Plant-based diet quality is associated with esophageal mucosa-associated microbiome profiles and disease severity in a high-risk Chinese population

Abstract

Background

Esophageal squamous cell carcinoma (ESCC) remains a major public health challenge in China. Diet is a modifiable risk factor, and the esophageal mucosa-associated microbiome may contribute to esophageal disease progression. However, population-based evidence integrating plant-based diet quality, esophageal microbiome profiles anddisease severity remains limited.

Methods

We conducted a cross-sectional study including 236 participants undergoing upper gastrointestinal screening in Linzhou, a high-risk region for ESCC. Dietary intake was assessed using a food frequency questionnaire, and plant-based diet indices, including the overall plant-based diet index (PDI), healthful plant-based diet index (hPDI) and unhealthful plant-based diet index (uPDI), were calculated. Esophageal mucosal swabs were profiled by 16S rRNA gene sequencing. Associations among diet indices, microbiome features, and esophageal disease severity were evaluated using regression and risk stratification models.

Results

Higher hPDI was inversely associated with esophageal disease severity, whereas higher uPDI were positively associated. Higher hPDI was associated with greater relative abundance of Rothia and Prevotellaceae-related taxa, whereas higher uPDI was associated with enrichment of Streptobacillus and Fretibacterium. Integrating dietary indices with microbiome features modestly but consistently improved risk stratification beyond epidemiological factors alone (e.g. AUC for the combined hPDI-microbiome model was 0.93, 95%CI: 0.89–0.98).

Conclusions

Healthful and unhealthful plant-based diet patterns showed opposite associations with esophageal disease severity and distinct mucosa-associated microbiome profiles, supporting an exploratory diet-microbiome framework for ESCC risk stratification.

Introduction

Esophageal cancer is a major global health burden, and China accounts for a substantial proportion of global esophageal cancer incidence and mortality. In 2022, China had an estimated 224,012 new esophageal cancer cases and 187,467 deaths, representing approximately 43.8% and 42.1% of the global totals, respectively. ESCC remains the predominant histological subtype in China [1]. Despite advances in treatment and population-based screening programs, the prognosis of ESCC remains poor because early-stage disease and precancerous lesions are often asymptomatic and difficult to detect [2]. Improving prevention and early risk identification, therefore, remains a public health priority, especially in high-risk regions.

Dietary factors are among the most consistently reported and potentially modifiable risk determinants for ESCC [3,4]. Previous epidemiological studies have linked low intake of fruits and vegetables, high consumption of pickled or highly processed foods and other unfavorable dietary habits to increased ESCC risk [5,6]. A recent large-scale prospective study in high-risk Chinese populations further demonstrated that unbalanced diets were associated with increased ESCC risk [7]. However, focusing on individual foods or nutrients may not fully capture the complexity of real-world dietary patterns. To address this limitation, plant-based diet indices have been developed to characterize overall diet quality by jointly considering healthful plant foods, unhealthful plant foods and animal-based foods [8]. The overall plant-based diet index (PDI) reflects greater intake of plant foods and lower intake of animal foods without distinguishing plant food quality. The healthful plant-based diet index (hPDI) assigns higher scores to nutrient-dense plant foods such as whole grains, fruits, vegetables, legumes, nuts and tea, whereas the unhealthful plant-based diet index (uPDI) assigns higher scores to less healthy plant foods such as refined grains, preserved or salted vegetables and added sugars. Therefore, hPDI and uPDI allow differentiation between plant-rich diets of higher and lower nutritional quality. These indices have demonstrated robust associations with health outcomes [9,10], yet their relevance to ESCC and its precancerous stages remains insufficiently explored.

In parallel, growing evidence indicates that the human microbiome plays an important role in cancer-related processes and is strongly influenced by diet [11,12]. In the context of ESCC, recent studies have also reported distinct microbial profiles associated with disease presence and progression [13], and microbiome-based models using oral or esophageal samples have shown potential for distinguishing ESCC from non-cancer states [14]. Importantly, the esophagus harbors a unique mucosa-associated microbial community that differs from fecal gut microbiota and is shaped by oral microbial influx, dietary exposures and local mucosal conditions [15]. Despite this, most diet-microbiome studies in ESCC have focused on gut microbiota or individual dietary behaviors, with relatively limited attention to comprehensive dietary patterns and esophageal mucosa-associated microbiome profiles.

Diet and the microbiome are therefore increasingly recognized as interconnected factors relevant to esophageal health. Diet quality may influence esophageal disease risk not only directly, but also indirectly by shaping local microbial ecological states along the oral-esophageal axis [16,17]. Xiao et al. investigated the impact of hazardous dietary habits on the microbial composition of esophageal precancerous lesions and found that such habits significantly affected the microbial profiles of affected patients [18]. A recent study found that different dietary habits lead to alterations in ESCC-related crucial microbiomes and functional pathways [16]. However, existing studies have largely examined dietary factors and microbial features in isolation, and their combined relevance to ESCC risk across the spectrum of esophageal disease severity remains underexplored at the population level.

Against this background, we conducted a population-based study in a high-risk region for ESCC to investigate the associations between plant-based diet quality, esophageal mucosa-associated microbiome profiles and esophageal disease severity. Specifically, we aimed to (1) examine the relationships between hPDI and uPDI and the prevalence of esophageal precancerous lesions and ESCC; (2) characterize diet-associated microbial configurations in esophageal swab samples; and (3) evaluate whether integrating dietary indices with microbiome features could improve risk stratification beyond established epidemiological factors. By integrating dietary assessment with esophageal microbiome profiling, this study seeks to provide a coherent framework for understanding how lifestyle-related exposures relate to esophageal microbial ecology and disease risk, with implications for ESCC prevention and risk assessment in high-risk populations.

Materials and methods

Study design and participants

This study was designed as a retrospective cross-sectional analysis based on the secondary use of questionnaire data, endoscopic/pathological information and esophageal swab specimens collected from participants undergoing upper gastrointestinal (UGI) cancer screening and related clinical evaluation in Linzhou, Henan Province, China, a well-recognized high-risk area for ESCC. For the present analysis, we retrospectively included 236 participants who underwent UGI cancer screening at Linzhou Cancer Hospital, during two discrete recruitment periods: August to September 2018 and January to March 2019. Participants were eligible for the present analysis if they met the following criteria: (1) aged 40-69 years at recruitment (the age range was selected in accordance with population-based UGI cancer screening practices in high-risk regions of China, where screening has generally targeted middle-aged and older adults at elevated risk for ESCC and its precancerous lesions); (2) permanent residents of Linzhou or neighboring high-risk areas; (3) completed the standardized epidemiological and dietary questionnaires; (4) underwent upper gastrointestinal endoscopy with histopathological confirmation; and (5) had esophageal swab samples available for 16S rRNA gene sequencing. Participants were excluded if they met any of the following criteria: (1) history of malignant tumors; (2) previous treatment for esophageal cancer or precancerous lesions; (3) incomplete key questionnaire, endoscopic, or pathological information; (4) insufficient sequencing depth or failure to pass microbiome sequencing quality control; (5) recent antibiotic use; (6) active inflammatory or infectious diseases; or (7) recent use of medications known to affect gastrointestinal microbial composition, such as proton-pump inhibitors. Based on pathological findings, participants were classified along the esophageal disease spectrum into normal esophageal mucosa, esophagitis, low-grade intraepithelial neoplasia (LGIN) and advanced esophageal neoplasia (AEN). AEN was defined as high-grade intraepithelial neoplasia (HGIN) or esophageal squamous cell carcinoma (ESCC). Trained interviewers collected standardized information on sociodemographic characteristics, lifestyle factors (including smoking and alcohol consumption), dietary habits, history of digestive system diseases and oral health conditions using structured questionnaires. A flow diagram of participant selection and study design is shown in Figure S1.

This study was approved by the Ethics Committee of the National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College (Approval No. 16-171/1250). All participants provided written informed consent prior to participation. All procedures were conducted in accordance with the Declaration of Helsinki.

Assessment of plant-based diet indices

Dietary intake over the past year was assessed using a validated semi-quantitative food frequency questionnaire (FFQ) administered by trained interviewers. The FFQ was adapted from dietary questionnaires previously used in large-scale Chinese population-based nutritional epidemiology and upper gastrointestinal cancer screening studies and was designed to capture habitual intake frequency and portion size of major food groups commonly consumed in the local population. The application of similar FFQ-based dietary assessment methods and plant-based diet index construction in Chinese populations has been described in previous studies [7]. Nevertheless, because FFQ data rely on participants' retrospective reporting of usual dietary intake, potential recall bias and measurement error cannot be fully avoided. For this analysis, the 16 food groups were selected based on the food items available in the questionnaire, local dietary characteristics and previously established plant-based diet index frameworks adapted to Chinese populations. The 16 predefined food groups were categorized into three groups based on nutritional quality following established definitions: (1) Healthful plant-based foods (whole grains, fruits, vegetables, legumes/soy products, vegetable oils, nuts, garlic and tea); (2) Unhealthful plant-based foods (refined grains, preserved/salted vegetables and added sugars); (3) Animal-based foods (meat, poultry, fish and seafood, eggs, dairy products and animal fats). This classification was used to characterize overall plant-based diet quality and should not be interpreted as evidence that any individual food item is independently protective.

Based on these categories, three plant-based diet indices were calculated: (1) Overall Plant-Based Diet Index (PDI): higher scores reflect higher intake of all plant foods (regardless of quality, both healthy and unhealthy) and lower intake of animal-based foods. Each food group was assigned a positive score (e.g. 5 for highest quintile of intake, 1 for lowest) if it was plant-based, and reverse-scored if animal-based; (2) Healthful Plant-Based Diet Index (hPDI): a variant of PDI that differentiates quality by giving positive scores for intake of healthful plant foods and negative (reverse) scores for unhealthy plant foods, while animal-based food groups were reverse-scored according to the standard plant-based diet index algorithm. A high hPDI indicates a diet rich in whole plant foods and limited in less healthy plants and animal products; (3) Unhealthy plant-based diet index (uPDI): the inverse of hPDI with respect to plant food quality [8]. Here, high intake of refined grains, sweets and other less healthy plant-based items yield higher scores, while higher consumption of whole plant foods yields lower scores. A high uPDI captures a plant-based diet pattern that is quantitatively plant-rich but nutritionally poor. Because the FFQ was designed primarily to capture habitual food group intake rather than detailed nutrient-level intake, total energy intake could not be estimated with sufficient precision. Therefore, PDI, hPDI and uPDI scores were not directly adjusted for total energy intake. The diet-related FFQ items, response categories, food group classification and scoring algorithm used to construct PDI, hPDI and uPDI are provided in Table S1 and Table S2.

For ease of interpretation, participants were categorized into high and low adherence groups for each plant-based diet index based on the median value when comparing microbiome measures. Continuous index scores were used in regression models assessing associations with esophageal disease severity. The reliability and validity of these indices in the Chinese dietary context have been supported by previous research [7].

Sample collection

Esophagus swab specimens were collected during routine upper gastrointestinal endoscopic examinations. To minimize contamination, esophageal swab specimens were collected before biopsy sampling and before any therapeutic manipulation. Sterile polyester-tipped swabs (Puritan Medical Products, USA) were introduced through the endoscope working channel under direct visualization, and operators were instructed to avoid unnecessary contact with oral secretions, gastric fluid and non-target mucosal surfaces. The swab was applied only after the endoscope had reached the target esophageal site. For participants with visible lesions, swabs were obtained directly from the lesion site using five looping brush passes. For participants without visible lesions, swabs were collected from the middle esophagus using the same procedure. The middle esophagus was selected as the standardized sampling site because it is anatomically comparable across individuals, is commonly involved in the field cancerization context of ESCC in high-risk regions and avoids arbitrary sampling from multiple non-lesional sites. Immediately after sampling, the swab head was cut into a sterile cryovial containing 1.5 mL of cell-preserving solution (PreservCyt®, Hologic, USA). Paired biopsy specimens from the same anatomical sites were obtained for histopathological diagnosis. All specimens were immediately transferred to −80 °C storage within 30 minutes and transported to the laboratory on dry ice.

All endoscopic examinations and swab collections were performed by trained endoscopists and project staff following a standardized operating protocol. Before study implementation, personnel received unified training on sampling site identification, swab handling, number of brush passes, sample labeling and storage procedures to reduce inter-operator variability.

DNA extraction, amplification and sequencing

DNA was extracted from esophageal swab samples using the QIAGEN PowerSoil DNA Isolation Kit (QIAGEN, Germany) according to the manufacturer's protocol. Extracted DNA was stored at −80 °C in Tris-EDTA buffer until further processing. Negative controls (reagent blanks without DNA template) were included throughout the extraction and amplification procedures to monitor potential contamination. The V4 hypervariable region of the bacterial 16S rRNA gene was amplified using the universal primer pair 515F (5′-GTGYCAGCMGCCGCGGTAA-3′) and 806 R (5′-GGACTACNVGGGTWTCTAAT-3′) [19,20], incorporating sample-specific barcodes. PCR reactions were performed in a total volume of 50 μL containing 1 μL of forward primer (10 μmol/L), 1 μL of reverse primer (10 μmol/L), 1 μL of template DNA, 4 μL of dNTPs (2.5 mmol/L), 5 μL of 10 × EasyPfu Buffer, 1 μL of EasyPfu DNA Polymerase (2.5 U/μL) and nuclease-free water to the final volume. The PCR thermal cycling conditions were as follows: initial denaturation at 95 °C for 5 min; 30 cycles of denaturation at 94 °C for 30 s, annealing at 60 °C for 30 s and extension at 72 °C for 40 s; followed by a final extension at 72 °C for 4 min. Amplicons were examined by agarose gel electrophoresis, and the expected amplicon size was approximately 300–350 bp. Purified amplicons were quantified using the Qubit dsDNA HS Assay Kit and pooled at equimolar concentrations. In the original sequencing dataset, 470 of 472 collected esophageal specimens were successfully amplified and sequenced, corresponding to a PCR amplification and sequencing success rate of 99.6%. Sequencing was performed on the Illumina MiniSeq platform using paired-end 2 × 150 bp chemistry. The present analysis included esophageal swab samples that passed amplification, sequencing and downstream quality-control procedures. All samples were sequenced in a single run to minimize batch effects and improve internal comparability across disease and dietary groups. However, we acknowledge that this design does not address inter-run reproducibility, and the observed microbial signatures require validation in independent cohorts and sequencing batches.

Sequence processing and taxonomic classification

Raw sequencing data were processed using QIIME2 (https://qiime2.org/) [21]. Quality control, denoising, chimera removal and feature table construction were performed using the DADA2 plugin [22]. Possible phiX reads and chimeric sequences were removed. Paired-end reads were truncated at 140 bp for both forward and reverse reads based on quality profiles and merged with a maximum mismatch of six bases. Low-abundance features (<5 total reads across all samples) were removed. To account for differences in sequencing depth, samples were rarefied to 1,000 reads per sample for diversity and compositional analyzes. Because the negative control was used primarily as a no-template control during amplification and did not show evident amplification, formal contaminant feature modeling based on negative-control sequencing abundance was not performed. Taxonomic classification was performed using a Naive Bayes classifier trained on the Greengenes 13_8 reference database at 99% similarity [23]. Microbial composition was summarized at the phylum and genus levels.

Statistical analysis

All statistical analyzes were conducted using R software (version 4.5.2). Alpha diversity was assessed using the Shannon index and observed operational taxonomic units (OTUs). Differences in alpha diversity across dietary index groups (categorized into low and high groups based on the median) and disease groups were evaluated using Wilcoxon rank-sum tests or Kruskal–Wallis tests, as appropriate. Beta diversity was assessed using unweighted UniFrac distances, and principal coordinates analysis (PCoA) was used for visualization. Differences in microbial community composition were tested using PERMANOVA (adonis2) with 999 permutations. Differentially abundant genera between groups were identified using Wilcoxon rank-sum tests based on relative abundance data, with multiple testing correction applied using the Benjamini‒Hochberg false discovery rate (FDR) method. Genera with FDR-adjusted P values < 0.05 were considered statistically significant. Effect sizes were quantified using log2 fold change (log2FC), calculated as the ratio of mean relative abundance between groups. Associations between plant-based diet indices and esophageal disease severity were evaluated using ordinal logistic regression models. Model 1 adjusted for age and sex. Model 2 further adjusted for smoking status, alcohol consumption, family history of cancer, intake of hot food, intake of fried food and teeth loss as an indicator of oral health status. These covariates were selected a priori based on previous literature and their potential associations with dietary patterns, oral-esophageal microbial profiles and ESCC risk. Esophageal pathological status was modeled as the ordered outcome reflecting disease severity and was therefore not included as a covariate. Participants with recent antibiotic use, active inflammatory or infectious diseases or recent use of medications known to affect gastrointestinal microbiota were excluded during eligibility screening. The proportional odds assumption was assessed and not violated. For risk stratification analyzes, logistic regression models were constructed using (1) dietary indices alone, (2) selected microbial features alone and (3) combined dietary and microbial features. Model performance was evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC). 10-fold cross-validation was applied for internal validation. All statistical tests were two-sided, and P < 0.05 was considered statistically significant.

Results

Baseline characteristics of participants

A total of 236 participants were included in the analysis. According to histopathological diagnosis, participants were classified into the normal group (n = 70), esophagitis group (n = 70), LGIN group (n = 70) and AEN group (n = 26; including 19 HGIN and 7 ESCC cases). Given the limited number of ESCC cases and our primary interest in identifying microbiome signatures associated with advanced lesions relevant to cancer prevention, HGIN and ESCC were combined and analyzed as AEN. Baseline characteristics differed across disease groups with respect to age, gender, education level, income, smoking status and dietary indices (Table 1). These imbalances indicated potential confounding and informed the selection of covariates in subsequent multivariable models, particularly age, sex, smoking status, alcohol consumption, family history of cancer, selected dietary habits and tooth loss.

Table 1.

Baseline characteristics of study participants according to esophageal disease status.

Variable	Overall N = 236	Normal N = 70	Esophagitis N = 70	LGIN N = 70	AEN N = 26	P
Age (years)	54.66 ± 7.30	51.89 ± 7.66	54.80 ± 7.16	56.21 ± 6.43	57.54 ± 6.85	<0.001
Gender						0.073
Male	89 (37.7%)	21 (30.0)	29 (41.4%)	24 (34.3%)	15 (57.7%)
Female	147 (62.3%)	49 (70.0)	41 (58.6%)	46 (65.7%)	11 (42.3%)
Marital status						0.466
Married/living with partner	208 (88.1%)	65 (92.9%)	61 (87.1%)	59 (84.3%)	23 (88.5%)
Divorced/seperated/widowed	28 (11.9%)	5 (7.1%)	9 (12.9%)	11 (15.7%)	3 (11.5%)
Education level						0.002
Primary school and lower	107 (45.3%)	30 (42.9%)	20 (28.6%)	43 (61.4%)	14 (53.8%)
Junior high school	98 (41.5%)	32 (45.7%)	36 (51.4%)	18 (25.7%)	12 (46.2%)
High school and above	31 (13.1%)	8 (11.4%)	14 (20.0%)	9 (12.9%)	0 (0.0%)
Occupation						0.836
Labor	203 (86.0%)	59 (84.3%)	62 (88.6%)	59 (84.3%)	23 (88.5%)
Others	33 (14.0%)	11 (15.7%)	8 (11.4%)	11 (15.7%)	3 (11.5%)
Income						0.057
<30,000 CNY	46 (19.5%)	5 (7.1%)	14 (20.0%)	19 (27.1%)	8 (30.8%)
30,000–50,000 CNY	120 (50.8%)	43 (61.4%)	36 (51.4%)	30 (42.9%)	11 (42.3%)
>50,000 CNY	70 (29.7%)	22 (31.4%)	20 (28.6%)	21 (30.0%)	7 (26.9%)
Smoking status						0.105
Yes	38 (16.1%)	11 (15.7%)	12 (17.1%)	7 (10.0%)	8 (30.8%)
No	198 (83.9%)	59 (84.3%)	58 (82.9%)	63 (90.0%)	18 (69.2%)
Drink status						0.784
Yes	7 (3.0%)	2 (2.9%)	3 (4.3%)	1 (1.4%)	1 (3.8%)
No	229 (97.0%)	68 (97.1%)	67 (95.7%)	69 (98.6%)	25 (96.2%)
Hot food						0.475
Yes	10 (4.2%)	3 (4.3%)	1 (1.4%)	4 (5.7%)	2 (7.7%)
No	226 (95.8%)	67 (95.7%)	69 (98.6%)	66 (94.3%)	24 (92.3%)
Fried food						0.017
Yes	9 (3.8%)	1 (1.4%)	0 (0.0%)	5 (7.1%)	3 (11.5%)
No	227 (96.2%)	69 (98.6%)	70 (100%)	65 (92.9%)	23 (88.5%)
Spicy food						0.371
Yes	47 (19.9%)	16 (22.9%)	17 (24.3%)	11 (15.7%)	3 (11.5%)
No	189 (80.1%)	54 (77.1%)	53 (75.7%)	59 (84.3%)	23 (88.5%)
Tooth loss						0.102
<3 teeth	99 (41.9%)	36 (51.4%)	26 (37.1%)	30 (42.9%)	7 (26.9%)
3-12 teeth	69 (29.2%)	23 (32.9%)	20 (28.6%)	17 (24.3%)	9 (34.6%)
>12 teeth	68 (28.8%)	11 (15.7%)	24 (34.3%)	23 (32.9%)	10 (38.5%)
Family history of cancer						0.987
Yes	152 (64.4%)	45 (64%)	44 (63%)	43 (69%)	20 (59%)
No	84 (35.6%)	25 (36%)	26 (37%)	19 (31%)	14 (41%)
PDI	52.02 ± 3.38	57.27 ± 1.34	54.63 ± 1.32	51.27 ± 0.87	48.20 ± 1.79	<0.001
hPDI	52.81 ± 2.99	55.77 ± 1.96	53.51 ± 1.53	50.77 ± 1.65	48.46 ± 1.50	<0.001
uPDI	50.62 ± 3.21	47.14 ± 2.29	50.36 ± 1.46	52.63 ± 1.45	55.31 ± 1.54	<0.001

Note. Values are presented as mean ± standard deviation (SD) for continuous variables and number (percentage) for categorical variables. Diet-related variables (except spicy food) refer to dietary frequency during the past year. PDI, overall plant-based diet index; hPDI, healthful plant-based diet index; uPDI, unhealthful plant-based diet index.

Associations between plant-based diet indices and esophageal disease severity

Associations between plant-based diet indices and esophageal disease severity were evaluated using ordinal logistic regression models (Table 2), with esophageal disease status modeled as an ordered outcome reflecting increasing pathological severity (normal, esophagitis, LGIN and AEN). Higher adherence to PDI and hPDI was inversely associated with esophageal disease severity, whereas higher uPDI scores were positively associated with more advanced disease stages. In the fully adjusted model (controlling for age, gender, smoking, drinking, family history and hot/fried food habits), a higher hPDI was significantly associated with a 17% lower odds of being classified into a more advanced disease (OR = 0.83, 95% CI: 0.76–0.91, P = 0.003). In contrast, the uPDI showed a significant positive association with disease severity (OR = 1.22, 95% CI: 1.11–1.34, P = 0.002 in the fully adjusted model). The overall PDI showed a weaker inverse association (OR = 0.93, 95% CI: 0.88–0.98, P = 0.009 in the crude model), which was attenuated and no longer statistically significant after multivariable adjustment. Sensitivity analyzes excluding AEN cases (restricted to normal, esophagitis and LGIN groups) yielded consistent results (Table S3).

Table 2.

Associations of plant-based diet indices with esophageal disease severity.

	Crude model			Model 1			Model 2
	OR	95% CI	P value	OR	95% CI	P value	OR	95% CI	P value
PDI	0.93	0.88, 0.98	0.009	0.94	0.89, 0.99	0.018	0.95	0.90, 1.01	0.073
hPDI	0.78	0.72, 0.84	<0.001	0.80	0.74, 0.87	<0.001	0.83	0.76, 0.91	0.003
uPDI	1.28	1.18, 1.40	<0.001	1.25	1.15, 1.37	<0.001	1.22	1.11, 1.34	0.002

Note. Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated using ordinal logistic regression models. Crude Model: not adjusted for covariates. Model 1: adjusted for age and gender. Model 2: adjusted for age, gender, smoking status, drinking status, family history of cancer, intake of hot food, intake of fried food and teeth loss.

Microbial diversity across plant-based diet patterns

We next examined whether esophageal mucosa-associated microbial diversity differed according to plant-based diet adherence. Participants were dichotomized into high and low groups based on the median values of PDI, hPDI and uPDI. As shown in Table 3, individuals with higher hPDI scores tended to exhibit modestly higher alpha diversity compared with those with lower hPDI scores, as assessed by the Shannon index and observed OTUs, although differences did not reach strong statistical significance. In contrast, participants with higher uPDI scores showed significantly lower observed OTUs, whereas Shannon diversity did not differ materially between groups. Beta diversity analysis based on unweighted UniFrac distances demonstrated subtle but statistically significant differences in microbial community composition between high and low hPDI and uPDI groups, as visualized by principal coordinates analysis (PCoA) (Figure 1; PERMANOVA P < 0.05). Similar patterns were observed in sensitivity analyzes excluding AEN cases (Figure S2).

Table 3.

Alpha diversity of the esophageal mucosa-associated microbiome according to plant-based diet indices.

		Shannon index	P	Observed OTUs	P
PDI	Low group	4.13	0.473	112.50	0.667
	High group	4.26		114.50
hPDI	Low group	4.10	0.651	112.00	0.220
	High group	4.28		121.00
uPDI	Low group	4.17	0.819	119.00	0.011
	High group	4.15		98.00

Note. Alpha diversity was assessed using the Shannon index and observed operational taxonomic units (OTUs). Participants were categorized into low and high groups based on the median value of each plant-based diet index.

Figure 1.

Beta diversity of the esophageal mucosa-associated microbiome according to plant-based diet indices. Principal coordinates analysis (PCoA) plots based on unweighted UniFrac distances showing differences in esophageal mucosa-associated microbial community structure between low and high groups of (a) PDI, (b) hPDI and (c) uPDI. Participants were dichotomized based on the median value of each index. Statistical significance was assessed using permutational multivariate analysis of variance (PERMANOVA).

Differential microbial taxa associated with plant-based diet patterns

Differential abundance analysis identified several genera that differed between participants with high and low adherence to plant-based diet indices (Figure 2). For hPDI, higher adherence was associated with increased relative abundance of Rothia and members of the Prevotellaceae family, whereas lower hPDI scores were associated with enrichment of Dialister and Peptostreptococcus (FDR-adjusted P < 0.05). For uPDI, higher scores were associated with increased relative abundance of Streptobacillus and Fretibacterium, while Campylobacter and Neisseria was significantly depleted in the high-uPDI group. Regarding the overall PDI, higher scores were associated with enrichment of Anaeroglobus, Actinobacillus and Alloprevotella, whereas Acinetobacter was more abundant among participants with lower PDI scores. These differential patterns remained largely consistent in sensitivity analyzes excluding AEN cases (Figure S3).

Figure 2.

Differentially abundant esophageal microbial genera associated with plant-based diet indices. Bar plots showing genera that differed significantly between low and high groups of (a) PDI, (b) hPDI and (c) uPDI based on relative abundance. Differential abundance was assessed using the Wilcoxon rank-sum test with false discovery rate (FDR) correction. Only genera with FDR-adjusted P values < 0.05 are shown. Bars indicate log2 fold change between high and low groups.

Microbial differences across esophageal disease groups

We further characterized differences in the esophageal microbiome across disease groups independent of dietary patterns. Overall, alpha diversity was modestly lower in AEN compared with normal controls. Median observed OTUs decreased from approximately 124 in the normal group to approximately 103 in the AEN group, while the Shannon index declined from approximately 4.48 to 4.16. Beta diversity analysis revealed significant differences in microbial community composition between AEN participants and individuals with normal esophageal mucosa (unweighted UniFrac PERMANOVA P = 0.047). Differential abundance analysis identified higher relative abundance of Dialister and Peptostreptococcus in AEN patients compared with normal group (Figure 3).

Figure 3.

Microbiome diversity and differentially abundant genera across esophageal disease groups. (a) PCoA plot based on unweighted UniFrac distances comparing esophageal mucosa-associated microbiome composition between normal participants and AEN participants. (b) Bar plot showing selected genera with significantly different relative abundance between the two groups.

Combined prediction model of ESCC and its precancerous lesions

To evaluate whether microbiome features provided complementary information beyond dietary indices, we constructed logistic regression models for risk stratification and compared their discriminative performance using ROC analysis (Figure 4). The outcome was defined as pathologically confirmed AEN (HGIN + ESCC), while the reference group consisted of participants with normal esophageal mucosa or esophagitis, excluding individuals with LGIN. Models based on dietary indices alone demonstrated moderate discrimination. Microbiome-only models incorporating selected genera (including Campylobacter, Streptobacillus, Neisseria, Rothia and Prevotella) showed comparable performance. Selected genera were chosen based on their differential abundance across diet groups and prior evidence linking them to esophageal microbial alterations. Notably, models integrating dietary indices and microbial features consistently achieved higher AUC values than models based on diet or microbiome alone. For example, the combined model incorporating hPDI, selected microbial features and covariates achieved an AUC of approximately 0.93, compared with approximately 0.85 for the diet-only model. Similar improvements were observed for uPDI-based and PDI-based models. The DeLong test confirmed that the combined models performed significantly better than diet-only models (P < 0.05). In addition, among non-AEN participants, similar risk stratification analyzes were conducted to distinguish LGIN from normal participants and esophagitis, and the corresponding ROC curves are presented in Figure S4.

Figure 4.

Receiver operating characteristic (ROC) curves for discriminating AEN from non-cancer groups using dietary indices and esophageal microbiome features. The outcome was defined as pathologically confirmed AEN (HGIN + ESCC), while the reference group consisted of participants with normal esophageal mucosa or esophagitis, excluding individuals with LGIN. ROC curves comparing the performance of models based on dietary indices alone, microbial features alone and their combination. Models were adjusted for demographic and lifestyle covariates. The area under the curve (AUC) was calculated to evaluate model discrimination.

Discussion

Esophageal cancer is a major public health burden in China, where ESCC represents the predominant histological subtype. Because early-stage ESCC and its precancerous lesions are often asymptomatic, timely identification of individuals at elevated risk remains challenging [24]. Although endoscopic screening has proven effective in reducing ESCC incidence and mortality in high-risk regions, its population-wide implementation is limited by cost, invasiveness and resource constraints [25]. Identifying modifiable risk-related features and improving risk stratification strategies are therefore critical for optimizing ESCC prevention and early detection efforts [26,27].

In this population-based study, we integrated plant-based diet indices with esophageal mucosa–associated microbiome profiles to explore their joint associations with esophageal disease severity. We observed that higher adherence to a healthful plant-based diet (hPDI) was associated with a lower prevalence of esophageal precancerous lesions and cancer, whereas higher adherence to an unhealthful plant-based diet (uPDI) was associated with increased disease prevalence. Beyond these epidemiological associations, plant-based diet patterns was linked to distinct microbial configurations in esophageal swab samples and the combination of dietary and microbial information improved risk stratification compared with epidemiological factors alone. Together, these findings suggest that diet-associated esophageal microbial profiles may represent an intermediate ecological layer linking lifestyle exposures with esophageal disease severity.

To our knowledge, this is among the first studies to examine plant-based diet indices in relation to ESCC and its precancerous stages. Our findings that higher hPDI scores are associated with lower disease prevalence, whereas higher uPDI scores are associated with higher prevalence, are consistent with evidence from nutritional epidemiology in other chronic diseases and gastrointestinal cancers [28,29]. A prospective cohort study in China demonstrated that balanced, plant-based diets could reduce upper gastrointestinal cancer risk, whereas an unhealthy PDI increased the risk [7]. Another cohort study of middle-aged UK adults suggested that greater adherence to the hPDI was associated with lower risks of cancer, with HRs (95% CIs) of 0.84 (0.78–0.91) for participants in the highest hPDI quartile compared with the lowest [30]. By contrast, higher uPDI scores were associated with higher risks of cancer [31]. Many dietary components of hPDI were similar to those of aMED and DASH, all of which emphasize a higher intake of healthful plant-based foods (e.g. vegetables, fruit, nuts, legumes and whole grains). These food groups provide dietary fiber, antioxidants and bioactive phytochemicals that have been associated with reduced inflammation and lower cancer risk [32,33], whereas plant-based diets dominated by refined grains and added sugars may lack these protective components and be associated with metabolic and inflammatory disturbances [34]. Prior studies in ESCC populations have shown that dietary patterns characterized by higher intake of fresh foods and lower intake of processed or pickled foods are associated with more favorable metabolic profiles and distinct microbial compositions [35]. Our results extend these observations to a broader population spanning normal esophagus, esophagitis, precancerous lesions and ESCC, emphasizing that diet quality, rather than plant food intake per se, may be particularly relevant for esophageal disease risk. However, residual confounding should be considered when interpreting these findings. Although we adjusted for major available demographic, lifestyle, dietary, family history and oral health-related factors, unmeasured or incompletely characterized factors may still influence the observed associations. Socioeconomic status may affect long-term dietary choices and health care access; culinary practices may be associated with both diet quality and ESCC risk; and oral health conditions may shape oral-esophageal microbial profiles. Because tooth loss was used only as a proxy indicator of oral health, residual confounding from oral hygiene behaviors, periodontal status, comorbidities and medication use cannot be fully excluded.

We further utilized esophageal swab samples to capture mucosa-associated microbial communities distinguished by different plant-based diet patterns. as the esophagus harbors a unique microbial ecosystem that is influenced by oral microbial influx, dietary exposures and local mucosal conditions [36]. Emerging evidence suggests that alterations in the esophageal microbiome are associated with mucosal inflammation and carcinogenesis, particularly in ESCC [37]. Within this context, we observed that hPDI and uPDI were associated with distinct esophageal microbial profiles, supporting the concept that diet quality shapes the local microbial ecology of the esophagus [38]. However, it is important to distinguish the ecological associations observed in this study from established biological mechanisms. The associations between dietary indices, microbial community structure and disease severity suggest biologically plausible links, but they do not demonstrate that specific dietary patterns directly caused microbial changes or that specific taxa mediated esophageal carcinogenesis. These findings do not imply a direct causal pathway but rather indicate that plant-based diet quality is associated with systematic differences in the esophageal mucosa-associated microbiome, which may align with different levels of disease severity.

Among the taxa associated with uPDI scores, Campylobacter emerged as a notable diet-associated feature in our analysis. Previous studies have reported enrichment of Campylobacter-related taxa in ESCC tissues, including studies integrating paired tumor and non-tumor samples with external validation datasets [39]. Although mechanistic studies in ESCC are limited, Campylobacter species are known to induce epithelial inflammation and disrupt mucosal barrier function at other mucosal sites [40]. In our study, Campylobacter was relatively enriched in the low-uPDI group. This discrepancy may reflect differences between diet-associated and disease-associated microbial patterns, sample type, disease composition, population characteristics and the limited taxonomic resolution of 16S rRNA sequencing. Therefore, Campylobacter should be interpreted cautiously and not classified as uniformly pathogenic or protective based on the present genus-level analysis. In addition, several anaerobic taxa commonly linked to oral and periodontal dysbiosis, including Peptostreptococcus, Dialister, Capnocytophaga, Atopobium, Filifactor, Fretibacterium and Treponema, were more prevalent in uPDI-associated profiles. While direct evidence linking each of these genera individually to ESCC remains limited, epidemiological and mechanistic studies have consistently associated poor oral health and chronic periodontal inflammation with increased risk of upper gastrointestinal cancers [41,42]. In this context, the enrichment of these taxa may reflect a broader inflammation-prone microbial ecological state along the oral-esophageal axis, which is biologically plausible in the progression of esophageal disease [43]. In contrast, higher hPDI scores were associated with enrichment of Rothia and relative preservation of Neisseria, two genera widely recognized as key nitrate-reducing bacteria in the oral cavity. These taxa play important roles in the enterosalivary nitrate–nitrite-nitric oxide pathway and are commonly considered markers of a health-associated oral microbiome [44]. Dietary intervention studies and systematic reviews have demonstrated that increased intake of nitrate-rich vegetables, an important component of healthful plant-based diets, is associated with higher abundance of Rothia and Neisseria in oral microbial communities [45]. Notably, recent studies examining oral and esophageal microbiota in ESCC patients have reported reduced abundance of nitrate-reducing taxa, including Neisseria- and Rothia-related lineages, compared with controls [46]. Within this framework, our findings suggest that adherence to a healthful plant-based diet may help maintain a more homeostatic oral-esophageal microbial ecosystem, potentially counterbalancing inflammatory processes associated with esophageal disease progression.

Beyond descriptive associations, we observed that integrating microbial features with established epidemiological factors improved the discrimination of esophageal disease severity compared with models based on epidemiological factors alone. As previous studies shown, the epidemiological factors enrolled in this study were multi-factorial responses and were important to the screening and risk predication of ESCC or microbiota composition in the gastrointestinal tract [47,48]. It follows that the role of epidemiological factors was definitely important, which may partly explain the good performance of epidemiological factors only in this study. However, the AUC of microbial targets only was not so high as epidemiological factors only. Although the predictive performance of microbiome-only models was modest, microbial information provided complementary value, supporting the concept that esophageal microbiota may capture biological variation not fully reflected by traditional risk factors. Moreover, the role of microbial biomarkers was non-negligible, which was more objective and quantifiable than epidemiological factors that were relevantly subjective and were easily influenced by lifestyle, diet and region [14,49]. Therefore, microbiome profiling should be viewed as complementary rather than substitutive to established epidemiological risk assessment. Traditional risk factors remain essential for population-level screening strategies, while microbial features may capture additional biological variation related to local mucosal ecology.

Importantly, our modeling approach was not intended to establish diagnostic biomarkers but rather to evaluate the incremental value of microbiome features for population-level risk stratification. Although the combined diet–microbiome models showed statistically significant improvement in AUC compared with diet-only models, the translational significance of this improvement should be interpreted cautiously. The present models were exploratory and internally validated only. Before any screening application, their performance must be confirmed in independent cohorts, with additional evaluation of calibration, reproducibility, cost-effectiveness and feasibility in field settings. Furthermore, implementation in screening settings also faces practical challenges. Microbiome-based risk stratification requires standardized sampling procedures, contamination control, sequencing or targeted detection platforms, reproducible bioinformatic pipelines and clear decision thresholds. Cost, turnaround time, inter-laboratory variability and population-specific microbial differences may further limit immediate application. Thus, the present findings should be considered a basis for future validation and assay simplification rather than direct evidence supporting near-term clinical implementation.

Limitations

Key strengths of this study include the inclusion of multiple stages of esophageal disease, the use of validated plant-based diet quality indices and the focus on mucosa-associated esophageal microbiota. However, several limitations warrant consideration. First, the cross-sectional design limits causal inference, and reverse causality cannot be excluded. Early esophageal lesions or related symptoms may have altered participants' dietary intake, appetite, swallowing behavior, or local microbial profiles, rather than dietary and microbial differences necessarily preceding disease progression. Although we conducted sensitivity analyzes excluding cancer cases, prospective studies are needed to clarify the temporal relationships among plant-based diet quality, esophageal microbial ecology and disease development. Second, although recent antibiotic use and medication exposure were considered during participant eligibility screening, detailed information on medication type, timing, duration and dose was limited. Therefore, residual effects of antibiotics, proton-pump inhibitors, or other medications that may influence microbial composition cannot be fully ruled out. Third, dietary intake was assessed using an FFQ, which may be affected by recall bias and measurement error because participants were asked to report their usual dietary intake over the previous year. In addition, total energy intake could not be precisely estimated from the available questionnaire, and the plant-based diet indices were therefore not directly energy-adjusted. Fourth, 16S rRNA sequencing provides limited taxonomic and functional resolution. Most microbial features were analyzed at the genus level, which may combine species or strains with different biological properties. Therefore, species- or strain-level effects and microbial metabolic pathways could not be assessed. Future studies using shotgun metagenomic sequencing, metatranscriptomics, metabolomics and targeted functional assays are needed to clarify the biological significance of these microbial patterns. Despite these limitations, our findings highlight the potential relevance of diet quality as a modifiable factor associated with esophageal microbial ecology and disease severity. In regions with a high burden of ESCC, improving plant-based diet quality, emphasizing whole, minimally processed plant foods, may contribute to risk reduction. Moreover, integrating dietary assessment with non-invasive microbial profiling could inform more targeted risk-stratification strategies to complement existing endoscopic screening programs. Further validation in independent cohorts will be critical to determine the utility of this approach for optimizing ESCC prevention and early detection.

Conclusion

In summary, this cross-sectional study showed that healthful and unhealthful plant-based diet patterns were differentially associated with esophageal disease severity and esophageal mucosa-associated microbiome profiles. Higher hPDI was associated with lower disease severity, whereas higher uPDI was associated with greater disease severity. A healthful plant-based diet was linked to microbial configurations characteristic of a more homeostatic oral-esophageal ecosystem, whereas an unhealthful plant-based diet was associated with microbial features overlapping with previously reported ESCC-related dysbiosis. Although causal inference is limited by the cross-sectional design and genus-level resolution, integrating dietary indices with esophageal microbiome features modestly improved risk stratification beyond epidemiological factors alone. These findings highlight the relevance of diet-associated microbial ecology along the oral-esophageal axis and provide a population-based framework for incorporating lifestyle and microbiome information into future ESCC risk assessment and prevention strategies.

Supplementary Material

Supplementary Material

Supplementary materials.docx

Funding Statement

This work was supported by the National Natural Science Foundation of China (82473704, 82273704), Noncommunicable Chronic Diseases-National Science and Technology Major Project (2024ZD0521001), National High Level Hospital Clinical Research Funding (2025-LYZX-R-A03), the Capital's Funds for Health Improvement and Research (2024-2G-40213), Beijing Nova Program (202504841003) and the Beijing Hope Run Special Fund of Cancer Foundation of China (ZZ2024A13).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data generated in this study are available upon request from the corresponding author.

Supplementary material

Supplemental data for this article can be accessed at https://doi.org/10.1080/20002297.2026.2682962.