Healthy plant-based diet might be inversely associated with gastric precancerous lesions: new evidence from a case-control study based on dietary pattern and fecal metabolic profiling

Abstract Preventing the progression of gastric precancerous lesions (GPLs) can reduce the morbidity and mortality of gastric cancer (GC). The preventive effect of a plant-based diet on cancers has been widely recognised. In this case–control study, 1,130 subjects were included using 1:1 propensity score matching for age and sex. Dietary habits, anthropometry and sample collection were conducted using standard and effective methods. Plant-based diet indices (PDIs) were calculated using a previously reported method. Faecal samples were analysed by untargeted metabolomics. Our study found that adherence to a healthy plant-based diet was inversely associated with the occurrence of GPLs. Metabolomic analysis identified six different metabolites correlated with GPLs, among which luteolin-related metabolites may be used as biomarkers of the association between PDIs and GPLs. In addition, the difference in N-acyl amides found in PDIs needs further verification. Our findings suggest that a healthy plant-based diet may have a protective effect against GPLs.


Introduction
Gastric cancer (GC) is the fifth most common cancer and the fourth leading cause of cancer-related deaths worldwide (Sung et al. 2021).Gastric precancerous lesions (GPLs) include irreversible chronic nonatrophic gastritis, chronic atrophic gastritis (CAG), gastric mucosal intestinal metaplasia (IM), and dysplasia (Correa 1992).Although the mechanism of GPLs developing into GC is not very clear (De Vries et al. 2008), the occurrence of GPLs are very common worldwide, especially in the East Asia (including China) (Du et al. 2014;Sipponen and Maaroos 2015), and they are irreversible (Qu and Shi 2022).In addition, there is a downward trend in the incidence of GC worldwide, but a striking increase in GPLs have been reported among young white Americans aged 25-39 years and Swedish adults aged 35-44 years (Anderson et al. 2010;Song et al. 2015), implying that the downtrend in GC may soon reverse (Song et al. 2015).Therefore, finding new preventive measures is important to effectively control the progression of GPLs to GC.
Genetic susceptibility, Helicobacter pylori infection, and an unhealthy lifestyle may be independent or combined factors influencing the development of GC (Chen et al. 2019;Acuna et al. 2023).Genetically predisposed individuals can partially offset their risk of GC by adhering to a healthy lifestyle (Jin et al. 2020), including healthy eating patterns and behaviours (Barone et al. 2022;Su et al. 2023).For example, a higher intake of healthy plant foods, such as vegetables and fruits, appears to reduce the risk of GPLs and GC (Kim et al. 2021;Toh and Wilson 2020).In recent years, the plant-based diet indices (PDIs) have received widespread attention because of their improvements in the limitations of the previous vegan diet, including the failure to exclude animal foods and distinguish the quality of plant foods (Martínez-González et al. 2014), and its value in predicting diseases has been reported (Satija et al. 2016;Kim et al. 2020;Zamani et al. 2020).However, previous studies have mainly focused on gastric cancer, and studies on the association between GPLs and plant-based diets based on dietary patterns are not available.
The development of high-throughput metabolomics has provided a new approach to nutrition research.As the final product of human cellular and microbial metabolism, faeces is one of the best biological materials to reflect information about physiological changes in the gastrointestinal tract (Zhao et al. 2021).Profiling faecal metabolomics has become an important way to explore the association among diet, human metabolism and gut microbiota composition in maintaining the healthy state of the organism, and helps to elucidate the physiological and biochemical mechanisms of dietary and metabolic changes in the body (Han et al. 2022).Metabolomic analysis is used to detect hundreds of low molecular weight components and metabolites downstream of the diet (Wang et al. 2022), by which objective biomarkers of dietary patterns can be efficiently identified and provide reliable evidence to explore the association between plant-based dietary patterns and GPLs risk (Barone et al. 2022).The combined use of the Food Frequency Questionnaire (FFQ) and metabolomic analysis partly compensates for the limitations, such as recall bias, of obtaining research information through purely self-reporting.
Considering the dietary habits and high prevalence of GPLs among Chinese residents, it is particularly important to evaluate the relationship between plant-based diet quality and GPLs as well as its potential metabolites to provide a multichannel solution for the prevention of GC in clinical practice.This study aimed to explore the relationship between PDIs and GPLs occurrence through an effective FFQ and to further explore the metabolite relationships using untargeted metabolomics, which provides an objective basis for evaluating the efficacy of PDIs on dietary patterns.

Study population
This case-control study was conducted randomly in five affiliated hospitals of Anhui Medical University in Hefei city, Anhui Province, from April 2019 to October 2021.Hefei,located in the middle of the Yangtze River Delta region in China,has a high incidence of GPLs and GC (Hu et al. 2018).Therefore, the results of this study have certain implications for the region and surrounding areas.Patients aged > 20 years who were diagnosed with GPLs by endoscopy and biopsy were included in the case group.Control subjects were recruited from natural populations who underwent physical examinations, including endoscopy at the same regional medical examination centre during the same period.Participants with a history of gastrointestinal surgery, malignancy, severe systemic mental illness, or special dietary behaviours due to hypertension or diabetes, pregnancy, or lactation were excluded from both the case and control groups.A total of 1,662 subjects met the inclusion criteria, including 920 and 742 subjects in the case and control groups, respectively.We excluded patients (n = 13) with incomplete FFQs, implausible baseline data (n = 3), and total energy intakes < 500 kcal/d or > 5000 kcal/d (Chen et al. 2015) (n = 45).Among the remaining participants, control and case study subjects were matched 1:1 by age (± 5) and sex using propensity score matching, and 1,130 subjects were included.A flowchart of the participant screening process is shown in Figure 1.The study protocol was approved by the Ethics Committee of Anhui Medical University (Approval Number: 20190292).This study was conducted in accordance with the Declaration of Helsinki.All participants were aware of this study's purpose and signed an informed consent form.

Dietary assessment and construction of PDIs scores
Dietary data for the past year were collected using a validated semiquantitative FFQ (SQFFQ), which was specially modified from the SQFFQ developed by Harvard University to incorporate the dietary habits of Chinese people.The questionnaire listed 116 commonly consumed food items from 5 major food groups, and respondents were asked to report the consumption frequency and the amount of single consumption of each of the food items, where the frequency of consumption was categorised into 9 levels, ranging from "not eaten" to "3 times a day".In calculating the daily intake of a food item, the frequency of consumption in the original questionnaire was converted to the daily consumption frequency according to the median value and then multiplied by the single serving size.Energy intake was calculated by multiplying the amount of food consumed by the energy of the edible portion of the food, which is available in the database of the sixth edition of the Standard Edition of the Chinese Food Composition Table .The food intake of each group was corrected by the residual method based on their total energy intake (Kim et al. 2020).
To minimise bias, the questionnaire was completed through face-to-face interviews with researchers who underwent rigorous and uniform training and assessment before they were allowed to participate in the survey.In addition, each researcher was provided with a specialised visual guide on food intake to help respondents answer the food consumption details as accurately as possible.In addition, each valid questionnaire was verified by strict and uniform standards, and double-blind data entry was adopted to ensure the high quality of the data.
The construction of three plant-based diet indices has been reported previously (Martínez-González et al. 2014;Satija et al. 2017).Briefly, we created 14 food groups based on nutrient composition and culinary similarity (Table 1).These foods were distinguished as healthy or unhealthy plant foods based on knowledge of the relationship between food and hypertension, type 2 diabetes (T2D), other outcomes (cardiovascular disease, certain cancers) and intermediate diseases (obesity, lipidemia, inflammation) (Satija et al. 2016).Based on previous research, we separated fried and instant foods (Sun et al. 2019) and fruit juices into unhealthy plant-based diet groups (Guasch-Ferre and Hu 2019).Considering that alcohol consumption has different associations with various health outcomes, we did not include it in the index, but a subsequent adjustment analysis was performed.To construct the overall plant-based diet index (PDI), healthy plant-based diet index (hPDI), and unhealthy plant-based diet index (uPDI), the subjects were ranked according to the quintile of residual in each food group and assigned forward or reverse assignments in turn.Forward assignment meant assigning 1, 2, 3, 4, and 5 sequentially from the lowest quintile to the highest quintile, and reverse assignment was the opposite.The three indices were created as follows: the PDI distinguished only plant food and animal food, namely, the sum of the positive score of plant food (did not distinguish healthy food from unhealthy food) and the negative score of animal food; the hPDI and uPDI distinguished the quality of plant food and reverse scored for animal food groups, but the former was positively scored for healthy plant-based foods and inversely scored for unhealthy plant-based foods; and the uPDI was opposite.The overall scores of the 14 food groups were finally added to obtain three types of indicators and used for subsequent analysis.

Diagnosis of gastric precancerous lesions and H. pylori infection
Endoscopy and biopsy have always been the gold standards for gastric cardia adenocarcinoma screening in high-risk areas in China (Gao and Fang 2015).Endoscopy was performed by experienced physicians to evaluate and grade gastric mucosal precancerous lesions according to the Chinese Society of Digestive Endoscopy criteria (Lin et al. 2016), and biopsies were used to determine their clinical diagnosis.The H. pylori IgG antibody detection kit-latex method (ABON, CAS) was used to diagnose H. pylori infection (Shatila and Thomas 2022).

Untargeted metabolomic analysis
Fifty-eight cases were randomly selected: a control group (normal, n = 20) and a case group (CPL, n = 38).Faecal samples were collected using sterile metabolic tubes and stored at −80 °C until testing.The faeces were then dissolved in an aqueous precooled methanol solution.After multiple homogenizations, the supernatant was extracted, sonicated in a water bath, centrifuged, and filtered through a 0.22-μm filter.The samples were calibrated by adding the internal standard 2-chlorophenylalanine.Ultrahigh-performance liquid chromatography (UPLC) tandem high-resolution mass spectrometry (Q-Exactive Plus, Thermo Fisher, USA) was used.Mobile phase A was acetonitrile/water (60:40 with 0.1% formic acid and 0.1% ammonium formate, v-v), and mobile phase B was isopropanol/ acetonitrile (90:10, 0.1% formic acid, 0.1% ammonium formate, v/v).In this experiment, two QC samples were prepared by extracting 10 μl of liquid from each sample before use of the machine and then evaluating the stability of the method by measuring the QC samples multiple times and evaluating the error between them.
Metabolism search software version 4.1 (Thermo Scientific, https://www.thermofisher.cn/) was used for peak identification, metabolic identification (secondary identification), peak extraction, and peak alignment, and quantification of metabolites with relative standard deviation (RSD) exceeding 30% and missing values exceeding 50% in QC samples were removed, and the remaining metabolites were normalised by the median peak area.P values were corrected using the false discovery rate (FDR) method.Data were preprocessed for pattern recognition and scaling using SIMCA-P 14.1 software (Umetrics, Umea, Sweden).
According to orthogonal partial least squares discriminant analysis (OPLS-DA), the projected variable importance (VIP) was obtained to explore the influence strength of the classification and discrimination between each metabolite and the groups.

Covariate assessment
The General Information Prediction test volume was used to collect basic information and demographic data on the study subjects through face-to-face interviews, including age, sex, education, occupation, annual per capita household income, marital status, family history, smoking status, and alcohol consumption status.All anthropometric measurements were performed by a trained surveyor.The research subjects wore light clothes and no shoes for measurement.Body weight (kg, continuous) was measured to the nearest 0.1 kg using a weight scale, and the height (cm, continuous) of the participants in a standing position was measured using a tape measure to the nearest 0.1 cm.Body mass index (BMI, kg/m 2 , continuous) was calculated as the weight in kilograms divided by the square of the height in metres.A brief International Physical Activity Questionnaire (IPAQ) was also used to assess the exercise intensity (MET) of each participant, and combined with personal BMI to calculate the daily physical activity level (kcal/day).

Statistical analysis
The research subjects were divided into three groups (Q1, Q2, and Q3) according to the tertiles of PDI, hPDI, and uPDI scores, respectively.The general characteristic distributions of the case and control groups and PDIs were expressed as the means (standard deviations, SDs) (continuous variables) and percentages (categorical variables), compared using one-way analysis of variance (ANOVA) and chi-square tests, respectively.Binary logistical regression analysis was used to examine the relationship between PDIs and GPLs, and three models were constructed by adjusting for different covariates, with the lowest tertile (Q1) in each model as the reference value.

Baseline characteristics of the subjects and nutritional characteristics of the dietary patterns
The average age of the subjects included in this study was 56.59 ± 8.13 years, and 45.49% were male.
Compared with the control group, the case group had a higher educational level, more alcohol drinkers, lower total energy intake, and higher number of HP infections (Table 2).In terms of dietary intake, the control group was more inclined to consume plant foods such as vegetables and fruits.The PDI and hPDI scores of the control group were significantly higher than those of the GPLs group, whereas the uPDI scores were the opposite.Compared with the Q1 group, the top PDI tertile group was more likely to have higher BMI scores, total energy intake, and physical activity levels.These differences were reversed for the uPDI (Table 3).The highest hPDI was associated with younger age, a larger proportion of females, and a lower prevalence of smoking.Compared with the low tertile groups, a higher PDI and hPDI were accompanied by greater whole grains, starches and potatoes, vegetables, fruits, beans, fungal algae, nuts, and fruit juices; however, the opposite was true for the intake of refined cereals, livestock, poultry, dairy, and fried and instant foods.The above phenomena were the opposite in the uPDI group.With a gradual  Q1 the score of various dietary patterns is located between 0% and 33.33% quantile; Q2 the score of various dietary patterns is located between 33.33% and 66.66% quantile; Q3 the score of various dietary patterns is located between 66.66% and 100.00% quantile.
increase in the uPDI score, the intakes of whole grains, starches and potatoes, vegetables, fruits, beans, livestock and poultry meat, aquatic animals, eggs, dairy, fungal algae, and nuts gradually decreased, but the intake of refined cereal showed the opposite trend.

Intestinal metabolic profiling in gastric precancerous lesions
Based on UPLC-MS to identify and quantify faecal metabolites, a total of 785 metabolites were identified in 58 faecal samples by positive and negative ionisation methods.The quality control (QC) results in the PLS-DA model diagram were relatively stable, which guaranteed the accuracy of the technology and instruments in the testing process (Supplementary Figure 1).According to the screening criteria(log 2 Fold change > 1 or < −1, p < 0.05, VIP > 1), 20 different metabolites between the case and control groups were screened.
After comparison with the HDMB, 6 metabolites were finally determined (Figure 4(A,B)).The metabolic differences between the case and control groups are shown in Table 4. Compared with the control group, the expression of glycitein, hesperetin, and luteolin 7-sulphate in the faeces of patients in the case group was reduced.These substances are markers of plant-based consumption.Simultaneously, elevated linoleic acid, 16-hydroxyhexadelanoic acid, and N-acetyl-DL-tryptophan levels were found in faecal samples of patients with precancerous gastric lesions (Figure 4(A,B)).KEGG enrichment analysis was performed on the metabolites of GPLs, and four metabolic pathways were changed (p < 0.05) (Figure 4(C)).They were D-arginine and D-ornithine metabolism (p = 0.0067), linoleic acid metabolism (p = 0.017), biosynthesis of unsaturated fatty acids (p = 0.042), and arginine and proline metabolism (p = 0.047).We used correlation analysis to analyse the above 6 different metabolites and PDIs.PDI and hPDI were positively correlated with luteolin 7-sulphate and N-acetyl-DL-tryptophan.At the same time, PDI and hPDI scores were inversely correlated with the metabolite 16-hydroxyhexadelanoic acid, while uPDI was positively correlated with the metabolite, but there was no significant difference (Figure 4(D)).

Effects of plant-based dietary patterns on intestinal metabolism
To verify whether a plant-based diet had an impact on metabolism in the body, we used the same screening criteria to identify 13 different metabolites related to the plant dietary indices, 5 metabolites in the PDI, 5 metabolites in the hPDI, and 8 metabolites in the uPDI.The different metabolites of the Q1 and Q3 group of PDIs are shown in Table 4.The overall expression levels of the representative differences in each group are shown in Figure 5.The expression of ursolic acid, linamarin, luteolin, and N-palmitoyl alanine in the faeces of patients with top PDI scores was significantly higher than that in the low tertile group, and this trend was also observed in the Q2 group.
The contents of caffeine acid 3-glucoside, N-palmitoyl alanine, luteolin, and linamarin in the Q3 group of the hPDI were also significantly higher than those in the Q1 group, but the difference in I-urobilinogen was the opposite.In the uPDI group, the expression of N-oleoyl tryptophan, N-stearoyl proline, linamarin, N-palmitoyl alanine, flazin, N-palmitoyl phenylalanine, and N-oleoyl methionine in the top tertile group was lower than that in the Q1 group, and the changes in 16-hydroxyhexadelanoic acid were the opposite.In the PDIs, differences in 5 N-acylamide substances were found, and the expression of these substances increased with an increase in plant-based dietary intake and vice versa, suggesting that plant-based dietary intake may affect the metabolism of N-acyl amide substances in the body.

Discussion
In our study, we observed that plant-based diet quality was associated with the GPLs.Greater adherence to a healthy plant-based diet was inversely associated with GPLs.In contrast, individuals who consumed an unhealthy plant-based diet mingt have a higher GPLs chance.This association persisted after adjusting for potential confounders.Faecal metabolomic analysis found 6 kinds of metabolites related to GPLs.At the same time, we found that luteolin and its metabolite luteolin 7-sulphate may be associated with a plant-based diet and GPLs, which provides theoretical support for in-depth exploration of the potential mechanism of the association between a plant-based diet and GPLs.Plant-based diets can promote disease prevention owing to their components.Higher intake of healthy plant-based foods, including vegetables, fruits, whole grains, and fungal algae, is accompanied by a high intake of vitamins, dietary fibre, and phytochemicals, which play a role in the development of gastric mucosal lesions through anti-inflammatory and antioxidant functions (Ma et al. 2008;Vincent et al. 2010).However, our results showed no association between the PDI and GPLs.Asian people include vegetables and grains in almost every meal.The overall intake of plant foods among Chinese people is high, but there are imbalances in dietary intake, such as excessive intake of refined grains and inadequate intake of fruits (Ma et al. 2023).Therefore, among those who already eat a plant-based diet, higher plant-based food intake may no longer cause clinically significant metabolic reactions (Kim et al. 2020), and the quality and balance of the diet deserve more attention.Notably, the association between plant-based dietary patterns and GPLs appears to be related to the quality of plant-based foods.Chronic diseases, including T2D (Satija et al. 2016), cardiovascular disease (Satija et al. 2017), and chronic kidney disease (Kim et al. 2019), are inversely associated with the hPDI and positively associated with the uPDI, which is consistent with the results of our study.These findings suggest that PDIs emphasise the impact of plant-based food quality on results.The above conclusions were also validated in a large prospective cohort study on the association between plant-based diet quality and overall mortality (Baden et al. 2019).After positive coding for fish, poultry, fermented dairy, low-fat dairy, and eggs, the HRs for the association between a 10-point increase in the hPDI and mortality risk remained unchanged, suggesting that changes in healthy and unhealthy plant foods, and not just the contribution of animal foods or a combination of the above factors, had a stronger impact on the outcomes.
To determine whether plant-based dietary patterns are associated with GPLs, it is necessary to control for the potential effects of other variables.Therefore, three regression models were created and stratified analysis was carried out according to HP infection.The results showed that the effect of diet scoring was very robust and not changed by these factors.When focusing on the intake of each food group in the PDIs, we found a phenomenon that did not match the original definition of the food groups.Aquatic animals and eggs as animal foods, and drinks as unhealthy plant foods were highly consumed in the group with high hPDI scores, and the opposite was true for the uPDI, suggesting the limitations of reflecting the association between dietary conditions and GPLs occurrence by a single food or food group.Likewise, when Baden et al. (Baden et al. 2019) individually excluded 18 food groups from their diet scores, the results remained unchanged, suggesting the importance of assessing the overall plant-based diet quality rather than specific foods.At the same time, in terms of the formulation of public health nutrition guidance, recommendations and guidance on dietary patterns are more in line with the lifestyle habits of residents in various regions, and are more likely to enhance the feasibility and adherence to nutrition guidance (Satija et al. 2017).
We screened 6 different metabolites between the case and normal control groups, suggesting that metabolic disorders had occurred in patients with GPLs.However, the specific intervention mechanism still needs to be further explored.Metabolomic studies focusing on gastric precancerous lesions are still relatively scarce.A study based on the different plasma metabolomics characteristics of GPLs in high-risk areas of GC reported that three metabolites (α-linolenic acid, linoleic acid, and palmitic acid) were significantly inversely associated with the risk of gastric lesion advancement (Huang et al. 2021), which is different from our research results.The reason may be due to dietary differences.Animal viscera, meat, and edible oils, including linseed oil, olive oil, and sunflower oil, are the main sources of linolenic acid.In our paper, the lower intake of plant foods in the case group was accompanied by an increase in animal foods intakes, which may cause an increase in linoleic acid content.Thus, differences in dietary patterns may have effects on the disease metabolic process.More population-based prospective studies are needed to examine the metabolic effects of dietary patterns on GPLs to identify new molecular features behind the progression of gastric lesions and GC occurrence to improve the risk assessment and early detection of GC.
Among the correlated analysis results of the above 6 different substances with PDIs, luteolin 7-sulphate, N-acetyl-DL-tryptophan, and 16-hydroxycetylanoic acid showed related trends, but 16-hydroxycetylanoic acid was not significantly different in terms of PDIs.N-acetyl-DL-tryptophan was positively associated with the hPDI in our assays.N-acetyltryptophan (NAT) is a metabolite called n-acyl-α amino acids that can function as a defense or signalling molecule (Doshi et al. 2021).NAT can quench reactive oxygen species (ROS), and increasing the concentration of NAT can afford a greater degree of protection against tryptophan oxidation (Dion et al. 2018).Luteolin and its derivative luteolin 7-sulphate are flavonoids found in different plants, such as vegetables, herbs, and fruits.As an anticancer drug, flavonoids can fight various types of human malignancies, including by inhibiting tumour growth and downregulating the expression of MMP9, Ki-67, and cMet to display antitumor effects in GC (Imran et al. 2019).Our results showed a positive correlation between the differentially abundant metabolite Luteolin 7-sulphate and the PDI in the case-control group, and luteolin was also positively correlated in the PDI and hPDI groups, which suggested consistency in the research results.Thus, luteolin and its metabolite luteolin 7-sulphate may be involved in the association between a plant-based diet and GPLs, which provides theoretical support for in-depth exploration of the potential mechanism of the association between plant-based diets and GPLs.
To further verify the credibility of PDIs from the perspective of biomarkers, we analysed the correlation between the detected metabolites and PDIs.Thirteen metabolites were identified, 5 metabolites in the PDI, 5 metabolites in the hPDI and 8 metabolites in the uPDI, and several have previously been reported as candidate biomarkers for foods and beverages.The hPDI emphasises the intake of healthy plant foods, but the uPDI is the reverse, so linamarin, which is the signature metabolite of potatoes (Zhou et al. 2019), was positively correlated with the PDI and hPDI but inversely correlated with the uPDI.Our study also found stable differences between several N-acyl amides with PDIs, which is novel.N-acyl amides have a variety of signalling functions in physiology (Battista et al. 2019), including cardiovascular activity, metabolic homeostasis, memory, cognition, pain, and motor control (Bradshaw and Walker 2005).N-acyl amides have also been shown to play a role in cell migration, inflammation, and certain pathological conditions, such as diabetes, cancer, neurodegenerative disease, and obesity (Grapov et al. 2012;Raboune et al. 2014;Cohen et al. 2017).We found that N-acyl amides were positively correlated with the PDI (N-palmitoyl alanine) and hPDI (N-palmitoyl alanine) but inversely correlated with the uPDI (N-oleoyl tryptophan, N-stearoyl proline, N-palmitoyl alanine, N-palmitoyl phenylalanine, and N-oleoyl methionine), suggesting a protective effect of a healthy plant-based diet on the organism and possibly considering N-acyl amides as a metabolic marker of a plant-based diet.Unfortunately, we did not find a statistically significant correlation between N-acyl amides and GPLs, so future studies are necessary to evaluate whether N-acyl amides represent a pathway linking plant-based dietary patterns with health or other diseases.
The main advantages of our research include that it is the first study to evaluate the relationship between PDIs and GPLs, the use of an FFQ combined with metabolomics to explore the potential mechanism of the association between PDIs and GPLs, and the use of an FFQ modified for Chinese dietary habits in conjunction with the Chinese Food Composition Table, which effectively overcame the mismatch caused by regional and cultural differences.However, this study also has some limitations.The main limitation is the inevitable selection and recall biases in case-control studies, but the application of metabolomic analysis might have verified the authenticity of the questionnaire relatively accurately and objectively and compensated for the lack of consistent testing of the questionnaire and the inevitable human errors.Ultimately, given the limited sample size and area of investigation, further large prospective studies are needed to examine the effects of plant-based diet indices on GPLs.
In short, our study found that plant-based diet quality was associated with GPLs.Adherence to a healthy plant-based diet was inversely associated with GPLs.Metabonomics analysis found significant differences between 13 plant-related metabolites and PDIs, and the results further affirmed the reliability and effectiveness of PDIs in this study.Most importantly, we determined that luteolin and its metabolite luteolin 7-sulphate may be associated with a plant-based diet and GPLs.If repeated, it may be considered as a biomarker of the plant-based dietary pattern of patients with GPLs.Further large prospective studies are needed to examine the effects of plant-based diets on GPLs.

Figure 1 .
Figure 1.flow chart of the screening process of the selection of eligible participants.

Figure 2 .
Figure2.crude and multivariable-adjusted ors and 95% cIs for GPls across tertiles of PdIs.crude: no adjustment for other variables.Model I: adjustment for gender, age, and total energy intake.Model II: further adjustment for education, per capita annual family income, occupation, marital status, family history, smoking status, alcohol status, physical activity level, and BMI based on Model I. Model III: further adjustment for HP infection based on Model II.Q1 the score of various dietary patterns is located between 0% and 33.33% quantile; Q2 the score of various dietary patterns is located between 33.33% and 66.66% quantile; Q3 the score of various dietary patterns is located between 66.66% and 100.00% quantile.(a) crude and multivariable-adjusted odds ratios and 95% cIs for GPls across tertiles of PdI.(B) crude and multivariable-adjusted odds ratios and 95% cIs for GPls across tertiles of hPdI.(c) crude and multivariable-adjusted odds ratios and 95% cIs for GPls across tertiles of uPdI.BMI: body mass index; GPls: gastric precancerous lesions; HP: Helicobacter pylori; hPdI: healthy plant-based diet index; PdI: plant-based diet index; Q, tertile; uPdI, unhealthy plant-based diet index.

Figure 3 .
Figure3.crude and multivariable-adjusted ors and 95% cIs for GPls across tertiles of PdIs stratified by H. pylori.crude: no adjustment for other variables.Model I: adjustment for gender, age, and total energy intake.Model II: further adjustment for education, per capita annual family income, occupation, marital status, family history, smoking status, alcohol status, physical activity level, and BMI based on Model I. Q1 the score of various dietary patterns is located between 0% and 33.33% quantile; Q2 the score of various dietary patterns is located between 33.33% and 66.66% quantile; Q3 the score of various dietary patterns is located between 66.66% and 100.00% quantile.(a) crude and multivariable-adjusted odds ratios and 95% cIs for GPls across tertiles of PdI stratified by H. pylori infection status.(B) crude and multivariable-adjusted odds ratios and 95% cIs for GPls across tertiles of hPdI stratified by H. pylori infection status.(c) crude and multivariable-adjusted odds ratios and 95% cIs for GPls across tertiles of uPdI stratified by H. pylori infection status.BMI: body mass index; GPls: gastric precancerous lesions; HP: Helicobacter pylori; hPdI: healthy plant-based diet index; PdI: plant-based diet index; Q: tertile; uPdI: unhealthy plant-based diet index.

Figure 4 .
Figure 4. Intestinal metabolic profiling in gastric precancerous lesions.(a) all differential metabolites in the case control group were screened according to predetermined screening criteria.(B) Heat map of the correlation between the identified 6 significantly different faecal metabolites and the case control group.colours denote the association directions (red, positive; blue, inverse) and magnitudes (the darker the colour, the stronger the magnitude).(c) the enrichment analysis of all differential metabolites in the case control group screened according to the aforementioned criteria.the differential metabolic pathways were screened according to p < 0.05.(d) Pearson correlation analysis between differential metabolites and PdIs in the case control group, red indicates positive correlation, blue indicates negative correlation, and the darker the colour, the stronger the correlation coefficient.use × to remove non-significant correlation values (p > 0.05).nc: normal control; PdIs: plant-based diet indices.

Figure 5 .
Figure 5. effects of plant-based dietary patterns on intestinal metabolism.the screening criteria are log2fold change > 1 or < -1, and p < 0.05.(a) expression levels of differential metabolites selected by PdI score.(B) expression levels of differential metabolites screened according to hPdI score.(c) expression levels of differential metabolites screened according to the uPdI score.Green indicates Q1, the score of various dietary patterns is located between 0% and 33.33% quantile; blue indicates Q2, the score of various dietary patterns is located between 33.33% and 66.66% quantile; red indicates Q3, the score of various dietary patterns is located between 66.66% and 100.00% quantile.hPdI: healthy plant-based diet index; PdI: plant-based diet index; Q: tertile; uPdI: unhealthy plant-based diet index.

Table 1 .
fourteen food groups were created based on the food nutrients and cooking similarities a .tomatoes, broccoli, cabbage, cauliflower, brussels sprouts, white radish, carrots, lotus root, celery, yellow or winter squash, eggplant or zucchini, spinach cooked, spinach raw, kale or mustard orchard greens, iceberg or head lettuce, romaine or leaf lettuce, celery, mushrooms, beets, alfalfa sprouts, garlic Starch and potato Potatoes, yams, purple sweet potatoes, sweet potatoes, and other sweet potatoes nuts nuts, Peanuts, melon seeds, pistachios, walnuts, pecan fruit, almonds, cashews, hazelnuts legumes Mung beans, red beans, dried beans, bean skin, tofu, and other dried bean products, edamame, green beans, peas, and other fresh beans fungal algae lave, laminaria japonica, mushrooms, shiitake mushrooms, and agaric fungus less healthy plant foods fruit juices apple cider (nonalcoholic) or juice, orange juice, grapefruit juice, or other fruit juice refined grains fine white rice, noodles, steamed bread, white bread, rice noodles, rice porridge, and millet porridge fried foods and instant foods french fries, fried bread stick, deep-fried dough cake, other fried foods, instant noodles, steamed stuffed bun, dumplings, and other instant foods animal foods dairy Skim low-fat milk, whole milk, cream, sour cream, ice cream, yogurt, whole milk powder, skim or low-fat milk powder egg eggs fish or seafood freshwater fish, marine fish, crab, eel, loach, freshwater shrimp, marine shrimp, and other fishes Meat Pork, beef, lamb, chicken, duck, pigeon, goose, animal offal, sausage, ham sausage, bacon a alcohol and coffee were not included in the plant-based dietary index, while later adjusted for in the multivariate analysis.
(Li et al. 2022)selected based on statistically significant differences in baseline data and factors identified in previous studies as influencing GPLs(Li et al. 2022).Furthermore, considering the effect of H. pylori infection on GPLs, we examined the potential associations between plant-based dietary patterns and GPLs by stratifying the H. pylori infection profile.
Model I was adjusted for age, sex, and total energy intake; Model II was further adjusted for education, annual per capita household income, occupation, marital status, family history, smoking status, alcohol consumption status, physical activity, and BMI based on Model I; and Model III was adjusted for H. pylori infection status based on Model II.http://www.hmdb.ca/),and meaningful variables were identified.Pathway analysis relied on the KEGG database (http://www.genome.jp/kegg/).Pearson's correlation analysis was used to evaluate the correlation between the results.GraphPad Prism 8.0 (GraphPad Software, USA) was used to analyse and visualise some parameters in this study.Data analyses were performed using Stata (version 16.0, Stata Corp), and a two-sided p value < 0.05 was considered statistically significant.

Table 2 .
demographic characteristics, dietary intake, and plant-based diet indices of the 1,130 cases and controls a .
b unit is 1 cnY.c BMI, Body mass index, calculated as weight divided by height squared.d the PdIs were expressed as median (min, max).

Table 3 .
demographic characteristics and dietary intake of the PdI, hPdI, and uPdI groups a .
BMI, body mass index; GPls, gastric precancerous lesions; PdI, plant-based diet index; Q, tertile; hPdI, healthy plant-based diet index; uPdI, unhealthy plant-based diet index.a unless otherwise noted, continuous-type variable values are expressed as mean (Sd) or median(P25, P75), and categorical variable values as percentages.PdI means total plant-based diet index; hPdI means healthy plant-based diet index; uPdI means unhealthy plant-based diet index.b unit is 1 cnY.c BMI; Body mass index, calculated as weight divided by height squared.d

Table 4 .
the identified differential metabolites between different groups.
PdI: plant-based diet index; Q: tertile; hPdI: healthy plant-based diet index; uPdI: unhealthy plant-based diet index; nc: normal control; fc: fold change; rt: retention time; M/Z: mass charge ratio.differences in the metabolic levels of different groups.the inspection method is an independent sample t-test.