Disruption of the oral microbiota is associated with a higher risk of relapse after allogeneic hematopoietic stem cell transplantation

Intestinal microbiota (IM) diversity and composition regulates host immunity and affects outcomes after allogeneic stem cell transplantation (allo-HSCT). We evaluated if the oral mucosa microbiota (OM) could impact the outcomes in patients who underwent allo-HSCT. Samples from the oral mucosa of 30 patients were collected at three time points: before the conditioning regimen, at aplasia, and at engraftment. We analyzed the associations of OM diversity and composition with allo-HSCT outcomes. Lower OM diversity at preconditioning was associated with a higher risk of relapse at 3 years (68% versus 33%, respectively; P = 0.04). Dominance (relative abundance ≥ 30%) by a single genus at preconditioning was also associated with a higher risk of relapse (63% versus 36% at 3 years, respectively; P = 0.04), as well as worse progression-free survival (PFS; 19% versus 55%, respectively; P = 0.01), and overall survival (OS) at 3 years (38% versus 81%, respectively; P = 0.02). In our study we observed that OM dysbiosis is associated with a higher risk of relapse and worse survival after allo-HSCT.

Bioinformatics pipeline. Reads were demultiplexed, and primer sequences were removed using the MiSeq Reporter software. Within the QIIME 2 framework 38 , using experiment-specific adaptive error models 39 , forward and reverse sequences were filtered for quality and bimeras, denoised, and merged into consensus sequences with the DADA2 pipeline 40 , generating unique amplicon sequencing variants (ASVs). ASVs were further filtered for chimeric sequences using the SILVA database 41 and UCHIME 42 . ASVs were taxonomically assigned using SILVA database and VSEARCH tool 43 . Statistical analyses. For alpha diversity analyses, the samples were rarefied to 12,500 reads before calculating the Shannon index, Simpson index, or the number of observed ASVs as bacterial diversity measures with the QIIME 2 q2-diversity plugin. Alpha diversity across groups was compared with the Mann-Whitney U test. OM diversity was classified based on the median Shannon index diversity measure across the study population at a given collection time point. Patients were classified as high diversity (above the Shannon index median) and low diversity (below Shannon index median). Fisher's exact tests and two-sided Student's t-tests were used to evaluate the associations between alpha diversity status and categorical and numerical clinical parameters, respectively. The relative abundance of each taxa was calculated with the QIIME 2 q2-taxa plugin. The taxa shown on relative abundance longitudinal plots are all those showing dominance (relative abundance ≥ 30%) in at least one study sample or relative abundance ≥ 5% in at least 25% of study samples. Differentially abundant genera across transplantation phases were identified using ANCOM test, with relative differences represented by the log-transformed average relative abundance fold change between groups. ANCOM W represents the proportion of null hypotheses rejected when sub-testing the differential abundance of a genus normalized by the abundance of each one of the genera in the dataset. W > 0.7 was considered as statistically significant. The relative abundance of a genus was considered to increase during allo-HSCT for a given patient when the relative www.nature.com/scientificreports/ abundance at engraftment was greater than at preconditioning and the final relative abundance was ≥ 0.1%. The probabilities of progression-free survival (PFS) and OS were calculated using the Kaplan-Meier method and compared using log-rank tests. Cumulative incidence rates were calculated for aGVHD, cGVHD, NRM, and relapse/progression. Ninety-five percent confidence intervals (95% CIs) were estimated using the Greenwood formula. Adjusted probabilities for outcomes after transplantation were estimated using the Cox proportional hazards method (PFS and OS) and Fine-Gray risk regression model (aGVHD, cGVHD, NRM, and relapse/ progression). The association between OM parameters and HSCT outcome was investigated in the final model after adjusting for the DRI. First-order interactions between OM parameters and each variable of interest were examined. The results are presented as relative risks of failure (adverse prognostic factors versus good prognostic factors), with 95% CIs and two-tailed P values. To examine the association between genus presence at preconditioning and relapse, only genera present in 25-75% of samples were evaluated, where presence was defined as relative abundance ≥ 0.1%. R software (version 3.5.0) and RStudio (version 1.2.5033) were used for statistical analyses. The statistical package cmprsk was used to evaluate relapse across groups with transplant-related death as the competing risk.
Ethics approval and consent to participate. The study was approved by the local ethics committee, according to the Declaration of Helsinki.

Results
Patient characteristics. Between January 2016 and April 2018, 30 patients who underwent allo-HSCT for hematologic malignancies and had oral mucosa samples collected were included in this study. The most common underlying diseases were acute myeloid leukemia and acute lymphoblastic leukemia (60%). Conditioning regimens and intensity, graft source, T-cell depletion, and other clinical characteristics are listed in Table 1. The underlying disease, disease status, and OM diversity at preconditioning are presented in Table S1. The median follow-up time for survivors was 41  months.

Microbiota dynamics analyses.
In total, 5,920,836 high-quality bacterial assigned sequencing reads were analyzed, representing 1723 unique ASVs. Out of the 90 samples sequenced, nine were excluded from diversity analyses owing to an insufficient number of high-quality reads (< 12,500 reads per sample, as determined using alpha diversity rarefaction curves) after the read-filtering steps employed in the pipeline. Therefore, adequate preconditioning samples were available for 27 of the 30 patients included in this study. The intrasample bacterial diversity (Fig. 1A) and richness (Fig. S1) of OM samples decreased significantly during the clinical course. This drop in diversity is associated with changes in taxa relative abundance during the same period (Fig. S2). Notably, all patients showed bacterial dominance by a single genus after preconditioning. In Fig. 1B, we show three representative patients with major dominance (relative abundance > 80%) by a single genus (Stenotrophomonas, Rothia, and Veillonella, respectively) at engraftment.
For a broader assessment of the relative abundance changes from preconditioning to subsequent transplantation phases, we employed the ANCOM test at the genus level. We observed statistically significant variations in the abundance of both opportunistic pathogenic and commensal genera (Fig. S3). From preconditioning to aplasia, there was a significant increase in the abundance of the potentially pathogenic genera Enterococcus and Lactobacillus, which were even more increased in the engraftment phase in terms of relative abundance fold change from preconditioning. Staphylococcus and Mycoplasma were other potentially pathogenic genera increased at engraftment. Contrarily, there was a significant decrease in the abundance of the commensal genera Haemophilus (at aplasia) and Gemella (at engraftment).
A global increase of potentially pathogenic genera occurs during allo-HSCT. However, evaluating each patient individually, we noticed irregular changes in the relative abundance of those same genera from preconditioning to engraftment. An increase in the relative abundance of Enterococcus, Lactobacillus, Staphylococcus, and Mycoplasma was observed in 32%, 40%, 56%, and 68% of patients (Fig. S4). Patients who presented an increase in Enterococcus relative abundance had a higher incidence of cGVHD when compared with patients without the increase of relative abundance (P = 0.03). No other associations between the increase in the relative abundance of potentially pathogenic genera and allo-HSCT outcomes was observed (Table S2).
Other relevant clinical parameters, such as conditioning intensity, underlying disease, and graft source, were not significantly associated with the risk of relapse (Fig. S5, Table S5).
Genus presence and transplant outcomes. As the genus level represents the most specific taxonomic level that still provides reliable taxonomic classification for V3-V4 amplicons, to further evaluate the association between preconditioning OM and transplant outcomes, we analyzed whether any non-core genus (those present in 25-75% of samples) was associated with a higher risk of relapse. In this exploratory analysis (without adjustment for multiple comparisons), of the 18 genera that matched the selection criteria tested in a univariate analysis for relapse (Fig. 4A, Fig. S6), only Solobacterium was significantly associated with lower relapse risk (9% versus 56%, respectively; HR 0.23, 95% CI 0.05-0.94, P = 0.04; Fig. 4B), and this association remained significant Mann-Whitney U tests were used with the preconditioning collection as the reference for comparisons. The boxes highlight the median values and cover the 25th and 75th percentiles, with whiskers extending to the more extreme value within 1.5 times the length of the box. Outliers are represented explicitly. Asterisks represent statistical significance: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. (B) OM genera relative abundance composition across transplantation phases for three representative patients showing the decrease in bacterial diversity. Only genera showing relative abundance ≥ 30% in at least one study sample or relative abundance ≥ 5% in at least 25% of study samples are shown. P preconditioning, A aplasia, E engraftment.    Fig. 4C). However, after adjusting for multiple comparisons using the Bonferroni correction, because of the limited statistical power of this study, the univariate association between Solobacterium presence and lower relapse risk lost significance (P = 0.72). The relative abundance of Solobacterium at preconditioning per patient is depicted in Fig. S7 Antibiotic use. From one week before until the first day of the conditioning regimen, 4 (13%) patients received antibiotics. From the first day of the conditioning regimen until engraftment, 28 (93%) patients received antibiotics: 20 (67%) used cefepime, 16 (53%) meropenem, 14 (47%) vancomycin, and four (13%) piperacillintazobactam. The use of these antibiotics were not associated with the risk of relapse (Fig. S5). We could not analyze the association between the use of antibiotics before transplant (30 days before starting the conditioning regimen) and OM bacterial diversity because of the small number of patients who used antibiotics at that time point.

Discussion
In this single-center observational study, we prospectively collected samples from the oral mucosa of patients who underwent allo-HSCT. To the best of our knowledge, this is the first study to evaluate the possible impact of the OM using ASVs on allo-HSCT outcomes. ASVs, which are read sequences denoised to single-nucleotide resolution, is a more reproducible and comprehensive technique with higher sensitivity and specificity than operational taxonomic units (OTU) in analyzing microbiota [44][45][46] . The OTU can identify bacteria at the genus level, while ASVs allow to distinguish bacteria at the species level, which could explain discrepancies between our findings when compared to previous studies. We observed that patients who presented low OM diversity or dominance of any genus before conditioning had a significantly increased risk of relapse. The dominance of any genus was also associated with worse PFS  www.nature.com/scientificreports/ of a single genus may denote a deeper immune imbalance and could represent a more sensitive predictor of alloHSCT outcomes when compared with OM diversity. The OM has different niches in the same environment and is as diverse as the IM. Previous studies evaluating OM and allo-HSCT have shown conflicting results, likely because of the use of low-resolution techniques for microbiota analysis and the small sample sizes. In one case series, there were no changes in OM after allo-HSCT. The most common oral organisms, e.g., Streptococcus, Gemella, and Veillonella, remained relatively stable after transplant 34 . However, another study showed a reduction in alpha diversity after allo-HSCT when compared with the pretransplant OM 35 , and this reduction was more pronounced in patients who developed oral mucositis 47 . Besides, we did not find any direct correlation between the use of antibiotics after conditioning and transplant outcomes, as other studies have shown for IM diversity 23,24 .
Recently, IM has attracted attention as a potential predictive marker for allo-HSCT outcomes. Previous studies have shown that low IM diversity is associated with a higher risk of mortality, but not with the risk of relapse [25][26][27] , diverging from our findings.
Higher risk of aGVHD in patients with low IM diversity 48 and a higher risk of transplant-related mortality attributable to GVHD 26 were also reported. In the oral mucosa samples analyzed in the current study, low OM diversity was associated with an increased risk of relapse but did not change the risk of mortality, aGVHD or cGVHD.
The dominance of a specific bacterial group in IM, Eubacterium limosum, has also been shown to be related to relapse and disease progression. In our series, the dominance of any genus was associated with a higher risk of relapse.
As opposed to what has been observed for IM 27 , all dominant genera at preconditioning reported herein are commensal organisms. Thus, it is unlikely that they all have detrimental roles in the allo-HSCT setting, being more plausible that the presence of dominance by any genus is a proxy for low diversity/dysbiotic OM.
Furthermore, the presence of Solobacterium in the OM before conditioning seems to have a protective effect against relapse. S. moorei, the only species in the Solobacterium genus, is normally associated with halitosis 49,50 and endodontic infection 51,52 . However, in the allo-HSCT scenario, the lack of Solobacterium could be a marker of dysbiosis, pretransplant disease status, or previous treatments. Alternatively, this genus may also play a role as an immune mediator by producing hydrogen sulfide 49 , a metabolite associated with decreased oxidative stress and increased sensitivity to antibiotics 53 . Although, the low overall Solobacterium relative abundance even in patients where it was present makes the latter alternative more unlikely, this finding need to be validated in future studies.
A previous study analyzed the tongue microbiota in patients who underwent alloHSCT and compared it with community-dwelling adults. AlloHSCT patients have a lower tongue microbiota alpha diversity when compared to community adults. Moreover, the presence of Staphylococcus haemolyticus or Ralstonia pickettii was associated with a higher risk of mortality. Nevertheless, no relationship was observed between alpha diversity of the tongue microbiota and incidence of transplant complications 46 . A study of salivary microbiota showed a reduction in alpha diversity during the course of transplantation. Again, no correlation between salivary microbiota diversity and alloHSCT outcomes was found 54 . The discrepancies between these studies and our findings may be related to different sites of sample collections, and different distinct microbiome analysis techniques.
Our study had several limitations of a relatively small and heterogenous single-center transplant cohort. However, as observed in studies of IM, in our series, OM showed a significant correlation with relapse and may also provide valuable information on host-related microbial dysbiosis, providing a simple, reproducible technique for collection and analysis prior to transplantation.
In conclusion, in the current study, we focused on preconditioning samples in order to identify potential clinical effects of OM on allo-HSCT outcomes and observer that lower OM diversity was associated with a higher risk of relapse after allo-HSCT and dominance by a single genus was associated with a higher risk of relapse and worse survival after allo-HSCT.
Prospective trials and validation cohorts are needed to confirm these findings and to test whether early interventions to correct OM dysbiosis or more aggressive strategies to prevent relapse in OM dysbiotic patients, such as early immunosuppression withdrawal, maintenance therapy, or prophylactic donor lymphocyte infusions, could improve the predicted adverse outcome.