The gut microbiome affects the responses of heart valve replacement patients to the anticoagulant warfarin

Backgrounds: Numerous algorithms based on patient genetic variants have been established with the aim of the GI and during However, approximately 35% of individual warfarin sensitivity remains unexplained. Gut microbiota composition related to the vitamin K generation should be taken into account. Methods : Faecal samples were collected from 200 inpatients undergoing heart valve replacement (HVR). Ultimately, faecal samples from 80 inpatients (27 low responders (LRs), 27 high responders (HRs) and 26 normal responders (NRs) were analyzed for microbiota composition through 16S rDNA sequencing. Fifteen samples (5 LRs, 5 HRs and 5 NRs) were also analyzed through metagenomic whole-genome shotgun (WGS) sequencing. We validated the results from 10 LRs, 10 HRs and 10 NRs. Results: Significant differences were observed in the diversity and composition of the gut microbiome among the three groups. The genera Bacteroides, Escherichia-Shigella and Klebsiella were enriched in the LRs, while the genus Enterococcus was enriched in the HRs. WGS sequencing indicated that the abundance of enzymes, modules and KO associated with bacterial vitamin K biosynthesis was significantly higher in LRs than in HRs or NRs. The 12 optimal microbial markers were identified through tenfold cross-validation with a random forest model. Conclusions: This study characterized gut microbiome in the different response to warfarin of HVR patients and suggested that gut microbiome might play an important role in the clinical warfarin anticoagulation therapy. warfarin functional genomic profiling of the gut microbiome via metagenomic whole genome shotgun (WGS) sequencing of 15 fecal samples (5 LRs, 5 NRs, 5 HRs) by MicroPITA analysis. In the validation phase, 30 LRs, 30 NRs and 30 HRs used to validate the diagnostic efficacy of the warfarin response classifier. HRV, heart valve replacement; INR, international normalized ratio; R, the difference rate between a patient’s actual warfarin dose and the theoretical warfarin dose calculated by the formula generated in this study; WSI, warfarin sensitivity index; MicroPITA, microbiomes: Picking Interesting Taxonomic Abundance.

The gut microbiome affects the responses of heart valve replacement patients to the anticoagulant warfarin Warfarin, a vitamin K antagonist, [1] is prescribed to nearly one million patients with prosthetic heart valves worldwide. [2] Although warfarin is highly effective in reducing thromboembolism, its use is limited by its narrow therapeutic index and the required frequent monitoring of the international normalized ratio (INR). [3] An INR of 1.5 to 2.5 is recommended in East Asian patients with heart valve replacement (HVR), consistent with the guidelines announced by the American College of Chest Physicians in 2008. [4] Overall, patients taking warfarin have significantly more bleeding events than patients taking other anticoagulants (P<0.001). [5][6][7] Numerous algorithms that are based on patient factors and genetic variants have been established by multivariate linear regression with the aim of reducing the risk of bleeding during the initial days of warfarin administration. [8] In 2007, we[9,10] created an algorithm for daily warfarin dose requirements so that appropriate warfarin maintenance doses could be calculated easily. [11] This tool has verifiably improved the decision-making process for warfarin dosing in HVR patients at our hospital, but approximately 35% of individual variability remains unexplained [10].
All types of vitamin K(VK) used by the human body are obtained from food or are synthesized by bacteria. [12] It has been stated that up to 50% of the human requirement for vitamin K is fulfilled by the intestinal production. [13,14] However, vitamin K concentrations in the human gut appear highly variable and are associated with gut microbiota composition. [15] To date, there has been no systematic study on the effect of bacterial VK on warfarin anticoagulation in HVR patients with a low dietary intake of vitamin K.
The aim was to establish possible differences in gut microbiota composition and functionality among warfarin low responder (LR), high responder (HR) and normal responder (NR) using next-generation metagenomics sequencing techniques. In the discovery phase, we characterized the gut microbiomes of different groups, analyzed the content of vitamin K in the stool with LC-MS/MS, and studied the mechanism through which the gut microbiome might influence the effects of anticoagulation therapy.
Furthermore, a validation cohort was used to evaluate the potential of the gut microbiome as a noninvasive biomarker, which showed the importance of gut microbiome in the clinical warfarin anticoagulation therapy ( Figure 1).

2.1Ethics and patients
Ethical permission for this study was obtained from the Health Authority Ethics Committee of the First Affiliated Hospital of Soochow University (Suzhou, Jiangsu, China). All patients provided written informed consent in accordance with the Declaration of Helsinki. Stable inpatients who had undergone HVR surgery and taken warfarin sodium tablets (Shanghai Xinyi Pharmaceutical Co., Ltd., Shanghai, China) during hospitalization from July 2017 to December 2018 were recruited. All patients were given written consent prior to their participation in the study.
The inpatients recruited for this study (Supplementary Table S1) were those who had first undergone heart valve replacement and had used second generation cephalosporin antibiotics during the operation to prevent infection Supplementary Methods 1.1 .

2.2Fecal sample collection and stool moisture measurement
Each inpatient was provided with fresh stool specimen collection systems during hospitalization and asked to collect all stools produced after cardiac valve surgery (Supplementary Methods 1.2). Five aliquots of 5 g were obtained from the homogenate and immediately stored at -80°C.
Stool moisture content was determined in duplicate from frozen homogenized faecal material (−80°C) as the percentage of stool mass loss after lyophilization.

2.416S rDNA microbial profiling analysis
The QIIME software suite (http://qiime.org/scripts/assign_taxonomy.html) and the related 16S database SILVA, V.128 (http://www.arb-silva.de) were used for taxonomic classification of Operational Taxonomic Units (OTUs). All OTUs for all samples in the discovery sets were collected (Supplementary A rarefaction curve was developed using the R language and was constructed using the number of sequences extracted and the diversity index of the corresponding OTUs. Based on the microbial profiles, we calculated the alpha diversity in the discovery phase to estimate the richness using the Chao index, and the diversity was calculated using the Shannon index and Simpson index. A large value of the Chao and Shannon indexes indicated a higher degree of diversity in the sample. These results were used to analyse effects of different phenotype factors. A Venn diagram can be used to represent the number of common and unique species (such as OTUs) among the LR, HR and NR groups. We created a Venn diagram using an R language tool.
A beta diversity analysis of interindividual variability was performed using the principal coordinate analysis (PCoA) method based on weighted UniFrac and Bray-Curtis dissimilarity at the genus level.

2.5Shotgun metagenomics analysis of fecal samples
We used microPITA (microbiomes: Picking Interesting Taxonomic Abundance) [16]  Bacterial DNA was extracted from stool samples as detailed in the Supplementary Methods 1.5. All predicted genes with 95% sequence identity (90% coverage) were clustered using CD-HIT [18] (http://www.bioinformatics.org/cd-hit/). The longest sequences from each cluster were selected as representative sequences to construct a non-redundant gene catalogue. After quality control, reads were mapped to the representative sequences with 95% identity using SOAPaligner [19] (http://soap.genomics.org.cn/), and the gene abundance in each sample was evaluated. Organism-specific gene hits were annotated with the Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology (KO) database. Comparisons of pathway enrichment across the LR, HR and NR groups showed differences in metabolic functions, especially menaquinone biosynthesis.

2.6UPLC-APCI-MS analysis for vitamin K 2 in fecal samples
Aliquots of plasma and fecal samples were freeze-dried to a constant weight and homogenized by using a mortar and pestle. 500 μL plasma sample was analyzed for vitamin K1 concentration, and approximately 40 mg dried fecal sample were analyzed for vitamin K2 content. Samples were measured using published methods with minor modifications [23,24]. Briefly, we changed the amount of protein precipitation and extraction agent as well as the proportion of mobile phase.

2.7OTU biomarker identification
We developed a random forest-based classification approach to perform faecal source identification [25] using microbial community data from the discovery cohort and validation cohort

2.8Statistical analysis
Continuous variables are expressed as the mean±SD. Discrete variables are expressed as percentages. The average relative abundances of specific taxa and KEGG functions are also expressed as percentages. The Kruskal-Wallis H test of variance was used to evaluate differences among the three groups. Continuous variables were compared between two groups using the Wilcoxon rank sum test. Statistical analyses were performed using SPSS V20.0 (IBM),considering P values<0.05 as statistically significant.

3.1Participant characteristics
The 200 enrolled HVR patients' demographics, indications for anticoagulation therapy, and genotypes are shown in Supplementary Table S1. According to the inpatients' warfarin doses and INRs, we found that 20% of the patients were oversensitive to warfarin and were prone to excessive anticoagulation, while 15% of the patients exhibited low sensitivity to warfarin. These groups of patients (20% and 15%) were defined as HR group and LR group, respectively. The others (65%) were defined as NR group. The warfarin sensitivity index (WSI) [26] was significantly different among the LR, HR and NR This finding indicated that at the beginning of anticoagulation therapy, three groups' clinical characteristics were all equivalent, but the difference of WSI among the three groups was still very significant (P<0.001).

3.2Stool moisture and form scale
Stool moisture and form are strongly associated with gut microbiota richness and composition. [27] In terms of stool moisture, there were no significant differences among LRs, HRs and NRs in the discovery phase (P=0.213), as shown in Table 1 and Supplementary Table S2. Moreover, all stool samples had a soft consistency. With regard to stool color, most of the stool samples were yellow, and there were no significant color differences among the stools LRs, HRs and NRs.

Biodiversity of fecal microbiota
To profile the differences in the structure of the gut microbiota among HVR patients, we performed Rarefaction analysis showed that the estimated OTU richness approached saturation in each group ( Figure 2A). The estimated sample coverage (Good's coverage) was approximately 0.999, and the correlation between duplicate samples was more than 99.5%, which indicated that the accuracy and reproducibility of the sequences were good ( Figure 2B). Compared with the NRs (n=26) and LRs (n=27), the HRs showed markedly reduced fecal microbial diversity as estimated by the Shannon index (P=0.0088 and 0.0060, respectively), while there was no significant difference between the LR group and the NR group in this respect (P=0.875) ( Figure 2C).
Moreover, a Venn diagram displaying the overlap between groups showed that 378 of the overall 660 OTUs were shared among the three groups. (Figure 2D, Supplementary Table S4). To measure the microbiome differences between samples, we calculated beta diversity using Bray-Curtis dissimilarity, and principal coordinate analysis (PCA) indicated a symmetrical distribution of the fecal microbial community among all samples ( Figure 2E).
Based on these data, we examined the diversity of the gut microbiomes in HVR inpatients and found that the alpha diversity of the gut microbiome was significantly higher in LRs (n=27) than in HRs (n=27) according to sobs, shannon, simposon, ace, chao indices (P=0.006, Supplementary Table S5).
To further explore these findings, we performed high-dimensional class comparisons through LEfSe (Supplementary Figure S1A, 1B). A representative cladogram of fecal microbiota structures and predominant bacteria showed the great differences in taxa among the LR, HR and NR groups (Supplementary Figure S1C). These data again demonstrated that there were differences in the abundance of bacteria in the fecal microbiome of the three groups with different responses to anticoagulation therapy; Escherichia-Shigella was enriched in the LR group, and Enterococcus was enriched in the HR group.

3.4Functional analysis of fecal microbiota through shotgun metagenomics
Next, we sought to gain further insight into the mechanism through which the gut microbiome may influence the response to anticoagulation therapy. We first conducted functional genomic profiling of  Figure 4B and Supplementary Table S10, the LR group contained higher level modules than the other two groups, and the levels of M00023, M00095, M00096 and M00116 were significantly increased.
The 16 KO functions from the available KEGG functions are equal to the enzymes listed above. The LR group showed a higher abundance of these K0s than the other two groups ( Figure 4C, Supplementary   Table S11).
Overall, the LR group had low selectivity for the anticoagulant warfarin, maybe because these patients are enriched with those enzymes, modules and KOs, which are all positive factors related to bacterial vitamin K biosynthesis; the reverse was found for the HR group.

3.6Identification of microbial OTU-based markers of HVR inpatient responses to anticoagulation therapy
To illustrate the diagnostic value of the fecal microbiome in determining responses to warfarin anticoagulation, we constructed a random forest classifier model that could specifically identify LRs, HR and NRs. Based on the 80 HVR patients in the discovery phase, bacterial genera corresponding to 50 OTU markers were selected as the optimal marker set (Supplementary Table S12). The cross-validation error curve (Supplementary Figure S3A) was calculated using the set of 50 identified optimal OTUs for both the discovery cohort and the validation cohort. The 12th solid point represents the point with the lowest error rate. Then, 12 genera were selected for the optimal marker set by random forest models.
In the discovery phase, the ROC curve between the LR and HR groups had an AUC value of 0.87 with a 95% confidence interval (CI) of 0.73 to 1 between LR and HR cohorts (Supplementary Figure S3B), that between the HR and NR had an AUC value of 0.78 with a 95% CI of 0.56 to 1 (Supplementary Figure S3C), and that between the LR and NR had an AUC value of 0.75 with a 95% CI of 0.56 to 0.91 (Supplementary Figure S3D). These data suggested that the random forest module based on microbial OTU markers had high diagnostic value for the LR, HR and NR groups.
In the validation phase, 30 LRs, 30 HRs and 30 NRs were used to validate the diagnostic efficacy for the LR and HR groups. The ROC curve between the LR and HR groups had an AUC value of 0.81 (95% CI: 0.62 to 0.99) (Supplementary Figure S3E), the AUC value between the HR and NR was 0.72 (95% CI: 0.48 to 0.92) (Supplementary Figure S3F), and the AUC value between the LR and NR was 0.69 (95% CI: 0.44 to 0.93), which indicated significant diagnostic potential for both groups.

Discussion And Conclusions
It has previously been suggested that the potential role of microbiota in vitamin K synthesis should be considered in selecting warfarin regimens [33]. Given the samples were from the first replacement surgery hospitalized patients taking warfarin anticoagulant therapy, vitamin k intake is limited, the diet daily intake of vitamin k and plasma concentrations of VK1 and MK4 were all have no significant difference in the three groups of patients, so this study focused on the intestinal bacteria groups for the synthesis of VK2 pathways and gene abundance.
Recent studies [15,32] have reported that several forms of vitamin K are synthesized by Bacteroides, Enterobacter, Veillonella, and Eubacterium lentum, which are typical members of the intestinal microflora. If the number of vitamin K-synthesizing bacteria in the microbiota increases for any reason, vitamin K levels will be elevated [33]. This study found the abundance of escherichia-shigella, which was positively correlated with vitamin K production, were significantly different among three groups (P<0.001) and showed LR group >NR group > HR group.
This study is the first report to illustrate gut microbial characteristics in HVR patients at