The experimental population consisted of obese women from Croatia aged 20 to 74 (N=70, median (age=45 (37, 55; 25th and 75th percentile). All participants were patients of the Croatian Obesity Treatment Referral Centre of the Division of Endocrinology, University Hospital Centre Zagreb. We focused only on obese women with BMI≥30 kg/m², who were subdivided into three classes according to BMI range (13). Association of each BMI class with hypertension classification, diabetes/prediabetes diagnosis and the amount of micronutrients in blood were observed. Fig. 1 shows the study design. In a priori power analysis, t-test was used and minimum sample size using G*Power v. 3.1.9.7 software (Heinrich Heine University, Duesseldorf, Germany; http://www.gpower.hhu.de/) was calculated (14). Accordingly, the minimum sample size with 0.80 prespecified effect size (d) to detect a difference of 0.8 standard deviation, 0.05 significance level (α) and 0.80 power level (1-β) was calculated to be 21 per group. Considering that, we decided to include 70 voluntary participants in this study as an appropriate sample size to have a possible significant difference between the two or three groups at the end, depending on the examined parameter.

This study included only females who were not taking any forms of supplements that contained vitamin D, B₁₂ or folic acid. Furthermore, exclusion criteria were pituitary and/or adrenal disease, untreated thyroid disease, prior myocardial infarction, stroke, transient ischemic attack or oncological disease. All analyses were performed during the first visit of obese patients to the Croatian Obesity Treatment Referral Centre (University Hospital Centre Zagreb, Croatia) without prior education in proper and balanced nutrition and controlled diet. Therefore, the diet of the patients was without accurate data on the composition and daily intake of calories. We included only female patients for several reasons: the Croatian Obesity Treatment Referral Centre has a much higher prevalence of obese female than male patients; gut microbiota differences between men and women, which can be influenced by the grade of obesity (15); the observed divergence could play a dominant role in defining gender differences in the incidence of metabolic and intestinal inflammatory diseases; studies in humans have produced conflicting results due to high inter-individual heterogeneity in age, diet and hormonal factors, and the largely unexplored influence of gender (16). Total body mass (kg), fat (%), fat mass (kg), muscle (%) and BMI (kg/m²) of the study participants were measured with the Tanita body fat scale (Tanita, Tokyo, Japan) using bioelectrical impedance analysis at the University Hospital Centre Zagreb.

The 25(OH)D values in serum below 50 nmol/L represented a deficiency, between 50.1 and 75 nmol/L were defined as an insufficiency, and anything above 75 nmol/L was considered as a sufficient level of serum 25(OH)D in a human adult (17). Folic acid values between 13.5 and 45.3 nmol/L are considered normal range and above 45.3 nmol/L are elevated folic acid concentrations. When folic acid concentrations in the blood are less than 6.8 nmol/L, it is considered a folic acid deficiency. Insufficient folic acid concentrations are between 6.8 and 13.4 nmol/L (18). It has been well-established that people with vitamin B₁₂ concentrations in blood between 200 and 350 pg/mL (147.6–258.3 pmol/L) have clear vitamin deficiency symptoms (19). Thus, in this study a cut-off value was 250 pmol/L. Fig. 1 shows the grouping of obese women according to vitamin concentrations.

Classification of obese women according to obesity-related disease was made using the following criteria: only diabetes/prediabetes diagnosis, only hypertension diagnosis, and without diseases (control group; Fig. 1).

For analytical procedures, saliva and blood samples of obese female patients were taken at the Division of Endocrinology, University Hospital Centre Zagreb, Croatia. Venous blood sampling (5 mL) was performed in VACUETTE^® tube CAT Serum Separator Clot Activator (Greiner Bio-One GmbH, Kremsmünster, Austria). The 25-hydroxy vitamin D, B₁₂ and folic acid concentrations were determined in the serum of fresh pooled blood samples. Furthermore, after rinsing the mouth with 15 mL of sterile saline solution to prevent the effect of previously eaten food on the microorganisms in the oral cavity, 2 mL of saliva samples were collected using sterilized centrifuge tubes (TPP, Trasadingen, Switzerland) and stored at −20 °C (20).

The isolation and identification procedures were previously described by Huđek et al. (20). Original saliva and a 1:1 dilution of saliva (0.5 mL) in phosphate buffer (pH=7.2) were inoculated on nutrient agar plates (de Man-Rogosa-Sharpe and Luria-Bertani). Furthermore, samples were incubated for two days under aerobic conditions in a thermostat at 37 °C. Microflex LT matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectrometer (Bruker Daltonik, Bremen, Germany) was used for the identification of colonies with different phenotypes. MALDI Biotyper 3.0 software package (Bruker Daltonik) with standard settings was used to process the recorded mass spectra.

In this work we isolated 118 different species of bacteria (5 different phyla: Firmicutes, Actinobacteria, Proteobacteria, Bacteroidetes and Fusobacteria) and 13 different species of fungi (2 different phyla: Ascomycota and Basidiomycota), 34 potentially pathogenic and 22 potentially probiotic bacteria (Table 1). The classification of probiotic bacteria was adapted according to Fijan (21) and pathogens according to Werth (22). For the statistical analyses, ’microbiota function’ presents the number of different species that fit into each of three categories – probiotic, pathogen or none. Bacteria were identified as binary variables (present/not present).

The laboratory where all analyses were performed has HAA accreditation (Croatian Accreditation Agency). The Roche Diagnostics total vitamin D assay as competitive electrochemiluminescence protein binding assay (ECLIA) was used for the quantitative determination of 25(OH)D, vitamin B₁₂ and folic acid in human serum samples (23, 24). Further, the ECLIA assay was used for the in vitro quantitative determination of vitamins on a Cobas 6000cee analyzer (Roche Diagnostics, Mannheim, Germany). The method used for serum 25(OH)D measurement was verified with an assay validated by the Vitamin D External Quality Assessment Scheme (DEQAS; http://www.deqas.org/) (25). The validation of methods used for folic acid and vitamin B₁₂ measurement was performed by DGKL (The German Society for Clinical Chemistry and Laboratory Medicine e.V.) at the Reference Institute for Bioanalytics in Bonn.

In this study, BMI (kg/m²) values positively correlated with a fat mass fraction (%), fat mass (kg) and age, while a negative correlation was demonstrated with serum 25(OH)D concentrations and muscle mass fractions (%) (Table 2 (27)). The same correlations among BMI, fat mass fraction (%), fat mass and muscle mass fraction (%) have been confirmed in other worldwide studies with obese individuals (28, 29). A statistically significant positive correlation was found between serum 25(OH)D and folic acid concentrations in the tested obese women (Table 2). According to a recent meta-analysis, the prevalence of vitamin D deficiency is 35% higher in people with obesity than in people with normal body mass (30). Through a complex chemical process, it is possible that excess body fat requires a surplus of vitamin D, which diminishes the amount available for other processes. Gominak (31) showed that vitamin D deficiency altered the intestinal microbiome, reducing vitamin B production in the gut. As well as for vitamin D, Kerns et al. (32) have confirmed that obese individuals have a deficiency in vitamin B₉ at a very high percentage (16-29%). Obese women in Croatia follow the worldwide trend of a correlation between higher obesity and lower 25(OH)D concentrations in blood (33).

In this study, we showed a statistically significant positive correlation between the BMI values and incidence of hypertension and diabetes/prediabetes diagnosis (Table 2). By dividing the tested obese women into 3 categories according to obesity-related diseases, a statistically significant difference in BMI values was determined between the obese women without any disease and the group of women with both hypertension and prediabetes/diabetes diagnosis (p<0.01; Fig. 2). As in our study, results from two national surveys have shown that the prevalence of hypertension and diabetes occurred across all ranges of BMI and increased with higher BMI (34). In vitamin-D-deficient patients with a diabetes diagnosis, supplementation of vitamin D increased serum 25-(OH)D, sirtuin 1, irisin and glycosylated haemoglobin concentrations, and these improvements may reduce insulin resistance. Furthermore, vitamin D can increase transport of glucose in the gut and insulin receptor gene expression in β-cells, while intestinal calcium absorption may serve as a stimulus for insulin release. Vitamin D deficiency, obesity and oxidative DNA damage are significant predictors of genomic instability (34). Vitamin D can control gene expression via genomic and epigenomic mechanisms, and that could be a reason why it has such wide-ranging non-skeletal health benefits (35). Epigenetic effect of 25(OH)D is linked to histone modifications, mainly acetylation. However, there is a study that showed that serum 25(OH)D could change the methylation ratio of the CYP2R1 gene, which is a potential candidate for predicting serum 25(OH)D variation (36).

There was a positive correlation between the number of oral Proteobacteria and Firmicutes bacteria (Spearman’s rho CC=0.284, p=0.017). Zeigler et al. (12) also reported bacteria from Proteobacteria phylum and individual Neisseria mucosa to be present in sixfold higher amounts in the saliva of obese adolescent subjects. Furthermore, Yang et al. (37) showed that Proteobacteria and Firmicutes bacteria were significantly associated with an increased obesity prevalence in obese individuals from the Southern Community Cohort Study, USA.

What should be emphasized first is that there is a positive correlation between the incidence of hypertension or prediabetes/diabetes and pathogenic bacteria in the oral cavity in obese women from Croatia (Table 2). In obese women with diabetes/prediabetes, there is a greater variety of fungi (number of different species) of the genus Candida than all 13 isolated fungi in this study (Mann-Whitney U=462.00, p=0.04). The identification results of the oral microbiota showed the predominance of Candida sp. in the Ascomycota phylum. The higher BMI values (if we compare only class 1 and class 3 obesity levels), depending on the presence of diabetes/prediabetes, positively affected the number of Ascomycota in the oral cavity (factorial ANOVA, F=4.091, p=0.021). Furthermore, a statistically significant difference in the number of Ascomycota among different obesity levels was seen only in the diabetes/prediabetes group (Tukey’s HSD post hoc multiple comparison for diabetes/prediabetes group: comparison between class 1 and class 2 (class 3) obesity levels, p=0.004; p=0.023). The predominance of Candida sp. in the saliva of women with diabetes/prediabetes can be explained by the fact that diabetes is a metabolic disorder that increases the likelihood of a fungal infection, especially that associated with Candida sp. pathogens, due to an immunosuppressive effect of the disease on the patient (38). Also, it has been proven that the fungi belonging to Ascomycota phylum produce biologically active secondary metabolites that impact diabetes and cardiovascular disorders (39). Because of the too general hypothesis, we had many post hoc observations like the present one, which had the potential impact on the quality of the work.

In general, in the saliva of obese women with concentrations of 25(OH)D in blood ≥50 nmol/L significantly more species of Actinobacteria were found than in those with lower values (p=0.029). In the study of Gominak (31), they hypothesized that the combination of serum 25(OH)D and vitamin B creates an environment that favours increase of the number of Actinobacteria, Firmicutes, Bacteroidetes and Proteobacteria (normal human microbiome). The same four bacterial species are found in every human with a healthy gut all over the world, so they are called the healthy foursome. The oral vitamin D supplementation in healthy volunteers has decreased the relative amount of pathogenic Escherichia spp., Shigella spp., Helicobacter spp. and Pseudomonas spp. (40). Long-term vitamin D deficiency may produce changes in the microbiota composition that promoted body mass gain and permanently changed environmental conditions that no longer favour the normal human microbiota. On the other hand, lower serum 25(OH)D values change the microbiota composition, reducing vitamin B production in the gut. With the decrease in Actinobacteria in the gut, the vitamin B₁₂ blood level was also lower. Therefore, vitamin B₁₂, as it exists in nature, may be produced by Actinobacteria (41).

Female individuals with concentrations of 25(OH)D in blood <50 nmol/L had a statistically significant increase of Candida sp. (p=0.022) and Lactobacillus sp. in the saliva (p=0.027). Since class 3 obesity was the most prevalent in the group with lower concentration in blood of 25(OH)D (<50 nmol/L), it can be concluded that the presence of Candida sp. and Lactobacillus sp. in the mouth of obese women is associated with vitamin D deficiency and morbid obesity. In our previous study (20), significantly more Lactobacillus sp. were observed in the saliva of obese women than in controls with normal body mass, which was explained by their increased consumption of snacks and food rich in sugar. Vitamin D has a signalling role in modulating the host immune system. Vitamin D/vitamin D receptor (VDR) signalling is a major contributor to the gut microbiome at the genetic, environmental and immune levels (42). VDR gene is the first gene identified as a factor that shapes the gut microbiome at the genetic level. VDR knock-out decreased the level of Lactobacillus and increased the level of Clostridium and Bacteroides in mice faeces (43). Furthermore, one of the roles of vitamin D in the human immune response is to induce an antimicrobial peptide called cathelicidin LL-37. Cathelicidin is a critical component for the body's ability to fight infections, such as candidiasis (44). This could be the reason for the increased oral amount of Candida sp. in the saliva of obese women with a vitamin D deficiency. In this study, the presence of Candida sp. was also positively associated with blood glucose concentrations (diabetes/prediabetes). Our results confirm those by Man et al. (45), who also showed that higher glucose concentration is directly related to Candida albicans growth, which may be associated with fungal infections that occur in non-controlled diabetic patients.

The mechanism of microbiota composition by which bacterial and fungal metabolisms impact the concentration of vitamin D in obese women is less known. Vitamin D can directly control the immune system by VDR signalling. On the other hand, some microbes like Lactobacillus reuteri increased 25(OH)D concentrations (46). Supplementation with Lactobacillus rhamnosus and Lactobacillus plantarum enhances the expression of VDR in intestinal epithelial cells (47). Khosravi et al. (48) showed that the increase of 25(OH)D in the blood had a positive effect on body mass loss, which is partly attributed to the anti-inflammatory effects.

Our study is subject to several limitations. We performed a preliminary study with only the female population involved. Defined selection reasons are given in the Methods, paragraph Subjects and study design. Briefly, differences in diet, age and hormonal factors may affect the development of obesity. In further research we plan to conduct a study on the male population. Since the initial hypothesis that changing trend of oral microbiota composition depends on micronutrient concentrations and/or presence of various obesity-related diseases in obese women in Croatia was not precise enough, manipulating a large amount of data generated a number of post hoc observations that contributed to the overall quality of the work. Therefore, this study can be defined in part as exploratory and hypothesis-generating.