Canada, USA, France, China	48
					AA vs. CRC	0.89	0.9	0.66
13	GMSM panel: eight gut microbiome-associated serum metabolites	Serum		Normal individuals: 72/53; Abnormal colorectal patients: 120/103	Abnormal vs. normal	0.92	84.9%	83.5%; PPV=0.91	China	49
					Adenomas vs. normal	0.84		63.2%; PPV=0.60
					Stage I/II CRC vs. normal	0.93		88.2%; PPV=0.65
					Stage III/IV CRC vs. normal	0.91		84.2%; PPV=0.80
14	Disease-specific mixtures of VOCs	Breath	Aeonose devices (The eNose Company)	CRC: –/70; AA: –/117; non-AA: –/117; hyperplastic polyp: –/15; Healthy: –/125	CRC	0.84	64%	95%	The Netherlands	50
					AA	0.73	59%	79%
15	Four compounds (acetone and ethyl acetate, ethanol and 4-methyl octane)	Breath	Sensor analysis and GC-MS	CRC: –/65; AA: –/22; Healthy: –/122	CRC	–	94%	85%	Israel	51
					AA	–	88%	100%
16	Ten discriminatory VOCs	Breath	GC-MS	CRC: 30; AA: 38; Non-AA: 130; Negative control (NC): 84	AA vs. NC	–	70%	79%	The Netherlands	52
					CRC and AA vs. NC	–	70%	77%
					CRC vs. NC	–	70%	80%
17	Propanal (threshold=28ppbv)	Breath	Selected Ion Flow-tube Mass Spectrometry	CRC: 50/25; positive controls with other conditions of the LGI tract: 50/23; negative controls with a normal LGI tract: 50/31	CRC vs. control	–	84.7%	83.3%	UK	53
18	Eight VOCs	Breath	ReCIVA Breath Sampler and GC-MS	Early‐stage CRC: 36; No cancer: 87	Early‐stage CRC vs. No cancer	0.985	94%	86%	Italy	54
19	pd-score: methylated DNA marker ITGA4 and Fusobacterium nucleatum (Fn) and Pepetostreptococcus anaerobius (Pa) and FIT	Fecal	qRT-PCR	CRC: –/67; AA : –/15; Healthy: –/35	CRC vs. Healthy	0.952	80.00%	94.03%	China	51
					Adenoma vs. Healthy	0.874	80.00%	80%
20	LUNAR-2: DNA methylation, mutations, and fragmentomic features of ctDNA	Blood: ctDNA	–	All: 614/434	Stage I/II	–	94%	88%	South Korea; Japan	55
					Stage III	–	94%	93%
21	PREEMPT CRC: tumor-derived and immune-derived signals from cfDNA, epigenetic, and protein biomarkers	Blood	WGS; WGBS; protein quantification	CRC: 43; colonoscopy-confirmed CRC-negative controls: 548	Early stage		90%	92%	CA	56
					Late stage		90%	84%

AUC indicates, area under the curve; CRC, colorectal cancer; FIT, fecal immunochemical testing; GMSM, gut microbiome-associated serum metabolite; GC-MS, gas chromatography-mass spectrometry; mNGS, metagenomic next-generation sequencing; NPV, negative predictive value; PPV, positive predictive value; SIFT-MS, selected ion flow-tube mass spectrometry; VOCs, volatile organic compounds.

DNA mutations

cfDNA derived from tumors carries the same DNA mutations found in tumor tissues and is commonly used in cancer companion diagnostics. Patients with early-stage cancer have lower ctDNA concentrations and tumor mutational burden than patients with advanced cancer57. The identification of mutations in early-stage cancer requires an approach with high sensitivity. Researchers have developed strategies for wet tests and bioinformatic analysis methods to lower the limit of detection or that block the wild-type effect58–61.

To improve the sensitivity of CRC mutation detection, Sun and colleagues (2021; DiaCarta Inc.) developed a novel xenonucleic acid-mediated qPCR clamping technology that can minimize wild-type background amplification. Based on this technology, the ColoScape (DiaCarta Inc.), an assay to detect nineteen mutations in four genes (APC, KRAS, BRAF, and CTNNB1) was developed. In patients with CRC, mutations were identified at a rate of 100% in cfDNA (N=137), which was higher than the 96.0% in FFPE samples (N=175). In patients with precancerous lesions, this assay has a 62.5% (95% CI: 35.8–83.7%) sensitivity for precancerous screening and a 95.2% (95% CI: 82.5–99.1%) specificity, with an area under the curve (AUC) of ~0.79. However, the data are limited to a small sample of 58 patients with precancerous lesions38.

This method may provide an alternative choice for mutation detection in ctDNA, but this approach still has limitations that may hamper its application, such as lower concentration of ctDNA in early-stage cancers and the variations in ctDNA affected by age and other factors.

mRNA and miRNA

It has been reported that tumor cells62 release cell-free RNA (cfRNA) and that this tumor-derived RNA can be detected in blood63,64. In addition, cfRNA signals may be amplified by overexpression of tumor-associated genes, which indicates that cfRNA may be used to detect cancer65. The mRNA panel CELTiC, which includes the TSPAN8, LGALS4, COL1A2, and CEACAM6 genes, was developed by Rodia and colleagues. in 2016. The research team analyzed the performance of the CELTiC panel in FIT-positive39 and FIT-negative40 participants. When the CELTiC panel was used in the FIT-positive group, the panel effectively distinguished false-positive FIT (NFIT) participants from healthy individuals with an AUC of 0.93 (sensitivity: 82%; specificity: 94%)39. This panel could identify healthy individuals from the NFIT individuals, to avoid the colonoscopy test. On the contrary, the CELTiC panel could discriminate the FIT true-negative participants (N=174) from those with NFIT, low risk lesions, as well as high-risk lesions or CRC patients (N=92) with an AUC of 0.87 (sensitivity: 83%; specificity: 76%)40. A clinical trial was designed to verify the performance of the CELTiC panel in the detection of CRC and its precursor lesions (NCT04980443).

miRNAs are small noncoding RNAs (18–24 nucleotides in length) that are critical for regulating cancer signaling66. Duran-Sanchon and colleagues reported an algorithm based on the abundance of two microRNAs (miR-421 and miR-27a-3p) in fecal samples combined with the FIT test, along with age and sex, called miRFec. This combination test could differentiate CRC patients from non-AA or normal colonoscopy with an AUC of 90%, and it could identify 97% of CRC patients with a specificity of 50%43. Recently, a novel liquid biopsy miRNA signature was developed to detect early-onset (younger than 50 y) colorectal cancer (EOCRC). This signature included four circulating miRNA markers (hsa-miR-513a-5p, hsa-miR-628-3p, hsa-miR-193a-5p, and hsa-miR-210) associated with cancer key genes. In the validation data including EOCRC (N=77) and nondisease control (N=65), the panel could identify stage I/II CRC patients with an AUC of 0.92 (95% CI: 0.84–0.96; sensitivity 0.92; specificity 0.80), and identify stage III/IV CRC patients with an AUC of 0.87 (95% CI: 0.79–0.92; sensitivity 0.79; specificity 0.86). This liquid biopsy miRNA panel showed exciting performance in EOCRC detection44.

In contrast, cell-free circulating microRNAs (cf-miRNAs) and exosome-associated microRNAs are the main variations of miRNA in the blood. Cancer-derived exosome microRNAs (CEx-miRNAs) are selectively released by tumor cells and are more stable67. Recently, the CRISPR/Cas-based platform was developed to detect an abundance of CEx-miRNAs, a CEx-miRNAs panel including miR-126, miR-1290, miR-23a, and miR-940 has the potential predictive ability for early-stage CRC. CRISPR/Cas13-based platforms are promising candidates for early, next-generation methods for early CRC diagnosis and prognostic prediction68.

Gut microbiome

The gut microbiome is increasingly being recognized for its influence on cancer. Some evidence supports the notion that generalized dysbiosis of gut microbiota may contribute to carcinogenesis in CRC69–71. First, the microbiota in adenoma or cancer is significantly different from that of the adjacent healthy mucosa72–74, and studies from preclinical models indicate that transplant of stool from patients with CRC can induce polyp formation, induce procarcinogenic signals, and alter the local immune environment in mice as compared with that from healthy controls75. Second, preclinical and clinical studies have suggested that dysbiosis of the gut microbiota may be an oncogenic driver in CRC. Some bacterial species can stimulate an inflammatory state that can promote carcinogenesis by induction of proinflammatory toxins (such as Bacteroides fragilis)76–78, increased production of reactive oxygen species79, and alterations in signaling pathways (Fusobacterium nucleatum)80 in human tumors and mouse models, or they may act to prevent antitumor immune functions (F. nucleatum)80,81. All above make gut microbiome to be potential biomarkers of CRC or adenoma detection.

Georg Zeller et al. 47 used metagenomic next-generation sequencing of fecal samples to identify taxonomic markers (two Fusobacterium spp., Porphyromonas asaccharolytica, and Peptostreptococcus stomatis) that distinguished patients with CRC from tumor-free controls in a cohort that included 53 patients with CRC, 42 with adenoma, and 61 healthy controls, in which the accuracy was 0.84 (AUC). In another study in China that enrolled 74 patients with CRC and 54 controls, 20 microbial gene markers that differentiated CRC and control microbiomes were identified by metagenomic next-generation sequencing, while four markers were validated in the Danish, French, and Austrian cohorts. Finally, targeted qPCR assays were employed in an independent Chinese cohort and two of these markers (F. nucleatum and Parvimonas micra) were used to accurately classify patients with CRC (AUC=0.84)45. In a meta-analysis, 1056 public fecal samples, including those from 217 patients with CRC, 306 with adenoma, and 252 healthy controls, were analyzed and 11 markers were identified to discriminate adenoma from controls (AUC=0.80)48. Moreover, a prospective randomized CRC screening trial (CRCbiome) was initiated to develop a microbiome-based classification algorithm to identify advanced colorectal lesions in screened FIT-positive participants82.

Gut microbiome detection using fecal samples may be a potential noninvasive method to screen patients and detect adenomas and CRC. However, large-scale randomized clinical trials and bench research are needed to determine the utility of gut microbiome detection in the screening and detection of adenomas and CRC.

Metabolites

Tumor growth may be accompanied by abnormal metabolic products, and these metabolites may be produced by both the human and gut microbiome83. Chen and colleagues constructed a model based on eight gut microbiome-associated serum metabolites that are significantly altered in patients with CRC and adenomas. The gut microbiome-associated serum metabolites model had an AUC of 0.92, with a sensitivity of 83.5%, a specificity of 84.9% in an independent validation cohort (including 103 patients with abnormal colorectal findings and 53 normal individuals), and a positive predictive value of 0.9149.

In addition, volatile organic compounds (VOCs) have recently been demonstrated to be promising biomarkers, and several studies have focused on VOCs in breath. Amal and colleagues identified four specific VOCs (acetone, ethyl acetate, ethanol, and 4-methyl octane) in breath specimens by gas chromatography-mass spectrometry. A sensor technology was employed to validate these four VOCs in a cohort containing 65 CRC patients, 22 AA or non-NAs, and 122 controls. The sensors could distinguish patients with AA from the controls with 100% sensitivity, 88% specificity, and 94% accuracy51. In 2020, a new platform termed electronic nose was developed, which may be more suitable for detecting VOCs in exhaled breath. In this study, 70 patients with CRC, 117 with one or more AAs, 117 with NAs, 15 with hyperplastic polyps, and 128 with normal colonoscopy findings were enrolled. The patients with CRC were distinguished from the controls with an AUC of 0.84, 95% sensitivity, and 64% specificity. Patients with AAs were distinguished from controls with an AUC of 0.73, 79% sensitivity, and 59% specificity50. The electronic nose was also used in a clinical trial (NCT03228095, N=3000) to reduce the number of unnecessary colonoscopies (Table 3).

Table 3.

The promising biomarkers under development in ongoing clinical studies.

Biospecimen	Devices/biomarkers	Conditions	Age (years)	Enrollment	Actual start date	NCT number
Blood	Circulating Tumor Cell and Circulating Tumor DNA	Colorectal cancer and adenomas	45–80	1000	15 December 2020	NCT05127096
Blood	LUNAR-2 Test: ctDNA DNA methylation, mutations, and fragmentomic features of ctDNA	Colorectal cancer, average-risk population	45–84	10 000	8 October 2019	NCT04136002
Blood	Freenome test: AI, machine learning; Multiomics: tumor-derived and immune-derived signals from cfDNA, epigenetic, and protein biomarkers	Colorectal cancer|polyp|adenoma; gastrointestinal tract cancers	45–85	35 000 finished (May, 2022)	20 May 2020	NCT04369053
Blood	CELTiC test: four mRNA (LGALS4, CEACAM6, TSPAN8, and COL1A2)	Colorectal carcinoma|polyp|adenoma; FIT-positive individuals	50–75	800	3 August 2021	NCT04980443
Stool	miRFec Test: two miRNAs (miR-421 and miR-27a-3p) expression and fecal hemoglobin concentration	Colorectal cancer, advanced adenomas or advanced serrated lesions	30–80	9670	23 March 2022	NCT05346757
Stool	EarlyTect : SDC2 Methylation	Colorectal cancer colon polyp	≥12	1112	7 June 2020	NCT04304131
Stool	methylation status of SDC2	Asymptomatic Chinese community population; colorectal cancer|neoplasm|advanced adenoma|adenoma serrated lesion	45–75	10 000	22 March 2021	NCT04786704
Stool	Fluorescent PCR: Multigene Methylation Detection	Colorectal cancer advanced adenocarcinoma hyperplastic polyp digestive system neoplasm	40–79	1273	26 January 2021	NCT04722055
Stool and blood	Stool DNA test; Blood mRNA test; FIT	Colorectal cancer|adenomatous polyp	45–80	1200	19 March 2018	NCT03411707
Breath sampling	Nanosensor-based devices (electronic nose): VOC detection in breath	16 diseases including: colorectal cancer|polyp|adenoma	≥18	3000	24 July 2017	NCT03228095
Breath sampling	Aeonose: VOCs in exhaled breath	Colorectal cancer|polyp of colon|colorectal neoplasms	50–75	750	25 May 2020	NCT04357158
Breath samples; fecal samples; blood sample	Breath sampling for VOC detection; Plasma/serum samples for group stratification; Fecal samples for occult blood testing and microbiota analysis; Tissue samples for specific VOCs analysis	Colorectal cancer|polyp of colon	≥18	3000	1 February 2022	NCT05173077
NA	Multiomics	Colorectal cancer	40–74	9000	21 July 2021	NCT04913233

AI indicates, artificial intelligence; FIT, fecal immunochemical testing; VOC, volatile organic compound.

In addition, other research teams also proved the applicability of VOCs in CRC early detection. Altomare et al.54 developed a breath sampler and found eight VOCs (Tetradecane, Ethylbenzene, Methylbenzene, 5,9‐Undecadien‐2‐one, 6,10‐dimethyl, Tridecane, Benzaldehyde, Dodecane, and Butyl hydroxytoluene) combined with ages would discriminate the early-stage CRC (N=36) from noncancers (N=87) with an AUC of 0.985, 86% sensitivity, and 94% specificity. In a multicenter prospective study, breath samples from 382 FIT-positive Dutch patients were collected. A biomarker profile was established based on ten discriminatory VOCs and this profile could discriminate CRCs (N=30) from controls (N=84) with a sensitivity and specificity of 80 and 70%, respectively52. Markar et al.53 identified propanal was significantly elevated in the breath samples of CRC patients, and it could distinguish CRCs from controls with a sensitivity of 83.3% and specificity of 84.7% in a prospective validation cohort.

The VOCs in breath samples may be a promising biomarker to screen patients with AA and CRC and distinguish them from healthy controls. There are still some limitations in the use of this technology. First, VOCs vary in different studies, and sample collection may be affected by many factors, such as coughing or talking. Second, sensor technology has some limitations, such as potential reproducibility issues, sensor drift, and the influence of alcohol.

Multiomics

The occurrence of CRC involves changes in multiple factors, including DNA, RNA, proteins, the microbiome, and metabolites. The combination of these biomarkers may have additional advantages in CRC screening or diagnosis. Considering cost, laboratory operations, and quality control, low-throughput multiomics may be a better option.

Cologuard (Exact Sciences) is the first FDA-approved multitarget stool DNA (MT-sDNA) test that is applicable to average-risk individuals aged 45 years and older12–14. The targets include methylation of the NDRG4 and BMP3 promoters, seven mutations in the KRAS gene, and FIT. The sensitivity for detecting CRC was 92%, but 42.4% for detecting advanced precancerous lesions12. However, while the Cologuard data are based on western populations, the ColoClear kit using the similar targets was validated in a Chinese cohort aged 40–74 years. The sensitivity of ColoClear for detecting CRC ranged from 91.94 to 95.5% and was 63.5% for detecting AA; the specificity ranged from 87.11 to 99.6%, the negative predictive value is 99.6%. Although both kits had a higher sensitivity than FIT, their lower specificity for early-stage cancer and precancer screening may limit their application. A registered trial (NCT04913233) of multiomics targets aimed to improve the specificity of CRC screening by greater than or equal to 15% without significantly reducing sensitivity is currently recruiting and plans to enroll 9000 participants.

Some researchers have also attempted to combine gut microbes and DNA markers in feces. Cao and colleagues constructed a pd-score model using methylated ITGA4 and bacteria [F. nucleatum (Fn) and Peptostreptococcus anaerobius (Pa)] by logistic regression. The pd-score detected CRC and AA with sensitivities of 94.03 and 80.00%, respectively84.

In addition to stool, tumor-derived biomarkers can also be identified in blood. The LUNAR-2 assay incorporates DNA methylation, mutations, and fragmentomic features of ctDNA, and a prospective registered trial (NCT04136002) to evaluate the test in an average-risk CRC screening cohort is ongoing. In a previous study, no significant difference was observed in the sensitivity of LUNAR-2 between patients with asymptomatic CRC (88%, 123/139) and those with symptomatic CRC (91%, 212/233; P=0.4). A lower sensitivity was observed in patients with stage I/II disease (88%, 211/239) compared with patients with stage III disease (93%, 182/195), at the specificity of 94%55. Another study used a machine learning platform to develop a blood test by combining tumor-derived and immune-derived signals from cfDNA and epigenetic and protein biomarkers. The test achieved a sensitivity of 92% for early-stage disease (N=17) and 84% for late-stage disease (N=11) with a specificity of 90%56. To achieve high sensitivity and specificity, the PREEMPT CRC study (NCT04369053) was performed, and according to the industry website, 35 000 participants were enrolled in 2022.

Due to shifts in the microbiome and metabolome that were observed in the very early stages of CRC development in a large cohort85, studies have combined metagenomic and metabolomic markers to develop screening assays. One trial (NCT05173077) that integrated VOC detection, microbiota analysis, and FIT is ongoing and plans to enroll 3000 participants.

Pan-cancer screening aids CRC screening

In recent years, many large-scale trials, such as CCGA, DECIPHER-Multi, and THUNDER, have focused on the development of biomarkers for pan-cancer screening and have published quite promising results in cancer detection and tissue of origin (TOO).

cfDNA methylation was used as a biomarker in the CCGA86 and THUNDER studies. CCGA study enrolled 2482 participants with cancer and 4207 without cancer and showed discrimination of cancers (12 types), including CRC, with a sensitivity of 67.3% (95% CI: 60.7–73.3%) for stage I–III disease and 39% (95% CI: 27–52%) for stage I disease; the specificity was 99.3% and the TOO localization was accurate in 93% of cases86. The THUNDER study is based on ELSA-seq technology and enrolled 473 patients with cancer and 473 age-matched patients without cancer in the validation cohort. In this trial, MCFBT-1 and -2 models were constructed, and while the specificity of both was higher than 95%, the sensitivity was lower for stage I cancers except liver cancer87.

On the contrary, Bao and colleagues established an ultrasensitive assay by incorporating five cfDNA fragment features in the DECIPHER-Multi-trial. 1214 participants were enrolled including 243 healthy volunteers; 86% of cancer patients (381 with primary liver cancer, 298 with CRC, and 292 with lung adenocarcinoma) had stage 0 or I disease. The model could discriminate CRC with a 94.6% sensitivity and a 95.0% specificity and demonstrated a 94.3% accuracy for predicting cancer origin88. Both the detection and origin classification power of this model are consistently high even when 1X sequencing coverage depth was used, which opens the door to screening applications for this tool.

However, the proportion of healthy participants in these three trials was much lower than that in the real world, and prospective studies should be performed to validate the application of these models in the screening of asymptomatic individuals.

Conclusion

Due to its noninvasive nature and convenient advantages, along with colonoscopy, FIT is recommended for CRC screening in most guidelines. However, the performance of FIT is affected by the cutoff of hemoglobin, with a sensitivity that ranges from 91 to 71% for CRC and from 40 to 25% for AAs89. The combination of FIT with other markers could improve screening performance. For example, the sensitivity for AA detection improved to 63.47% when FIT was combined with detection of DNA methylation and mutations89. However, the performance of FIT and other biomarker combination in CRC screening still need to be verified in large cohort. With the development of experimental technology and bioinformatics, it is expected that ideal early screening markers will be developed.

Due to the location of colorectal tumors, stool can be used for CRC screening. Current commercial kits, including single-gene (SDC2) methylation and low-throughput multiomics panels (Cologuard, ColoClear), mainly use DNA methylation and mutations as primary markers. However, the performance of these kits in tumor screening is limited, especially for early-stage CRC and AAs. In addition, feces also contain intestinal microorganisms and metabolites, and their changes can reflect the health status of the body, which provides a new approach to CRC screening.

Blood can be collected in a minimally invasive manner and is widely used for disease detection. ctDNA in blood carries tumor information and is used for cancer screening. However, the low abundance of ctDNA in patients with early-stage disease is a significant challenge. Researchers are developing new lab techniques to increase the sensitivity of detection. For example, XNA-mediated qPCR was developed to detect precancerous lesions with a 62.5% sensitivity and a 95.2% specificity38. The progression of artificial intelligence has provided tools for mining new biomarkers. In one study, machine learning discovered cfDNA fragment features from whole-genome sequencing data, which could distinguish CRC patients from healthy individuals37,88. In other studies, machine learning has driven the construction of multiomics models. In the PREEMPT CRC trial, cfDNA, epigenetic and protein biomarkers were combined, both sensitivity and specificity were higher than 90% for early-stage CRC screening56. In the LUNAR-2 trial, DNA methylation, mutations, and fragmentomic features of ctDNA in blood were combined, 94% specificity and 88% sensitivity were observed for stage I/II CRC patients. These indicate the value of multiomics for early-stage CRC screening. In addition to ctDNA, blood mRNA and miRNA are also developed for early CRC detection and have demonstrated good performance.

In recent years, sensor technology has made great strides and is used to detect VOCs in breath samples. A trial has attempted to combine breath, stool, and blood samples for early CRC detection (NCT05173077), the results may provide some evidence for the combination of several noninvasive tests. Finally, the progression of pan-cancer screening technology including improvements in specificity and TOO accuracy provide a novel method for CRC screening.

In the end, these biomarkers are promising. In the future, it is hoped that there will be an easy, low-cost, and accurate way to stratify populations with average-risk, and identify patients with colorectal or adenoma requiring further treatment.

Ethical approval

Not applicable.

Sources of funding

This work was supported by Key R&D Plan of Jiangsu Provincial Department of Science and Technology (BE2020723).

Author contribution

D.S. and J.B. conducted background research and designed the outline; F.W., Y.Z.J., Q.R.H., and T.T.S. completed the main text part; W.L.D., R.D., and C.Z. completed the introduction and conclusion parts; Q.X. and C.Q. completed the figure and tables.

Conflicts of interest disclosure

The authors have declared no conflicts of interest.

Research registration unique identifying number (UIN)

1. Name of the registry: not applicable.

2. Unique identifying number or registration ID: not applicable.

3. Hyperlink to your specific registration (must be publicly accessible and will be checked): not applicable.

Guarantor

Jun Bao, Department of Medical Oncology, Jiangsu Cancer Hospital & Jiangsu Institute of Cancer Research & The Affiliated Cancer Hospital of Nanjing Medical University, No. 42, Baiziting, Nanjing, China. baojun@jszlyy.com

Data statement

Not applicable.