Circulating microbiome DNA features and its effect on predicting clinicopathological characteristics of patients with colorectal cancer

Abstract

Background

Colorectal cancer (CRC) presents a complex tumor microenvironment influenced by genetic and microbial factors. Microbial DNA from the gut and tumor microenvironment can translocate into the bloodstream, forming a circulating microbiome associated with prognosis and clinicopathological features. This study investigates the peripheral venous blood microbiome in CRC patients using 2bRAD-M sequencing and evaluates its clinical significance.

Methods

Peripheral venous blood samples from 29 CRC patients (19 males, 10 females; mean age 57 years) and 10 healthy controls were analyzed to assess microbial diversity. Additionally, 20 tumor tissue samples from CRC patients were examined via RT-qPCR to validate blood-tumor microbial correlations. Statistical analyses evaluated associations between microbial abundance and clinical features, including metastasis and PD-L1 Combined Positive Score (CPS). Comparative analyses between CRC patients and healthy controls were performed to identify disease-specific microbial signatures.

Results

A total of 270 microbial species were identified, with dominant phyla including Actinomycetota, Bacillota, Bacteroidota, and Pseudomonadota. Bosea lupini was significantly associated with metastasis stage (p = 0.034), while Mycobacterium tuberculosis (p = 0.022), Porphyromonas pasteri (p = 0.017), and Bosea lupini (p = 0.045) correlated with CPS. Microbes such as Bosea lupini, Ralstonia mannitolilytica, and Porphyromonas pasteri suggested potential tumor-derived translocation into the bloodstream.

Conclusion

This study identifies a distinct peripheral venous blood microbiome in CRC patients, highlighting specific microbes associated with clinicopathological features and disease progression. These findings suggest the potential of blood microbiomes as noninvasive biomarkers for CRC prognosis and therapeutic targets, warranting further investigation in larger cohorts.

Graphical Abstract

Introduction

Colorectal cancer (CRC) can be considered a marker of socioeconomic development; as the global economy improves, the incidence of CRC rises annually. Currently, CRC accounts for over 1.9 million new cases each year, with deaths exceeding 900,000. Projections indicate that the global burden of CRC will continue to escalate [1]. CRC is a disease linked to intestinal stem cells, and its fundamental etiology remains unclear. In addition to genetic predisposition, statistical models suggest that 70–90% of CRC risk can be attributed to environmental factors such as diet, gut microbiota, and pathogenic infections [2]. Additionally, gut microbiota have been found to cross the epithelial barrier and enter the systemic circulation, reaching organs such as the liver and lungs, potentially triggering metabolic and immune responses in these organs, thereby influencing tumor progression and treatment outcomes [3]. While some microorganisms have been implicated in CRC development through mechanisms such as inducing chronic inflammation, producing genotoxic substances, and modulating immune responses [4, 5], the complexity and individual variability of microbiomes pose challenges in linking microbial composition with disease outcomes.

Traditional sequencing technologies have limitations, including low resolution, which constrain the ability to analyze microbiomes at the species level. Although metagenomics offers a powerful method for species-level analysis, it is hindered by high computational costs, complex analytical requirements, signal-to-noise issues, and host contamination [6]. These challenges are exacerbated when dealing with human samples that have limited microbial reads. Even with comprehensive metagenomic databases, researchers must determine whether detected signals are off-target or biased [7].

We employed 2bRAD-M sequencing and novel computational analysis methods to identify microbes in the blood of CRC patients and validated the presence of these microbes in paired tumor tissues using reverse transcription quantitative Polymerase Chain Reaction (RT-qPCR). This study proposes a feasible approach for identifying blood microbiota. Additionally, we calculated the correlation between microbial abundance and clinicopathological characteristics, providing preliminary evidence for CRC prognosis analysis through blood microbiome profiling.

Methods

Study population

This study recruited 29 patients diagnosed with CRC. Peripheral venous blood samples were collected from all participants, and tumor tissues were obtained from 20 patients. Tumor tissues from 9 patients were unavailable due to clinical diagnostic and therapeutic use (4 cases) or degradation and contamination caused by prolonged storage or poor preservation (5 cases). As a control group for blood microbiome analysis, peripheral venous blood samples were also collected from 10 healthy individuals with no history of malignancies, severe organ dysfunction, or recent use of antibiotics, antifungal, or antiviral drugs.

To minimize confounding factors affecting blood microbiome composition, the following inclusion and exclusion criteria were applied:

Inclusion criteria:

Confirmed CRC diagnosis.

No history of other malignancies or severe organ dysfunction (e.g., heart, liver, or kidney failure).

Exclusion criteria:

Autoimmune diseases or use of immunosuppressive medications.

Antibiotic, antifungal, or antiviral drug use within 4 weeks.

Recent chemotherapy, radiotherapy, or immunotherapy.

Severe liver or kidney disease, chronic infections, substance abuse, or heavy alcohol consumption (defined as >14 units/week for women and >21 units/week for men, based on the World Health Organization Global Status Report on Alcohol and Health, with 1 unit defined as 8 grams of pure alcohol).

Detailed demographic information for CRC patients, including comorbidities, treatment history, and lifestyle factors, is provided in Supplementary Material 2.

Sample collection

Peripheral venous blood was collected from the above patients. Each blood sample aliquot was immediately stored at – 80 °C and kept for no longer than 6 months until further analysis. 20 paired tumor tissues out of above 29 patients were obtained and then during surgery or biopsy were also promptly stored at – 80 °C. All surgical procedures were conducted under aseptic conditions.

DNA processing, library construction and sequencing for 2bRAD-M sequencing

The 2bRAD-M library preparation was primarily based on the original protocol developed by Wang et al. [8], with minor modifications. DNA ranging from 1 pg to 200 ng was digested using 4 U of the enzyme BcgI (New England Biolabs) for 3 h at 37 °C. Following digestion, adaptors were ligated to the DNA fragments. The ligation reaction consisted of combining 10 µl of digested DNA with 10 µl of a ligation master mix, which included 0.2 µM of each of two adaptors and 800 U of T4 DNA ligase (New England Biolabs). The ligation was performed at 4°Cover 12 h.

The ligation products were then amplified, and the Polymerase Chain Reaction (PCR) products were resolved on an 8% polyacrylamide gel. Bands approximately 100 bp in size were excised from the gel, and the DNA was eluted into nuclease-free water over 6–12 h at 4 °C. Sample-specific barcodes were introduced during PCR using platform-specific barcode-bearing primers. Each 20 µl PCR reaction included 6 µl of gel-extracted PCR product, 0.2 µM of each primer, 0.3 mM dNTPs, 1 × Phusion HF buffer, and 0.4 U of Phusion high-fidelity DNA polymerase (New England Biolabs). The PCR products were purified using the QIAquick PCR Purification Kit (Qiagen) and subsequently sequenced on the Illumina Nova PE150 platform.

The 2bRAD-M procedure was performed at Qingdao OE Biotech Co., Ltd. (Qingdao, China). Details of the bioinformatics analysis for 2bRAD-M are provided in Supplementary Material 1 [9, 10].

Quality control

Rigorous quality control measures were applied to ensure the integrity of the data, including adherence to the “RIDE” checklist [11]:

Sample Randomization: Samples and treatments were randomized during collection and processing to prevent batch effects and daily variation.

Uniform Processing: The same researchers, reagents, and equipment were used for all samples to reduce variability.

Clean Environment: Samples were processed in a low-contamination environment, with surfaces and equipment treated using 3% sodium hypochlorite solution and UV radiation.

Negative Controls: Each sampling, extraction, and amplification batch included negative controls (sampling blanks, DNA extraction blanks, and no-template amplification) to assess contamination.

Physical Isolation: Library preparation and DNA extraction were conducted in separate rooms to minimize cross-contamination.

Dual Indexing: Non-redundant dual indexing was used to prevent index swapping during sequencing.

Sequencer Maintenance: NaOCl and maintenance washes were performed between sequencing runs to reduce cross-contamination.

Reads level decontamination (RLD)

To remove host contamination, a read filtering method called RLD was used [12]. In this method, the number of reads in the target sample is defined as T, and the number of reads in the negative control is defined as N. The contamination level, denoted as D, can be calculated using the formula:

$$D=N*\frac{T}{T+N}$$

This approach ensures the accuracy and reliability of the sequencing data, making it suitable for downstream analysis and interpretation.

Diversity analysis

Alpha diversity was calculated using the Chao1 index, which estimates species richness by considering low-abundance species, the Shannon index, which accounts for species richness and evenness, and the Simpson index, which reflects overall community diversity. Beta diversity was assessed using the Euclidean distance, representing dissimilarity between microbial communities in a multidimensional space. These calculations were performed using the “vegan” package in R, and the results were visualized as Principal Coordinate Analysis (PCA) scatter plots.

Kyoto encyclopedia of genes and genomes (KEGG) functional prediction

Microbial community functional profiles were predicted from 16S rRNA gene sequences using PICRUSt2 (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States). KEGG Orthology (KO) functions were categorized into three hierarchical levels (Level 1, Level 2, and Level 3). Differential analysis of KEGG pathways was conducted using the Wilcoxon rank-sum test for pairwise comparisons and the Kruskal–Wallis test for multiple groups. Only statistically significant pathways (p < 0.05) were reported and visualized to illustrate functional differences between groups.

RT-qPCR assays

RNA was extracted from primary colorectal tumor tissues of 20 CRC patients using TRIzol reagent (Invitrogen) following the manufacturer’s protocol. The extracted RNA was assessed for concentration and purity using a Nanodrop spectrophotometer. Complementary DNA (cDNA) was synthesized using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems).

RT-qPCR was performed in a 20 µl reaction volume containing SYBR Green Master Mix (Applied Biosystems), specific primers for targeted microorganisms (e.g., Bosea lupini, Ralstonia mannitolilytica, Porphyromonas pasteri, and Mycobacterium tuberculosis), and 10–50 ng of cDNA template. Cycling conditions were as follows: an initial denaturation at 95 °C for 10 min, followed by 40 cycles of denaturation at 95 °C for 15 s, annealing at 60 °C for 30 s, and extension at 72 °C for 30 s. Melting curve analysis was performed to confirm product specificity.

Negative controls and technical triplicates were included to ensure reliability. RT-qPCR data were log-transformed to reduce sparsity and facilitate comparison with sequencing data. Quantitative results are presented as mean ± standard deviation (SD), consistent with statistical analysis methods described below.

Statistical analysis

All statistical analyses were conducted using R software. Binary logistic regression was applied to evaluate associations between microbial abundance and clinical or pathological features, while linear regression was used for continuous variables. To ensure comparability and reduce data sparsity, sequencing abundance and RT-qPCR-derived data were log-transformed. RT-qPCR results were reported as mean ± standard deviation (SD), whereas other data were presented as raw values unless otherwise noted. Correlation coefficients were calculated with corresponding p-values, and a significance threshold of p < 0.05 was applied.

Results

Clinical pathological characteristics of the included patients

A total of 29 patients were included in this study, comprising 19 males and 10 females, with an average age of 57 years. Among these cases, 12 patients were diagnosed as rectum cancer, 3 diagnosed as left-sided colon cancer, and 14 diagnosed as right-sided colon cancer. All the pathologies were confirmed adenocarcinomas. Tumor staging, based on the 8th edition of the American Joint Committee on Cancer (AJCC) staging system, ranged from stage IIa to IVc. Distant metastases were identified in 18 patients and involved the lungs, liver, and peritoneum, while regional lymph node involvement (N stage) was assessed within the colorectal cancer lymphatic drainage areas. The remaining 11 patients showed no evidence of distant metastases. Detailed clinical characteristics of the patients, including tumor location, stage, pathological subtype, and metastatic sites, are provided in Supplementary Material 2.

Species diversity of the blood microbiome

A total of 29 peripheral venous blood samples were analyzed, with raw read counts per sample ranging from 5,920,689 to 8,974,394. After applying rigorous quality control measures, reads containing over 8% N bases or with more than 20% of bases below Q30 were excluded. This process yielded clean tags ranging from 4,342,508 to 6,698,198 per sample. Clean reads were mapped to the 2bRAD-M database for microbial annotation. A secondary library was constructed from potential microbial genomes, and clean reads were re-mapped to this library to calculate the relative abundance of microbial species. Microbial nomenclature and sources were annotated using publicly available databases and BacDive [13].

The analysis identified 270 distinct microbial species across 29 samples, distributed among 15 phyla, 19 classes, 39 orders, 66 families, and 121 genera. The number of species detected per sample ranged from 9 to 135. Cluster analysis based on species abundance organized the data, and a heatmap (Fig. 1A–D) illustrates the microbial distribution across samples.

Fig. 1

Analysis of microbial diversity in serum samples. A–D Heatmaps of microbial abundance at the species level in 29 serum samples, hierarchically clustered and ordered based on the Euclidean distance of microbial abundance. E–G Alpha diversity analysis: Includes Chao1 and Shannon indices, represented by curves for each sample. The x-axis indicates sampling depth, and the y-axis displays the indices, quantifying microbial richness and diversity. Specaccum curves show the cumulative number of species identified relative to the number of samples. Plateauing curves suggest adequate sampling to represent species diversity. H, I PCA of microbial data: Principal Component Analysis plots illustrating sample variations based on microbial composition. Points represent individual samples, with closer points indicating similar microbial profiles. Confidence ellipses are included to demonstrate the variability and clustering of samples

Alpha diversity, evaluated using Chao1 and Shannon indices alongside species accumulation curves, showed a plateauing trend (Fig. 1E–G), indicating sufficient sequencing depth and sampling effort. Beta diversity, assessed via PCA (Fig. 1H, I), revealed that most samples clustered within a single confidence ellipse, demonstrating high similarity in microbial community composition across the samples.

Composition of the blood microbiome

Fifteen phyla were identified across all samples, with the predominant phyla including Actinomycetota (16.30%), Bacillota groups A, B, and C (7.41%, 10.37%, and 7.41%, respectively), Bacteroidota (27.78%), and Pseudomonadota (21.85%). At the genus level, 121 genera were identified, with Prevotella (16.53%), Actinomyces (10.74%), and Centipeda (8.26%) being the most abundant.

The top 30 microbial taxa at the phylum class, order, family, and genus levels are presented In Supplementary Fig. 1. At the species level, the top 30 species were analyzed and grouped based on pathological features, including lung, liver, and peritoneal metastasis, as well as the absence of distant metastasis (Fig. 2A). Cluster analysis of microbial species and patient pathological characteristics is visualized in a heatmap (Fig. 2B).

Fig. 2

Correlation of the Top 30 Microbial Abundances at the Species Level in Serum with Distant Tumor Metastasis. A This bar graph illustrates the proportional abundance of microbes in the samples, categorized by the status of tumor metastasis and specific metastatic sites. B Heatmaps depict microbial abundances with clustering of samples. Each sample is labeled according to the status and site of tumor metastasis, effectively illustrating the outcomes of the clustering analysis

Figure 3 compares the top 30 differentially abundant microbial species between healthy controls and CRC patients. Healthy controls exhibited higher relative abundances of species such as Pelomonas sp.003963075, while CRC patients showed enrichments in Aspergillus nidulans, Sphingomonas paucimobilis, and Bacillus A cereus. Relative abundances are shown in Fig. 3A, and a heatmap (Fig. 3B) highlights distinct clustering patterns between the two groups. These findings suggest significant differences in blood microbiome composition between healthy individuals and CRC patients, pointing to potential disease-associated microbial signatures.

Fig. 3

Differentially abundant microbial species in peripheral venous blood between healthy controls and CRC patients. A Stacked bar plot showing the relative abundances of the top 30 microbial species across individual samples. B Heatmap depicting clustering patterns of the top 30 microbial species between healthy controls and CRC patients, grouped by metastatic status

Beta diversity, indicator analysis, and functional prediction

Compositional and functional differences in the blood microbiome between healthy controls and CRC patients are presented in Fig. 4. The Bray–Curtis distance heatmap (Fig. 4A) revealed distinct clustering patterns: healthy controls formed a cohesive cluster with uniform microbial profiles, while CRC samples showed heterogeneous clustering partially corresponding to metastatic sites, such as lung, liver, and peritoneal metastases. PCA plots (Fig. 4B) further highlighted the separation between healthy controls and CRC patients, with the first two principal components (PC1: 23.2%, PC2: 15.2%) clearly segregating the groups, indicating significant beta diversity differences.

Fig. 4

Beta diversity, indicator species, and functional predictions of blood microbiomes in healthy controls and CRC patients. A Bray–Curtis distance heatmap of microbial community clustering patterns. B PCA plots showing beta diversity differences between groups. C Indicator species analysis identifying microbial species associated with healthy controls and CRC patients. D KEGG functional predictions illustrating differences in enriched pathways between healthy controls and CRC patients

Indicator species analysis (Fig. 4C) identified Pelomonas sp.003963075 as predominantly abundant in healthy controls, while Streptococcus pneumoniae and Klebsiella pneumoniae were enriched in CRC samples. These species may serve as potential microbial biomarkers distinguishing the groups.

KEGG functional predictions (Fig. 4D) revealed significant enrichment of pathways related to membrane transport, biosynthesis of secondary metabolites, and metabolic pathways in CRC patients compared to healthy controls, suggesting a potential microbiome-mediated influence on CRC pathogenesis.

Statistical analysis of microbial abundance and clinical correlations

The relationships among the top 30 microbial species from CRC patients’ peripheral blood and their clinical features are visualized in a Sankey diagram (Fig. 5A). This diagram illustrates the distribution of microbial species across pathological groups, including metastatic sites and Combined Positive Score (CPS) of PD-L1 classifications, highlighting specific microbial correlations with clinical characteristics. For example, Streptococcus pneumoniae and Bacillus A bombysepticus were predominantly found in non-metastatic cases, suggesting distinct microbiome patterns.

Fig. 5

Microbial associations with pathological features of serum samples visualized through Sankey diagrams. These diagrams depict the relationships between microbial abundance and pathological characteristics across various taxonomic levels, emphasizing the strength of these associations as indicated by the thickness of the lines, which represent the significance of the p-values. A Taxonomic hierarchy of the top 30 abundant microbes in serum samples, ranging from Kingdom (L1) to Species (L7). B Associations between the top 30 abundant microbes and pathological features of the samples. C Statistically significant microbial-pathological associations across all serum samples

Binary logistic regression was used to evaluate the association between microbial abundance and clinical features, including age, gender, primary tumor location, N stage, M stage, and sites of distant metastasis. Linear regression analyzed correlations with CPS of PD-L1. No significant associations were observed for age, gender, primary tumor location, or N stage. However, a trend towards significance was noted for M stage, with Bosea lupini showing statistical significance (p = 0.034). Furthermore, Mycobacterium tuberculosis (p = 0.022), Porphyromonas pasteri (p = 0.017), and Bosea lupini (p = 0.045) were significantly correlated with CPS. Other microbes exhibited trends of difference but did not reach statistical significance when analyzing metastatic sites (Fig. 5B).

Bias in microbial selection and comprehensive microbial analysis plan

The initial selection of the top 30 microbes, based on abundance, introduced potential biases, such as overrepresentation of highly abundant microbes in single samples or exclusion of low-abundance but significant microbes. To address this, we expanded the analysis to include all samples for a more comprehensive assessment of the microbial community.

This expanded analysis identified additional microbes associated with specific clinical outcomes. For liver metastasis, UBA3263 sp.001689615 (p = 0.045), Parabacteroides goldsteinii (p = 0.040), and Stenotrophomonas indicatrix (p = 0.045) were significant. For peritoneal metastasis, Muribaculum gordoncarteri (p = 0.041) and Ralstonia mannitolilytica (p = 0.023) were identified. Associations with CPS included Veillonella atypica (p < 0.01), Amulumruptor sp.001689515 (p = 0.047), UBA7173 sp.002491305 (p < 0.01), Ralstonia mannitolilytica (p = 0.017), Acinetobacter guillouiae (p < 0.01), and Stenotrophomonas maltophilia (p < 0.01) (Fig. 5C).

Microbial abundances were grouped based on pathological features (Figs. 6A, C). In Fig. 6A, a comparison between patients with and without distant metastases revealed a significant decrease in Bosea lupini abundance in the distant metastasis group, suggesting a potential association with metastatic status. Subgroup analyses (Fig. 6C) categorized patients into non-metastatic, lung metastasis, liver metastasis, and peritoneal metastasis groups. Microbes such as UBA3263 sp.001689615, Parabacteroides goldsteinii, Stenotrophomonas indicatrix, Muribaculum gordoncarteri, and Ralstonia mannitolilytica were positively correlated with liver and peritoneal metastases, indicating their potential roles in metastatic progression.

Fig. 6

Visual representations of microbial abundance and diversity across different pathological features in CRC patients. A Pyramid plot representing the abundance of Bosea lupini, grouped by “No distant Metastasis” and “Distant Metastasis Present,” showing a significant decrease in Bosea lupini abundance in the distant metastasis group. B PCA scatter plot based on Euclidean distances for “No distant Metastasis” and “Distant Metastasis Present” groups, illustrating distinct microbial abundance patterns. C Expression levels of Muribaculum gordoncarteri, Parabacteroides goldsteinii, Ralstonia mannitolilytica, Stenotrophomonas indicatrix, and UBA3263_sp001689615 in groups defined by peritoneal metastasis, liver metastasis, no distant metastasis, lung metastasis, and multi-organ metastasis excluding peritoneal metastasis (Mlb stage). D Detailed PCA scatter plot based on Euclidean distances for groups with “Non-liver or peritoneal metastasis,” “Liver metastasis,” and “Peritoneal metastasis,” highlighting distinct clusters and variations in microbial community composition. E Line plot showing the trend of microbial abundance correlating with CPS, with CPS values scaled down by a factor of 100 for visualization purposes due to the large disparity in values

PCA scatter plots based on Euclidean distances (Figs. 6B, D) showed distinct microbial abundance patterns among groups, with the non-metastatic group displaying high within-group similarity. Linear correlations between microbial abundance and CPS (Fig. 6E) highlighted marked trends, especially in sample S24 (CPS = 20). Notably, Bosea lupini and other key microbes demonstrated significant correlations with CPS, supporting the potential link between microbial profiles, clinical status, and pathological features.

Correlation of microbial presence in blood and colorectal tumor tissues: RT-qPCR analysis of key microorganisms

RT-qPCR assays quantified seven key microorganisms in primary colorectal tumor tissues from 20 CRC patients. Among these, Bosea lupini, Ralstonia mannitolilytica, Porphyromonas pasteri, and Mycobacterium tuberculosis were highly abundant in tumor tissues, suggesting potential associations with tumor biology or patient clinical status. Notably, Bosea lupini was consistently expressed across nearly all tumor samples (mean ± SD: 16.85 ± 1.80, Fig. 7A). Figure 7B highlights significant differences in the abundance of key microorganisms in peripheral blood between healthy controls and CRC patients.

Fig. 7

Correlation of microbial presence in blood and colorectal tumor tissues through RT-PCR analysis. A RT-qPCR quantification of seven key microorganisms in primary colorectal tumor tissues from 20 CRC patients. B Bar plot highlighting significant differences in the abundance of key microorganisms in peripheral blood between healthy controls and CRC patients. C–F Scatter plots illustrating correlations between microbial abundance in the blood and RT-qPCR quantification in tumor tissues for selected microorganisms. G–I Microorganisms with low detection rates in blood samples were excluded from statistical analysis

Correlation analysis revealed significant positive relationships between microbial abundance in blood and tumor tissues for Bosea lupini (Pearson’s r = 0.58, p = 0.007), Ralstonia mannitolilytica (Pearson’s r = 0.79, p < 0.001), Porphyromonas pasteri (Pearson’s r = 0.47, p = 0.034), and Mycobacterium tuberculosis (Pearson’s r = 0.80, p < 0.001) (Fig. 7C–F). Due to the low detection rates of Acinetobacter guillouiae, Stenotrophomonas maltophilia, and Veillonella atypica in blood samples (only 2 cases detected each), statistical analysis was not conducted for these microorganisms (Fig. 7G-I).

These findings suggest that certain microorganisms detected in the blood may originate from tumor tissues, potentially linking them to tumor biology or patient clinical status.

Discussion

Previous studies have shown that CRC tumors and surrounding mucosa share similar microbial compositions, but differences in abundance suggest tumor-associated microbes are distinctive [14] Our findings reinforce that analyzing gut microbial abundance alone may overlook dynamic changes driven by tumor evolution, such as the “hit-and-run” mechanism, where initial disease-triggering microbes are replaced as the tumor microenvironment evolves [15]. Using metagenomics, we identified microbes in blood, a low-biomass environment, by applying the 2bRAD-M method combined with the RLD algorithm to reduce host contamination and noise [12, 16]. Future refinements in sequencing accuracy could further enhance microbial detection and reduce host-derived interference. RT-qPCR analysis further validated significant correlations between microbes detected in blood and tumor tissues, ruling out contamination during sample collection.

In healthy individuals, microbial translocation from other body sites into the bloodstream is typically sporadic and transient [17]. However, studies have shown that patients with specific diseases, long-term organ or tissue damage, or compromised immune systems are more susceptible to specific microbial infections. This increased susceptibility likely stems from less strictly regulated microbiome niches in these hosts compared to healthy individuals [18]. In CRC patients, distinct peripheral venous blood microbiome profiles have been observed compared to healthy controls [19]. Robertina Giacconi and colleagues proposed that these blood microbes may originate from the gut [20]. While this “gut-origin” hypothesis provides a plausible explanation, our findings suggest a more complex relationship. We identified significant correlations between microbial abundance in the blood of CRC patients and clinical events such as tumor CPS status and organ metastasis. These findings indicate that the presence of microbes in the blood may not be coincidental but rather actively influenced by tumor-associated changes in the host microenvironment.

The “leaky gut” hypothesis posits that microbial translocation results from gut barrier disruption during tumor progression [21]. However, emerging evidence suggests translocation may also occur without observable gut barrier damage, driven by tumor microenvironment changes [22, 23]. This highlights the active role of tumor biology in reshaping microbiome niches, supporting our hypothesis that intratumoral microbes migrate into the bloodstream through selective niche adaptation.

Microbiome niches are determined by host-provided resources and conditions, allowing specific microbes to thrive and influence disease progression [24, 25]. Typically, physical niches are colonized by a consistent set of species or strains that combine nutritional and physiological conditions to support microbial survival [26]. We hypothesize that the migration of intratumoral microbes into the bloodstream in CRC results from the active selection of microbiome niches, a process potentially contributing to tumor metastasis and immune expression events.

Previous studies have observed associations between specific microorganisms and CRC pathological characteristics. For example, Fusobacteria are enriched in CRC tumors, with the degree of enrichment linked to liver metastasis events [27, 28]. Lactococcus and Fusobacterium are more abundant in cancerous tissues, while Pseudomonas and Escherichia-Shigella are reduced [29]. Additionally, Fusobacterium nucleatum enhances tumor invasion and migration, modulates immune responses, and may potentiate immunotherapy efficacy by inducing inflammation [30, 31]. These findings suggest that specific microbes actively shape the metastatic and immune landscape of CRC.

In our study, microbes such as Bosea lupini, Parabacteroides goldsteinii, and Ralstonia mannitolilytica showed significant correlations with metastatic events and CPS scores, indicating their involvement in CRC progression. We also observed upregulation of Fusobacteria and Escherichia at the genus level in peripheral venous blood of CRC patients. Although these findings did not show consistent correlations with pathological characteristics, they suggest a trend in microbiome niche alteration among CRC patients. Increased nutrients (e.g., nitrates) or intestinal inflammation may support the survival of specific microbes, ultimately reshaping microbiome niches [32]. Tumor-resident microbes may influence the premetastatic microenvironment, predisposing target organs to microbial colonization before the arrival of tumor cells. This phenomenon could facilitate liver metastasis in CRC, as highlighted in previous studies [33]. Together, these findings emphasize the dynamic interplay between microbial niches and CRC progression, underscoring the need for further research into the mechanisms underlying these associations.

Despite these insights, our study has limitations. The small sample size (29 CRC patients) limits the generalizability of the findings, and the cross-sectional design precludes causal inferences. Larger, longitudinal studies are needed to validate these results and explore dynamic changes in circulating microbiomes during CRC progression. Additionally, mechanistic studies in cell and animal models are required to elucidate how these microbes influence tumor biology and immune responses.

Conclusions

This study underscores the potential role of circulating microbes in CRC progression, demonstrating that microbial DNA in the blood of CRC patients correlates with metastasis and immune responses. Our findings suggest that the peripheral venous blood microbiome actively interacts with tumor biology and could serve as a promising noninvasive biomarker for CRC prognosis and a potential therapeutic target. While the study is limited by its small sample size and cross-sectional design, it lays a foundation for future research to validate these observations in larger cohorts and to explore the underlying mechanisms through functional and mechanistic experiments.