Dental caries as a risk factor for bacterial blood stream infection (BSI) in children undergoing hematopoietic cell transplantation (HCT)

Subject enrollment and sample collection

Study patients: The University of California San Francisco granted approval to carry out this study within its facilities under IRB protocol # 15-18297. Forty-one of 43 pediatric HCT recipients, age 8 months to 25 years, treated in the University of California, San Francisco (UCSF) Benioff Children’s Hospital and approached for participation in the study, were enrolled between the years 2016 and 2019. Inclusion criteria were HCT recipients who could cooperate with study procedures, and parents and patients willing to participate and sign informed consent and assent forms. Exclusion criteria were patients or parents unable to understand or participate in study procedures, or allergies to multiple hygiene and cosmetic products. Written consents by parents and assent by children were obtained for all subjects.

Dental caries risk was determined by retrospective examination of dental clearance records. Past caries experience is a robust predictor of the disease (Philip, Suneja & Walsh, 2018), and we classified participants with dental clearances stating “clear/no caries detected/minimal decay”, or extractions of third molars with no other mention of dental caries or restorations, as low caries risk (LCR). Subjects with dental clearances stating “treatment for multiple dental caries” or indications of untreated decay 2 months prior to admission for transplant, were classified as high caries risk (HCR).

All patients were administered prophylactic levofloxacin from Day-2 until neutrophil recovery (Alexander et al., 2018). Mucositis was assessed from the medical records with daily assessments by nurses and physicians recorded according to the WHO classification standards. Bacterial blood stream infections (BSI) were defined as any positive bacterial culture of (central line) blood samples during the transplantation hospital stay.

Oral samples were collected by using a cotton swab to collect saliva and plaque, by rubbing the cotton swab across the tooth surfaces and under the tongue. All samples were stored at −80 °C until they were submitted for 16S rRNA sequencing.

16S rRNA sequencing and bioinformatics

Oral samples were taken from the entire cohort and sent to uBiome for sequencing. Thirty-one fastq files were sent back to us from uBiome.

Samples were processed and sequenced by uBiome (San Francisco, CA, USA) as reported by Almonacid et al. (2017). Briefly, the V4 variable region of the 16S rRNA genes was amplified by PCR using universal primers (515F: GTGCCAGCMGCCGCGGTAA and 806R: GGACTACHVGGGTWTCTAAT). The primers contained Illumina tags and barcodes. PCR products were pooled, column-purified, and size-selected through microfluidic DNA fractionation. Sequencing was performed in a pair-end modality on the Illumina NextSeq 500 platform rendering 2 × 150 bp pair-end sequences. The sequences obtained from uBiome included primers within the 150 bp sequence, and when the primers were removed, the remaining sequences were up to 125 bp in length. With this length of sequence, there were not overlapping sequences greater than 10 base pair, which is not ideal and therefore lead us to use only forward reads for our analysis.

Bioinformatic analyses were done using QIIME2 (version 2021.4) and R programming language (version 4.1.2), on forward reads of the sequencing data. Reads containing more than two expected errors and more than eight consecutive same nucleotide repeats were discarded, and the raw sequences were then truncated to 125 pb, denoised, chimera filtered, clustered into sequence variants. The taxonomy was assigned to operational taxonomic units (OTUs; 97% clustering) using the HOMD (Human Oral Microbiome Database version15.2) database with QIIME2 pipeline default settings. After analyzing rarefaction curves and the sampling depth that would retain the most feature in QIIME pipeline we have left eight samples for analyses. Feature table, rooted tree and taxonomy files were then imported into R for diversity, PCoA and differential expression analysis. Phyloseq package (version 1.30.0) was used for diversity and PCOA analyses. We further evaluated the differentially expressed taxa with “The Linear Discriminative Analysis (LDA) Effect Size (LEfSe) program of the Galaxy environment”.

Statistical analysis

Analysis of BSI as the dependent or the outcome variable of interest, was constructed as a ‘yes/no’, and was compared in the two groups using a Fisher’s Exact test. A correlation matrix was done comparing BSI, caries risk, age, gender, HCT-type (autologous or allogenic), cell type (peripheral or bone marrow), length of time for mucositis, and highest mucositis grade. A multiple regression analysis was done to further investigate the effect of age and mucositis, on caries risk and BSI. The specific logistic regression model fitted to the data was: Logit (BSI) = $b_0+b_1{X}_1+b_2{X}_2+b_3{X}_3$_. All data analyses were conducted using R version 4.1.2. Effect Size (LEfSe) was performed using the Galaxy web application using default settings. Statistical significance was defined as p ≤ 0.05.

Demographics

In our study group, 27 of 41 subjects (65.9%) were classified as LCR and 34.1% were HCR. There were no significant differences (p < 0.05) between the groups in relative to age, gender, diagnosis, stem cell source, or maximum mucositis grade (Table 1).

Incidence of BSI

All BSI incidences were between post-HCT and prior to ANC (absolute neutrophil count) recovery/engraftment. Among them 57% were allogeneic HCT recipients (see Table 2). High caries risk was significantly associated with subsequent BSI during transplantation (p = 0.035) (Fig. 1). Correlation analysis of other possible risk factors for BSI, showed only BSI to be significantly positively correlated with caries risk, (p < 0.05) (Fig. 2). Multiple regression analysis of the association of BSI with age, caries risk, and mucositis grade, showed an even more significant association between BSI and caries risk (p = 0.02) (Table 3).

Microbiome analysis

We compared the oral microbiome of a combination of saliva and plaque collected from four HCR and four LCR subjects following pretransplant conditioning (see Table 4). 16S rRNA sequencing yielded mean number of 88,610.875 reads per sample (range 40,520.0–187,555.0) after quality control processes. Raw sequencing files were deposited in GEO with submission number: GSE199991. Clustering was carried out at 97% identity to yield 478 distinct OTUs with a length of 125 bp. The HOMD framework allowed 98% of the OTUs to be assigned to uniquely identifiable genus-level groups. 454 OTUs classified as bacteria remained after filtering, and were assigned as Eukaryotes, Archaea, chloroplasts, and mitochondria. The most abundant phylum, family, and genus, across all samples at >1% abundance, are shown in Fig. 3, and summarized in Table 5, which shows the most abundant top four taxa across all samples.

Following conditioning, the alpha diversity in the LCR group was significantly greater than in the HCR low caries risk group (p < 0.05) (Fig. 4A). Beta diversity metrics showed no clear clustering by community composition by principal coordinate analysis (Figs. 4B–4D).

We next performed LEfSe and DESeq2 analyses to identify individual microbial taxa that differed between groups. Both LEfSe and DESeq2 were performed from phylum to species level (Fig. S2). For the comparison HCR vs LCR, in DESeq2 analysis, 22 taxa were enriched in the LCR group, while eight taxa were enriched in the HCR group (Fig. S2). LEfSe showed 56 taxa enriched in the LCR group while two taxa were enriched in HCR group (Fig. 5, Fig. S3.4). Fifteen taxa were enriched in the LCR group, order Lactobacillales and class Bacilli were the taxa significantly enriched in HCR caries group in both analysis (Table 6).

While differences were only significant at the order and class level, relative abundances of Streptococcus and Lactobacillus at the genus level showed relatively high levels of Streptococcus in 3 of the 4 HCR subjects, and Lactobacillus in the 4^th HCR subject. Genus Streptococcus and Lactobacillus were at lower levels or undetectable in the LCR subjects (see Fig. 6).