Early childhood caries (ECC) is referred as “the presence of one or more decayed (non-cavitated or cavitated lesions), missing (due to caries), or filled tooth surfaces” in any primary tooth in a child aged 71 months or younger (Drury et al., 1999; Selwitz et al., 2007). ECC is the most common chronic childhood disease (Bagramian et al., 2009) which affects 23% of preschool children in the United States (Dye et al., 2015) and over 60% of children in China (Hu et al., 2011). This disease is a critical public health concern because its characteristics of wide-ranging, rapid-progressing, severe, and difficult to treat. ECC could produce kinds of adverse effects on children both physically and psychologically, posing great economic burden on families and the society (Cummins, 2013).

The current recommendation of care for ECC includes restoration and extraction of carious teeth, topical application of fluoride, and oral hygiene instructions. The incidence of ECC was significantly higher in children with a previous history of the disease (Javed et al., 2017). It was reported that a significant proportion (24–52%) of children treated for ECC will develop new caries lesions within 6 months after the initial treatment (Graves et al., 2004; Amin et al., 2010; Berkowitz et al., 2011). Thus, the management after treatment and prediction of ECC recurrence are of particular clinical significance. Traditional predictors for caries relapse after dental treatment included socio-economic status, general health status, parental compliance of dietary, and oral hygiene instructions, follow-up and recall appointment intervals, and exposure to topical or systemic fluoride. However, these traditional predictors remain controversial (Berkowitz et al., 2011; Amin et al., 2015) with certain limitations (Tellez et al., 2013). Novel approaches based on best evidence are needed to improve the prevention of ECC recurrence.

Currently, classifying and predicting host states based on human microbiota has become one key goal of human microbiome projects worldwide (Knights et al., 2011; Human Microbiome Project Consortium, 2012). Researchers have increasingly agreed with the hypothesis from an ecological perspective to explain the mechanism of caries occurrence and development (Marsh, 2006, 2010; Yang et al., 2012; Jiang et al., 2014), and syntrophic and competitive interactions dominate in oral ecology (Horner-Devine et al., 2007; Lamont et al., 2018). Under this condition, next generation sequencing technology (NGS) provides an important technical means for the study of oral microbial community (Siqueira et al., 2012), and has been used for caries prediction conforming to the ecological plaque hypothesis. As a representative sample of oral ecology, saliva is a non-invasive source containing complex genetic information of human and oral microbes (Baum et al., 2011; Zhang et al., 2016) used extensively in the early diagnosis of dental caries (Vukosavljevic et al., 2011), and its potential value in caries prediction was recently reported (Teng et al., 2015).

Although several recent studies focused on the variation of the oral microbiota during the course of ECC onset or progression to build the biomarker methods in preventing ECC, little is known about ECC recurrence at present: what variation happens on the oral microbiota during the course of ECC recurrence or whether ECC recurrence reflects a particular infection that might be detected at the initial phase. The answer would be helpful for taking appropriate preventive measures at an appropriate time, as well as monitoring and predicting the development of ECC recurrence after the initial treatment.

Thus, our research group conducted a longitudinal study investigating the salivary proteomic profiles in the development of ECC on preschool children aged 3 years old in one kindergarten in Beijing, China (Ao et al., 2017). We utilized NGS to investigate the variation of the salivary microbiota during the course of ECC recurrence, whilst to explore the potential microbial indicators of ECC recurrence, so as to explore if certain microbiota at the post-treatment baseline could predict which children were more likely to develop recurrent ECC, which would potentially facilitate improvements in the strategy for ECC prevention.

ECC is a critical public health concern and the poor clinical outcomes owing to high incidence of recurrence after ECC treatment exacerbates its challenge (Almeida et al., 2000). Technical development for risk assessment and early prediction of ECC recurrence has become both significant and urgent (Selwitz et al., 2007). A number of previous studies have reported traditional predictors of caries relapse after dental treatment, but some of them remain controversial (Berkowitz et al., 2011; Amin et al., 2015) with apparent limitations in the existing methods (Tellez et al., 2013). Novel approaches based on best evidence are needed to improve the prevention of recurrence of this disease.

With the rise of the ecological plaque hypothesis for dental caries, study of the composition and characteristics of oral microbiome from an overall perspective has become the focus of research. According to recent advances in understanding the role of the oral microbiome in health and disease, the development of oral microbiota in preschool children appears to be driven in part by changes in the relative abundance of taxa rather than the acquisition/loss of certain taxa (Yang et al., 2012; Jiang et al., 2014; Zaura et al., 2014). This suggests that measures on features and dynamics of microbiome could be appropriate to define probable risk, onset, progression and recurrence of ECC (Hemadi et al., 2017). As the collection of saliva is quick, non-invasive, and easy to perform in home care, methods based on salivary microbiota can potentially serve as a targeted, sensitive, and patient-friendly measure for risk assessment and detection of ECC (Kaczor-Urbanowicz et al., 2017). In the present study, in order to ascertain whether the microbial composition of saliva after treatment could help us predict which children had the highest risk for the recurrence of ECC, we developed a method which had predictive potentiality for ECC recurrence based on salivary microbiota. The identification of children with high risk of relapse at an early post-therapy stage would be helpful for taking appropriate preventive measures before recurrent lesions occur.

In the sequencing analysis, the rarefaction curves (Figure S1B) and the specaccum curve (Figure S1C) confirmed that sequencing depth of all the samples and the sample size was reasonable. The comparison of Shannon microbial diversity indices (Figures 1C,D) showed no statistically significant differences in both cross-sectional and longitudinal comparisons. Similar to our observations, some previous studies had also shown no significant difference in the α-diversity of those with and without caries (Griffen et al., 2012; Xu et al., 2014; Johansson et al., 2016; Zhou et al., 2017), however some other studies had shown that increased α-diversity favored healthy status (Gross et al., 2010; Belstrom et al., 2014; Xiao et al., 2016). The contradiction amongst the results was brought by the sampling methods, analysis methodologies, or ethnic groups. Therefore, further studies are needed to verify this issue. In the present study, 13 known bacterial phyla and 85 genera derived from 3,650,544 sequences were detected from 84 samples, which suggested that this sequencing analysis could provide the details of the oral microbiota. The level of microbiome diversity in the present study was consistent with sequencing studies in other population with similar conditions (Jiang et al., 2014, 2016). Besides, the results of β-diversity comparison in our study (Figure 2) indicated that saliva from children with and without ECC exhibited similar microbial community structure.

PCoA (Figure 3A) and heatmap plots (Figure 3B) demonstrated that microbial community structure could distinguish the ER and NER groups at baseline. As expected, the saliva microbiome features may serve as a microbial indicator to define susceptibility of ECC recurrence. In the specific comparison between the ER and the NER group at baseline, a group of f bacterial genera, including Fusobacterium, Prevotella, Leptotrichia, and Capnocytophaga, showed statistical differences in abundance. Based on their tendencies between different time points, it was inferred that higher abundance of Fusobacterium, Leptotrichia, and Capnocytophaga, as well as lower abundance of Prevotella, were potential indicators of the recurrence of ECC. In previous studies, Fusobacterium, Leptotrichia, and Capnocytophaga were also reported to be enriched under the status of caries occurrence (Kanasi et al., 2010; Ling et al., 2010; Jiang et al., 2013). But intriguingly, enriched Prevotella had been reported to be associated with ECC (Palmer et al., 2010; Tanner et al., 2011), however in our present study the lower abundance of Prevotella favored ECC recurrence, indicating that oral microbiome might play a different role in the courses of onset and relapse of caries. Furthermore, studies showed that Prevotella species were not similarly distributed between caries and caries-free population, whilst several species of Prevotella (Prevotella spp., Prevotella shahii, Prevotella pallens) presented at a much higher level in the caries-free group than the caries group (Yang et al., 2012; Jiang et al., 2016; Gomez et al., 2017). Therefore, species-level and potentially even strain-level resolution might be important for the issue. Prevotella did not fit the classical characteristics of cariogenic organisms as acid-tolerant or even acid-sensitive (Simon-Soro and Mira, 2015). Thus, whether Prevotella play an active role or are just bystanders in cariogenesis still needed to be further explored.

Receiver-operating characteristic (ROC) analysis was proven to be effective in a previous study for establishment of the predictive model (Teng et al., 2015). In the present study, the AUC of Fusobacterium, Prevotella, Leptotrichia, Capnocytophaga and their combination for prediction for ECC recurrence were 0.857, 0.833, 0.786, 0.833, and 0.952, respectively. These values were considerably higher when compared with a predictive study on the onset of ECC (Teng et al., 2015). Further studies on the probable mechanism of the pathogenesis and progression of ECC recurrence might yield better understanding of the nature.

Some limitations should be considered for the present research. First, only 28 children were recruited in this study, which sample size might not be so sufficient for a predictive study to move further to conduct subgroup analysis with regards to the status of ECC recurrence. Second, the sequencing technology used in this study was based on Illumina MiSeq PE250 platform, which had difficulty to provide effective resolution below the bacterial genus level. In our present study, the OTUs were not aligned well at species level (the unidentified OTU percent at species level ranged from 62.2 to 89.5%), thus these species-level data were not provided to avoid potential bias. For in-depth understanding the etiology of caries, it is expected that further studies on oral microbiota could utilize novel techniques to perform analyses at the species level. Third, dental caries is a complicated multifactor disease, which is caused not only by microbial infection, but also associated with many other factors from the host and environment. Although these factors were not fully considered in this study, we were recruiting children from the same kindergarten who had the same regular diets and oral health instructions, and they shared similar living environments. This could considerably reduce the probable impacts resulted by some other background factors. Fourth, oral microbiome was reported to be influenced by a number of hereditary or environmental factors (Li et al., 2014; Gomez et al., 2017; Gao et al., 2018), but the present study only covered the population of Chinese preschool children in Beijing, which should be interpreted with caution when considering other ethnic groups. Fifth, the pathogenic mechanism associated with the verified microbial indicators of ECC recurrence is still unrevealed. Further studies are needed to explore the specific pathogenic mechanism and signal transduction pathways in the recurrence of ECC.

In addition, for now there is no study using advanced technology (high-throughput approach and mass spectrometry) to investigate the profile of saliva microbiome and proteome at the same time for prediction of dental caries. As previously reported, physiological state of the human body could be detected by monitoring changes in the composition of saliva (Banderas-Tarabay et al., 2002; Van Nieuw Amerongen et al., 2004). Proteins are also the principal ingredient in saliva (Chiappin et al., 2007). Proteins and peptides contained in saliva are important to maintain oral health and homeostasis, since the severity and frequency of oral disease are often associated with the qualitative and quantitative changes of salivary proteome/peptideome (Dawes et al., 2015). Caries risk assessment could be conducted by investigating the bacterial abundance, protein identity and concentration. Further work with both salivary microbiome and proteome/peptideome considered is expectable to fill this gap and contribute to our knowledge on the recurrence of ECC. Moreover, longitudinal monitoring and surveillance could help understand how salivary microbiome and proteome/peptideome changed in ECC recurrence after management and interventions of the disease. In the future, once the systematic biomarkers for the ECC recurrence were verified, the findings could push forward the translation of rapid chair-side test for clinical practice or self-monitoring for home care.

In conclusion, salivary microbiome exhibited potentiality for predicting the recurrence of ECC at an early phase. Fusobacterium, Prevotella, Leptotrichia, and Capnocytophaga were associated to the pathogenesis and progression of ECC recurrence. The method using relative abundance of these four bacterial genera and their combination to predict the recurrence of ECC showed satisfactory accuracy. These findings might facilitate more effective strategy to be taken in the management of the recurrence of ECC.

The raw sequencing data of this study are available in the NCBI Sequence Read Archive with accession number SRP162752.

CZ, CY, SA, XYS, and SZ contributed to the concept, design, data acquisition, analysis, and interpretation and drafted and critically revised the manuscript. XRS and FC drafted and critically revised the manuscript. All of the authors have read and approved the final manuscript.