Abstract
Background Acute Graft-versus-Host Disease (aGvHD) is the main complication of allogeneic hematopoietic stem cell transplantation (allo-HSCT). Gut microbiota composition has been associated with aGvHD, however, the specific composition of the intestinal ecosystem involved in the pathologic process at the time of aGvHD onset is not yet fully known.
Methods A relevant series of allo-HSCT microbiome data sets were combined and reanalyzed, with focus on species level changes in the microbiota, using state-of-the-art and in part proprietary 16S analysis routines.
Results Immunosuppressive conditioning was found to impact gut microbiome (GM) composition. GM composition was found highly correlated with aGvHD status at onset (R2=0.4). The species diversity, average biological safety level (BSL), gram staining balance and the strict anaerobe proportion at aGvHD onset indicate that a moderate level of gram positive strict anaerobe cleansing with antibiotics reduces the risk of aGvHD. Butyrate producers were found associated with non-GvHD controls, as well as the probiotic species and IL-10 inducer Bifidobacterium longum. Among species associated with mortality and aGvHD, we found Th1 and Th17 eliciting bacteria. We formulated a probiotic composition putatively competing with mortality and aGvHD species, from correlation analysis in a large superset of gut samples (n=17,800) from clinical- and crowd-sourced studies.
Implications To restore the Treg/Th17 balance, probiotic supplementation with B. longum eventually combined with Bacillus clausii, prebiotics and vitamin B9 is a treatment option. Simultaneous use of vancomycin, metronidazole, ceftazidime and rifaximin for GM control is a possibility, because of antibiotic resistance of B. longum.
Introduction
Allogeneic Hematopoietic Stem Cell Transplantation (allo-HSCT) is a curative option for many patients with high-risk hematopoietic malignancies and hematological disorders. The success of allo-HSCT can be hampered by a process in which donor-derived T cells recognize host healthy tissue as non-self, causing an immune-mediated complication known as acute Graft-versus-Host Disease (aGvHD), predictive for morbidity and mortality of the patients. Even among transplants sourced from HLA-matched siblings, aGvHD occurs in 40% of recipients and reaches 50-70% with unrelated donor HSCT, yet less than half of the patients who develop aGVHD experience a response [Magenau et al. 2016]. Bacterial LPS has been proposed as a triggering factor for aGvHD, based on murine models [Hill et al. 1997]
In patients undergoing allo-HSCT, antibiotics are routinely prescribed to reduce the risk of opportunistic infections. Antibiotics can profoundly impact gut microbiome (GM) composition and it has been shown broad spectrum antibiotics used after allo-HSCT increase the risk of aGvHD in mice and humans [Shono et al. 2016, Weber et al. 2019]. On the other hand, gut decontamination (GD), i.e. the complete suppression of the GM, has been shown to prevent aGvHD following allo-HSCT, but is difficult to achieve [Beelen et al. 1999, Vossen et al. 2014]. As a matter of fact, no standardized protocol for prophylactic and peri-transplant antibiotic treatment has been established as standard of care across transplantation centers.
It has been reported that autologous fecal micro-biota transfer (auto-FMT) can in some instances restore the baseline GM after antibiotics courses applied during the allo-HSCT [Taur et al. 2018]. Of note, this baseline GM is potentially already impacted by chemotherapy [Montassier et al. 2014]. Hence some centers practice probiotics and 3rd party FMT reconditioning of GM once the antibiotics treatment has been ended [Vossen et al. 2014].
A lower GM diversity index at the time of engraftment (neutrophil recovery) has been associated in multiple studies with increased incidence of intestinal aGvHD disease and transplant related mortality (TRM) [Pamer et al. 2014, Taur et al. 2014, Golob et al. 2017, Han et al. 2018, Galloway-Peña et al. 2019]. Enterococci have been proposed as a landmark for aGvHD whether induced by antibiotics or not [Holler et al. 2014], whereas intestinal Blautia has been found associated with reduced death from aGvHD [Jenq et al. 2015]. It has also been reported that at the time of engraftment, decreased Lachnospiraceae and Ruminococcaceae and increased Enterobacteriaceae correlate with aGvHD development and a lowered Treg/Th17 ratio [Han et al. 2018].
We combined a series of relevant microbiome studies, available in the form of raw 16S data, in order to increase statistical power and investigate detailed baseline and aGvHD onset microbiome composition in relation to therapeutic outcome.
Materials & Methods
The materials summarized in Table 1 have been made available in the form of raw 16S data sets in the short read archive. Five of the six collected datasets have been made available as part of scientific publication: PRJEB23820 [Biagi et al. 2019], PRJEB16057 [Liu et al. 2017], PRJNA528754 [Galloway-Peña et al. 2019], PRJNA592853 [D’Amico et al. 2019] and PRJNA491657 [Taur et al. 2018].
Data analysis
Amplicon Sequence Variants (ASVs) were generated with the R Bioconductor package dada2, version 1.12.1 with recommended parameters [McMurdie, Paul J et al. 2016], involving quality trimming, discarding of sequences with N’s, assembly of forward and reverse sequences and contamination and chimera removal. Further analysis involved multiple alignment with mafft, version 6 [Katoh et al. 2009] and approximately-maximum-likelihood phylogenetic tree generation with FastTreeMP, version 2 [Price, Morgan N et al. 2010].
Taxonomic classification of ASVs was performed by cur|sor, an in-house program using random forest based supervised learning on the Ribosomal Database Project (RDP) release 11.5. Resulting classifications are available from https://github.com/GeneCreek/GvHD-manuscript in the form of R data objects.
Random forest survival analysis was carried out with the R package ranger, version 0.12.2.
aGvDH status regression analysis, with relative abundance of taxa resolved at the species level as independent variables, was performed using the R package relaimpo [Groemping, Ulrike [2006].
Results
Overall GM composition evolution across allo-HSCT
Data set SRP162022 comprises patients undergoing allo-HSCT (n=736) of which 14 received an auto-FMT 49 days after stem cell infusion, Fig. 1. Recovery of Shannon species diversity seems to be boosted through auto-FMT, whereas strict anaerobes were not recovered. The antimicrobial resistance of the microbiome decreased, as reflected by the average biological safety level (BSL), which means that a less pathogenic GM was recovered through auto-FMT.
Influence of conditioning on the GM
Dataset ERP017899 contains immunosuppressive conditioning intensity specification and baseline GM samples of 41 patients who underwent allo-HSCT. The conditioning was qualified as low, intermediate or high with few cases of high conditioning. We regrouped the intermediate level with the high level and tested various GM composition covariates comparing the two regimes. The Shannon species richness and the strict anaerobe proportion reached significance, Fig. 2.
Furthermore, the use of the immunosuppressant cyclosporine, which in the dataset is mutually exclusive with the use of tacrolimus, seems to favor the growth of gram-positive bacteria, Fig. 3.
Baseline GM composition and survival
Dataset ERP017899 contains baseline GMs and survival information for up to two years of 41 patients who underwent allo-HSCT. In brief, we fitted random survival forests (RSF) using log scaled relative abundances of taxa resolved at the species level as independent variables. Variable importance and associated p-values were estimated using permutation. We retained p<0.05 taxa as selected variables and the experience was repeated 50 times. Final models were built using 20 fold cross validation and a subset of taxa selected at least 10 times. Model performance was assessed using the integrated Brier score (IBS) and the best performing model was retained. This model used 11 taxa, Fig. 4 and had an IBS of 0.176. All selected taxa had higher average relative abundance at baseline in patients who subsequently deceased during the two year censoring time span. Of note is the position of the probiotic species Bifidobacterium bifidum.
aGvHD onset and GM composition
We combined data sets SRP243841, ERP105598 and SRP234378 into a 68 patients and 80 samples dataset comprising samples taken at aGvDH onset or beyond with matched controls. There is a significant difference in Shannon species diversity between the two patient groups, Fig. 5. Compositional differences could also be documented at the species level. Using linear regression and relative importance analysis, we found that 12 taxa plus added Shannon diversity index accounted for 39.8% (adjusted R2) of the aGvHG/non-aGvHD variability, Fig. 6. The linear model performs on the dataset with a receiver operating characteristic (ROC) area under the curve (AUC) of 0.915, supplemental Fig. S1.
Paradoxically, although not reaching significance, the strict anaerobe proportion of non-GvHD controls seems even lower at the time of aGvHD onset as compared to the baseline, Fig. S2. Non-GvHD controls have a significantly lower proportion of gram positive bacteria as compared to baseline, Fig. S3. Another remarkable feature of the aGvHD control GM is the higher average biological safety level (BSL) which reaches significance at aGvHD onset, Fig. S4.
In silico screening of the allo-HSCT GM
To investigate if qualified presumption of safety (QPS), other than Bifidobacterium longum are of therapeutic interest, we computed pairwise χ2 tests for all QPS with mortality and aGvHD case associated species reported above, Fig. 7. We used over 17,800 samples from several tens of clinical and crowd-sourced studies as the basis for the test. Supplemental Fig. S5 provides the same analysis for control associated species and Fig. S6 provides pairwise χ2 tests results between QPS species, the former to assess respect of control associated species and the latter to assess compatibility of QPS species for combination therapy. Bifidobacterium longum and Bacillus clausii both are in exclusion with a number of case-associated species and are found in mutual inclusion (Fig. S6), hence we predict the species are compatible and can be used in a combination therapy.
Discussion
Metabolites and aGvHD
It has been shown in a mouse model that indoles derived from intestinal microbiota act via type I interferon signaling to limit GvHD [Swimm et al. 2018]. The relation between indole and GvHD has been established in humans as well, through the quantification of an indole derivative, 3-indoxyl sulfate (3-IS), in urine [Weber et al. 2015]. Folate (vitamin B9) is known to enhance the survival of Foxp3+ Threg cells, thus diminishing intestinal inflammation through stabilization of the intestinal Threg population [Kinoshita et al. 2012]. Bifidobacteria from human origin are folate producers [D’Aimmo et al. 2012, Sugahara et al. 2015].
Furthermore, GM production of butyrate benefits intestinal barrier function and decreases GvHD in a mouse model [Mathewson et al. 2016]. Notwithstanding this, it has been shown once aGvHD of the gut has occurred, butyrogens may impair recovery and lead to chronic GvHD [Golob et al. 2019].
Allo-HSCT conditioning and the GM
Our analysis of overall GM evolution across allo-HSCT shows that at baseline, i.e. after immunosuppressive conditioning, GM composition is already significantly altered, with notably decrease of strict anaerobes among which most butyrate producers are found. At the same time, gram positive bacteria seem to be boosted by the immuno-suppressive conditioning. Cyclosporine for instance seems to boost gram positive species, whereas its alternative Tacrolimus does not.
Auto-FMT and allo-HSCT
If auto-FMT is to be practiced, the question arises when to source the samples. Sourcing after the initiation of immunosuppressive conditioning would avoid restoring butyrate producing species, which could be the safest option. Auto-FMT from our results not only restores a certain level of species diversity but also lowers the relative abundance of pathobionts as indicated by the average biological safety level.
antibiotics use and the GM
Three features in our analysis point to a different use of oral antibiotics in aGvHD controls as compared to aGvHD cases: the lower proportion of gram positive and strict anaerobes and the higher proportion of BSL 2 organisms, all at aGvHD onset. This tends to credit the thesis that antibiotic gram positive anaerobe cleansing lowers the risk of aGvHD. On the other hand, higher species diversity is associated with aGvHD controls, which does not plead for a high overall GM cleansing level.
T cells in aGvHD
Type 1 or Tc1/Th1 maturation is recognized as the dominant pattern in aGvHD and is linked to severe GI tract pathology. Th2 differentiation causes predominantly pathology in pulmonary, hepatic, and cutaneous tissues [Henden and Hill 2015]. Also, Treg/Th17 decrease has been associated with aGvHD and proposed as a biomarker [Liu et al. 2013, Han et al. 2018]. Regulatory T-cell therapy for GvHD has been suggested [Heinrichs et al. 2016, Elias and Rudensky 2019].
Mortality associated species
The H2S producer and oral commensal Atopobium parvulum is the top mortality associated species. It has been found involved in oral infections [Downes et al. 2001]. Importantly, it is a pre-dominant member of the adherent microbiome of pediatric IBD patients, associated with disease severity and promotes colitis in Il10-/- mice whilst eliciting a IL-17 response [Muehlbauer et al. 2013]. The probiotic bacterium Bifidobacterium bifidum induces Th17 differentiation [López et al. 2011]. Likewise, Prevotella species are associated with a Th17 response [Larsen 2017]. Prevotella melaninogenica in particular, elicits a IL-17B response through outer membrane vesicles [Yang et al. 2019].
Clostridium aldenense has been associated with a cytokine IL-7 response [Luna et al. 2017], i.e. generic development and maintenance of T cells. Likewise, Phascolarctobacterium faecium has been found to correlate positively with immune response, i.e. IFNγ+CD8+ T cells in a mouse model [Tanoue et al. 2019].
Acidaminococcus fermentans has been studied for its anaerobic glumate/glutamine fermentation. Glutamine has been shown to induce Treg cells in a mouse aGvHD model [Song et al. 2013] and hence its catabolism would work against the induction of a Treg response.
Streptococcus sanguinis induces a Th1 response in peripheral blood mononuclear cells [Yanagihori et al. 2006].
All in all, the mortality associated species are associated with immune activation and a Th17/Th1 response.
aGvHD and non-GvHD associated species
The position of the commensal Bacteroides thetaiotaomicron as an aGvHD associated species is intriguing. Although commensal, its mucolytic activity makes polysaccharides available for other organisms, including pathogens [Curtis et al. 2014].
The aGvHD associated species Bacteroides nordii elicits a macrophage activator INFγ response [Schirmer et al. 2016].
As stated above, Prevotella species are associated with a Th17 response, with a pro-inflammatory role for the aGvHD associated species Prevotella copri in rheumatoid arthritis [Scher et al. 2013].
The position of the human pathobiont Bacteroides vulgatus on the side of non-GvHD controls is a paradox. The bacterium activates NF-κB in a human gut epithelial cell line [Ó Cuív et al. 2017]. NF-κB is required for the development of both anti-inflammatory Treg and proinflammatory Th17 cells [Ruan and Chen 2012] and its systemic inhibition has been proposed as a strategy to prevent GvHD [Vodanovic-Jankovic et al. 2006].
The non-GvHD control associated Flavonifactor plautii has been shown to induce Treg cells in mice [Ogita et al. 2020] and, importantly, the white blood cell growth factor GM-CSF in humans [Luna et al. 2017]. The non-GvHD control associated Bacteroides uniformis induces IL-10 and Treg in PBMCs and LPMCs [Neff et al. 2016]. Lastly, the non-GvHD control associated Bifidobacterium longum elicits in majority IL-10 [López et al. 2011].
Modulating the allo-HSCT GM
From both our direct association and in silico allo-HSCT GM screening results, Bifidobacterium longum seems to present potential to modulate the allo-HSCT GM. To increase putative case associated species exclusion, B. longum could be combined with another probiotic species, Bacillus clausii, with which it is found associated in vivo more frequently than respective prevalences predict. B. clausii presents further interest in that it has been reported to increase IL-10 in allergic children [Ciprandi et al. 2004].
A selection of a high folate producer for B. longum could be privileged for increased Treg stimulus. Combination of B. longum with a bifidogenic prebiotic seems indicated, either choosing or avoiding the butyrigenic inulin-type fructans (ITF) or arabinoxylanoligosaccharides (AXOS) [Rivière et al. 2016], depending on the patient’s aGvHD status.
In vitro metronidazole resistance among Bifidobacteria has been reported [Charteris et al. 1998, Moubareck et al. 2005]. Bifidobacteria are only moderately susceptible to the anti-gram positive vancomycin [Lim et al. 1993, Charteris et al. 1998, Kheadr et al. 2007] and to the anti-anaerobe ceftazidime but susceptible to amoxicillin [Charteris et al. 1998]. Of note, a rifaximin resistant strain of B. longum W11 has been documented and is commercially available [Graziano et al. 2016]. Prophylactic use of rifaximin has been shown to lower IL-6 levels [Qayed et al. 2013], preserve urinary 3-IS levels and increase survival as compared to ciprofloxacin/metronidazole [Weber et al. 2016]. B. clausii has a wide spectrum of natural antibiotic resistance, including macrolides, betalactams, aminoglycosides, metronidazole, cefuroxime, ceftriaxone, cefotaxime and cephalosporin [Abbrescia et al. 2014].
Conclusions
From our results, it appears aGvDH risk is reduced by GM cleansing with targeted antibiotics. Auto-FMT is capable of accelerating the recovery of GM diversity and replacement of pathobionts selected by antibiotics. The Threg/Th17 balance could explain why we see certain species associated with aGvHD or TMR and others with non-GvHD controls. From this perspective, and because of in silico prediction of exclusion of aGvHD associated species, it seems of interest to supplement B. longum as a prophylactic, especially if antibiotics use is limited to vancomycin, metronidazole, ceftazidime and rifaximin. Prebiotics and vitamin B9 could further enhance the treatment, just like the association with another probiotic species, B. clausii.
Data Availability
Data and ananlysis scripts have been made available under Github.
Declarations
Conflict of interest
The authors have no conflict of interests related to this publication.
Authors’ contributions
Study design, data collection, data analysis and writing of the manuscript (ML); data analysis and writing of the manuscript (MD).
Acknowledgements
The authors received no financial support for this study. The authors acknowledge the contributions to the Short Read Archive made by the respective institutions and acknowledge scientific journals for enforcing this practice.