Article Text
PDF
Abstract
Objective Gut microbiome composition is associated with multiple diseases, but relatively little is known about its relationship with long-term outcome measures. While gut dysbiosis has been linked to mortality risk in the general population, the relationship with overall survival in specific diseases has not been extensively studied. In the current study, we present results from an in-depth analysis of the relationship between gut dysbiosis and all-cause and cause-specific mortality in the setting of solid organ transplant recipients (SOTR).
Design We analysed 1337 metagenomes derived from faecal samples of 766 kidney, 334 liver, 170 lung and 67 heart transplant recipients part of the TransplantLines Biobank and Cohort—a prospective cohort study including extensive phenotype data with 6.5 years of follow-up. To analyze gut dysbiosis, we included an additional 8208 metagenomes from the general population of the same geographical area (northern Netherlands). Multivariable Cox regression and a machine learning algorithm were used to analyse the association between multiple indicators of gut dysbiosis, including individual species abundances, and all-cause and cause-specific mortality.
Results We identified two patterns representing overall microbiome community variation that were associated with both all-cause and cause-specific mortality. The gut microbiome distance between each transplantation recipient to the average of the general population was associated with all-cause mortality and death from infection, malignancy and cardiovascular disease. A multivariable Cox regression on individual species abundances identified 23 bacterial species that were associated with all-cause mortality, and by applying a machine learning algorithm, we identified a balance (a type of log-ratio) consisting of 19 out of the 23 species that were associated with all-cause mortality.
Conclusion Gut dysbiosis is consistently associated with mortality in SOTR. Our results support the observations that gut dysbiosis is associated with long-term survival. Since our data do not allow us to infer causality, more preclinical research is needed to understand mechanisms before we can determine whether gut microbiome-directed therapies may be designed to improve long-term outcomes.
- INTESTINAL MICROBIOLOGY
- LIVER TRANSPLANTATION
Data availability statement
Data are available on reasonable request. The raw microbiome sequencing data and basic phenotypes used in this study are available at the European Genome-Phenome Archive under accession numbers EGAD00001008907 (https://ega-archive.org/datasets/EGAD00001008907), EGAS00001006257 (https://ega-archive.org/studies/EGAS00001006257) and EGAS00001006258 (https://ega-archive.org/studies/EGAS00001006258). Due to patient confidentiality, the clinical datasets associated with the metagenomic datasets are available on request to the University Medical Centre Groningen. Access to this clinical dataset requires a minimal access procedure consisting of a request per email (datarequest.transplantlines@umcg.nl) for a data access form. A response will be provided within two working weeks. This access procedure is to ensure that the data are being requested for research/scientific purposes only and thus comply with the informed consent signed by TransplantLines participants, which specifies that the collected data will not be used for commercial purposes.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Current literature suggests that a gut microbiome signature might be associated with mortality risk in the general population.
A higher diversity of gut microbiota is associated with lower mortality in allogeneic hematopoietic-cell transplantation recipients.
Liver and kidney transplant recipients suffer from gut dysbiosis and a previous analysis with a relatively low number of events showed that gut dysbiosis is associated with mortality after transplantation.
WHAT THIS STUDY ADDS
Across kidney, liver, heart and lung transplant recipients, we identified two overall microbial community variation patterns that are associated with all-cause mortality independent of the type organ transplant, and specifically to death from malignancy and infection.
Multiple indicators of gut dysbiosis predict all-cause mortality and death by cardiovascular diseases, malignancy and infection.
We find multiple bacterial species associated with all-cause and cause-specific mortality. Using three different methods, we identify multiple bacterial species (shared between different analytical approaches) that are associated with an increased or decreased risk of mortality following solid organ transplantation.
Using a machine learning algorithm, we identified a balance of 19 bacterial species that was associated with all-cause mortality.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Our results support emerging evidence showing that gut dysbiosis is predictive of long-term survival, indicating that gut-microbiome targeting therapies might improve patient outcomes.
Introduction
Gut dysbiosis, while not clearly defined, is a condition typically characterised by the growth of pathogens at the expense of commensal bacteria when compared with healthy gut microbiomes. A dysbiotic gut microbiome has been observed in many diseases, including inflammatory bowel disease, obesity, diabetes mellitus and cancer.1–4 Different population-based cohort studies report a large overlap in microbial associations with general health, which suggests a common dysbiotic signature in compromised health.5–7
Recent evidence suggests that these dysbiotic signatures are not only associated with an individual’s health at the time of sampling but also predictive of long-term survival. The gut microbiome, characterised by shotgun metagenomics of stool samples, was associated with mortality in a well-characterised population-based study consisting of 7211 Finnish adults with a follow-up of 15 years.8 This study found that members from the Enterobacteriaceae family were associated with an increased mortality risk.8 While studies linking gut dysbiosis with patient survival in specific disease populations are scarce, the relationship has been studied extensively in allogeneic haematopoietic-cell transplantation where, for example, a lower alpha diversity is associated with an increased mortality risk.9 10 We recently reported, in the setting of solid organ transplantation, that the extent of gut dysbiosis (in terms of each recipient’s microbiome distance to the centroid of the general population) in both liver and kidney transplant recipients (KTR) was associated with higher all-cause mortality risk.11 However, the number of events in this study was relatively small due to the limited follow-up time.
In the current study, we present results from an in-depth survival analysis of both all-cause and cause-specific mortality in a large population of solid organ transplant recipients (SOTR). In this study, we have 3.7 times more samples and 2.7 times more events from liver, kidney, heart and lung transplantation recipients compared with our previous analysis. Because this population of SOTR is characterised by polypharmacy, multimorbidity and the prevalence of gut dysbiosis is high compared with the general population,11 12 it represents an ideal model to study the relationship between gut dysbiosis and long-term survival. With metagenomes from this unique population of SOTR (n=1337) together with metagenomes from the general population (n=8208), we analysed the relationship between the gut microbiome and all-cause and cause-specific mortality. These findings are of interest to the transplantation community but also to a broader readership as it increases our general understanding of the relationship between the gut microbiome and long-term health outcomes.
Results
Characteristics of SOTR
In total, 1337 SOTR who provided a faecal sample at a variable time after transplantation were included in the TransplantLines Biobank and Cohort study. Of these, 766 were KTR, 334 were liver transplant recipients (LTR), 170 were lung transplant recipients (LuTR) and 67 were heart transplant recipients (HTR). The average age (±SD) of all recipients was 57 (±13.0) years, 784 recipients (59%) were male, and the average time since transplantation was 7.6 (±8.0) years across all organ types (figure 1). During the follow-up period (min=2.8 years and max=6.5 years), a total of 162 participants (88 KTR, 33 LTR, 35 LuTR and 6 HTR) died (online supplemental table 1). Of these, 48 (28%) died due to infection-related mortality, 38 (23%) due to cardiovascular-related mortality, 38 (23%) due to malignancy-related mortality and 40 (25%) died from other causes (online supplemental table 1).
Supplemental material
Cohort overview. (A) Overview of the number of recipients per transplantation type, including the number of deaths. (B) The number of males and females and (C) age at the time of inclusion per transplantation type. (D) The number of deaths as a function of years since transplantation.
In the first part of the analysis, we computed for each transplant recipient’s gut microbiome, multiple indicators of gut dysbiosis: Shannon diversity index, the distance to the average microbiome composition of the general population, and richness of antibiotic resistance genes (ARGs) and virulence factors (VFs). With multivariable Cox regression including age, sex, body mass index (BMI) and years since transplantation, we subsequently investigated the relationship between each of these indicators and recipient all-cause and cause-specific mortality. We analysed all-cause mortality for each transplantation type separately. However, due to a limited number of events per organ type, we performed the cause-specific analysis on all SOTR pooled. In the second part of the analysis, we aimed to identify bacterial species that individually or jointly predicted mortality. We investigated the relationship between transplant recipient mortality and (1) each species’ CLR (centred log-ratio) transformed relative abundance, and (2) each species’ CLR transformed relative abundance in the general population (grouped as lower (<25% percentile) or a higher (>75% percentile) in the general population). Lastly, we applied a machine learning algorithm to identify the balance that best predicts mortality.
Overall community variation is associated with mortality risk
We first performed a principal component analysis (PCA) on species-level CLR transformed relative abundances (this produces the Aitchison distance which is the gold standard for compositional data). On all SOTR pooled, we observed a positive significant relationship between principal component (PC) 1 and all-cause mortality (hazard ratio (HR) 1.32, 95% CI 1.13 to 1.54, false discovery rate (FDR)=5.8×10−4) and a negative relationship between PC3 and all-cause mortality (HR 0.80, 95% CI 0.68 to 0.93, FDR=4.0×10−3; figure 2A,B). PC3 was also associated with a lower mortality for KTR and HTR (KTR: HR 0.73, 95% CI 0.59 to 0.89, FDR=2.2×10−3; HTR: HR 0.34, 95% CI 0.12 to 0.95, FDR=0.03; figure 2C). This suggests that bacterial species with high loadings on PC1 are associated with increased mortality while species with high loadings on PC3 are associated with decreased mortality. Analysing this further, we found that the five species that exhibited the largest positive loadings onto PC1 (and therefore associated with mortality) were Ruminococcus gnavus, Clostridium clostridioforme, Clostridium symbiosum, Hungathella hathewayi and Clostridium innocuum (figure 2E), and the five species that exhibited the largest positive loadings onto PC3 (and therefore inversely associated with mortality) were Bifidobacterium adolescentis, Dorea longitena, Bifidobacterium longum, Collinsella aerofaciens and Eubacterium rectale (figure 2F). PC2, PC4 and PC5 were not significantly associated with mortality (FDR>0.10; online supplemental figure 1, online supplemental table 2). In the cause-specific analysis, we found that PC1 was related to death from malignancy and infection (malignancy: HR 1.64, 95% CI 1.20 to 2.24, FDR=2.0×10−3; infection: HR 1.42, 95% CI 1.06 to 1.89, FDR=0.01; figure 2A,B and D), and PC3 with a decreased risk of death from malignancy (HR 0.68, 95% CI 0.49 to 0.94, FDR=2.0×10−2). Finally, mortality status significantly explained variation in microbiome species composition (PERMANOVA: R2=0.1%, p=4.1×10−3; PERMANOVA stratified by transplant organ type: R2=0.1%, p=0.01).
Supplemental material
Forest plot showing results from a Cox regression analysis performed on PC1 and PC3 for all-cause and cause-specific mortality, respectively. (C) PCA plot with dots representing recipients and different colours transplantation type. Circles and crosses represent centroids of recipients that are alive and dead at the time of the follow-up, respectively. (D) PCA with dots representing recipients, colours different causes of death, and crosses centroids. (E, F) The top 10 species with the largest positive and negative loadings on PC1 and PC3, respectively. I.: Intestinimonas. (G) All-cause and cause-specific per gut dysbiosis indicator. PCA, principal component analysis.
Gut dysbiosis indicators are associated with all-cause and cause-specific mortality
In the cause-specific analysis performed on all SOTR, we found that the Shannon diversity index was related to death from malignancy (HR 0.71, 95% CI 0.54 to 0.93, FDR=0.01). We then calculated the distance from each transplant recipient’s gut microbiome to the average composition of the general population and performed a multivariable Cox regression. This analysis revealed that the distance to the general population was significantly associated with a higher all-cause mortality risk in all SOTR (HR 1.29, 95% CI 1.11 to 1.51; FDR=9.3 x 10−4; figure 2G, online supplemental figure 2). In the cause-specific analysis, the distance to the general population was related to death from infection (HR 1.46, 95% CI 1.11 to 1.93, FDR=7.2×10−3), malignancy (HR 1.39, 95% CI 1.02 to 1.89, FDR=0.03) and cardiovascular disease (HR 1.36, 95% CI 1.01 to 1.87, FDR=0.04; figure 2G). Finally, we found that harbouring a higher richness of ARGs and VFs were associated with an increased all-cause mortality risk (ARGs: HR 1.27, 95% CI 1.09 to 1.47; FDR=2.2×10−3, figure 2G; VFs: HR 1.14, 95% CI 1.01 to 1.27; FDR=0.03; figure 2G) and with death from infection in the cause-specific analysis (ARGs: HR 1.45, 95% CI 1.10 to 1.90, FDR=8.0×10−3; VFs: HR 1.28, 95% CI 1.09 to 1.51, FDR=3.0×10−3; figure 2G). We did not observe a significant association between the Shannon diversity index and all-cause mortality when we analysed recipients together or stratified by transplantation type (FDR>0.05, online supplemental tables 3 and 4). Taken together, lower Shannon diversity, a more dissimilar composition compared with the general population, and higher richness of ARGs and VFs, were associated with increased mortality.
Multiple species are associated with mortality
We identified a total of 23 (16%; FDR<0.10) species whose CLR transformed relative abundance were associated with all-cause mortality in multivariable Cox regression models adjusting for age, sex, BMI and years since transplantation (online supplemental table 5). When we analysed all SOTR, we found four Clostridium species (C. innocuum: HR 1.35, 95% CI 1.18 to 1.54, FDR=0.001; C. clostridioforme: HR 1.34, 95% CI 1.16 to 1.54, FDR=0.003; C. symbiosum: HR 1.35, 95% CI 1.17 to 1.57, FDR=0.003; and C. bolteae: HR 1.32, 95% CI 1.157 to 1.52, FDR=0.004) that were positively associated with all-cause mortality (figure 3A; online supplemental table 5) and death from infection in the cause-specific analysis (online supplemental table 6; figure 3B). Other species that were associated with an increased mortality risk included H. hathewayi (HR 1.29, 95% CI 1.12 o 1.50, FDR=0.01), Veillonella parvula (HR 1.29, 95% CI 1.12 to 1.49, FDR=0.01) and R. gnavus (HR 1.26, 95% CI 1.09 to 1.46, FDR=0.03; figure 3A; online supplemental table 5). In the cause-specific analysis, we found that the abundance of H. hathewayi (HR 1.48, 95% CI 1.18 to 1.92, FDR=0.08) and V. parvula (HR 1.57, 95% CI 1.21 to 2.03, FDR=0.04) was related to death from infection (online supplemental table 6; figure 3B), and that the abundance of R. gnavus was associated with death from malignancy (HR 1.83, 95% CI 1.34 to 2.49, FDR=0.04; online supplemental table 6; figure 3B). We also identified multiple species that were associated with a lower mortality risk when we analysed all SOTR. For example, butyrate producers Eubacterium hallii (HR 0.75, 95% CI 0.63 to 0.89, FDR=0.01), Firmicutes bacterium CAG 83 (HR 0.77, 95% CI 0.66 to 0.90, FDR=0.01), Gemmiger formicilis (HR 0.77, 95% CI 0.66 to 0.89, FDR=0.01) and Faecalibacterium prausnitzii (HR 0.83, 95% CI 0.73 to 0.95, FDR=0.05) were associated with a lower mortality risk (figure 3A; online supplemental table 5). Other commensals that were associated with a lower mortality risk included Adlercreutzia equolifaciens (HR 0.77, 95% CI 0.66 to 0.91, FDR=0.02), Prevotella copri (HR 0.77, 95% CI 0.65 to 0.91, FDR=0.03), Asaccharobacter celatus (HR 0.79, 95% CI 0.67 to 0.93, FDR=0.04), D. longicatena (HR 0.80, 95% CI 0.68 to 0.93, FDR=0.04), Bifidobacterium adolescentis and B. longum (HR 0.80, 95% CI 0.68 to 0.95, FDR=0.07 and HR 0.83, 95% CI 0.72 to 0.96, FDR=0.08, respectively). Lastly, to test the generalisability of our findings, we analysed mortality data from the general population. This analysis replicated the protective association between all-cause mortality and two Bifidobacterum species (B. adolescentis (HR=0.95, p=2.0×10−16) and B. longum (HR=0.93, p=2.0×10−16), see the Methods section). We could not, however, replicate any of the other associations we found in the transplantation cohort.
Multiple species are associated with all-cause and cause-specific mortality. (A) A forest plot showing species with abundances that were associated with all-cause mortality in the multivariable Cox regression models. HRs with 95% CIs are plotted together with the false discovery rate (FDR)-corrected p values. (B) Forest plots showing species with abundances that were associated with cause-specific mortality in multivariable Cox regression models. HRs with 95% CIs and FDR-corrected p values are shown.
We took this analysis one step further by categorising transplantat recipients based on whether their species’ CLR transformed relative abundances were outside of their average range in the general population (higher in GP: >75% percentile; lower in GP: <25% percentile), and if this binary stratification led to stronger associations with mortality. This measure is similar to our dysbiotic indicator ‘distance to the average of the general population’ but instead gives an indication of the extent of gut dysbiosis per species as opposed to the level of microbiome composition. When we analysed all SOTR, we found several bacterial species that were also associated with mortality in the previous analysis (11/16 and 22/23 species at an FDR<0.05 and FDR<0.1, respectively) but with stronger HRs (at an FDR<0.05; figure 4). For example, the four Clostridium species (C. innocuum, C. clostridioforme, C. symbiosum and C. bolteae) that were positively associated with all-cause mortality in the previous Cox regression models exhibited up to 1.8 times higher HRs in this analysis (range of increase: min=1.3, median=1.35, max=1.8; figure 4; online supplemental table 7).
Kaplan-Meier curves showing the mortality probabilities of recipients compared with the general population. We categorised whether each species’ CLR transformed relative abundance in the transplant recipients were outside of its ‘average’ range in the general population (higher (>75% quantile) or lower (<25% quantile)). Bar plots depict the percent of deceased patients in the lower (<25% quantile; Q1) and higher (>75% quantile; Q4) group, respectively. (A) Transplant recipients with a high abundance (Q4 in the general population) of the focal species exhibited a decreased mortality risk compared with recipients in the Q1 group. (B) Transplant recipients with low abundances (Q1 in the general population) of the focal species exhibited a decreased mortality risk compared with recipients in the Q4 group. CLR, centred log-ratio.
Relationship between polypharmacy, the gut microbiome and mortality
To analyse the relationship between polypharmacy, the gut microbiome and mortality, we performed a series of analyses. First, we analysed the relationship between polypharmacy and mortality using a multivariable Cox regression correcting for age, sex and years since transplantation. Prior to this analysis, we defined different medication regimens by clustering recipients based on the types of immunosuppressants and antibiotics they were taking (this analysis identified seven medication regimens; see the Methods section). In this analysis, we did not find any significant association between any of the medication regimens and mortality (p>0.05). Second, we analysed the relationship between species’ CLR transformed relative abundances and medication use by including medication regimens, age, sex and years since transplantation as independent variables in multivariable linear models. From these models, we computed post-hoc constrasts comparing all medication regimens with each other. Of 672 tested combinations, 151 were significant (FDR<0.05, online supplemental figure 3A,B). Lastly, we performed a mediation analysis to test whether the relationship between each of the seven medication regimens and mortality was mediated by any of the 32 bacterial species we found to be associated with mortality in the sections above. However, we did not observe any significant mediation effect of any of these species (p>0.05; online supplemental figure 3C).
The functional potential of the gut microbiome is associated with mortality
We identified a total of 4 (3%) metabolic pathways (MPs), 20 (28%) VFs and 0 (0%) ARGs that were significantly associated with all-cause mortality in multivariable Cox regression models adjusting for age, sex, BMI and years since transplantation (FDR<0.10; online supplemental tables 10-13). Peptidoglycan biosynthesis I (HR 0.74, 95% CI 0.63 to 0.88, FDR=0.042), uridine monophosphate biosynthesis (HR 0.76, 95% CI 0.66 to 0.89, FDR=0.042), coenzyme A biosynthesis II (HR 0.78, 95% CI 0.68 to 0.90, FDR=0.042) and L-histidine biosynthesis (HR 0.78, 95% CI 0.66 to 0.92, FDR=0.086) were associated with decreased all-cause mortality risk (online supplemental table 8). In the cause-specific analysis, glycolysis I (HR 0.88, 95% CI 1.48 to 3.96) and II (HR 0.90, 95% CI 1.47 to 4.08) from fructose 6-phosphate were associated with a lower risk of death from infection (FDR<0.10; online supplemental table 9).
We found multiple VF genes that were associated with all-cause mortality (FDR<0.10; online supplemental table 10). For instance, 12 genes coding for adherence and invasion (VF0221, VF0404 and VF0506, FDR<0.10; online supplemental table 11), 7 genes coding for iron uptake (VF0123, VF0136, VF0227, VF0228 and VF0229, FDR<0.10; online supplemental table 10) and 1 gene coding for type II secretion system (VF0333) were significantly associated with increased all-cause mortality risk. Results for the cause-specific analysis can be found in online supplemental table 11.
Lastly, we found 32 ARGs that were associated with all-cause mortality with a p<0.05. Of these, 6 were associated with death from cardiovascular diseases, 10 were associated with death from infection, and finally, 10 were associated with death from malignancy (FDR>0.10; online supplementaltables 12-13). However, none of these 32 ARGs were significantly associated with mortality after multiple testing correction (FDR>0.05).
The gut microbiome as a predictive biomarker of mortality
Finally, to identify a predictive biomarker, we used the machine learning algorithm CoDaCoRe. This algorithm identifies the balance that is most predictive of the outcome, in this case, death (dead/alive at the time of censoring) in all SOTR (see the Methods section). We first applied this machine learning algorithm on a training set (80%) and then evaluated model performance on a testing set containing the remaining number of patients. The training and testing set had an equal proportion of events. This algorithm identified a balance consisting of 19 species that had a classification accuracy of 88% in the test set (Area Under the Curve (AUC) in test set=0.68; figure 5A,B). In this balance, the numerator set—or species whose geometric mean is predictive of death consisted of eight species (Bacteroides eggerthii, B. fragilis, H. hathewayi, C. bolteae, C. clostridioforme, C. symbiosum, Ruminococcaceae bacterium D16 and Escherichia coli; figure 5A; online supplemental table 14), and the denominator set—or species whose geometric mean is predictive of mortality consisted of 11 species (B. adolescentis, B. longum, Adlercreutzia equolifaciens, A. celatus, P. copri. E. hallii, Anaerostipes hadrus, Coprococcus comes, D. longicatena, Fusicatenibacter saccharivorans and G. formicilis; figure 5A; online supplemental table 14). All of these species’ CLR transformed relative abundances were individually associated with mortality in our previous analysis (FDR<0.10; figure 3A; online supplemental table 5). Finally, we tested whether the identified balance also could predict mortality in a multivariable Cox regression adjusting for age, sex, BMI and years since transplantation. We found that this balance indeed was associated with an increased all-cause mortality risk (HR 1.74,95% CI 1.48 to 2.04, p=1.60×10−11, figure 5C). When we stratified transplant recipients based on whether they harboured lower or higher than the median value of this balance, we found that the HR increased by almost a factor of 1.4 (HR 2.40, 95% CI 1.71 to 3.35, p=3.40×10−7, figure 5D).
Machine learning algorithm identifies a log-ratio predictive of mortality. (A) Species included in the identified CoDaCoRe ratio. (B) Receiver operating characteristic (ROC) curve demonstrating discriminative power of the balance identified by CoDaCoRe in the testing set. (C) The relationship between the HR and the balance for light blue area representing the 95% CI of the HR. (D) Kaplan-Meier curves for recipients harbouring lower (orange) and higher (blue) the median log-ratio value across all SOTR. AUC, Area Under the Curve; B, Bifidobacterium; F, Fusicatenibacter; R, Ruminococcaceae; SOTR, solid organ transplant recipient.
Next, we analysed whether we could confirm these findings in the general population cohort that was also included in this study. We performed a multivariable Cox regression based on the predictive balance identified by CoDaCoRe but in the general population cohort. In total, 11 of the 19 species that were included in the balance were present in the general population with a prevalence above the minimum of 20% (online supplemental figure 4, see the Methods section). In the general population, the calculated balance was not significantly associated with mortality (HR 1.06, 95% CI 0.99 to 1.15, p=0.11). Because the population of SOTR is different from the general population11, we selected individuals from the general population (1) with the 10% lowest and 10% highest gut microbiome health index6 scores, and (2) with the 10% highest and 10% lowest Aitchison distances to the average of the transplantation cohort. We did not find a significant association between the balance and all-cause mortality in neither subset of the general population (1: HR 1.00, 95% CI 0.86 to 1.17 p=0.98; 2: HR 1.17, 95% CI 0.99 to 1.34 p=0.07).
Discussion
In the current study, we report an in-depth analysis of the gut microbiome in relation to both all-cause and cause-specific mortality in a population of SOTR from the TransplantLines cohort and biobank study.13 We observed gut microbial signatures associated with both all-cause and cause-specific mortality, especially death from infection. The distance of the gut microbiome to general population controls, resistome and VF richness were associated with a higher mortality risk. We found a consistent mortality-related gut microbial signal consisting of previously disease-associated species. Interestingly, we discovered that if the abundance of a species among SOTR is outside what is considered the ‘normal’ range in the general population, it also predicts mortality. Overall, our results show that gut dysbiosis-related signatures are associated with mortality across different types of organ traplantation.
Diversity analysis was partly consistent with previously reported results. Patients with a lower Shannon diversity had 29% higher risk of malignancy-related mortality. However, we did not observe a significant relationship between the Shannon diversity and all-cause mortality in the pooled SOTR analysis or stratified by organ-type. Thus, we were unable to confirm previously reported associations between Shannon diversity and mortality in haematopoietic stem cell and liver transplantation recipients.10 11 Similar to the gut microbiome mortality analysis performed in the FINRISK study, we observed a significant relationship between all-cause and cause-specific mortality and patterns representing overall microbiome community variation.8 For example, SOTR that were 1 SD higher than average in PC1 had a 32% higher mortality risk, while SOTR 1 SD higher in PC3 had a 20% lower mortality risk. We observed a consistent gut microbial signal in all of the performed mortality analyses.
Many of the species that were associated with mortality in our study have previously been associated with disease in the general population.5 Specifically, we found that abundances of several clostridium species, including C. clostridioforme, C. symbiosum, C. bolteae, and V. parvula and R. gnavus were associated with higher mortality in SOTR and with multiple diseases in the general population. In contrast, P. copri, D. longicatena and F. prausnitzii were associated with lower mortality in SOTR and with general health in the general population.5 Furthermore, how different the gut microbiome of SOTR are compared with the general population is associated with all-cause-related, infection-related, cardiovascular-related and malignancy-related mortality. This relationship was consistent with individual bacterial species in our study, but when compared with the FINRISK study, we only replicated the finding of a lower mortality risk associated with a higher abundance of F. prausnitzii in our SOT population.8 However, in the FINRISK study, Cox regression analysis was performed on the genus level and SHOGUN was used for taxonomical classification while we performed our analysis on the species level and we used MetaPhlAn for taxonomic profiling, potentially limiting the comparability between the two studies.14 Thus, further research using diverse populations and standardised methodology is needed to test whether our findings generalise to broader populations.
We observed that a lower abundance of G. formicilis, F. bacterium CAG 83, E. hallii and F. prausnitzii - four butyrate producing species 15–18 - is associated with increased mortality. Butyrate is a short-chain fatty acid with a broad utility in the gut, including maintenance of the intestinal barrier, anti-inflammatory effects, anti-obesity and anti-oxidant functions.19 20 It was previously observed that KTR and LTR have a lower abundance of butyrate producing bacteria compared with general population controls and that a lower abundance of butyrate producing bacteria is associated with a lower health-related quality of life in KTR.11 21–23 We now find a higher mortality risk for SOTR with a lower abundance of butyrate producing bacteria. These results suggest that reduced butyrate levels could potentially have a direct role in mortality for SOTR. A lower abundance of butyrate producing bacteria was associated with increased occurrence of graft-vs-host disease and transplantation-related mortality in HSCT recipients.24 We did not observe any associations between mortality-related and butyrate-related metabolic pathways. Interestingly, we observed that H. hathewayi, R. gnavus and R. D16 were associated with increased mortality. These three bacterial species have previously been associated with maintaining gut barrier function. For example, R. gnavus is able to digest glycans and mucins through glycosidase enzymes which is important for maintaining barrier integrity, however, to much mucus degradation can impair the mucus layer and intestinal integrity.25 H. hathewayi can degrade glycosaminoglycans which have a role in maintaining gut barrier function and it has been shown previously that excessive degradation of glycans can diminish gut barrier integrity and increase susceptibility to infections.26 Multiple VF genes were associated with a higher mortality which in combination with potential loss of barrier integrity could pose great risks for SOTR. A loss of gut barrier integrity due to a potentially lower butyrate production and increased mucus and glycan degradation could play a role in the observed association between species with these functions and mortality. However, additional analysis should be performed to confirm these hypotheses. Our results warrant further studies into the role of butyrate producing and mucus degrading bacteria and mortality in SOTR. The use of butyrate producing probiotics may represent a promising avenue to improve outcomes for SOTR.27
Strengths of the current study include a large sample size of SOTR and the availability of a large control group from the same Dutch population. With this dataset, we were able to pinpoint a gut microbiome mortality-related signature in a group of SOTR with a high prevalence of gut dysbiosis. A limitation of the current study is its observational nature, which means causality or directionality of the reported association cannot be inferred. While reverting the observed gut dysbiosis may improve survival after transplantation, it is also possible that the associations we observe, rather than being causal, results from indirect effects such as poor overall health. Future studies should try to identify which factors influence the gut microbiome mortality-related signature we observed in the current study. Another limitation of the current study is that samples were not obtained at uniform time points after transplantation due to the cross-sectional nature of our cohort. In the cause-specific mortality analysis, we were also limited by power, and because we used a broad classification for malignancy, we could not perform a tumour specific analysis. Lastly, to capture the dynamic (i.e., time-dependent) nature of the gut microbiome, future studies would also benefit from a longitudinal design.
This study highlights a gut microbiome mortality-related signature in a population of SOTR with a high prevalence of gut dysbiosis. These findings are of interest to the transplant community as well as our general understanding of the relationship between the gut microbiome and health. Our results support emerging evidence showing that gut dysbiosis is associated with long-term survival, indicating that gut-microbiome targeting therapies might improve patient outcomes although causal links should be identified first.
Methods
Study design
All SOTR cross-sectional gut microbiome data from the TransplantLines Biobank and Cohort study (Trial registration number NCT03272841) was included.13 The TransplantLines study has been previously described in detail and aimed to include all potential adult SOTR and kidney donors at the University Medical Center Groningen (UMCG), The Netherlands, starting from June 2015.13 We included 1337 faecal samples from SOTR and 8208 subjects from the Dutch Microbiome Project were included to quantify the extent of dysbiosis and per species dysbiosis analysis.5 In the TransplantLines project, one sample per individual was included as a baseline sample at varying time after transplantation. Faecal samples from TransplantLines and DMP were collected using the same procedures and processed with the same DNA extraction protocols (see below).
Clinical data
In the TransplantLines study, every transplant recipient was asked to fill in questionnaires and blood, urine and faecal samples were collected. A detailed description, including details regarding the rationale of the study design, inclusion/exclusion criteria and randomisation of the TransplantLines study, is given by Eisenga et al.13 In the current study, the primary outcome was overall survival. Clinical records were assessed to verify if a participant was alive or deceased, using a censoring date of 1 January 2022. If a patient was deceased, we assessed the cause of death and classified the cause of death into cardiovascular, infection, malignancy or other related mortality categories. At the time of sampling, there were no patients with acute rejection. We have collected mortality data for the 8208 individuals of the general population for which we have gut microbiome data available. By linking the data to the wide Statistics Netherlands (CBS) registry, we observed in total, 139 individuals who have died in the follow-up period since sample collection until the censoring date of 1 May 2023. No data regarding the cause of death were available.
Patient and public involvement
Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Sample selection and gut microbiome data generation
Faecal sample collection and subsequent processing
Patients were asked to collect a faecal sample the day prior to the study visit. A FecesCatcher (TAG Hemi VOF, Zeijen, The Netherlands) was sent to the patients at home. Faeces were collected and stored in appropriate tubes and frozen at home (at −18°C) immediately after collection. Frozen faecal samples were collected by UMCG personnel and stored at −80°C until DNA extraction.
DNA extraction
Microbial DNA was extracted using QIAamp Fast DNA Stool Mini Kit (Qiagen, Germany) according to the manufacturer’s instructions. The QIAcube (Qiagen, Germany) automated sample preparation system was used for this purpose. Library preparation was performed using NEBNext Ultra DNA Library Prep Kit for Illumina for samples with total DNA amount lower than 200 ng, as measured using Qubit 4 Fluorometer while samples with DNA yield higher than 200 ng were prepared using NEBNext Ultra II DNA Library Prep Kit for Illumina. Libraries were prepared according to the manufacturer’s instructions. Metagenomic shotgun sequencing was performed using Illumina HiSeq 2000 sequencing platform and generated approximately 8 Gb of 150 bp paired-end reads per sample (mean 7.9 gb, st.dev 1.2 gb). Library preparation and sequencing were performed at Novogene and MGI.
Metagenomic data processing
Illumina adapters and low-quality reads (Phred score <30) were filtered out using KneadData (V.0.5.1).28 Then Bowtie2 (V.2.3.4.1)29 was used to remove reads aligned to the human genome (hg19). The quality of the reads was examined using FastQC toolkit (V.0.11.7). Taxonomy alignment was done by MetaPhlAn3 (V.3.7.2)29 30 with the database of marker genes mpa_v20_m200. Metacyc pathways were profiled by HUMAnN2 (V.0.11.1).31 Bacterial VFs and ARGs were identified using shortBRED (shortbred_identify.py (V.0.9.5) (51) and shortbred_quantify.py tool (V.0.9.5)) against VFs database of pathogenic bacteria (19 August 2016) database (http://www.mgc.ac.cn/VFs/main.htm) and comprehensive antibiotic resistance database (V.1.1.1) (https://card.mcmaster.ca/) separately. Samples were further excluded in case of an eukaryotic or viral abundance >25% of total microbiome content or a total read depth <10 million. In total, we identified 1132 taxa (17 phyla, 27 class, 52 order, 98 family, 231 genera and 705 species). After filtering for a prevalence of 10% and relative abundance threshold of 0.01%, 141 species and 149 MPs were left. Hereafter, total-sum normalisation was applied. For VFs and ARGs, we only included genes with a prevalence of 10%. Analyses were performed using locally installed tools and databases on CentOS (release 6.9) on the high-performance computing infrastructure available at UMCG and University of Groningen (RUG). An example of scripts used for microbiome processing is available at https://github.com/GRONINGEN-MICROBIOME-CENTRE/TransplantLines.
Statistical analysis
Because of the compositional nature of the metagenomic sequencing data,32 we transformed our data using the CLR transformation prior to all of the analyses, apart from computing alpha diversity and running CoDaCoRe. We analysed polypharmacy by using unsupervised cluster analysis (we used vegdist from the vegan R package with ‘Jaccard’ dissimilarity to perform hierarchical cluster analysis with Ward D agglomeration from the hclust function from the stats R package) to identify seven different medication regimens community of transplantation patients. In the case of the PCA analysis, with CLR-transformed relative abundances as the input, the samples space is the so-called Aitchison geometry with the distance between samples characterised by the Aitchison distance.33 We calculated the Shannon Diversity Index using the vegan 34 package in R. To analyse the association between dysbiosis indicators and mortality, we used multivariable Cox proportional hazard models as implemented in the function coxph() in the R package survival (V.3.5-5). We included sex, BMI and years since transplantation as independent variables. We used the cox_wrapper function from Salosensaari et al to analyse the relationship between CLR transformed species abundances and mortality.8 We used the survreg() and mediate() function from the mediation package in R to perform multivariable mediation analysis for Cox regression. The adonis2() function from the vegan package was used to perform distance-based analysis for mortality. Mortality status, survival time, age and time since transplantation were included in the model and we assessed the marginal effects of the terms for all SOTR and stratified by transplantation type with 9999 permutations. To further analyse the relationship between dysbiosis and the gut microbiome, we used gut microbiome data from the general population and reclassified CLR transformed species abundance for SOTR according to quantiles of the general population. Finally, we applied CoDaCoRe35 to identify the bacterial log-ratio or so-called balance that best predicts mortality in our SOTR population using mortality status as the dependent variable. We set lambda=0 and the maximum for base learners to 1. The most predictive ratio was then calculated per SOTR and Cox regression analysis was performed on this ratio. FDR was applied as a correction for multiple testing in all analysis.36
Data availability statement
Data are available on reasonable request. The raw microbiome sequencing data and basic phenotypes used in this study are available at the European Genome-Phenome Archive under accession numbers EGAD00001008907 (https://ega-archive.org/datasets/EGAD00001008907), EGAS00001006257 (https://ega-archive.org/studies/EGAS00001006257) and EGAS00001006258 (https://ega-archive.org/studies/EGAS00001006258). Due to patient confidentiality, the clinical datasets associated with the metagenomic datasets are available on request to the University Medical Centre Groningen. Access to this clinical dataset requires a minimal access procedure consisting of a request per email (datarequest.transplantlines@umcg.nl) for a data access form. A response will be provided within two working weeks. This access procedure is to ensure that the data are being requested for research/scientific purposes only and thus comply with the informed consent signed by TransplantLines participants, which specifies that the collected data will not be used for commercial purposes.
Ethics statements
Patient consent for publication
Ethics approval
All participants signed an informed consent form prior to sample collection. TransplantLines (METc 2014/077) and Lifelines (METc 2017/152) were approved by the local institutional ethics review board (IRB) from the UMCG. Both studies adhere to the UMCG Biobank Regulation and are in accordance with the World Medical Association (WMA) Declaration of Helsinki and the Declaration of Istanbul.
Acknowledgments
We would like to thank all participants from the TransplantLines- and Lifelines cohort and biobank study. We would like to thank the Centre for Information Technology of the University of Groningen (RUG) for support and for providing access to the Peregrine high-performance computing cluster and the Genomic Coordination Center (UMCG and RUG) for support and for providing access to Calculon and Boxy high-performance computing clusters. We thank the MMHP project and MGI for sequencing of the samples in this study.
References
Footnotes
X @CasSwarte, @jingyuan_fu, @HarmsenHermie, @WeersmaLab, @WeersmaLab
JRB and RKW contributed equally.
Collaborators TransplantLines investigatorsC. Annema, F. A. J. A. Bodewes, M. T. de Boer, K. Damman, A. Diepstra, G. Dijkstra, C. S. E. Doorenbos, M. F. Eisenga, M. E. Erasmus, C. T. Gan, A. W. Gomes Neto, E. Hak, B. G. Hepkema, F. Klont, H. G. D. Leuvenink, W. S. Lexmond, G. J. Nieuwenhuis-Moeke, H. G. M. Niesters, L. J. van Pelt, A. V. Ranchor, J. S. F. Sanders, M. J. Siebelink, R. J. H. J. A. Slart, D. J. Touw, M. C. van den Heuvel, C. van Leer-Buter, M. van Londen, E. A. M. Verschuuren & M. J. Vos.
Contributors HB, JCS, SJLB, RKW and JRB conceived and designed the project. JCS, SJLB and RKW obtained funding. TJK, JCS, HB, EAMV, TCG and SJLB collected samples. JCS and JRB analysed and interpreted data. JCS, JRB, SJLB and RKW drafted the manuscript. JCS, RG, RKW and JRB revised the manuscript. SZ, LMN, RG, TJK, VEDM, KD, EAMV, TCG, JF, AZ, HJMH and HB were involved in critically reviewing the manuscript. JCS, SJLB, JRB and RKW accept full responsibility for the work and conduct of the study. All the authors approved the final version of the manuscript.
Funding This study was funded by Dutch Ministry of Economic Affairs and Climate Policy(Not Applicable).Seerave Foundation(Not Applicable). Chiesi Farmaceutici(PA-SP/PRJ-2020-9136). Astellas BV(Not Applicable). Dutch NWO/TTW/DSM(project number 14939). EU Horizon Europe Program(miGut-Health 101095470). Netherlands Organization for Scientific Research(Not Applicable).
Competing interests The TransplantLines Biobank and Cohort study received funding from Astellas BV (TransplantLines Biobank and Cohort study) and Chiesi Pharmaceuticals BV (PA-SP/PRJ-2020-9136) and was cofinanced by the Dutch Ministry of Economic Affairs and Climate Policy by means of the PPP allowance made available by the Top Sector Life Sciences and Health to stimulate public–private partnerships. Sequencing of the kidney part of the TransplantLines cohort was funded by a grant from the Dutch NWO/TTW/DSM partnership programme Animal Nutrition and Health (project number 14939) to SJLB, RKW is supported by the Seerave Foundation, the Netherlands Organization for Scientific Research (NWO) and the EU Horizon Europe Programme grant miGut-Health: personalised blueprint of intestinal health (101095470). JF is supported by the Dutch Heart Foundation IN-CONTROL (CVON2018-27), the ERC Consolidator grant (grant agreement No. 101001678), NWO-VICI grant VI.C.202.022, the AMMODO Science Award 2023 for Biomedical Sciences from Stichting Ammodo and the Netherlands Organ-on-Chip Initiative, an NWO Gravitation project (024.003.001) funded by the Ministry of Education, Culture and Science of the government of The Netherlands.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.