Gut microbiota signatures and predictive model of KPS in advanced colorectal cancer patients.

Introduction
Colorectal cancer (CRC), a prevalent malignant tumor of the digestive tract, has a historically higher incidence in developed European and American nations [1]. The epidemiological burden is particularly pronounced in China, where rising living standards and changing dietary patterns have driven a continuous increase in CRC incidence. Currently, CRC is the second most prevalent malignant tumor in China, exceeded only by lung cancer [2].
Current standard therapies for CRC, primarily surgical resection and combined chemoradiotherapy, are associated with considerable systemic toxicity and adverse effects. Given these therapeutic challenges, comprehensive pretreatment assessment of patients’ functional capacity is essential. The Karnofsky Performance Status (KPS) scale is the internationally established metric for this evaluation. In clinical practice, KPS scores ≤ 80 correlate with both diminished treatment response and reduced chemotherapy tolerance, while patients maintaining scores > 80 typically retain functional independence with minimal symptoms. Evidence suggests that targeted interventions aimed at elevating KPS scores are associated with better therapeutic tolerance and quality of life outcomes.
An intricate symbiotic relationship exists between the gut microbiota and the host, in which microbial communities play fundamental roles in host nutrient assimilation, immune regulation, and metabolic homeostasis [3]. The colorectal region harbors an exceptionally dense microbial population, estimated at approximating 3 × 1013 bacterial cells [4]. Importantly, bidirectional interactions between colorectal microorganisms and epithelial cells modulate essential physiological processes including energy acquisition, metabolic signaling, tissue homeostasis, and immune responses [5]. Dysbiosis-driven inflammation and carcinogen generation are well-established promoters of colorectal carcinogenesis [6], while restoring microbial equilibrium holds therapeutic potential for mitigating CRC progression [7].
To date, no study has elucidated the relationship between gut microbiota composition and KPS scores in patients with colorectal cancer. This study addresses this critical knowledge gap by performing comparative 16 S rRNA sequencing analysis of gut microbiota in advanced CRC patients stratified by KPS score (high vs. low functional status). The aim is to identify gut microbiota signatures associated with KPS status, thereby informing future hypotheses on microbiota-targeted interventions.

Methods

1. Clinical patient enrollment
Between December 1, 2023, and December 2024, seventy-three patients with advanced colorectal cancer who attended Zunyi Hospital of Traditional Chinese Medicine were recruited and divided into a high-KPS group (Group A, KPS score ≥ 80) and a low-KPS group (Group B, KPS score < 80) according to the KPS score. All enrolled patients were under active systemic treatment for advanced CRC at the time of fecal sample collection, consistent with contemporary standard regimens that primarily involve combination chemotherapy (e.g., FOLFOX, FOLFIRI, CAPEOX) with or without targeted agents (e.g., bevacizumab, cetuximab) or immunotherapy, along with appropriate supportive care, as per current clinical guidelines [8]. The study was approved by the Ethics Committee of Zunyi Hospital of Traditional Chinese Medicine (approval number: ZYZYY-03-001), adhering to the Declaration of Helsinki, and all participants provided written informed consent.
Inclusion criteria for patients with advanced colorectal cancer were as follows: (1) diagnosis of advanced colorectal cancer meeting the criteria of the Chinese Protocol of Diagnosis and Treatment of Colorectal Cancer (2023 edition) [8], confirmed by imaging and pathology. (2) age ≥ 18 years, with no restriction on gender; (3) expected survival ≥ 3 months; (4) Patients were informed about the study and signed the informed consent form.
Exclusion criteria included: (1) severe cardiovascular or cerebrovascular diseases, hepatic or renal insufficiency, or other systemic diseases; (2) other concurrent malignant tumors or autoimmune diseases; (3) pregnant or lactation; (4) use within the past 2 months of any of the following: antibiotics; probiotics, prebiotics, or other known microbiota-modulating agents. (5) any other condition deemed by the investigators to make the patient unsuitable for enrollment.

2. Data collection and laboratory examination
Basic information including gender, age, height, and weight was recorded for all subjects. Body mass index (BMI) was calculated. Additionally, for colorectal cancer patients, data on disease onset, duration, tumor stage, and KPS score were collected. Laboratory blood tests, including biochemistry, complete blood count, coagulation profile, and thyroid function, were also performed.

3. Stool sample collection and 16 S rRNA gene sequencing
Fecal samples were collected from each participant, placed into a sampling tube containing 4 mL of storage solution at room temperature, thoroughly mixed, and transported to Xiamen Treatgut Biotechnology Co., Ltd. Genomic DNA was extracted using the QIAamp Fast DNA Stool Mini Kit (QIAGEN, Germany). DNA concentration and purity were measured with a Multiskan™ GO microplate spectrophotometer (Thermo Fisher Scientific, USA), and integrity was verified by agarose gel electrophoresis. The V4 region of the bacterial 16S rRNA gene (515F-806R) was amplified by PCR using indexed primers: forward 5’‑GTGCCAGCMGCCGCGGTAA‑3’ and reverse 5’‑GGACTACHVGGGTWTCTAAT‑3’. PCR products were purified with the AxyPrep™ PCR Cleanup Kit (Axygen, USA). Library quality was assessed using a Qubit® 2.0 fluorometer (Thermo Fisher Scientific, USA) and a QSEP100 system (Bioptic, China), and final sequencing was performed on an Illumina NovaSeq platform.

4. Bioinformatics analysis
Alpha diversity metrics, including Observed Species, Shannon index, Simpson index, ACE, Chao1, and evenness (J) were calculated using QIIME2 software on the OTU abundance table for each sample. Beta diversity analysis was performed using weighted and unweighted UniFrac distance matrices, and differences in microbial community structure were visualized via Principal Coordinate Analysis (PCoA), Principal Component Analysis (PCA), and Non‑Metric Multidimensional Scaling (NMDS). Microbial taxa exhibiting significant abundance differences between groups were identified using Linear Discriminant Analysis Effect Size (LEfSe) with an LDA score threshold of 2.

5. Random forest modeling
Microbial markers that differed significantly between high-KPS group (Group A) and low-KPS group (Group B), as identified by LEfSe analysis, were used as input features. A random forest model (Group A vs. Group B) was built from genus-level relative abundance data using the “randomForest” package in R, with the aim of discriminating between the groups and identifying potential diagnostic biomarkers. Model performance was evaluated through ten‑fold cross‑validation repeated 10 times. The area under the curve (AUC) and receiver operating characteristic (ROC) curves were generated with the “pROC” package, and results were visualized using the “ggplot2” package.

6. Statistical analysis
Continuous variables that followed a normal distribution are presented as mean ± standard deviation (mean ± SD) and were compared between groups using the independent‑samples t-test. The Kruskal-Wallis test was applied to assess differences in alpha-diversity indices between groups. Differences in beta diversity were tested using permutational multivariate analysis of variance (PERMANOVA). The Wilcoxon rank‑sum test was used to compare microbial abundance across taxa, with P‑values adjusted for multiple testing via the Benjamini-Hochberg false discovery rate (FDR) correction. Predictive modeling was performed with the randomForest package (v4.7‑1.1). Statistical significance was defined as a two‑sided P‑value < 0.05.

Results

1. Baseline patient characteristics
This study included 73 patients diagnosed with CRC, 47 in the high‑KPS group (Group A; 23 males and 24 females) and 26 in the low‑KPS group (Group B; 13 males and 13 females). The median age was 62 years, ranging from 37 to 88 years, while the median BMI in the high-KPS group (Group A) was 22.89, varying between 14.87 and 30.42, and the median BMI in the low-KPS group (Group B) was 21.68, varying between 14.87 and 29.76. The median disease duration was 12 months in both groups; a summary is provided in Table 1.

2. Differences in the structure of gut microbiota in advanced colorectal cancer patients with different KPS scores
The distribution of gut microbiota OTUs was analyzed using a Venn diagram (Fig. 1A). There were 766 OTUs shared between the two groups, 272 OTUs unique to high-KPS group (Group A), and 221 OTUs unique to low-KPS group (Group B), indicating that gut microbiota structure differs between CRC patients with different KPS scores. These unique OTUs may be associated with patient functional status, potentially through effects on gut metabolism or motility. Microbiota composition profiles also differed between groups (Fig. 1B). In high-KPS group (Group A), the relative abundances of Escherichia-Shigella, Parabacteroides and Klebsiella were lower at the genus level.

Furthermore, differences were observed at multiple taxonomic levels (Fig. S1A-E). In high-KPS group (Group A), at the phylum level, the relative abundance of Proteobacteria was lower, while the relative abundances of Actinobacteria and Fusobacteria were higher. At the class level, the relative abundances of Gammaproteobacteria and Bacilli were lower, whereas the relative abundance of Negativicutes was higher. At the order level, the relative abundances of Enterobacteriales and Lactobacillales were lower, while the relative abundance of Bifidobacterales was higher. At the family level, the relative abundance of Enterobacteriaceae was lower, whereas the relative abundance of Prevotellaceae was higher. At the species level, the relative abundances of unclassified Escherichia-Shigella and unclassified Parabacteroides were lower.

3. Differences in gut microbiota diversity between CRC patients with high- and low-KPS scores
No significant differences were observed in α-diversity indices (including ACE, Chao1, Observed species, Shannon, Simpson, and evenness J) between the high-KPS and low-KPS groups (Fig. 2A), indicating comparable within-sample microbial richness and evenness. The ANOSIM, which represents the similarity analysis between groups, revealed a significant difference between groups (R = 0.092, p = 0.0376; Fig. 2B). To further evaluate inter‑individual differences, β‑diversity was assessed using unweighted UniFrac distances and visualized via 3‑dimensional principal coordinate analysis (PCoA). The first two PCoA axes explained 16.94% (PCoA1) and 7.92% (PCoA2) of the variance, with a significant separation between groups (p = 0.012). Principal component analysis (PCA) yielded similar results, where PC1 and PC2 accounted for 13.4% and 8.02% of the variance, respectively (p = 0.008; Fig. 2C). Although statistically significant, the modest proportion of variance explained by the leading axes suggests that group differences, while detectable, are embedded within substantial interpersonal microbial variability.

4. Differential genera of gut microbiota in CRC patients with different KPS scores
The top 10 differentially abundant genera are displayed in a box plot (Fig. 3A). Compared to high-KPS group (Group A), low-KPS group (Group B) had a significantly lower abundance of Bifdobacterium, Prevotella_2, Lachnospira, Anaerostipes, Lachnospiraceae_ND3007_group, Dorea, and Ruminococcaceae_UCG-013, and a significantly higher abundance of Enterococcus, Sphaerochaeta, and Hungatella.

We used linear discriminant analysis (LDA) combined with circle plots of LDA value distributions to identify key biomarkers in gut microbial communities with different KPS scores. The histogram of LDA value distributions (Fig. 3B) demonstrated significantly different taxa with scores greater than a set threshold of 2 in the LDA analysis. Potential biomarkers in the high-KPS group included Bifidobacterium at the genus level, Bifidobacteriaceae at the family level, and 23 other taxa; the low-KPS group was characterized by 12 taxa, such as Enterococcaceae at the family level, Enterococcus at the genus level, and Sphaerochaeta at the genus level. A higher LDA score indicates a greater contribution of the taxon to distinguishing the two groups.

5. Differential metabolic pathways in the gut microbiota of advanced colorectal cancer patients with different KPS scores
Differential analysis of metabolic pathway prediction among different groups using 16 S sequencing data enabled us to understand the differences in metabolic pathways of functional genes of microbial communities between groups. KEGG pathway enrichment analysis showed that, compared with high-KPS group (Group A), low-KPS group (Group B) showed significant enrichment in a total of 9 metabolic pathways, including drug metabolism-cytochrome P450, metabolism of xenobiotics by cytochrome P450, tryptophan metabolism, and others (Fig. 4A). COG analysis revealed 20 metabolic pathways with significant differences between the two groups (Fig. 4B). Among these, pathways such as the ABC-type proline/glycine betaine transport system (COG4176) and Peroxiredoxin (COG0450) were significantly enriched in patients with low KPS scores, whereas COG categories including uncharacterized conserved protein (COG2966) were enriched in patients with high-KPS scores.

6. Correlation analysis between the gut microbiota and clinical indicators in CRC patients
Next, Pearson correlation analysis was used to examine the correlation between gut microbiota and blood indices in CRC patients (Fig. 5). Albumin (ALB) level showed a significant negative correlation with the genera Tyzzerella_3 and Bifidobacterium, Hungatella was significantly positively correlated with CA199, and Ruminococcaceae_UCG‑008 was significantly positively correlated with several routine blood indicators, including BAS, LY, MONO, and NEUT.

7. A predictive model for high- and low-KPS scores in CRC patients based on differential bacterial genera
A random forest model was constructed using the 10 differential bacterial genera (LDA ≥ 2) as input features to stratify patients by KPS status. Feature importance was evaluated based on the mean decrease in the Gini coefficient (Fig. 6A). Anaerostipes, Enterococcus, and Bifidobacterium were identified as the most influential genera. Given their established roles in key gut processes, such as short‑chain fatty acid synthesis, immune regulation, and pathobiont control, these genera may have substantial relevance to the clinical outcomes examined here. The model demonstrated high performance in the validation set (AUC = 0.992) and retained predictive ability in the test set (AUC = 0.750), with corresponding ROC curves shown in Fig. 6B.

Discussion
The KPS score serves as a critical prognostic indicator in oncology, with lower scores (KPS ≤ 80) strongly correlating with diminished treatment response and reduced chemotherapy tolerance. This comparative analysis of gut microbiota structure between CRC patients with high- and low-KPS scores was conducted by examining taxonomic composition and diversity. Using 16 S rRNA gene sequencing and bioinformatics analysis, we identified significant beta-diversity separation between the two groups, while alpha-diversity indices showed no statistically significant difference. Differential abundance analysis revealed a distinct microbial signature: the high-KPS group (Group A) was enriched in beneficial genera such as Bifidobacterium, Prevotella_2, Lachnospira, and Anaerostipes, whereas the low-KPS group (Group B) showed a marked increase in Enterococcus. Functional prediction based on KEGG and COG databases further indicated that low-KPS group (Group B) exhibited significant enrichment in pathways related to xenobiotic metabolism (e.g., cytochrome P450) and oxidative stress response (e.g., peroxiredoxin), which are often associated with tumor progression and inflammation. These structural and functional disparities in the gut microbiota were leveraged to develop a random forest model using the top differential genera, which demonstrated high predictive accuracy (AUC = 0.992) for stratifying patients by KPS status, demonstrating its potential as a clinical stratification tool. Collectively, this integrated analysis transcends descriptive profiling toward a more functional and predictive understanding of the microbiota‑status association in CRC.
Prior research has established significant gut microbiota diversity differences between CRC patients and healthy individuals [9]. Aligning with these findings, our analysis reveals distinct microbial diversity patterns between high- and low-KPS CRC patients, suggesting that disease-mediated health status deterioration may involve microbiota alterations. Consistent with reports of reduced Bifidobacterium abundance in CRC versus healthy controls, we observed significantly higher abundance of SCFA-producing probiotic bacteria (including Bifidobacterium, Lachnospira, Anaerostipes, Lachnospiraceae_ND3007_group, and Ruminococcaceae_UCG-013) in high-KPS patients [10–13]. High-KPS group (Group A) exhibited enrichment of SCFA-producing bacteria. This observation is consistent with the documented role of SCFAs in inhibiting CRC metastasis [14] and suppressing gut inflammation [15], which may be associated with a better functional status. The significant reduction of Hungatella in low-KPS patients is an intriguing finding. Although the role of Hungatella in cancer progression is not well-defined, its reported associations with various pathological conditions, including psychiatric disorders [16] and chronic kidney disease [17], suggest it may be a broad indicator of compromised host health. In our study, its positive correlation with the tumor marker CA199 [18] specifically suggests a potential link to CRC-associated pathophysiology. We therefore hypothesize that the co-occurrence of reduced Hungatella, elevated CA199, and poor performance status may reflect a shared underlying state of systemic inflammation or dysbiosis in advanced CRC. This observation generates a hypothesis that warrants direct validation in future CRC-focused studies.
KEGG metabolic pathway analysis revealed that drug metabolism-cytochrome P450 (CYP450), Metabolism of xenobiotics by cytochrome P450, two pathways related to CYP450 metabolism, were significantly enriched in patients with low-KPS scores, and the CYP450 metabolism The fatty acid epoxides produced are hydrolyzed by soluble epoxide hydrolases (sEH) to dihydrometabolites (e.g., DHETs), which typically have proinflammatory or low biological activity properties. sEH has been shown to promote CRC progression [19]. In addition, COG metabolic analysis revealed that Peroxiredoxin (PRDX) metabolic pathway was significantly enriched in patients with low-KPS scores, and PRDX was similarly associated with tumor progression and poor prognosis [20]. The enrichment of CYP450 and PRDX metabolic pathways in low-KPS patients suggests a potential link between these microbiota-associated functions and patient performance status, but further studies are needed to confirm this. The KEGG and COG pathway predictions are inferred from 16 S rRNA gene sequencing data rather than derived from direct metagenomic or metatranscriptomic analysis. Consequently, these results should be interpreted as hypotheses regarding potential metabolic differences that require future validation through direct functional profiling.
This study has several limitations. First, the sample size, particularly of the low-KPS subgroup, is relatively small, which may affect the generalizability of the findings and is a common challenge in initial exploratory microbiome studies. Consequently, the high discriminatory performance of our model in internal cross-validation (AUC = 0.992) must be interpreted with caution due to the risk of overfitting. This result primarily underscores the associative strength of the selected microbial signature within our cohort rather than its proven generalizability. Second, the single-center design (Zunyi Hospital) may limit the extrapolation of our results to broader populations with different demographic, dietary, or clinical management backgrounds. Third, while a formal a priori power calculation was not feasible due to the absence of preliminary effect size data for microbiota-KPS associations in CRC, the stringent participant criteria aimed to minimize confounding variability. Notably, despite the sample size and single-center setting, we observed statistically significant separations in β-diversity, consistent differential abundance of key genera, and a high-performance predictive model. Furthermore, potential variations in diet and specific treatment details among participants, which were not accounted for, might have introduced additional confounding. To address these limitations and advance toward clinical translation, future multi-center studies with larger, prospective cohorts are essential to validate these associations, elucidate potential causal mechanisms, and externally test the predictive model to determine its robustness and clinical applicability.
In summary, comparative analysis of gut microbiota structure and composition in distinct CRC patient subgroups stratified by KPS revealed novel associations between specific microbial genera and patient functional status, rather than causal relationships. We identified both potentially beneficial and pathogenic bacterial signatures associated with clinical outcomes in colorectal cancer. These findings provide a rationale for future longitudinal studies to explore the potential causal links between gut microbiota dynamics and CRC progression. Furthermore, the random forest model constructed from differential microbial features demonstrates potential utility in the exploratory assessment of functional status in CRC patients, warranting further validation before clinical application.

Supplementary Information