Fish oil lipid-lowering effect prediction model based on gut microbial features and construction method therefor

Abstract

Disclosed in the present invention are a fish oil lipid-lowering effect prediction model based on gut microbial features and a construction method therefor. The method comprises: acquiring fish oil intervention treatment medical record data of patients with T2D and HTG, and collecting fecal samples and peripheral blood samples; extracting gut microbial DNA from the fecal samples, using Shotgun metagenomic sequencing to obtain gut microbial genomic data, and performing statistical analysis to obtain microbial feature parameters related to fish oil intervention treatment for the patients; detecting TG data before and after intervention for the patients to estimate a gut microbial variable difference between a TG-responsive group and a non-responsive group; and constructing a prediction model for gut microbial baseline variables, using a repeated iterative feature method to obtain an optimal AUC value, and extracting a gut microbial feature having the maximum contribution to the TG-responsive group and an optimal random forest model TG-responsive probability threshold. In the present invention, the prediction model constructed on the basis of the gut microbial features can effectively determine the lipid-lowering effect of fish oil on the patients with T2D and HTG.

Description

Prediction model of fish oil lipid-lowering efficacy based on intestinal microbial characteristics and its construction method Technical Field

[0001] The present invention belongs to the field of drug efficacy prediction, and particularly relates to a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics and a construction method thereof. Background Art

[0002] Fish oil (FO), rich in marine omega-3 polyunsaturated fatty acids (PUFAs), such as docosahexaenoic acid (DHA; C22:6ω3) and eicosapentaenoic acid (EPA; C20:5ω3), has been shown to lower triglyceride (TG) levels. The UK government recommends that all adults consume 6.5% of their energy from PUFAs and recommends eating a portion of oily fish (providing approximately 0.45 g / day of long-chain omega-3s) weekly, with the exception of pregnancy and lactation. The Dietary Guidelines for Americans also encourage the selection of seafood that is higher in EPA and DHA and lower in methylmercury. The American Diabetes Association also supports a Mediterranean diet rich in polyunsaturated fatty acids, as well as dietary patterns that reduce saturated fat intake and increase omega-3 fatty acid intake to improve lipid management.

[0003] Existing evidence indicates that the intestinal flora is closely related to the absorption and metabolism of dietary components and profoundly affects human physiological metabolism, including lipid metabolism. The Lifelines DEEP study showed that the intestinal flora has a significant impact on changes in blood lipids and identified a series of bacterial groups associated with TG levels. At the same time, an increasing number of clinical studies have pointed out that the baseline intestinal flora structure is related to the degree of metabolic benefit after dietary intervention. The effects of omega-3 fatty acid dietary supplements on the intestinal flora have mainly been studied in various rodent models and have not yet been systematically evaluated in large-scale human populations. Unlike the results of mouse studies, studies on healthy or overweight / obese groups have observed that omega-3 fatty acid supplements cause limited changes in human intestinal microorganisms, and no consensus has been reached among different research groups. In addition, few studies have elucidated the heterogeneous effects of intestinal flora on the TG-lowering efficacy of fish oil intervention in patients with type 2 diabetes (T2D) and hypertriglyceridemia (HTG), based on which the present invention is proposed.

[0004] Summary of the Invention

[0005] The main purpose of the present invention is to provide a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics, which is constructed by analyzing the species and functional composition of the human intestinal flora, to solve the problem in the existing technology of the lack of effective prediction of the lipid-lowering efficacy of fish oil intervention in patients with T2D and HTG.

[0006] Another object of the present invention is to provide a method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] The present invention provides a method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics, comprising the following steps:

[0009] S1. Obtain the patient's medical records of fish oil intervention treatment, and collect stool samples and peripheral blood samples at baseline and several follow-up points;

[0010] S2. Extracting intestinal microbial DNA from the stool sample, using shotgun metagenomic sequencing to obtain human intestinal microbial metagenomic data, and performing preprocessing and statistical analysis on the data to obtain microbial characteristic parameters associated with the patient's fish oil intervention treatment;

[0011] S3. detecting triglyceride (TG) data of the patients at baseline and at several follow-up points, and dividing the patients into a TG response group and a non-response group;

[0012] S4. Based on the characteristic parameters associated with the fish oil intervention treatment of the patient described in step S2 and the TG data described in step S3, using a logistic regression model to estimate the differences in intestinal microbial variables between the TG response group and the non-response group at baseline and several follow-up points, including intestinal microbial species and intestinal microbial functional pathways;

[0013] S5. Use a random forest model to construct a prediction model for intestinal microbial baseline variables, use repeated iterative features to obtain the optimal AUC value, and extract the intestinal microbial features that contribute most to the TG response grouping and the optimal random forest model TG response probability threshold;

[0014] S6. The intestinal microbial characteristics are incorporated into a linear regression model to evaluate the relationship between the predicted TG change value and the actual TG change value of the intestinal microbial baseline variables.

[0015] Preferably, in step S1, the patient is a T2D patient combined with HTG with stable blood sugar control.

[0016] Preferably, in step S3, patients whose fasting TG decreased by ≥30% at week 12 are classified as the TG response group, and patients whose TG decreased by ≤10% are classified as the non-response group.

[0017] Preferably, in step S4, the intestinal microbial species is selected from one or a combination of two or more of Rosoburia sp. CAG471, Eubacterium ramulus, Dorea formicgenerans, Fusicatenibacter saccharivorans, Coprococcus comes, Gemmiger formicilis and Roseburia hominis.

[0018] Preferably, in step S5, the intestinal microbial characteristics include L-histidine degradation III, thiamine diphosphate biosynthesis pathway III, Hungatella effluvii, Fusicatenibacter saccharivorans, Eubacterium ramulus, Dorea formicigenerans and Ruminococcus torques.

[0019] Preferably, in step S5, based on the intestinal microbial characteristics, the TG response probability of the prediction model of the intestinal microbial baseline variables for the TG response group and the non-response group is calculated, and the TG response probability when the sum of specificity and sensitivity reaches the maximum is the optimal TG response probability threshold.

[0020] More preferably, in step S5, the optimal random forest model TG response probability threshold is 0.549.

[0021] The present invention also provides a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics, which is obtained by any of the methods for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics, wherein:

[0022] Gut microbial characteristics included L-histidine degradation III, Superpathway of thiamin diphosphate biosynthesis III, Hungatella effluvii, Fusicatenibacter saccharivorans, Eubacterirum ramulus, Dorea formicigenerans, and Ruminococcus torques;

[0023] Based on the intestinal microbial characteristics, the TG response probability of the TG response group and the non-response group is obtained by the fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics. When the TG response probability is greater than the optimal random forest model TG response probability threshold, fish oil has a good lipid-lowering effect on T2D patients with stable blood sugar control and HTG.

[0024] Preferably, the optimal random forest model TG response probability threshold is 0.549.

[0025] Compared with the existing technology, the beneficial effect of the present invention is that the fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics in the present invention effectively solves the problem in the existing technology of the lack of effective prediction of the lipid-lowering efficacy of fish oil intervention in patients with T2D and HTG. The prediction model constructed based on seven intestinal microbial characteristics can effectively determine the lipid-lowering efficacy of fish oil in patients with T2D and HTG with stable blood sugar control, providing important reference value for clinical treatment and diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a comparison of the intestinal microbial diversity of the response group (R) and the non-response group (NR) in the embodiment; wherein, A: comparison of α diversity (Shannon index at the species level) between the R group and the NR group, using the Wilcoxon rank sum test; B: violin plot of the inter-individual differences in intestinal microbial composition (Bray-Curtis distance at the species level) between the R group and the NR group.

[0027] Figure 2 is a forest plot of the difference in species richness between the R group and the NR group at week 0, week 4, or week 12 in the embodiment; wherein the risk ratio and 95% confidence interval (CI) of each species are estimated according to the logistic regression model and expressed as effect size (Y-axis), yellow dots indicate effect size <1 and P <0.05, representing species with higher abundance in the NR group; cyan dots indicate effect size >1 and P <0.05, representing species with higher abundance in the R group.

[0028] Figure 3 is a receiver operating characteristic (ROC) curve in the embodiment, showing the prediction of TG response groups (R group and NR group) using a random forest (RF) model based on baseline clinical phenotype (n = 16, red), fasting serum lipids (n = 721, blue) and intestinal microbial variables (n = 365, green, including species and functional pathways), showing the area under the curve (AUC) and its 95% confidence interval.

[0029] FIG4 is a directional SHAP value diagram of the relationship between TG changes (target variables) and seven selected microbial features with the highest predictive AUC in the embodiment; wherein the X-axis shows the SHAP value of each variable in each sample, and the average absolute SHAP value is shown on the left, indicating the feature importance of each feature.

[0030] FIG5 is an explanation of the TG changes (target variable) in the fish oil intervention group (N=117) and the placebo group (N=114) based on the linear regression model constructed based on seven selected microbial features in the embodiment, as shown by the coefficient of determination (R 2 ) estimation, the scatter plot shows the predicted TG change value (fitted value, Y-axis) and the actual TG change value (X-axis) of each person, and the Spearman correlation coefficient and P value between the predicted and actual values ​​are calculated and displayed. DETAILED DESCRIPTION

[0031] To make the technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Example 1

[0033] 1. Study population

[0034] The study population in this example is based on the OCEAN study. The OCEAN (Effect of Omega-3 Fatty Acids on HTG in Patients with T2D) study is a phase IV, multicenter, randomized, double-blind, placebo-controlled trial designed to compare the effects of fish oil supplementation or corn oil placebo (ClinicalTrials.gov ID, NCT03120299). Subjects were enrolled from April 25, 2017, to March 5, 2021. The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine. All participants provided written informed consent before enrollment.

[0035] Eligible subjects were patients with T2D and HTG who had stable glycemic control but were not receiving lipid-lowering therapy. The trial included four screening / baseline visits (weeks -6, -2, -1, and 0) and three treatment visits (weeks 4, 11, and 12). During the screening phase, fasting blood samples were collected at visits 2 (week -2) and 3 (week -1) to measure serum fasting triglyceride concentrations, and the mean baseline fasting triglyceride level was determined as the average of these two values. If the mean TG values ​​at visits 2 and 3 did not meet the inclusion criteria, an additional visit was allowed (visit 3.1, one week after visit 3), and the mean fasting TG level was replaced by the mean TG values ​​at visits 3 and 3.1.

[0036] After completing a 6-week screening period, eligible patients were randomly assigned to receive 12 weeks of either 4 g / day of FO capsules (OMEGATREASURER, GOWELL, Chengdu, China) or 4 g / day of placebo capsules (corn oil). Each FO capsule contained 900 mg of omega-3 fatty acids, including 400 mg of EPA and 320 mg of DHA. Participants and investigators were masked to treatment group assignment to ensure impartiality of the evaluation.

[0037] During each visit, participants completed questionnaires assessing their acceptance of treatment, adherence to interventions, physical activity levels, high-fat diet scores, and any adverse events experienced. Fasting and postprandial blood samples (0, 4, and 12 weeks) were collected at each visit for biochemical parameters including lipid profiles, blood glucose, and glycation. Fecal samples were collected at 0, 4, and 12 weeks for gut microbiome analysis.

[0038] 2. Shotgun metagenomic sequencing and gut microbiome analysis

[0039] Fecal samples were self-collected from patients with T2D at baseline and after 4 and 12 weeks of supplementation with FO or corn oil placebo. Microbial DNA was extracted from 758 fecal samples using the Mag Pure Fast Stool DNA KFKitB.33 kit. Shotgun metagenomic 100-bp paired-end (PE) sequencing was performed on the MGI platform (based on DNA nanospheres, Shenzhen, China). Fastp (version 0.20.1, default parameters) was used to filter low-quality reads, and Bowtie2 (version 2.4.2, default parameters) was used to filter human sequences (database hg38). On average, 70.11 million (±20.80 million) high-quality non-human sequencing reads were generated per sample after quality control. Microbial taxonomic analysis of high-quality non-human sequences was generated at the species level using MetaPhlAn3 (version 3.0.4) using default parameters, and 952 microbial species were identified. HUMAnN2 (v0.11.1) software was used for functional pathway analysis of metagenomic samples, and 392 pathways were identified. Given the high sparsity of known microbial data, rare microbial variables with low occurrence rates (758 metagenomic samples had an occurrence rate less than 20%) were excluded, resulting in a final set of 238 species and 260 pathways for the OCEAN study.

[0040] 3. Statistical analysis

[0041] Patients with a ≥30% reduction in fasting TG at week 12 were classified as the responder group (R, 38.5%, 45 / 117), and patients with a ≤10% reduction in TG were classified as the non-responder group (NR, 27.4%, 32 / 117). These 32 groups of responder and non-responder groups matched for baseline TG levels were analyzed. After adjusting for age and sex, a logistic regression model was used to estimate the differences in intestinal microbial variables (species and functional pathways) between the responder and non-responder groups at baseline, week 4, and week 12. A BH-adjusted P value <0.05 was considered significant, and a BH-adjusted P value ≥0.05 and a P value <0.05 were considered to be a statistically significant trend.

[0042] The predictive performance of three different types of baseline variables was evaluated: clinical variables (age, sex, triglycerides, TC, HDL-C, LDL-C, ApoB, FPG, 30-min PG, 2-h PG, HbA1c, and different antidiabetic drug types: metformin, sulfonylureas, AGI, DPP4-I, and SGLT-2), lipid metabolites (n = 721), and gut microbial variables (n = 498, 238 species and 260 functional pathways). Random forest models (leave one out cross-validation (LOOCV), R package "randomForest", v4.7-1.1) were used to evaluate the predictive performance of each model. For each model, an iterative feature selection process was applied to obtain the best AUC value, and the feature variables with the greatest contribution were extracted for each type of model (number of iterations = 100, with a maximum number of iterations set to 20 to avoid overfitting). Among all iterative models, the model with the highest predictive performance was determined, with an average AUC of 0.77 (Figure 3).

[0043] This model was selected for further analysis, including seven specific microbial signatures: L-histidine degradation III, Superpathway of thiamin diphosphate biosynthesis III, Hungatella effluvii, Fusicatenibacter saccharivorans, Eubacterium ramulus, Dorea formicigenerans, and Ruminococcus torques. The probability of triglyceride (TG) response was calculated for fish oil responders and non-responders using a random forest discriminant model based on these seven microbial signatures. The optimal threshold for TG response probability was determined when the sum of specificity and sensitivity reached the maximum. SHAP value plots were used to estimate the contribution of the features in the model. The seven selected microbial signatures were incorporated into a linear regression model using the glm function (R package "stats," v4.1.0) to estimate the explained variance, explained degree, and predicted value (fitted value) of TG response in the fish oil and placebo groups, respectively. Spearman rank correlation analysis was performed to assess the correlation between the model-predicted values ​​and the actual TG reduction.

[0044] 4. Results

[0045] A comparative analysis of 32 pairs of patients in the response group and the non-response group matched for baseline TG levels showed that at baseline, there were no significant differences in anthropometric indicators, diabetes duration, medication use, blood glucose levels, TG and other lipid components between the two groups (P>0.05, Table 1).

[0046] Table 1

[0047] Further comparison of the gut microbial species composition between the two subgroups at baseline and at both follow-up points revealed no statistically significant differences in α- and β-diversity between the two groups at baseline (P>0.05, Figure 1). However, the relative abundance of 49 gut species was significantly different between the two groups at at least one time point (P<0.05; Figure 2). Notably, the abundance of seven species was lower in the responder group than in the non-responder group at all time points (P<0.05), including Rosoburiasp CAG471, Eubacterium ramulus, Dorea formicgenerans, Fusicatenibacter saccharivorans, Coprococcus comes, Gemmiger formicilis, and Roseburia hominis. Most of these species were negatively correlated with triglyceride levels (P<0.05 after BH correction).

[0048] To evaluate the predictive performance of various baseline variables for the magnitude of fish oil-induced TG reduction, a RF model was constructed to screen features and then evaluated using the LOOCV method. During model construction, repeated iterative feature selection was used to identify the most relevant variables for predicting TG response. Figure 3 shows the prediction of TG response (responder vs. non-responder group) by RF models constructed based on baseline clinical phenotype (n = 16, red), fasting serum lipid metabolites (n = 721, blue), and gut microbial species and functional composition (n = 365, green). The area under the curve (AUC) and its 95% confidence interval (CI) are shown. The results showed that the baseline gut microbiota exhibited superior performance in distinguishing the responder and non-responder groups (AUC = 0.77, 95% CI: 0.65-0.89) compared with the clinical phenotype (AUC = 0.53, 95% CI: 0.39-0.68) and lipid species (AUC = 0.58, 95% CI: 0.44-0.72). The highest AUC (AUC = 0.77) was obtained by using seven selected gut microbial features, including L-histidine degradation III, Superpathway of thiamin diphosphate biosynthesis III, Hungatella effluvii, Fusicatenibacter saccharivorans, Eubacterium ramulus, Dorea formicigenerans, and Ruminococcus torques ( Figure 3 ).

[0049] As shown in Figure 4, the SHAP value plot shows the direction of the relationship between TG changes (target variables) and the seven selected microbial features with the highest prediction AUC. The X-axis shows the SHAP value of each variable for each sample. The larger the absolute SHAP value, the higher the contribution of the feature.

[0050] As shown in Figure 5, the linear model constructed using seven microbial features explained 18% of the variation in TG changes in all patients in the fish oil intervention group (P = 0.004), and the Spearman correlation coefficient between the predicted and actual TG changes was 0.43 (P = 2.09E-7). The model had no predictive ability for TG changes in the placebo group (explanatory power 2%, P = 0.97; Spearman correlation coefficient = 0.07, P = 0.46).

[0051] The above is a preferred embodiment of the present invention, but the present invention should not be limited to the contents disclosed in this embodiment. Therefore, any equivalent or modified implementations that do not depart from the spirit disclosed in the present invention fall within the scope of protection of the present invention.

Claims

1. A method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics, comprising the following steps: S1. Obtain the patient's medical records of fish oil intervention treatment, and collect stool samples and peripheral blood samples at baseline and several follow-up points; S2. Extracting intestinal microbial DNA from the stool sample, performing shotgun metagenomic sequencing to obtain human intestinal microbial metagenomic data, performing preprocessing and statistical analysis on the data, and obtaining microbial characteristic parameters related to the patient's fish oil intervention treatment; S3. detecting the TG data of the patients at baseline and at several follow-up points, and dividing the patients into a TG response group and a non-response group; S4. Based on the characteristic parameters associated with the fish oil intervention treatment of the patient described in step S2 and the TG data described in step S3, using a logistic regression model to estimate the differences in intestinal microbial variables between the TG response group and the non-response group at baseline and several follow-up points, including intestinal microbial species and intestinal microbial functional pathways; S5. Use a random forest model to construct a prediction model for intestinal microbial baseline variables, use repeated iterative features to obtain the optimal AUC value, and extract the intestinal microbial features that contribute most to the TG response grouping and the optimal random forest model TG response probability threshold; S6. The intestinal microbial characteristics are incorporated into a linear regression model to evaluate the relationship between the predicted TG change value and the actual TG change value of the intestinal microbial baseline variables.

2. The method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics according to claim 1, characterized in that: In step S1, the patient is a T2D patient combined with HTG with stable blood sugar control.

3. The method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics according to claim 1, characterized in that: In step S3, patients whose fasting TG decreased by ≥30% after 12 weeks of fish oil intervention were classified as the TG response group, and patients whose TG decreased by ≤10% were classified as the non-response group.

4. The method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics according to claim 1, characterized in that: In step S4, the intestinal microbial species is selected from one or a combination of two or more of Rosoburia sp. CAG471, Eubacterium ramulus, Dorea formicgenerans, Fusicatenibacter saccharivorans, Coprococcus comes, Gemmiger formicilis and Roseburia hominis.

5. The method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics according to claim 1, characterized in that: In step S5, the intestinal microbial characteristics include L-histidine degradation III, Superpathway of thiamin diphosphate biosynthesis III, Hungatella effluvii, Fusicatenibacter saccharivorans, Eubacterirum ramulus, Dorea formicigenerans and Ruminococcus torques.

6. The method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics according to claim 5, characterized in that: In step S5, based on the intestinal microbial characteristics, the TG response probability of the prediction model of the intestinal microbial baseline variables for the TG response group and the non-response group is calculated, and the TG response probability when the sum of specificity and sensitivity reaches the maximum is the optimal TG response probability threshold.

7. The method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics according to claim 6, characterized in that: In step S5, the optimal random forest model TG response probability threshold is 0.

549.

8. A fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics, characterized by: The method for constructing a fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics according to any one of claims 1 to 7 is obtained, wherein: Gut microbial characteristics included L-histidine degradation III, Superpathway of thiamin diphosphate biosynthesis III, Hungatella effluvii, Fusicatenibacter saccharivorans, Eubacterirum ramulus, Dorea formicigenerans, and Ruminococcus torques; Based on the intestinal microbial characteristics, the TG response probability of the TG response group and the non-response group is obtained by the fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics. When the TG response probability is greater than the optimal random forest model TG response probability threshold, fish oil has a good lipid-lowering effect on T2D patients with stable blood sugar control and HTG.

9. The fish oil lipid-lowering efficacy prediction model based on intestinal microbial characteristics according to claim 8, characterized in that: The optimal random forest model TG response probability threshold is 0.549.

Images