Enrichment of sulphate-reducers and depletion of butyrate-producers may be hyperglycaemia signatures in the diabetic oral microbiome

ABSTRACT Objectives This study aimed to investigate oral microbial signatures associated with hyperglycaemia, by correlating the oral microbiome with three glycaemic markers. Potential association between clinical parameters and oral bacterial taxa that could be modulating the hyperglycaemic microbiome was also explored. Methods Twenty-three individuals diagnosed with type 2 Diabetes Mellitus (T2D) and presenting periodontitis were included, as well as 25 systemically and periodontally healthy ones. Fasting blood glucose, glycated haemoglobin, salivary glucose, periodontitis classification, caries experience and activity and salivary pH were evaluated. The V4 region of the 16S rRNA gene was amplified from total salivary DNA, and amplicons were sequenced (Illumina MiSeq). Results Hyperglycaemia was correlated with proportions of Treponema, Desulfobulbus, Phocaiecola and Saccharimonadaceae. Desulfobulbus was ubiquitous and the most enriched organism in T2D individuals (log2FC = 4). The Firmicutes/Bacteroidetes ratio was higher at alkali salivary pH than acidic pH. In the network analysis, Desulfobulbus was clustered in a negative association with caries-associated and butyrate-producing bacteria. Conclusion The salivary microbiome is shaped by systemic hyperglycaemia, as well as changes in the salivary pH, which may be linked to local hyperglycaemia. The enrichment of predictive biomarkers of gut dysbiosis in the salivary microbiome can reflect its capacity for impairment of hyperglycaemia.


Introduction
Several oral manifestations of Diabetes Mellitus (DM) 30 can be explained by the hyperglycaemia state that directly favours the enrichment of microbial pathogens, promoting damage of cellular function, and consequently, local inflammatory responses. This occurs due to interactions between the increased concentration of advanced glyca-35 tion end-products and the increased proinflammatory © cytokine levels. A well-established oral manifestation of DM is periodontitis, which may also impair the systemic glycaemia control [1,2]. A reduced salivary flow is also commonly observed in individuals suffering from this 40 metabolic disorder [3,4]. The oral health is deprecated, essentially when the glycaemic levels are uncontrolled [5,6]. The poor glycaemic control can make adults with type 2 DM (T2D) more prone to dental caries, although the reasons behind this association are not yet explained 45 [7]. A potential hypothesis is that the hyperglycaemia may increase the glucose levels in the saliva of patients, changing the oral microbial environment and promoting salivary acidification [8].
A growing body of literature recognizes the impor-50 tance of salivary glucose as a biomarker of blood glucose levels [9][10][11][12]. Salivary glucose may be accountable for reducing the pH of the oral cavity, since oral bacteria can use glucose as a substrate in fermentative pathways, releasing acids as final metabolites. If these changes in 55 the availability of metabolic substrates linger, the socalled 'dynamic stability stage' of the oral microbiome can be lost [13]. The acidification would hamper acidogenic bacterial growth, shifting the ecological balance of the microbiota [8,14]. Furthermore, individuals with 60 uncontrolled DM frequently present ketoacidosis increasing the ketone bodies (acetone, acetoacetic acid, and β-hydroxybutyric acid) in blood and urine [15], and probably in saliva. The potentially altered pH [4] can represent a selective pressure over the diabetic oral 65 microbiome. If the pH-balance of the microbial community is disrupted by severe environmental pressures, the microbiome may collapse into an 'acidogenic stage' ( © increase in the acidogenic microorganisms) that initiates dental caries or into an ''inflammatory stage" (increase of 70 inflammophilic anaerobic microorganisms) leading to periodontitis [13]. The impact of DM on the salivary microbial biodiversity has been investigated [14, [16][17][18][19]. Bacteroidota and Proteobacteria are enriched in the 75 salivary microbiome of DM patients, somehow reflecting the pattern seen in the gut microbiome [20], and suggesting a potential correlation of gut and oral microbiomes in diabetic subjects. Indeed, the imbalance observed in the gut microbiota might 80 be a main contributor of local and systemic diseases [18,21]. Microbial communities along mucosal surfaces throughout the digestive tract are hypothesized as risk factors for impaired glucose regulation. Since some gastric bacteria are introduced through the oral 85 cavity, it is possible that a decreased salivary pH due to hyperglycaemia may act as a filter to inhibit replenishment of gastric Bacteroidota, while more easily transmitting gastric Firmicutes [8]. This can also be explained by the communication through 90 secondary metabolites of different microbiota of the human body [22]. Nevertheless, it is not yet clear if dysbiosis in the oral microbiome is a typical feature of hyperglycaemia and a potential contributor to progression of hyperglycaemia itself. Microbial metabo-95 lites from the oral cavity could serve as crosstalk mediators between host and microbiome, impacting glucose metabolism. So far, the oral microbial signatures associated with hyperglycaemia, as well as its oral manifestations, are still undetermined. 100 The oral © dysbiosis-diabetes relationship is to be elucidated. A fundamental need is understanding how the oral microbiome shifts from homeostatic to dysbiotic condition, altering the oral health status of DM patients. Understanding this process would allow 105 more efficient treatments for oral manifestations of DM. This study aimed to investigate oral microbial signatures associated with systemic and local hyperglycaemia, by correlating the oral microbiome with three glycaemic biomarkers (glycate haemoglobin © 110 (HbA1c), fasting blood glucose © (FBG) © and salivary glucose). We also aimed to explore a potential association between biological markers and oral bacterial taxa that could be modulating the hyperglycaemic microbiome, such as the salivary pH, T2D and peri-115 odontitis diagnosis, and levels of dental caries.

Ethics
The study was approved by the Research Ethics Committee of the School of Health Sciences of the 120 University of Brasilia (process number 87962818.4.0000.0030) in accordance with the declaration of Helsinki. All patients signed a formal consent form and received basic dental treatment.
Healthy participants received oral hygiene instruction 125 and professional prophylaxis.

Study design, setting and participants
This study was nested in a cross-sectional study [23], and it was performed and reported following the STROBE checklist [24]. Eligible individuals 130 (>18 years old) were enrolled in the Diabetes Dental Clinic at the University Hospital of Brasilia (Federal District, Brazil). Patients were recruited from June 2018 to March 2020.
Individuals with and without a diagnosis of T2D 135 were included in order to ensure a broad range of glycaemic levels. Cases of T2D were defined through a previous medical diagnosis. All patients in this group were using hypoglycaemic medication, either insulin or another hypoglycaemic drug. Only indivi-140 duals diagnosed with any level of periodontitis were included in this group to assure its homogeneity (interdental clinical attachment loss © ≥3 mm detectable at © ≥2 non-adjacent teeth) [25]. Another group of systemically and periodontally healthy individuals 145 was included (named as no-T2D), which were selected among family members and other individuals under treatment at the university clinics. All individuals, either T2D or no-T2D, went through blood and saliva glucose levels measurements (as 150 described below). Individuals with type 1 DM were excluded, as well as those with severe systemic comorbidities, pregnant or puerperia, transplanted patients, individuals with a history of epilepsy, or with systemic conditions that may influence the phy-155 siology of the salivary gland, such as hypothyroidism, radiotherapy or chemotherapy treatment that preceded 3 months.
Based on a pilot study [26], a minimum number of 14 samples is required to detect a correlation of 0.7 160 with power of 80% in an alpha of 5% (Fisher's Z test) between bacteria taxa and clinical parameters. For a mean difference of 2 (standard deviation of 1.5 and 3.1) in the Firmicutes/Bacteriodota ratio between T2D and no-T2D microbiomes, a minimum of 48 165 samples is required, to which was added a loss rate of 10%, resulting in a sample size of 52 individuals.

Salivary sampling
Stimulated and passive salivary flow samplings were performed in the morning (8-10 am) to minimize the 170 effect of circadian rhythms. Individuals were asked to refrain from drinking, eating, and performing physical activities at least 2 h before salivary collection. © The salivary collection was carried out for 5 min of passive drooling. Upon collection, 500 μL of the saliva samples 175 were aliquoted into microtubes and pellets stored at −20°C until further DNA extraction and sequencing.
Stimulated saliva was also collected for 5 min, while participants chewed a rubber device (1 x 1 cm, free of flavor). They were tied to a piece of dental floss so that 180 there was no danger of swallowing by the patient during chewing.

Glycaemic markers
Fasting blood glucose (FBG) (hexokinase method; mg/dL) and glycated haemoglobin (HbA1c) 185 (turbidimetric inhibition immunoassay; %) tests were carried out © at the university's © partner-certified centre of diagnosis (Sabin labs, Brasilia -Distrito Federal, Brazil). Salivary glucose was measured from the stimulated saliva using the Labtest Glucose liquiform® 190 kit (Labtest Diagnóstica S.A -Minas Gerais, Brazil), according to the manufacturer's instructions with an adaptation for the saliva volumes, as follows: after centrifugation, 150 µL of the supernatant was added to 500 µL of the kit reagent 1 (phosphate buffer 30 195 © mmol/L, pH 7.5; phenol 1 mmol/L; glucose oxidase 12,500 U/L; peroxidase 800 U/L; 4-aminoantipyrine 290 mol/L; azide sodium 7.5 mmol/L; and surfactant). A glucose standard was added to the experiment. After homogenization and incubation at 37°C for 200 10 min, 250 µL of the reaction was transferred to the 96-wells plate, in duplicates, and read at 505 nm. Blood and glucose levels were analysed as continuous variables, and the salivary glucose was also categorised as high ( © ≥0.35 mg/dL) and low ( © -205 <0.35 mg/dL), according to © the data distribution.

Sucrose frequency intake
A 24-h diet recall was performed to determine the frequency of sucrose intake.

Periodontitis classification
210 All patients underwent periodontal examination and evolution of panoramic x-rays. The stage and extension of the periodontitis were then classified by the same examiner, with broad experience as a periodontist, based on the International Classification of Periodontal 215 Diseases [2017,25].

Dental caries detection
Dental caries examinations were performed by trained and calibrated dental students (Kappa > 0.7), as described elsewhere [23]. Briefly, the presence 220 of caries was observed and recorded by thorough dental examination under artificial light, in a supine position, using clinical mirrors, WHO probes, and tooth isolation with cotton rolls. After tooth cleaning and drying, the visual-tactile inspection was per-225 formed to record active and inactive coronal caries lesions, based on the Nyvad criteria [27]. Caries activity (the number of surfaces with either noncavitated or cavitated caries) and the traditional DMFS (WHO criteria; at the cavity level, representing 230 the past caries experience) were evaluated.

Salivary pH
The salivary pH was tested on the stimulated saliva using the pH-Fix® indicator strips (Macherey-Nagel GmbH & Co. KG-Düren, Germany). After 1 min of 235 immersion in the saliva, the result was compared to the standard table, as indicated by the manufacturer. The buffering capacity was used for adjustment in the multivariate analysis. It was measured from 1 mL of stimulated saliva; then 3 mL of 0.005 M hydrochloric 240 acid was added, and the pH was measured with an indicator strip after 2 min.

265
The amplicon sequence variants (ASVs) were generated through the DADA2 pipeline v.1.12.1 [28] in R version 3.6.1 [29]. Reads were trimmed in 15 nt on left © side, and the identified Phi-x sequences were removed. Datasets were filtered allowing 270 a maximum of two expected bases errors per read, N called bases were not permitted, and reads were truncated at 220 nt for the forward and 200 nt for the reverse fragments. Error rates were estimated using a training set of reads and 275 inferred to the whole dataset, and sequences were denoised. Denoised reads were © merged, and chimeras identified by method consensus were removed. Qualified sequence variants had an average length of 246 bp and were assigned using the 280 Silva v.138 database [30]. Before performing downstream analysis, ASVs assigned to Eukaryote, Chloroplast, and Mitochondria were removed using the Phyloseq package (version 1.34.0) [31]. The Spearman's correlation was performed to 285 determine the correlation between © the taxa and explanatory variables using the Microbiome R package (version 1.12.0) [32]. © The taxa presenting a mean relative abundance higher than 0.001% © and significant association (p < 0.01) to the vari-290 ables tested were plotted in a heatmap. The Canonical Correlation Analysis (CCA) was performed using the vegan R package (version 2.5-7) [33] and plotted using the ggrepel package (version 0.9.1) [34]. © The significance of correlation between 295 canonical axes and explanatory matrix was tested with 10,000 permutations. The alpha diversity was estimated for a dataset of sequence variants rarefied to 35,000 sequences per sample, by the rarecurve function from the vegan 300 R package (version 2.5-7) [33]. The Shannon's index, Chao1ʹs index © and the Pielou index of samples were determined using the Microbiome R package (version 1.12.0) [32], for univariated comparison between groups © and bivariate comparisons between 305 groups and pH ( © alkali -pH 8, © neutral -pH 7 © or © acidic -pH 6) or salivary glucose ( © ≥0.35 or © -<0.35 mg/dL). The square-root transformed relative abundances of sequence variants combined at the genus level (or the highest taxonomic level anno-310 tated) were used to build matrices of similarity based on the Bray-Curtis dissimilarity. The ordination distance was plotted in a non-metric multidimensional scaling (nMDS) using the Phyloseq package (version 1.34.0) [31]. 315 Microbial taxa with differential abundance between groups were identified by DESeq algorithm with the Benjamini-Hockenberg (BH) correction test. Results were obtained using DESeq2 package (version 1.30.1) [35]. 320 Mean and standard © deviations were calculated for clinical parameters. The relative abundances at different taxonomic levels were used to evaluate comparisons within and between groups regarding the salivary pH and salivary glucose. Pearson's cor-325 relation, Mann © -Whitney and Kruskal © -Wallis nonparametric tests were applied for data comparison using SPSS (SPSS Inc. version 26, Chicago, IL).
Network analysis was performed for © the modified centered log ratio (mclr) normalized data, using the 330 spring association method from the NetCoMi package (version 1.0.2) [36]. Differences into the taxa association between sample groups were tested with the cluster fast greedy method.

Clinical characteristics
Saliva samples were obtained from 52 individuals who underwent dental © examinations. Samples from four individuals were excluded from analysis due to missing data. From the remaining sample, 23 individuals had 340 a clinical diagnosis of T2D, from which 10 were insulin users and the remaining used other hyperglycemic medication. The same individuals had periodontitis: n = 6 had stage 4 generalized periodontitis, n = 11 had stage 4 localized periodontitis; n = 6 had stage 3 345 localized periodontitis. Their HbA1c and FBG levels varied © from 6 to 14.2% and 47 to © 310 mg/dL, respectively, confirming a great range of glycaemia levels. Twenty-five other individuals were systemically and periodontally healthy, all of them presenting HbA1c 350 lower than 6% and FBG lower than 120 mg/dL. Glycaemic markers, either from saliva or blood, were significantly higher in T2D individuals. However, the local hyperglycaemia varied more, and six subjects had T2D diagnosis but salivary glucose levels were below 355 0.35 mg/dL, while seven subjects had salivary glucose © >0.35 mg/dL and had no diagnosis of T2D (Table 1). Patients with T2D (average age = 58 ± 8) were slightly older than patients with no-T2D (average age = 43 ± 13) (p < 0.001). Besides their hyperglycaemic state, their 360 frequency of 24 h-sucrose intake was higher than for no-T2D, similar to their caries experience (DMFS). This pattern was not observed for active caries (active D-S component). The salivary pH and the salivary glucose had a weak negative correlation (r = −0.3; 365 p = 0.04).

Sequencing output
The dataset from saliva samples sequencing, after screening and optimization, resulted in 2,281 ASVs. Seventy-eight ASVs belonged to the Archaea domain 370 and 2,203 ASVs to the Bacteria domain. Archaea represented 0.01% of the reads, and 33 samples presented at least one taxon belonging to the Archaea domain. The overall salivary microbiota was composed © of 33 phyla, 61 classes, 130 orders, 194 families, 332 genera © and 407 375 different taxa annotations. A total of 47 samples were included in downstream analyses after quality checking.

Correlation of taxa with glycaemic markers and clinical parameters
There were 119 taxa significantly correlated (p < 0.05) 380 to at least one of the analysed clinical parameters, including the three glycaemic markers © and 38 out of 119 taxa with p-value < 0.01 (see Supplementary Table 1 for © correlation values). What stands out from this result is the positive correlation between both blood glucose 385 levels (FBG and HbA1c) with Treponema, Desulfobulbus © and Phocaicola. The salivary pH had the highest number of negatively correlated taxa: Actinobacillus, Haemophilus, Kingella, Mannheimia, Neisseriaceae, Prevotellaceae UCG.004, T2WK15B57, 390 and TM7a. Abiotrophia and Oceanivirga were positively correlated © while Desantisbacteria was negatively correlated with both caries variables (active caries extent and DMFS, representing caries activity and past caries experience, respectively), with strength of correlations 395 between 0.4 and 0.6 ( Figure 1(a)).

Diversity and relative abundances of the salivary microbiome
The diagnosis of T2D was used to compare diversity 410 and relative abundances, so as the salivary pH © that represented the clinical parameter with the highest significance in the CCA multivariate analysis. The salivary glucose was also tested, as the glycaemic marker that varied most independently of the T2D 415 diagnosis, and represented the local hyperglycaemia.
There was no difference for alpha-diversity in the salivary microbiome regarding the diagnosis of T2D (Figure 2 Table 2), although both groups presented differences 420 in clinical characteristics that should substantially shape the oral microbiome (age, hyperglycaemia, diagnosis of T2D, periodontitis, caries experience). The alpha-diversity was calculated for samples rarefied at 35,000 reads. Two samples (one with no-T2D 425 and another with T2D) were removed from the set of analysis due to lower counts of reads (Supplementary Figure 2). A borderline result showed higher diversity in individuals without a diagnosis of T2D when they had low salivary glucose (Figure 2(a)).
© Twenty-three out of 33 phyla presented very low abundance and prevalence. The average number of phyla was 22 in T2D and 25 in no-T2D samples.
445 Regarding the archaeal content, the Euryarchaeota phylum (which includes methanogenic organisms) was detected in only two samples, while Halobacterota, Chrenarchaeota © and Nanoarchaeota were more prevalent archaeal phyla, but representing no more than 450 0.001% of the total reads (Supplementary Figure 3). The 10 most abundant phyla presented no differences in their relative abundance regarding the diagnosis of T2D (Figure 2(c)). The increase of the salivary pH was followed by a clear reduction of Proteobacteria © and an 455 increase © in Firmicutes and Actinobacteriota. The relative abundance of Actinobacteriota was also affected by the salivary glucose (Figure 2(c)), which was confirmed when the taxa average differences were compared. The Actinobacteriota phylum showed higher abundance in 460 the group of individuals with salivary glucose © <0.35 mg/ dL than in the group with salivary glucose © ≥0.35 mg/dL (17.6 ± 6.8% vs. 13.1 ± 5.1%; p = 0.01) © )( Table 4, p = 0.08). The Firmicutes/Bacteroidota ratio significantly increased with the alkalinisation of the salivary 465 pH (p = 0.03). The opposite was observed for Proteobacteria that seemed to be increased in abundance through saliva acidification (pH 6; p = 0.003). Bacteroidota (p = 0.02), andSpirochaetota (p = 0.01) were in low abundance, butwere most likely affected 470 by the salivary pH. Regarding the genus level, it is worth to mention that Veillonella was enriched in the acidic saliva (p = 0.01). These estimations were performed based on non-parametric calculations, considering the small size of the saliva samples at acidic and alkali pHs 475 (Table 4).
As a complementary analysis, the average abun-505 dances of deliberately selected genera (usually linked to dental caries) were compared to test the hypothesis of the enrichment of typical acidogenic/acidophilic    taxa due to the hyperglycaemia state. No differences were observed for the five acid-related genera in T2D 510 and no-T2D samples (from 390 taxa from 47 samples) ( Figure 5(a), Supplementary Table 3). Interestingly, the same was observed for the proteolytic pathobiont taxa, except for Treponema, that was significantly more abundant in T2D samples 515 ( Figure 5(b)).

Network analysis
The microbial network © profiles of samples from individuals diagnosed with T2D and no-T2D considered 110 taxa that were prevalent in at least 10 out of 47 samples 520 (Figure 6). Distance of centralities (degree, betweenness, eigenvector, and closeness) were tested using the Jaccard index. Degree (p < 0.05), betweenness (p < 0.01), and eigenvector (p < 0.001), but not for closeness (p > 0.1) presented significant differences between groups 525 (Supplementary Table 4). Keystone taxa were not identified by the centrality values.
The node pairs were connected by the shortest path in T2D samples compared to those in no-T2D, indicating that the T2D microbiota taxa had better 530 interconnected clusters. This can be particularly relevant for taxa nodes associated with Acidovorax (sp82) and Acinetobacter (sp32) (dark red cluster in T2D) © or for nodes associated with Actinomyces (sp6), Granulicatella (sp7) © and Solobacterium (sp8) (red 535 cluster in T2D) © and for nodes associated with Eikenella (sp104) and Haemophilus (sp5) (orange cluster in T2D) (see Supplementary Table 5 for the © network taxa ID annotation).
In the T2D, Veillonella (sp9), Bifidobacteria (sp175), and Scardovia (sp108), amongst others (light blue cluster) are clustered in direct or indirect negative association with Desulfobulbus. This indicates an important 560 opposition to the caries-associated and butyrateproducing bacteria.
A central role of Veillonella (sp9) confirmed its importance in the T2D microbiome, when combined with the previously described higher relative abun-565 dances in acidic than alkali pHs. Veillonella appears as a connector node linking three clusters in positive © association with Dialister (sp210) (light blue cluster), with Haemophilus (sp5) (orange cluster) © and with Prevotella (sp11) (gold cluster). In the no-T2D  570 samples, we cannot see Veillonella as a connector node taxon, but it is clustered with direct or indirect positive association with Haemophilus (sp5), Dialister (sp210) © and Prevotella (sp11). It suggests Veillonella as a multifunctional taxon capable of maintaining the 575 bacterial network structure of the oral microbiota in T2D samples. Butyrivibrio (sp102) was differentially more abundant in no-T2D samples © and also seemed to have a role as cluster connector within the T2D samples. In no-T2D, it 580 was negatively linked to the Anaerovoracacea family XIII UCG-001 (sp468), while in T2D it comprised the same cluster of Prevotella (gold cluster) and positively associated with Atopobium (sp21) (red cluster), Novosphingobium (sp754) (dark red cluster), and 585 Catonella (sp122) (gold cluster). Another differentially abundant bacterium © Leptotrichia (sp20) was enriched in no-T2D, and in the network analysis was positively and directly associated with Stomatobaculum (sp38) in both groups. Differentially abundant bacteria in T2D, 590 Tannerella (sp90), was positively associated with Johnsonella (sp55) © and clustered with Selenomonas (sp52), Prevotellaceae YAB2003 group (sp49), Alloscardovia (sp136) © and Peptostreptococcales-Tissierellales W5053 (sp411) (blue cluster). Tannerella is 595 known as an abundant component in periodontitis sites, and it may reflect its importance as © a link in the network of T2D, where most individuals had stages 3/4 localised periodontitis.
Pondering the strongest positive links in the T2D, 600 some correlations may be highlighted: Pseudoramibacter (sp222) with Parascardovia (sp211) © and Bifidobacterium (sp175) with Scardovia (sp108). Pseudoramibacter represents a very common microorganism in root canals, while all others belong to the 605 family Bifidobacteriaceae. Meanwhile, strong positive association in no-T2D samples included the fermenter and nitrate-reducer Actinobacillus (sp44) and Haemophilus (sp5), a bacterial genus found in all oral cavity sites of patients. Also, in no-T2D samples, there 610 was a positive association of Desulfobulbus (sp147) and the infection-associated Peptostreptococcus (sp26).

Discussion
Some oral microbial signatures such as Desulfobulbus were associated with all hyperglycaemia indices 615 tested. Members of this genus were favoured by the hyperglycaemia state showing significant positive correlation with FBG and HbA1c levels. Desulfobulbus was ubiquitous and highly abundant in individuals with T2D. Moreover, Desulfobulbus seemed to be an 620 antagonist to caries-associated and butyrateproducing bacteria. The increase © in blood glucose levels was significantly linked to the increase in Treponema, Phocaiecola and Saccharimonadaceae species. On the other hand, Actinobacteriota may be negative association between Desulfobulbus and butyrate-producing bacteria from the Bifidobacteriaceae family in the diabetic oral microbiota. Locally, mem-685 bers of the Bifidobacteriaceae family are strongly associated with caries as their fermentation end products are mainly organic acids. These metabolic products can reduce the pH, leading to critical acidity levels for tooth demineralisation, indicating 690 a potential link to the increased prevalence of caries in diabetic individuals [7]. Organic acids can be converted into short-chain fatty acids (SCFAs) by butyrate-producing bacteria through cross-feeding interactions [45]. Systemically, the SCFAs including 695 butyrate serve as key mediators of microbial-host signalling and are linked to a better insulin response [46]. For instance, Bifidobacterium spp. have antiinflammatory properties and protect the epithelial barrier by reducing lipopolysaccharides and the tri-700 methylamine N-oxide (TMAO) influx into the blood [21]. Our results on the salivary microbiota are in line with the decreased butyrate-producing bacteria levels in the gut microbiome of DM individuals [47]. This is the case of the butyrate-producer Butyrivibrio, 705 belonging to the Clostridiales order, that was significantly depleted in the T2D samples and presented a central role in the network analysis. Changes in the SCFAs metabolism in the diabetic gut microbiome © are linked to the enrichment of Bacteriodota, which 710 was also found © to be enriched in our data. A biomarker of this event is the Firmicutes/ Bacteriodota ratio. It has been positively linked with blood glucose levels [46], and we showed its significant reduction in the salivary microbiota among indi-715 viduals with lower salivary pHs.
Other oral phyla were reduced in the hyperglycaemia state. Actinobacteriota was significantly lower in salivary glucose © >0.35 mg/dL, while Proteobacteria (Methyloversatillis and Brevundimonas genus) were 720 the most significantly depleted organisms in T2D samples. Many pathobionts are members of the Proteobacteria phylum, and a high proportion of such organisms may have a pro-inflammatory effect in diabetic subjects. This observed imbalance in the 725 microbial composition can be a result of the gut dysbiosis © and potentially impair the insulin resistance. Leptotrichia was likewise associated with samples without T2D © and is a representative of the core microbiome, present in almost all individuals [48,49], 730 as a bridge between early and late colonizers within oral biofilms. New studies investigating microbial functions are necessary to explain connections with Stomatobaculum, as observed in our network analysis, independently of the © sample group status. 735 The altered blood sugar status can disrupt homeostasis, providing a more profound change on the microbiota profile particularly when combined with periodontitis [19,50]. It is essential to point out that genera strongly associated with periodontitis, 740 Tannerella and Treponema, demonstrated a connection with increased pH, the diagnosis of T2D, and the blood glycaemic levels. TM7, Neisseriaceae [G-1] bacterium HMT-174 (F0058) © and Tannerella demonstrated a positive correspon-745 dence with HbA1c, FBG, and salivary glucose in the CCA multivariate analysis. Additionally, Tannerella had several links in the T2D-associated microbial network. Since all diabetic individuals included in the present study were also diagnosed with some 750 level of periodontitis, it was not possible to clarify if the higher level of some periodontal © -related taxa was influenced by the T2D condition or by periodontitis, although the periodontitis extent was included in the PERMANOVA multivariate test, showing no signifi-755 cant impact in the analysis. Despite this potential limitation, it is important to highlight both the results of the Spearman correlation, as well as of the canonical correlation analysis, demonstrating that all the glucose parameters profoundly impacted the salivary 760 microbiota changes. Hence, it is the glycaemic status rather than the T2D diagnosis that perhaps should be considered a biomarker related to salivary dysbioses.
Another obvious factor influencing the oral ecosystem of individuals with T2D was the significant 765 salivary dysfunctions, such as pH changes. Goodson et al. evaluated changes in abundance of some bacterial species in the saliva of adolescents with high concentrations of salivary glucose, showing that the higher the salivary © glucose, the lower the pH of the 770 saliva [8], which we confirmed here. As glucose is a well-known energy source for many oral bacteria, changes in its concentration would lead to reduced overall bacterial diversity, favouring acidic and acidogenic bacterial species. We showed an enrichment of 775 the Abiotrophia and Oceanivirga in the oral microbiota with the increase of active caries extent and DMFS. For instance, Oceanivirga has been found in pharyngeal infections [51], while Abiotrophia was enriched in adolescents from a community with 780 high caries prevalence when compared to the ones from a low caries prevalence community [52]. Meanwhile, Desantisbacteria was negatively correlated with both caries variables and significantly affected by the salivary pH. Indeed, the salivary pH 785 had the highest number of negatively correlated taxa and significantly changed the beta diversity and the CCA multivariable analysis. This confirmed the relevance of the pH changes in the diabetic microbiome, even though the results did not directly change the 790 proportion of the typical acidogenic microbiota regarding the diabetes status. Furthermore, there was a central role of Veillonella spp. in the bacterial network of the diabetic salivary microbiome (Figure 6 © ), and they were significantly enriched in 795 the acidic saliva ( Table 4). Members of this genus are linked to the classical Socransky's purple-complex [53], and their lactate metabolism might facilitate the pH neutralization in biofilms [54]. They have been already related to hyperglycaemia elsewhere [55] and 800 dental caries [56].
© These characteristics might explain the significant higher proportions of these organisms in lower pH environments. Furthermore, they are health-associated organisms in periodontal sites [57].
Our results confirmed the importance of analysing 805 not only the main taxa present but also the microorganisms in low abundance, as these may be impacted by clinical parameters. Current research on the salivary microbiome has mainly been restricted to the identification of the most abundant 810 microbiota associated with health or disease. We believe that this strategy could cause an incomplete misunderstanding of the ecology and environment as metabolic functions exerted by low-abundant microorganisms can be linked to the dysbiotic microhabi-815 tats in a sort of 'butterfly effect' [58]. This can be clearly observed by the inclusion of Desulfobulbus in the analysis, even at very low relative abundance.
Although representing a minority taxon, its ubiquity and association with clinical parameters were found 820 to be consistent. Besides, the network analysis indicated its important role in the microbiome, as discussed above.
The analysis comparing the diagnosis of T2D vs. no-T2D should be interpreted with caution, consid-825 ering confounding aspects affecting all differences between samples. Those factors represent additional selective pressure over the microbiota composition in the diabetic group. Also, in general, the dichotomization of individuals did not separate the ones with 830 controlled from uncontrolled glucose levels, those with long-term diagnostic of DM and use of hypoglycaemic drugs were not taken © into account either.
To overcome this issue, we performed several analyses without considering the diagnostic of T2D, 835 instead taking into account the glycaemic status using HbA1c, FBG and salivary glucose as continuous variables. Salivary glucose showed some influence in the salivary microbiome, and this trend should be further studied using a more sensitive test for salivary 840 glucose measurement. Although it is not possible to confirm that salivary glucose plays an important role in disturbing the microsystem, perhaps it favours microorganisms that influence the pH balance. In this case, the salivary pH would be indirectly influ-845 enced by the salivary hyperglycaemia, although it is not the microbiota typically acidogenic that © is enriched in the hyperglycaemic state. Other clinical parameters, not evaluated here, could also be involved in the imbalance of the diabetic micro-850 biome, such as smoking and adiposity. Future perspectives in this field include the development of a longitudinal study to confirm these associations, © and hence the potential of targeting the oral microbiome as an approach to detect and treat T2D. 855 In conclusion, the salivary microbiome was shaped by systemic hyperglycaemia, as well as changes in the salivary pH, which may be linked to local hyperglycaemia. Locally, these changes might be related to the oral manifestations of T2D, including their higher 860 caries experience. Systemically, the enrichment of predictive biomarkers of gut dysbiosis in the salivary microbiome can reflect its capacity of impairment of the hyperglycaemia. More than leading to local changes in the oral cavity, the oral microbiome may 865 harbour important biomarkers for the early diagnosis of T2D due to the enrichment of sulphate-reducers and depletion of butyrate-producers. In the context of the integrated hypothesis of caries and periodontal diseases [13], due to the link of sugar-driven hyper-870 glycaemia and inflammation in periodontal tissues, there is a potential to control caries and periodontal diseases by stabilization of blood sugar levels.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Q2
Author's contributions 890 CPVL: Contributed to conception, design, data acquisition and interpretation, drafted and critically revised the manuscript; DCG: Contributed to conception, design, data acquisition and interpretation, drafted and critically revised the manuscript; MCMG: Contributed to conception, 895 design, data acquisition and interpretation, and critically revised the manuscript; LPS: Contributed to conception, design, data acquisition and interpretation, and critically revised the manuscript; PCK: Contributed to data acquisition and critically revised the manuscript; TD: Contributed 900 to conception, design, data acquisition and interpretation, and critically revised the manuscript; LGAB: Contributed to conception, design, data acquisition and interpretation, performed statistical analysis, drafted and critically revised the manuscript; NDT: Contributed to conception, design, 905 data acquisition and interpretation, performed statistical analysis, main supervision, drafted and critically revised the manuscript.