Abstract Oral mucositis (OM) is a common debilitating dose-limiting toxicity of cancer treatment, including hematopoietic stem cell transplantation (HSCT). We hypothesized that the oral microbiome is disturbed during allogeneic HSCT, partially accounting for the variability in OM severity. Using 16S ribosomal RNA gene sequence analysis, metabolomic profiling, and computational methods, we characterized the behavior of the salivary microbiome and metabolome of 184 patients pre- and post-HSCT. Transplantation was associated with a decrease in oral α diversity in all patients. In contrast to the gut microbiome, an association with overall survival was not detected. Among 135 patients given methotrexate for graft-versus-host disease prophylaxis pre-HSCT, Kingella and Atopobium abundance correlated with future development of severe OM. Posttransplant, Methylobacterium species were significantly enriched in patients with severe OM. Moreover, the oral microbiome and metabolome of severe OM patients underwent distinct changes post-HSCT, compared with patients with no or mild OM. Changes in specific metabolites were well explained by microbial composition, and the common metabolic pathway was the polyamines pathway, which is essential for epithelial homeostasis. Together, our findings suggest that salivary microbial composition and metabolites are associated with the development of OM, offering new insights on pathophysiology and potential avenues of intervention.

RS and AE, and OK and AN, had an equal contribution Background: Oral mucositis (OM) is a debilitating complication of allogeneic hematopoietic stem cell transplantation (HSCT) propagated by an inflammatory response. While certain patient and transplantation-related characteristics increase the risk of OM, there is marked variability in OM severity across recipients even of the same treatments. Given the substantial perturbation of the gut microbiome following transplantation and its role in facilitating immune and metabolic functions, we hypothesized that the oral microbiome is also disturbed during transplantation and may explain population variation in OM severity. Methods: In this single-center observational study, we prospectively collected weakly saliva samples from adult patients undergoing allogeneic HSCT and 20 healthy control volunteers. A total of 625 saliva specimens were collected starting from 7 days before HSCT to 34 days post-transplant. Patients gave a median of 4 samples (range 1-7). Samples underwent 16S rRNA gene sequencing (V4 region) on the Illumina platform. Pre-transplantation samples were defined as those collected between days -7 and -1. Sixty pre- and post-HSCT samples underwent mass spectrometry-based metabolomic profiling (Metabolon, Durham, NC). For all patients, OM was graded according to the Common Terminology Criteria for Adverse Events (CTCAE) v4.0, with grade 3-4 defined as severe mucositis. Results: In this largest analysis of the oral microbiome in transplant recipients, 184 patients were included. Transplantation indications varied, the leading being AML (47%). Myeloablative conditioning regimens were used in 70% of patients. Methotrexate (MTX) was given as graft-versus-host disease prophylaxis in 74% of patients. Median time to OM development was 7 days (IQR 5-9); 58% and 43% of patients developed grade 2-4 and grade 3-4 OM, respectively. Pre-transplantation oral α-diversity, as measured by the Shannon Index, was similar to that of healthy controls (p=0.460), but later decreased, reaching a trough on day 12 followed by an increase to levels that were lower than baseline at day 34 (Fig. 1A). Indeed, there was a 24% reduction in peri-engraftment α-diversity relative to pre-transplant (p<0.001). Among patients with grade 2-4 OM, the reduction in α-diversity over time was more pronounced compared to transplant-recipients had no or only mild OM (grade 0-1) (Fig. 1B). Since MTX is an established risk factor for OM and is known to induce microbiota alterations in the gut, we restricted the analysis hereon to the 135 patients exposed to MTX. Interestingly, using linear discriminant analysis effect size (LEfSe), Rothia, Kingella, and Atopobium were enriched in pre-transplant samples of patients who went on to develop severe OM. In a pair-wise analysis of samples collected on days -7 to -1 and 7 to 13, the decrease of α-diversity and increase in β-diversity was more pronounced in patients with grade 3-4 OM compared to those with 0-1 OM (Fig. 1C). In samples collected between days 7 to 13, saliva communities of patients with severe mucositis were more likely to harbor Methylobacterium while those in patients with grade 0-1 had a higher abundance of Treponema and the genus TG5 (Fig. 1D). Finally, metabolic profiling of saliva collected pre and post HSCT showed marked changes before and after transplantation in patients developing mucositis (Figure 1E). Microbiome-associated metabolites enriched in patients developing mucositis included histidine, phenylalanine, tyrosine, and tryptophan. Conclusions: We demonstrate that the oral microbiota is disrupted during allogeneic HSCT. Microbiota injury is more profound in patients who developed oral mucositis. Our findings indicate that there may be a distinct microbiological and metabolic signature in patients with mucositis, with some changes preceding the clinical phenotype. These may serve as potential intervention points to treat and prevent mucositis. Figure 1 Disclosures No relevant conflicts of interest to declare.

No abstract

Survival analysis (SA) prediction involves the prediction of the time until an event of interest occurs (TTE), based on input attributes. The main challenge of SA is instances where the event is not observed (censored). Censoring can represent an alternative event (e.g., death) or missing data. Most SA prediction methods suffer from multiple drawbacks that limit the usage of advanced machine learning methods: A) Simplistic models, B) Ignoring the input of the censored samples, C) No separation between the model and the loss function, and D) Typical small datasets and high input dimensions. We propose a loss function, denoted suRvival Analysis lefT barrIer lOss (RATIO), that explicitly incorporates the censored samples input in the prediction, but still accounts for the difference between censored and uncensored samples. RATIO can be incorporated with any prediction model. We further propose a new data augmentation (DA) method based on the TTE of uncensored samples and the input of censored samples. We show that RATIO drastically improves the precision and reduces the bias of SA prediction, and the DA allows for the inclusion of high-dimension data in SA methods even with a small number of uncensored samples.