An observational case-control study was conducted between January 2017 and March 2019 at The First Affiliated Hospital of Soochow University. Stool samples were collected prior to colonoscopy. All patients with CRC (37 males and 29 females) and CRA (66 males and 62 females) were first diagnosed via colonoscopy screening, and the diagnosis was later confirmed by pathology. The pathological diagnosis was performed by two professionals. Inclusion criteria were as follows: i) Age ≥18 years old; and ii) colonoscopy. The exclusion criteria for all participants included the use of the following medicines: Antibiotics within 1 month of study participation, non-steroidal anti-inflammatory drugs or probiotics. Individuals who reported chronic bowel disorders, food allergies or dietary restrictions were also excluded from the study. Additional exclusion criteria for patients with CRC included chemotherapy or radiation treatment prior to surgery. All patients were categorized according to histopathological features on the basis of the TNM classification of malignant tumors after surgery (26). A total of 83 healthy subjects (43 males and 40 females) were selected as controls by volunteering during a physical examination, and none of the healthy subjects had gastrointestinal tract disorders or any antibiotics treatments in the 3 months before sample collection. The clinical variables included age, sex and BMI (kg/m²). All 277 participants (age range, 26–88 years) had been local residents of Suzhou city for >5 years prior to the study. In total, one fecal sample was self-collected prior to bowel preparation the day before colonoscopy from each patient or healthy subject. Samples were transported to the laboratory within 24 h after collection.

All individuals provided written informed consent prior to participating in the study. All procedures were performed in accordance with, and were approved by, the ethical standards of the institutional and/or the national research committee [the Ethics Committee of The First Affiliated Hospital of Soochow University; approval no. + 056 (2016)], and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Stool samples were immediately frozen in liquid nitrogen and stored at −80°C. DNA was extracted using a TIANamp Stool DNA kit (Tiangen Biotech Co., Ltd.) according to the manufacturer's protocols (27). The integrity of DNA was measured via 2% (w/v) agarose gel electrophoresis. Purified nucleic acids were quantified using a Qubit 3.0 instrument (Thermo Fisher Scientific, Inc.), and stored at −80°C. Nucleic acids were extracted from all stool samples in a single batch by one operator to avoid inter-batch variation.

All reactions were performed in a 96-well optical PCR plate. Each reaction contained 40 ng extracted fecal DNA, 250 nM primers and 2X ChamQ Universal SYBR qPCR Master Mix (Vazyme Biotech Co., Ltd.) in 20 µl reaction volume. Amplification and detection of DNA was performed with the Applied Biosystems 7500 Fast Real-Time PCR System (Applied Biosystems; Thermo Fisher Scientific, Inc.) with the following reaction conditions: Initial denaturation at 50°C for 2 min and 95°C for 2 min, followed by 40 cycles at 95°C for 15 sec and 60°C for 1 min. The primers sequences (V3-V4) were as follows: P. micra forward, 5′-GTCACTACGGAAGAATTTGTC-3′ and reverse, 5′-GGCTTGAGCGATAATAACTTC-3′; and total bacterial DNA forward, 5′-GTGSTGCAYGGYTGTCGTCA-3′ and reverse, 5′-ACGTCRTCCMCACCTTCCTC-3′. Each sample was assayed three times. Results were analyzed using 2^−ΔΔCq method (28).

A systematic PubMed search with the terms 16S, colorectal cancer or adenoma and gut microbiome was performed to identify studies involving 16S rRNA sequencing of stool samples from patients with CRC or CRA, and healthy controls. Available data was only found in four studies: Zeller et al (29) (accession no. ERP005534); Zackular et al (8) (http://www.mothur.org/MicrobiomeBiomarkerCRC); Baxter et al (15) (accession no. SRP062005); and Mori et al (30). All four datasets were obtained from samples from patients with CRA or CRC, and healthy subjects as controls (Table I).

During data processing, short overlapping forward and reverse reads from the same fragment were joined together using PANDAseq (v0.21.1) to form overlapping sequences of the V3-V4 16S region (31). After joining, the resulting fragments were trimmed using Trimmomatic (v0.30) (32). The average probability of a base being called in error was <0.01, and the minimal length of a fragment was 100 bp. Next, the chimeric sequences were removed using Vsearch (v1.9.6) (33). The samples were uniformly subsampled at a rarefaction level of 30,000 sequences per sample, to mitigate bias of the analyses due to differences in sampling depth. Samples with <30,000 reads were removed, and a collection of sequences suitable for further Quantitative Insights Into Microbial Ecology (QIIME v1.9) analysis was thus obtained (34). The sequences were then clustered into operational taxonomic units (OTUs) using a 99% similarity cutoff, and the relative abundances were calculated for the OTUs in each sample. The OTUs were classified using the assign_taxonomy.py script in QIIME using UCLUST (v1.2.22) (35) as an assignment method, and the Silva 99% OTU database, which was a modified version of Silva v132 (36), with the removal of uncultured or unclassified entries and the addition of extra entries from CORE database (37).

All statistical analyses were conducted in R software (version 2.15.3; R Foundation for Statistical Computing). For the qPCR method, the abundance (A) of P. micra in a sample was calculated as the ΔCq relative to the total bacterial DNA in the sample, and the relative abundance was calculated as ln (A × 10⁹+1). For 16S rRNA sequencing data, the A of an OTU in a sample was calculated as the ratio of the sequence count of the OTU relative to the total number of sequences in the sample, and the relative abundance of the OTU was determined as ln (A × 10⁶ +1) (38). Differentially abundant OTUs were selected with the Wilcoxon rank-sum test for the comparison of healthy subjects with the CRC or CRA group (39). The Benjamini-Hochberg procedure was used to calculate the false discovery rate according to the adjustment of the P-values obtained from the Wilcoxon rank-sum test.

Comparisons between groups were performed with unpaired Student's t-tests and χ² tests for quantitative and categorical variables, respectively. Variables that followed a Gaussian distribution were compared with one-way ANOVA. The correlation between the quantity of the P. micra and the characteristics of patients such as age, BMI, sex and tumor progression were calculated using Kendall, Pearson or Mann-Whitney analysis. As the majority of the datasets did not meet the assumptions of a normal distribution, non-parametric Dunn's tests with Kruskal-Wallis tests or the Mann-Whitney U test were used, where applicable. P<0.05 was considered to indicate a statistically significant difference. Meta-analysis was performed using the meta for package (v2.4-0) (40). The DESeq2 package (v1.28.1) was used to conduct the difference analysis on the OTUs of crc2, crc4, crc45 and crc49, and the log2 fold change of each OTU in each sample was obtained (41). Receiver operating characteristic (ROC) curves were drawn using the pROC package (v1.16.2) (42). Other diagrams were generated using the ggplot2 (v3.3.2) and ggpubr packages (v0.4.0) (43,44).

To investigate the associations between P. micra and CRA and CRC, qPCR was performed to detect the relative abundance of P. micra in the fecal samples of 277 subjects (including 83 healthy controls, 128 patients with CRA and 66 patients with CRC) recruited from Suzhou (Table II). The results demonstrated that the relative abundance of P. micra in patients with CRC was significantly higher compared with the healthy controls and CRA (CRC vs. control, P=8.2×10⁻¹¹; CRC vs. CRA, P=4.9×10⁻⁸; Fig. 1A), while the relative abundance in patients with CRA was not different from the healthy controls (P=0.2; Fig. 1A). Then, 80% of the samples were used as the training set and the rest as the test set to establish a prediction model, and it was found that CRC samples could be distinguished from healthy control samples with an area under the curve (AUC) of 0.867 and a cutoff of 8.589 (Fig. 1B). The model performed well for the test set with an FPR (false positive rate) of 0.053 and an FNR (false negative rate) of 0.3 (Table III). However, the CRA samples were poorly distinguished from the healthy control samples (AUC, 0.554 at a cutoff of 8.311) with an FPR of 0.105 and an FNR of 0.615 (Fig. 1C; Table III). These results suggested that P. micra may serve as a diagnostic marker for CRC, but not for CRA.

It was also identified that P. micra was predominantly enriched in stages I/II and III/IV of CRC compared with the healthy controls (Fig. 2A), and the relative abundance of P. micra was not affected by the sex, BMI or the site of cancer origin (right and left) of the patients but was affected by age when a Pearson correlation was used (Fig. 2B and C; Table IV).

The association between P. micra and CRC was analyzed in four public datasets. A total of 596 samples, including 158 CRC, 253 CRA and 185 healthy control samples, were included in the analysis after quality filtering. Compared with the healthy controls, the relative abundance of P. micra in patients with CRC was significantly higher in all four datasets (crc2, P=0.001; crc4, P=0.023; crc45, P=0.0001; crc49, P=0.036) but was not different in patients with CRA (P=0.18-0.94) (Fig. 3A). Furthermore, there were significant increases in the fold changes in the relative abundance of P. micra in the CRC group compared with the healthy control group in all four datasets, while there were few changes between the healthy control and CRA groups (Fig. 3B and C).