Type 1 diabetes, periodontal health, and a familial history of hyperlipidaemia is associated with oral microbiota in children: a cross-sectional study

Seventy-six children with T1D were recruited over a twelve month period at the Pediatric Diabetes Clinic at the Women’s and Children’s Hospital, Adelaide, Australia between 2018 and 2019, under ethics approval (Women’s and Children’s Health Network Human Research Ethics Committee; HREC/17/WCHN/165). Informed written consent was obtained by parents/guardians, or from the children themselves if they were 16 years or older. Children between 8 and 18 years of age with previously diagnosed T1D through detectable islet cell autoantibodies were consecutively recruited and underwent dental examination (Table 1). This age range was selected to include children in their mixed and permanent dentitions with at an appropriate level of dental development and compliance levels to allow for thorough periodontal examination. The exclusion criteria included a diagnosis of diabetes other than T1D, inadequate English skills to understand the information sheet, participants with a fever or infection, participants with diabetic ketosis, or those taking antibiotics at the time of dental examination. Parents confirmed a positive blood test and requiring management for those with a family history of hyperlipidaemia.

Information on familial, medical, and dental history were provided by the parents, guardians, or participants, which included current home oral hygiene practices (including frequency of brushing, type of toothpaste, and regular flossing; Additional file 1: Table S1). Dental and periodontal examination was performed by a single trained and calibrated dentist. Periodontal risk markers were measured on six teeth (fully erupted first permanent molars in each quadrant, the right maxillary central incisor and left mandibular central incisor), whereby each periodontal parameter was recorded on six points per tooth (mesio-buccal, buccal, disto-buccal, mesio-lingual, lingual, disto-lingual) by the same operator. Periodontal risk markers including plaque index, gingival index, bleeding on probing, and pocket depth, were evaluated and described further in [13]. For this study, pocket depth (PD) was used as the metric to characterise periodontal risk and is defined as no periodontal PD ≥ 3 mm or periodontal PD ≥ 3 mm. To account for periodontal PD being a potential confounding variable, periodontal risk marker status was separated into no periodontal risk markers (no periodontal pockets with depth ≥ 3 mm; individuals = 38) or periodontal risk markers (at least one periodontal pocket with depth ≥ 3 mm; individuals = 34), prior to assessing the association between the oral microbiota and a family history of hyperlipidaemia. This created four groups; (1) no periodontitis without a family history of hyperlipidaemia (n = 19); (2) no periodontitis with a family history of hyperlipidaemia (n = 19); (3) with periodontitis and without a family history of hyperlipidaemia (n = 23); (4) with periodontitis and with a family history of hyperlipidaemia (n = 11). Body mass index (BMI) z-score was calculated using the Children’s Hospital of Philadelphia Research Institute pediatric z-score calculator (https://​zscore.​research.​chop.​edu/​). Due to these samples being a convenience sample, sample size was dependent on the recruitment of cases over a twelve month period.

Gingival (plaque) samples were collected by a practitioner wearing a clean mask and gloves. More specifically, a swab was passed against the gingival margin between the tooth and gingiva from distal to medial on the buccal surface of the lower left first permanent molar. The gingival swab was collected at the time of dental examination and was immediately stored in sterile tubes at − 80 °C.

Samples were transported to a dedicated low-biomass microbiome laboratory at the University of Adelaide for DNA extraction (Qiagen Dneasy Powersoil kit) and 16S ribosomal RNA library preparation (see for [13] for detailed methods). DNA sequencing was completed using a 150 paired-end kit on an Illumina MiSeq at the Australian Cancer Research Foundation, Cancer Genomics Facility (Adelaide, Australia). As part of this process, extraction blank controls (EBCs) and no template controls (NTCs) were also processed to identify background laboratory contamination.

As the data for this study was publicly available, demultiplexed DNA sequences were downloaded from the QIITA repository (Study ID 13235). Data processing followed the same pipeline as [13]. Briefly, in QIIME2 ([14]; v2019.7), sequences were joined using vsearch [15] before undergoing quality assessment and denoising at 250 bp with Deblur [16]. A SEPP insertion tree [17] was created, and sequences were assigned using the SILVA database (v132; 16S rRNA gene 515–806). Using decontam [18], contaminant amplicon sequence variants (ASVs) were identified from EBCs and NTCs (prevalence threshold set to 0.6) and were removed from gingival samples. Samples with low sequencing depth (< 5000 sequences) or incomplete metadata, and ASVs with < 11 sequences per ASV were also removed.

Microbial diversity and statistical analyses were performed in QIIME2. At a rarefied depth of 5000 sequences, diversity (alpha diversity) was calculated using observed ASVs (richness) and Faith’s phylogenetic diversity [19] metrics. Significant associations between diversity and categorical metadata were tested using a Kruskal–Wallis [20] pairwise statistical test, whereby FDR adjusted p-values (q-values) were reported. Similarly, differences in bacterial composition diversity were calculated using unweighted UniFrac [21] and Bray–Curtis metrics at a rarefied depth of 5000 sequences. Both PERMANOVA [22] and adonis [23] tests were used to determine statistical differences between groups and the variation that a metadata category contributes to the composition, respectively. We also explored confounding factors between HbA1c, periodontal PD, and a family history of hyperlipidaemia using standard linear regression models in R (version 3.6.1) with the ‘lm’ function. The approached used to explore confounding relationships between these factors was to drop insignificant three- and two-way interactions in a stepwise approach until all significant interactions were identified.

Co-correlations between genera were determined by exporting biom tables from QIIME2 and importing them into MEGAN6 [24]. The taxon chart tool with the co-occurrence function was used, and equal numbers of samples were compared in each instance (n = 11 per group). In MEGAN6, ASV data was collapsed to the genus taxonomic level and visualised as co-occurrence plots using the Pearson correlation test with the following parameters: observed in 80% to 100% of samples and had an edge threshold of 80%.

Using decontam, 54 ASVs were identified as contaminants from EBCs and NTCs (Additional file 2: Table S2) and removed from the biological samples. From 76 samples, three individuals had incomplete metadata, while one sample had a low sequencing depth; these four samples were discarded from all downstream analyses. In the remaining 72 samples (Table 1), 905 ASVs were identified from 2,567,677 sequences (range sequences/sample = 5354–94,295; median = 33,920).

A total of 12 phyla were detected from 72 gingival samples. The most abundant phyla (> 2% mean relative abundance) were Proteobacteria (43.3%), Firmicutes (30.7%), Bacteroidetes (9.7%), Actinobacteria (8.3%), and Fusobacteria (6.3%) (Additional file 1: Fig. S1). From 131 total identified genera, 12 genera were highly abundant (> 2% relative abundance): Haemophilus (19.2%), Streptococcus (16.3%), Neisseria (10.3%), Veillonella (9.9%), Aggregatibacter (5.3%), Fusobacterium (3.8%), Actinobacillus (3.2%), Actinomyces (2.8%), Leptotrichia (2.5%), Corynebacterium (2.4%), Prevotella (2.0%), and Rothia (2.0%).

Using linear regression models, HbA1c, periodontal PD, and a family history of hyperlipidaemia were intially independent of one another and not interacting with one another (Additional file 1: Table S3). Subsequently, we removed any insignificant three-way and two-way interactions, and determined that glycated haemoglobin influenced the oral microbiota independently of periodontal PD and a family history of hyperlipidaemia (Additional file 1: Table S4), and that periodontal PD was confounded by a family history of hyperlipidaemia. As a result, two groups were created to account for periodontal risk factors when investigating a family history of hyperlipidaemia: no periodontal risk markers (no periodontal pockets with depth ≥ 3 mm; individuals = 38) or periodontal risk markers (at least one periodontal pocket with depth ≥ 3 mm; individuals = 34) for the following analyses. Overall, there were four groups assessed; (1) no periodontal risk markers without a family history of hyperlipidaemia (n = 19); (2) no periodontal risk markers with a family history of hyperlipidaemia (n = 19); (3) with periodontal risk markers and without a family history of hyperlipidaemia (n = 23); (4) with periodontal risk markers and with a family history of hyperlipidaemia (n = 11).

Children with no periodontal risk markers and a family history of hyperlipidaemia had significantly lower oral microbial diversity compared to those without a family history of hyperlipidaemia (Kruskal–Wallis; phylogenetic: H = 11.174, q = 0.001; richness: H = 8.616, q = 0.003; Fig. 1A, B). However, this was not in the case in children with periodontal risk markers; there were no significant differences between children with or without a family history of hyperlipidaemia (Faith’s PD: H = 0.009, q = 0.927; richness: H = 0.049, q = 0.825; Fig. 1C, D).

We measured compositional variation using unweighted UniFrac and found a significant difference in phylogenetic composition between children with no periodontal risk markers with or without a family history of hyperlipidaemia (unweighted UniFrac; PERMANOVA: pseudo-F = 4.281, q = 0.001; adonis: R2 = 0.117, q = 0.02; Fig. 2A). However, there were no significant differences for weighted non-phylogenetic composition between these groups (Bray–Curtis: pseudo-F = 1.164, q = 0.265; Fig. 2B). In children with high-risk periodontal markers, we saw no compositional differences between children with or without a family history of hyperlipidaemia (PERMANOVA; unweighted UniFrac: pseudo-F = 0.743, q = 0.725; Bray–Curtis: pseudo-F = 0.948, q = 0.511; Fig. 2C, D).

Using the Pearson metric, we identified distinct networks of co-occurring genera between children with no periodontal risk markers and a family history of hyperlipidaemia or children without a family history. In children with no periodontal risk markers and without a family history of hyperlipidaemia, four networks were observed (Additional file 1: Fig. S2A), simplified as: (1.) Campylobacter and Fusobacterium; (2.) Rothia and Corynebacterium; (3.) Alloprevotella, Porphyromonas, and Capnocytophaga; and (4.) Streptococcus, Gemella, and Bergeyella. In children with no periodontal risk markers and a family history of hyperlipidaemia, there were only two co-occurrence networks: (1.) Corynebacterium and Capnocytophaga; and (2.) Fusobacterium and Leptotrichia (Additional file 1: Fig. S2B).

In children with periodontal markers and no family history of hyperlipidaemia, a single network of six species commonly associated with periodontal disease (Prevotella, Porphyromonas, Fusobacterium, Campylobacter, Capnocytophaga, and Leptotrichia) was detected (Additional file 1: Fig. S2C). However, children with periodontal risk markers and a family history of hyperlipidaemia maintained four distinct networks: (1.) Actinomyces and Kingella; (2.) Cardiobacterium and Bergeyella; (3.) Rothia and Gemella; and (4.) Fusobacterium, Leptotrichia, Corynebacterium, and Prevotella (Additional file 1: Fig. S2D).