Plasma methylation markers for early colorectal cancer detection

Abstract

BACKGROUND

Colorectal cancer (CRC) is a leading cause of cancer-related mortality worldwide. Early detection significantly improves patient outcomes, but current screening methods have limitations in terms of sensitivity, specificity, and patient compliance. Circulating cell-free DNA in the bloodstream, including methylated DNA fragments, offers a promising noninvasive alternative for cancer detection.

AIM

To identify a high-specificity plasma methylation panel for early noninvasive CRC detection.

METHODS

We conducted a pilot recruiting CRC patients, benign gastrointestinal lesion cases and healthy volunteers. Genomic DNA from paired tumor-paracancerous tissues and white blood cells, plus plasma cell-free DNA, was extracted. We used multiplex probe melting assay to screen markers with minimal leukocyte methylation background, followed by quantitative real-time polymerase chain reaction verification. Independent in silico sequencing data validated the panel’s discriminatory power. We further analyzed associations between panel positivity and clinicopathological features, as well as its diagnostic specificity across other gastrointestinal malignancies.

RESULTS

The analysis revealed a three-marker methylation panel comprising IKZF1, ITGA4, and C9orf50, with a sensitivity of 81.03% and a specificity of 100% for healthy individuals, 85.71% for benign diseases, and an overall specificity of 95.83% in distinguishing CRC cases from controls in plasma samples. The area under the receiver operating characteristic curve for this panel was 0.8962 (95%CI: 0.8339-0.9585; P < 0.0001). The positivity of the panel was associated with the size and advancement of the tumor and the clinical specificity, especially in asymptomatic gastrointestinal patients who showed good potential for CRC diagnosis.

CONCLUSION

The three-marker methylation panel offers high diagnostic accuracy and could be integrated into routine clinical practice, such as screening and auxiliary diagnosis of CRC.

Key Words: Colorectal cancer; Cell-free DNA; DNA methylation; Multiplex probe melting assay; Plasma biomarker; Early detection

Core Tip: This study developed a three-marker plasma DNA methylation panel – comprising IKZF1, ITGA4, and C9orf50 – specifically designed for early detection of colorectal cancer. Leveraging multiplex probe melting assay screening, the panel effectively suppresses confounding methylation signals derived from white blood cells. It demonstrates superior diagnostic performance for colorectal cancer compared to the clinically established SEPT9 methylation blood test, achieving higher sensitivity at matched specificity and exhibiting strong potential as a noninvasive complementary diagnostic tool.

INTRODUCTION

Colorectal cancer (CRC) ranks third in incidence rates and second in mortality worldwide[1]. In China, CRC has the second highest incidence among tumors[2], with its mortality rate continuously increasing, particularly within high-income provinces[3]. Screening and early diagnosis of cancer within the appropriate population can significantly diminish the incidence and mortality rates associated with this affliction[4]. Screening guidelines for CRC recommend colonoscopy and fecal occult blood test/fecal immunochemical test as first-tier recommendations[5]. However, its invasiveness, dietary constraints, and troublesome bowel preparation lead to a low compliance rate for colonoscopy[6]. Other noninvasive fecal tests, such as fecal immunochemical test and Cologuard, are disfavored by test takers due to sampling issues, which is preferable for blood-based detection[7]. Therefore, the employment of noninvasive and accurate blood tests for the early detection of CRC and adenoma can enhance patient compliance and is highly demanded.

Abnormal regulation of gene expression via DNA methylation is a pervasive characteristic during tumorigenesis[8], and there have been pertinent investigations on CRC detection via blood by measuring the methylation level of cell-free DNA (cfDNA) released by tumors[9]. However, only one commercial product (Epi proColon) utilized blood-based cfDNA methylation detection for CRC diagnosis based on the methylation detection of the SEPT9 promoter region[10]. The overall performance, particularly sensitivity in detecting early-stage tumors, is quite restricted[11]. Furthermore, the marker exhibits some detectable capabilities in detecting other tumors[12,13], particularly gastrointestinal tumors such as gastric cancer[14]. However, its clinical specificity is insufficient for its utilization as a screening or diagnostic tool for CRC.

Due to limitations in blood collection volume and the relatively low abundance of tumor-derived cfDNA in the blood, current blood-based detection methods aim to improve overall detection performance primarily through multitarget and multiomics approaches that complement each other[15]. For CRC detection in blood, besides improving overall performance, especially sensitivity, it is essential to have good clinical specificity to better meet the requirements of screening or auxiliary diagnostic scenarios. This study aimed to identify markers that minimize background noise within cfDNA, thereby improving sensitivity for early CRC detection while demonstrating robust clinical specificity and enabling their application in the blood-based diagnosis of CRC.

MATERIALS AND METHODS

Study population

This study was approved by the Ethics Committee of Zhujiang Hospital, Southern Medical University (approval No. 2017-PTWK-004) and conducted in accordance with the principles of the Declaration of Helsinki. Twenty paired CRC cancerous and paracancerous tissues were collected at the time of surgery at Zhujiang Hospital, Southern Medical University between November 2021 and April 2022. All tissue samples underwent histopathology review by an expert pathologist before DNA extraction and were stored at -80 °C until use. Fifty-one white blood cell (WBC) samples (buffy coat) from patients without a history of cancer at any site were collected.

A total of 273 patients and healthy individuals who gave informed consent were enrolled at Zhujiang Hospital, Southern Medical University between March 2022 and August 2023. Blood (10 mL) was drawn from each subject with a 10 mL K2EDTA tube (BD; cat. 367525) and stored at 4 °C. Plasma was separated within 4 hours after receiving whole blood samples using a two-centrifugation step protocol and immediately stored at -80 °C until use. Plasma volume ranged from 4.0 mL to 6.0 mL. Repeated freezing and thawing of plasma was avoided. The normal cohort included hospitalized patients who had physical examinations without any benign, precancerous, or cancerous lesions confirmed by colonoscopy. The benign and malignant cohort included patients with benign, precancerous, or cancerous lesions. All resected lesions underwent histopathology review for pathology confirmation.

DNA extraction, bisulfite treatment, and methylation analysis

Genomic DNA (gDNA) from tissue samples was extracted by HiPure Universal DNA Kit (Magen Biotech; cat. IVD3018) and WBC was extracted by HiPure Blood DNA Mini Kit (Magen Biotech; cat. D311102) according to the manufacturer’s protocol. cfDNA was isolated by the HiPure Circulating DNA Midi Spin Kit D (Magen Biotech; cat. D318203D) according to the manufacturer’s protocol. The gDNA or cfDNA concentration was measured by the Qubit^TM dsDNA HS Assay Kit (Thermo Fisher Scientific; cat. Q32854) and the cfDNA quality was examined using the High Sensitivity DNA Kit (Bioptic; cat. C105105) on a Qsep100 Bioanalyzer Instrument.

Bisulfite conversion was performed using EZ-96 DNA Methylation-Lightning^TM MagPrep (Zymo Research; cat. D5046) according to the manufacturer’s protocol. DNA was first denatured at 98 °C for 8 minutes and treated with sodium bisulfite at 54 °C for 1 hour. DNA was collected with MagBinding beads, desulfonated by adding desulfonation buffer, and incubated at room temperature for 17 minutes. Bisulfite-converted DNA was purified by beads and eluted with 25 μL elution buffer.

A multiplex probe amplification (MPA) assay was performed on bisulfite-converted WBC (buffy coat) gDNA, as described previously[16], to quickly screen methylation-clean markers in WBC DNA. In short, for each target, a probe and a partially complementary oligonucleotide (PCO) to the probe were designed. Within one channel, different pairs of probe and PCO had different melting temperatures by varying the length of different PCOs. After the amplification and melting analysis phase, a Ct from one channel for aggregate signals and a melting profile for qualitatively identifying each target were obtained. Two to four markers with the same dye were added to one channel. All MPA reactions were performed in technical duplicates to ensure result reliability. An MPA reaction on 50 ng bisulfite-converted WBC gDNA in 7 μL elution buffer mixed with 10 μL EpiTect MethyLight Master Mix (Qiagen; cat. 59496) and 3 μL CRC 14-gene panel (primer, probe, and PCO sequence shown in Supplementary Table 1) was performed in a real-time thermocycler (Hongshi Medical; cat. SLAN-96P) under the following cycling conditions: (1) Denaturation at 95 °C for 2 minutes; and (2) 50 cycles (95 °C for 15 seconds, 60 °C for 1 minute). Postamplification melting profile analysis was performed under the following conditions: (1) After the last cycle of polymerase chain reaction (PCR); (2) Heating at 72 °C for 15 seconds; (3) Cooling to 25 °C; and (4) Slowly increasing the temperature to 75 °C. Fluorescence emission data were continually collected during the increasing temperatures.

Bisulfite-converted gDNA of 20 paired cancerous and paracancerous tissues from CRC was examined by a methylation-specific quantitative real-time PCR (qPCR). Each sample was assayed in technical duplicates; the average Ct value from valid replicates was adopted for subsequent calculation. A reaction involving 50 ng bisulfite-converted DNA in 7 μL elution buffer mixed with 10 μL EpiTect MethyLight Master Mix (Qiagen; cat. 59496) and 3 μL oligos for each marker (primer and probe sequences shown in Supplementary Table 2) was performed in a real-time thermocycler (Hongshi Medical; cat. SLAN-96P) under the following cycling conditions: (1) Denaturation at 95 °C for 2 minutes; and (2) 50 cycles (95 °C for 15 seconds, 60 °C for 1 minute). All results were normalized by an internal control (ACTB). Ct (Ct marker - Ct ACTB) was calculated to represent the normalized methylation level of each marker in a tissue sample, and Ct values for reactions returning no amplification signals were set to 25.0, a number greater than all calculated Ct values. The reactions were performed in a real-time thermocycler (Hongshi Medical; cat. SLAN-96P).

Bisulfite-converted cfDNA was further analyzed by a CRC 8-gene panel. In short, a strategy of multiplexing PCR was performed, followed by methylation-specific qPCR quantification for candidate methylation markers. In some channels, two targets were combined to enhance sensitivity, as reported previously[17], to overcome the limited quantity of cfDNA while diminishing the number of reaction tubes to two tubes in detecting eight methylation markers simultaneously, as reported previously[18]. Pre-amplification and subsequent methylation-specific qPCR steps for cfDNA samples were both run in technical duplicates. A multiplex eight-marker targeted preamplification of methylated regions was performed on 23 μL bisulfite-treated DNA using 25 μL Q5U Master Mix (NEB; cat. M0597) and 2 μL CRC 8-gene outer panel (primer and probe sequence shown in Supplementary Table 3) in a thermal cycler (Thermo Fisher; cat. 4484073) under the following cycling conditions: (1) Denaturation at 98 °C for 10 seconds; (2) 15 cycles (98 °C for 10 seconds, 60 °C for 1 minute); (3) Final extension at 65 °C for 5 minutes; and (4) Hold at 4 °C. The amplified products were diluted four times before use, and an 8 μL diluted amplified product was used as template for methylation-specific qPCR in a real-time thermocycler (Hongshi Medical; cat. SLAN-96P) using 10 μL Luna Universal qPCR Master Mix (NEB; cat. M3003) and 2 μL CRC 8-gene inner panel (primer and probe sequence shown in Supplementary Table 3) under the following cycling conditions: (1) Denaturation at 95 °C for 30 seconds; (2) 40 cycles (95 °C for 15 seconds, 60 °C for 30 seconds); and (3) With fluorescent signals collected at the annealing/extension step (60 °C for 30 seconds). The interpretation method for this assay is as follows: Confirm whether the Ct value of ACTB is < 26, otherwise, it is invalid. The positive cutoff values for various methylation markers are as follows: (1) CHST11 + ZEB2 ≤ 40; (2) IKZF1 ≤ 40; (3) ITGA4 ≤ 30; (4) C9orf50 ≤ 40; (5) VAV3-AS1 + SFMBT2 ≤ 40; and (6) SEPT9 ≤ 40. When any marker is determined as positive, the sample is considered positive.

The independent validation of the marker set using Gene Expression Omnibus data

The methylation data for CRC and corresponding adjacent tissues were retrieved from the Gene Expression Omnibus database under accession number GSE124600. The data processing pipeline is as follows: (1) The raw FASTQ files were initially downloaded; and (2) Adapter sequences within these files were removed using Trim Galore. Subsequently, the cleaned reads were aligned to the hg19 reference genome using the bwameth. The evaluation of the alignments was assessed with Qualimap. The methylation status of the three markers was determined by calculating the ratio of reads mapping to these regions relative to the total reads mapping across the human genome, expressed as methylated alleles per million mapped reads. The diagnostic efficacy of the marker panel was then rigorously evaluated through receiver operating characteristic (ROC) analysis, utilizing the methylation values in combination for all three markers within the context of CRC and adjacent tissues. This approach allowed for a comprehensive assessment of the marker panel’s performance in distinguishing between cancerous and non-cancerous tissues.

Statistical analysis

Statistical analysis was performed using SPSS version 21.0 (IBM, Armonk, NY, United States) or GraphPad Prism version 8.0 (GraphPad Software, San Diego, CA, United States). The diagnostic values of the panel were estimated using ROC curves. The relationship between the panel positivity and clinicopathological characteristics was assessed using the χ² test. All tests were two-tailed, and P value of < 0.05 was considered significant. The differences in methylation levels were analyzed by the Mann-Whitney U test, for which P value of < 0.05 was considered statistically significant. The Ct (Ct paracancer - Ct cancer) of marker ≥ 4 claimed that the marker was discriminative between the cancerous and paracancerous tissues. Two-by-two contingency tables were generated to calculate diagnostic performance metrics for the methylation panel, and raw counts are presented in Supplementary Table 4.

RESULTS

Patients and samples

Fifty-one WBC samples and 20 pairs of CRC and paracancerous tissues were collected to screen high signal-to-noise markers for plasma-based noninvasive CRC detection. A total of 250 patients and controls, including 117 CRC patients, five patients with advanced adenoma, 60 benign patients with hyperplastic polyps, tubular adenoma, villous adenoma, inflammatory ulcers, juvenile polyps, or serrated adenoma, and 68 normal patients without any benign, precancerous, or cancerous lesions confirmed by colonoscopy, were enrolled for panel development and validation. Relevant clinical characteristics of the cohort, including age, gender, histology, and tumor staging (American Joint Committee on Cancer criteria), are listed in Table 1. Precancerous patients with stage I + II cancers comprised 50.8% of the total CRC patients. Twenty-three plasma samples from gastrointestinal diseases were enrolled, including one gastrointestinal stromal tumor, one gastric benign disease, two esophageal cancers, and 19 gastric cancers for the clinical specificity test.

Table 1 Demographic and clinical characteristics of the study cohort.

Item	Control		Colorectal neoplasms
	Normal (n = 68)	Benign (n = 60)	Advanced adenoma (n = 5)	CRC (n = 117)
Age	56 (24-64)	60 (17-77)	67 (48-86)	64 (36-90)
Gender
Male	40	36	1	76
Female	28	24	4	41
Stage
I	-	-	-	19
II	-	-	-	38
III	-	-	-	36
IV	-	-	-	22
Unknown	-	-	-	2
Other gastrointestinal diseases
	Benign	Gastrointestinal stromal tumor	Esophageal tumor	Gastric tumor
Total (n)	1	1	2	19

Tumor stage was only applicable to patients with colorectal cancer tumor. “-” indicates not applicable. CRC: Colorectal cancer.

Identifying DNA methylation biomarkers suitable for blood-based CRC detection

Because most cfDNA stems from WBCs, it is critical to eliminate methylated signals in tumor DNA that overlap with those in WBCs, as in the presence of strong background signals, it is difficult to distinguish the small proportion of signals originating from the tumor. Therefore, the goal is to enhance the signal-to-noise ratio by identifying candidate tumor markers without such potential confounding signals from WBCs, thus providing a better opportunity to detect weak signals from the tumor, especially in early-stage plasma testing. Thus, to screen markers with a high signal-to-noise ratio within cfDNA, MPA assay, a medium-throughput qPCR method, was employed to quickly assess the methylation level of 16 markers within 51 WBC gDNA. Markers that exhibited conspicuous signals within WBCs were removed, such as markers BCAN and BMP3, where a strong signal was present in most WBC samples (Figure 1 and Table 1). Meanwhile, marker SDC2, frequently employed in fecal detection of CRC, can also be detected in 11 of these 51 WBCs and should be removed for plasma-based CRC detection, suggesting that the markers for detecting CRC using blood and feces cannot be arbitrarily interchanged. Different regions of SEPT9 show very different methylation signal patterns in the WBC context and indicate that specific regions within a particular gene are more critical as markers. The signals of the markers in WBCs are listed in Table 2, and those markers with interfering signals in WBCs were abandoned in the subsequent analysis.

Figure 1 Example of multiplex probe amplification assay. A: Graphic presentation of the amplification plot and melting curve showed that the signal was from BCAN; B: Graphic presentation of the amplification plot and melting curve showed that the signal was from BMP3; C: Graphic presentation of the amplification plot and melting curve showed that the signal was from BCAN plus BMP3.

Table 2 Signal of markers detected in 51 white blood cell samples.

Marker	Detected	Ratio
VIM	0	0%
VAV3-AS1	0	0%
ZEB2	2	3.9%
IKZF1	1	2%
KCNQ5	1	2%
NDRG4	21	41.2%
ITGA4	0	0%
C9orf50	0	0%
BCAN	35	68.6%
BMP3	40	78.4%
SDC2	11	21.6%
SEPT9-1	41	80.4%
SEPT9-2	2	3.9%
SFMBT2	0	0%
CHST11	0	0%
ZNF671	4	7.8%

In the remnant markers, 20 pairs of CRC and paracancerous tissues were investigated to detect cancer-specific markers. In Table 3, overall excellent discriminative capability was significantly evident, with SEPT9-2 manifesting the least discriminative abilities in six pairs, but all 20 pairs of cancerous and paracancerous tissues had detectable signals with SEPT9-2, suggesting that it was not just a cancer-specific marker but also had tissue methylation background. VAV3-AS1 exhibited discriminative capabilities across 10 pairs of samples and did not detect any signal in five cancer tissues. CHST11 was investigated across 11 groups of samples and was not detected in five cancer tissues. SFMBT2 and ZEB2 were detected across 14 pairs of samples and IKZF1 and ITGA4 across 15 pairs of samples, whereas C9orf50 exhibited discriminative capabilities within 18 pairs of samples, as listed in Table 3. The five most discriminative markers resulted in no more than three tumor tissues being undetected, and the combination of IKZF1, ITGA4, and C9orf50 can discriminate all 20 pairs of tissue samples.

Table 3 Discrimination of candidate markers for colorectal cancer in cancer and paracancerous tissues.

Marker	Detected	Discrimination (%)	P value
CHST11	11	55%	0.049
C9orf50	18	90%	< 0.0001
SFMBT2	14	70%	< 0.0001
ZEB2	14	70%	< 0.0001
IKZF1	15	75%	0.0064
VAV3-AS1	10	50%	0.0589
ITGA4	15	75%	< 0.0001
SEPT9-2	6	30%	< 0.0001

Developing and validating a panel for early CRC detection in blood

This section describes analyses based on two non-overlapping plasma cohorts with distinct objectives: (1) A training set cohort for evaluating candidate markers and selecting the optimal marker combination; and (2) An internal validation cohort from the same center for verifying the performance of the fixed final three-marker panel without additional marker screening.

Considering the clean methylation level of all eight discriminative markers in WBC testing and for improving the test sensitivity in the blood, eight markers were transferred into two tubes for qPCR testing under circumstances where only four fluorescence channels were available within a single tube. The eight markers were separated into two tubes, with some channels containing two markers. One channel combined ZEB2 with CHST11; another channel had VAV3-AS1 with SFMBT2; a combination of ZEB2 with CHST11, IKZF1, and VAV3-AS1 with SFMBT2 was created, in conjunction with an internal control gene ACTB within a single tube, and another tube contained ITGA4, C9orf50, and SEPT9-2 with ACTB.

The primary objective of this training set analysis was to evaluate the diagnostic performance of multiple candidate methylation markers and identify an optimal marker combination for CRC detection, for which 68 normal samples, 28 benign samples, and 58 malignant samples were investigated. ZEB2 with CHST11, IKZF1, ITGA4, C9orf50, and SEPT9-2 showed 100% specificity, whereas VAV3-AS1 with SFMBT2 exhibited a specificity of 92.65% in 68 normal samples. The high specificity of the selected markers in normal samples echoed previous findings on WBCs. Among the 28 benign samples, the specificity of ZEB2 with CHST11 and IKZF1 reached 100%, whereas the specificity of ITGA4 and C9orf50 was 92.86%. Furthermore, SEPT9 exhibited a specificity of 96.43%, and VAV3-AS1 with SFMBT2 had 67.86%. Compared to the marker specificity of plasma sample testing in blood from normal individuals, samples from those with benign diseases showed an overall decline in specificity, especially VAV3-AS1 with SFMBT2 combination. This might be because they all had a high methylation level in the tissue context (Figure 2). The benign disease status means there was some tissue damage, and the tissue may release cfDNA into the circulation. In 58 malignant plasma samples, the sensitivity was 29.31% for ZEB2 with CHST11, 55.17% for IKZF1, 55.17% for VAV3-AS1 with SFMBT2, 67.24% for ITGA4, 58.62% for C9orf50, and 24.14% for SEPT9. To better evaluate the detection sensitivity in early-stage samples, the sensitivity of these markers was examined in three precancerous lesions, 12 stage I CRC samples, and 17 stage II CRC samples. The outcome is illustrated in Figure 3A, with less sensitivity in ZEB2 with CHST11 and SEPT9 within early-stage plasma samples. VAV3-AS1 with SFMBT2 exhibited a high degree of redundancy, among other markers. Due to its poor specificity, upon close examination, the combination of IKZF1, ITGA4, and C9orf50 as the detection panel was deemed optimal. The panel’s overall specificity combining IKZF1, ITGA4, and C9orf50 was 100% for normal individuals, 85.71% for benign diseases, and 95.83% for control samples, and the overall sensitivity was 81.03%, with an area under the ROC curve (AUC) of 0.8962 (95%CI: 0.8339-0.9585, P < 0.0001; Figure 3B). Complete two-by-two contingency tables for different subgroups are provided in Supplementary Table 4.

Figure 2  Normalized methylation levels of eight candidate markers in paired colorectal cancer and paracancerous tissues.

Figure 3 Performance of the candidate markers in colorectal cancer detection. A: Sensitivity and specificity of the candidate markers in the training set; B: Receiver operating characteristic curve of the training and validation sets; C: Receiver operating characteristic curve of the three-group comparison in all samples.

To further substantiate the efficacy of this panel, validation within an independent cohort was conducted, including 32 patients with benign diseases and 64 with CRC. The panel had an overall specificity of 87.5% in the benign samples and a sensitivity of 79.69% in the CRC samples, with an AUC of 0.875 (95%CI: 0.8065-0.9435, P < 0.0001; Figure 3B). The outcomes of the independent validation cohort are akin to those of the training set, with a combined result yielding specificity of 100% for normal patients and 86.67% for benign patients, whereas the overall sensitivity was approximately 80.33%. The AUC for the normal vs cancerous disease patients was 0.9016 (95%CI: 0.8574-0.9459; P < 0.0001), the AUC for the benign disease patients vs malignant patients was 0.8749 (95%CI: 0.8254-0.9244; P < 0.0001), and the AUC for normal and benign patients vs malignant patients was 0.8891 (95%CI: 0.8444-0.9338, P < 0.0001; Figure 3C). With the advancement of the stage, the overall sensitivity of the panel increased from 40% for advanced adenomas, 63.16% for stage I, 81.6% for stage II, and 86.1% for stage III to 100% for stage IV.

To externally validate the performance of the selected marker panel, we conducted a comprehensive search for in silico data encompassing regions within the 3 amplicons and samples with a size exceeding 50. Given that the CpG sites within these amplicons were beyond the detection range of both the HumanMethylation450 BeadChip (450K) and the EPIC BeadChip, we were unable to utilize the extensive datasets available in public repositories using these platforms. Consequently, we obtained sequencing data employing a methylation pattern-enrichment methodology. The data were subsequently analyzed under the premise that reads in the methylation pattern-enrichment sequencing library were proportional to hypermethylation levels. For each sample, the cumulative normalized reads of the marker panel were employed for ROC analysis. The resulting AUC for the comparison between cancerous tissue and adjacent paracancerous tissue was 1.000 (95%CI: 1.000-1.000, P < 0.0001), indicating that the marker panel is highly effective and can be employed as a CRC-specific marker for tumor detection.

The correlation between the positivity rate of the final panel and clinicopathological characteristics was further explored. There was no significant association between the panel positivity rate, gender (P = 0.8148), and tumor location (P = 0.0721), whereas a significant correlation was observed for age (P = 0.0285). The positivity rate of the panel was significantly higher in patients with more advanced TNM stages than in patients with less advanced stages (stage I: 63.16%, stage II: 81.58%, and stage III-IV: 91.38%; P = 0.0173). However, the positivity rate of the panel was not significantly related to vascular invasion (P = 0.2880) or nerve invasion (P = 0.2136). Further analysis showed that panel positivity was also significantly greater in patients with more advanced T, N, and M stages (P = 0.0028, P = 0.0382, P = 0.0122, respectively). As circulating tumor DNA (ctDNA) is released from the tumor, tumor size impacted detectability. The positivity rate of the panel in patients with tumor size ≥ 5 cm was markedly higher than that in patients with tumor size < 5 cm (≥ 5 cm: 95.92% and < 5 cm: 66.15%, P < 0.0001; Table 4).

Table 4 Relationship between the methylation panel and pathological characteristics of patients with colorectal neoplasms (advanced adenoma and colorectal cancer), n (%).

Variables	Panel positive (n = 98)	Panel negative (n = 24)	P value
Gender			0.8148
Male	61 (79.22)	16 (20.78)
Female	37 (82.22)	8 (17.78)
Age (years)			0.0285
< 50	10 (66.67)	5 (33.33)
50-60	23 (69.70)	10 (30.30)
> 60	65 (87.84)	9 (12.16)
Location			0.0721
Colon	39 (81.25)	9 (18.75)
Rectosigmoid transition	35 (89.74)	4 (10.26)
Rectum	24 (68.57)	11 (31.42)
Stage			0.0173
I	12 (63.16)	7 (36.84)
II	31 (81.58)	7 (18.42)
III-IV	53 (91.38)	5 (8.62)
Unknown	2 (100.00)	0 (0.00)
T stage			0.0028
T1-T2	13 (56.52)	10 (43.48)
T3	64 (88.89)	8 (11.11)
T4	17 (85.00)	3 (15.00)
Unknown	2 (100.00)	0 (0.00)
N stage			0.0382
N0	43 (74.14)	15 (25.86)
N1	35 (85.37)	6 (14.63)
N2	16 (100.00)	0 (0.00)
Unknown	2 (100.00)	0 (0.00)
M stage			0.0122
M0	73 (77.66)	21 (22.34)
M1	21 (100.00)	0 (0.00)
Unknown	2 (100.00)	0 (0.00)
Vascular invasion			0.2880
Absent	45 (77.59)	13 (22.41)
Present	44 (86.27)	7 (13.73)
Unknown	9 (69.23)	4 (30.77)
Nerve invasion			0.2136
Absent	27 (77.14)	8 (22.86)
Present	64 (84.21)	12 (15.79)
Unknown	7 (63.64)	4 (36.36)
Size			< 0.0001
< 5 cm	43 (66.15)	22 (33.85)
5 cm	47 (95.92)	2 (4.08)
Unknown	8 (100.00)	0 (0.00)

Advanced adenomas do not have a TNM stage; therefore, the TNM staging analyses (T, N, and M stages) were based on the 117 colorectal cancer cases only.

To further evaluate the clinical applicability of this three–marker panel, we additionally calculated positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (LR) and negative LR across different control subgroups, together with their corresponding 95%CIs. In the overall cohort consisting of 117 CRC patients and 128 combined controls (68 healthy individuals + 60 benign gastrointestinal patients), the panel achieved a PPV of 94.83% (95%CI: 88.34%-97.85%), an NPV of 83.64% (95%CI: 76.10%-89.00%), positive LR of 19.43 (95%CI: 9.87-38.26) and negative LR of 0.198 (95%CI: 0.13-0.30). When only healthy volunteers served as controls, the PPV reached 100% (95%CI: 94.81%-100.00%) with an NPV of 73.91% (95%CI: 62.75%-82.57%). For benign gastrointestinal disease controls alone, PPV was 92.45% (95%CI: 83.20%-96.84%) and NPV was 68.42% (95%CI: 55.76%-78.74%).

Clinical specificity analysis in gastrointestinal diseases

Most gastrointestinal tumors possess a common DNA methylation marker, such as SEPT9. Among these, CRC and gastric cancer share the most common biomarkers. Thus, to assess the clinical specificity of the current panel in CRC detection, a total of 23 samples from gastrointestinal diseases were collected, encompassing one with gastrointestinal stromal tumor, 1 with benign gastric disease, 2 with esophageal cancers, and 19 gastric cancers, with a distribution of three stage I, six stage II, and 10 stage III. This panel called five samples positive, all of which were gastric cancers. Among these positive samples, one sample was at stage II and four at stage III. The overall clinical specificity was 78.26%.

DISCUSSION

CRC is a leading cause of cancer-related deaths worldwide, and its incidence and mortality rates are increasing in China due to westernized lifestyles and dietary changes. Screening of asymptomatic populations and early diagnosis of cancer patients are key to the prevention and control of CRC. However, current screening methods have some drawbacks, such as the cumbersome preparation and invasiveness of colonoscopy and patient noncompliance with fecal sampling, resulting in the under-recognition of populations who should be screened or tested for CRC, which is not conducive to the prevention and control of the disease. A blood test that is easy to use, has high compliance, and can be used for screening or early diagnosis, followed by colonoscopy confirmation, is a strategy that can better facilitate CRC detection. However, the first blood-based CRC detection product (Epi proColon) did not meet the requirements due to limited performance. This study reported the performance and potential applications of a blood-based three-marker panel consisting of IKZF1, ITGA4, and C9orf50 in CRC detection.

The IKZF1 gene encodes the Ikaros protein, a transcription factor involved in regulating normal lymphopoiesis and tumorigenesis. It interacts with several oncogenic pathways commonly deregulated in CRC. For example, IKZF1 can affect the Wnt/β-catenin signaling pathway[19], which is frequently activated in CRC and associated with increased cell proliferation and resistance to apoptosis[20]. IKZF1 may indirectly promote tumor growth and invasion by affecting these pathways. The ITGA4 gene encodes the 4-subunit of integrins, which are cell surface receptors involved in cell-cell and cell-extracellular matrix interactions[21]. Integrins are crucial in various cellular processes, including adhesion, migration, proliferation, and survival. The ITGA4 signaling pathway has been implicated in tumor progression and metastasis in CRC[22]. C9orf50 has been confirmed as a new regulator of immune evasion. Its essence is a disordered protein, which can drive liquid phase separation, promote adaptor assembly and maintain the accuracy of RNA splicing. Inhibition of C9orf50 will enhance the immunogenicity of tumor cells and enhance T cell infiltration in poorly infiltrated tumors, thereby inhibiting tumor growth[23]. At present, there are few studies on C9orf50. Some preliminary studies have found that its methylation can be used as a potential marker for the diagnosis of CRC, but its function and mechanism in CRC have not been clarified, which still needs further exploration. In previous studies, these markers are closely associated with abnormal changes in certain signaling pathways, and eight studies have suggested that detecting their methylation can be used to detect CRC[18,24-30]. However, comparing previous studies to this study, there are several differences. First, some previous studies have mainly focused on fecal samples, where methylation detection of ITGA4 and C9orf50 have demonstrated good performance[26,31]. However, developing a blood test with higher compliance significantly differs from developing fecal-based products. In plasma, the proportion of tumor DNA released from the tumor is even smaller and more fragmented, and the amount of cfDNA in plasma is low, with a large proportion coming from WBC. The cfDNA from WBC forms the background signal for marker detection. Therefore, a detection method with a higher signal-to-noise ratio is required to improve the detection of weak signals. This involves selecting areas with low background methylation levels for testing, which also can distinguish between tumor and nontumor tissues. In contrast, testing fecal samples only requires selecting regions that differentiate between tumor and nontumor tissues. Second, in some plasma-based studies, the sensitivity is insufficient. For example, the sensitivity of ITGA4 in detecting CRC was 36.84%, with 80.85% specificity in healthy controls[30]. In another study, the AUC of C9orf50 was 0.7, with moderate sensitivity and specificity[24]. The sensitivity of Colvera comprising IKZF1 with BCAT1 ranged from 62% to 77%, with a deteriorated sensitivity of 6% in advanced adenoma samples[27-29,32]. To improve the detection performance in plasma, a screening of markers was conducted, starting with the analysis of regions with a high signal-to-noise ratio in methylation. This allows for detecting the true-positive signals from potentially small amounts of tumor-released DNA rather than being obscured by background signals, which would otherwise fail to distinguish between cancer and noncancer patients. A more complementary set of markers for the combination was selected to enhance sensitivity, especially for early-stage patients where ctDNA is present in very low proportions. Among the 12 advanced adenomas and stage I samples, 7 cases were identified by a single marker, with four cases from IKZF1, two cases from C9orf50, and one case from ITGA4. Third, in some previous studies, plasma-based tests have shown very good performance[18,25], but their controls were often healthy populations, lacking a benign disease control group. The specificity of this group is very important in screening, especially in auxiliary diagnostic scenarios. In this study, a specificity of 86.5% was achieved in benign diseases. The efficacy of the marker panel in detecting CRC was rigorously validated through an external assessment using in silico sequencing data, ensuring the reliability and generalizability of our findings. In clinical practice, in addition to using the marker panel, it is also recommended that doctors integrate other examination results and patients’ symptoms to evaluate the cancer risk of patients. For high-risk patients, further colonoscopy is recommended to reduce unnecessary colonoscopy caused by false positives as far as possible. Clinical specificity in other tumors is also crucial for single-cancer detection, but these studies rarely address these aspects. Many methylation markers are common among gastrointestinal tumors[33,34], like SEPT9 methylation in the diagnosis of CRC and gastric cancer, leading to misidentification[14]. The marker IKZF1 has been proven to have good clinical specificity in breast and prostate cancers[35] but has detectable capabilities in esophagogastric cancers in combination with BCAT1[36]. The marker C9orf50 has also shown detectable capabilities in gastric cancer in another panel[34]. The clinical specificity of this panel for CRC detection was evaluated, primarily focusing on gastric cancer, as these two cancers are always misidentified with each other in some studies[33,34,37]. This panel has a specificity of 78.26%, and the positive call is often shown in symptomatic stage III gastric cancer samples. In order to improve the specific diagnostic efficacy of this methylation detection panel for CRC in future work, we will further verify the discrimination ability of this panel by expanding the sample cohort. At the same time, we should continue to explore more simple and effective marker panel.

In addition to the study of performance metrics, such as sensitivity and specificity, this study also attempted to analyze the correlation between the positivity rate of this panel and clinicopathological characteristics to see if there is any bias. Clinicopathological characteristics associated with the positivity rate of the panel analyses showed that the panel’s detection capabilities were closely related to tumor progression and size, as the release of ctDNA into the circulation had been mathematically shown to be closely related to tumor size and stage[38,39]. This study only missed two samples among tumors of > 5 cm, with one of them being an adenoma with high-grade intraepithelial neoplasia. The positivity rate was not significantly associated with gender, location, vascular invasion, or nerve invasion, but was significantly associated with age, which may be due to most older patients being diagnosed at a later stage in this cohort.

It is important to screen or diagnose cancers with high clinical specificity, particularly for single-cancer screening or diagnosis. Of the 19 evaluated gastric cases, 5 positive samples were identified, among which four samples were stage III gastric cancer. Given that patients frequently do not present with symptoms in the early stage, only one sample was identified as positive, indicating that this panel possesses notable clinical specificity in screening or diagnosing CRC. The discrepancy between this study and others in gastrointestinal disease detection may be due to the selected region of the markers for inquiry, as the regions from the same gene show a different methylation level, like the above result of two regions of SEPT9 testing in WBC.

This study has limitations. First, this work is a single-center retrospective pilot investigation enrolling subjects only from Zhujiang Hospital, Southern Medical University. The sample composition, demographic distribution, and clinicopathological profiles of participants may carry inherent institutional and regional selection bias, which could restrict the generalizability of our three-marker methylation panel. Although an independent internal validation cohort was incorporated to assess diagnostic performance, all specimens were collected from the same medical center; this cohort cannot be recognized as rigorous external cross-center validation. Second, the overall sample size of this pilot trial remains relatively limited, which cannot comprehensively reflect the panel’s diagnostic specificity across a broader spectrum of digestive tract neoplasms. Third, the current sample size of advanced adenoma in this study is small, which is insufficient to evaluate the screening ability of adenoma stage. Given that the adenoma detection rate is an important indicator in CRC screening, we will expand the sample size in the follow-up study to provide data support for the screening evaluation of adenoma stage. Besides, the detection sensitivity for precancerous advanced adenomas and stage I early CRC is still suboptimal, and further optimization of marker combination or experimental workflow is required to improve the identification rate of minimal early lesions.

Notwithstanding the above drawbacks, this research only serves as an exploratory pilot study to screen and verify the high signal-to-noise methylation panel for plasma-based early CRC detection. We have initiated a prospective multi-center clinical trial across multiple medical institutions in Guangdong Province to expand the sample pool, enroll more diverse populations and patients with various gastrointestinal benign and malignant lesions, and conduct standardized external cross-center validation of the IKZF1/ITGA4/C9orf50 methylation panel. In follow-up work, we will further enlarge the external validation cohort covering different geographic regions, optimize the multiplex qPCR detection system, and incorporate additional complementary methylation biomarkers to elevate the early-detection sensitivity while retaining favorable clinical specificity, so as to fully validate the clinical translational value of this blood-based assay for CRC screening and auxiliary diagnosis.

In summary, based on blood data that showed specificity in the normal population with a rate of 100% and a specificity of 86.67% among those with benign diseases and in a cohort with > 50% early-stage samples, including precancerous, stage I + II samples, the overall sensitivity reached 80.33%, with 40% at the precancerous stage and 63.16% at the stage I. Although the performance of this panel is somewhat subpar compared to stool DNA tests, such as Cologuard[40], it is superior to Epi proColon or carcinoembryonic antigen[10,41,42]. Additional markers can be integrated to improve sensitivity, and the interval between tests may be shortened, such as the situation in the MRD test, demonstrating that sensitivity increases by combining more test results[43]. Meanwhile, panel specificity was high, resulting in a relatively low false-positive rate even in that case. These findings indicate the potential of this panel in detecting CRC in the screening or auxiliary diagnosis scenario.

CONCLUSION

This study established a three-gene plasma methylation panel (IKZF1/ITGA4/C9orf50) with low leukocyte background via MPA screening. The panel achieves favorable sensitivity and high specificity for CRC, and its diagnostic efficacy was verified by independent plasma cohort and public sequencing data. Though limited by small precancerous lesion samples, this noninvasive blood assay exhibits promising translational value for early CRC screening and auxiliary diagnosis. Larger multi-center cohorts will be recruited in follow-up work to further validate and optimize its clinical performance.