Using next‐generation sequencing to detect oral microbiome change following periodontal interventions: A systematic review

Abstract Objectives This systematic review was to evaluate the change of oral microbiome based on next‐generation sequencing (NGS)‐metagenomic analysis following periodontal interventions among systematically healthy subjects. Materials and Methods A structured search strategy consisting of “metagenomics” and “oral diseases” was applied to PubMed, EMBASE, and Web of Science to identify effective papers. The included studies were original studies published in English, using metagenomic approach to analyze the effectiveness of periodontal intervention on oral microbiome among systematically healthy human subjects with periodontitis. Results A total of 12 papers were included in this review. Due to the heterogeneity of selected study, quantitative analysis was not performed. The findings as to how alpha diversity changed after interventions were not consistent across studies. Six studies illustrated clear separation of microbial composition between dental plaque samples collected before and after intervention using principal coordinates/component analysis. The most commonly detected genera before intervention were Porphyromonas, Treponema, Tannerella, and Prevotella, while Streptococcus and Actinomyces usually increased and became the dominant genera after intervention. Correlation network analysis revealed that after intervention, the topology of network was different compared to the corresponding pre‐interventional samples. Conclusion Existing evidence of metagenomic studies depicts a complex change in oral microbiome after periodontal intervention.


| INTRODUC TI ON
Periodontitis is one of the most common oral diseases in the world.
The global age-standardized prevalence of severe periodontitis (SP) in 1990 to 2010 was 11.2%, and the incidence was at 0.7% (Kassebaum et al., 2014). It is a main reason for tooth loss, incurring great expense on dental prosthetic treatment (Petersen & Ogawa, 2005). The etiology of periodontitis is complex, which starts from polymicrobial infection caused by intraoral biofilm, and is accompanied by host inflammatory response, leading to severe destruction of periodontal apparatus (connective tissue, alveolar bone, periodontal ligament).
Around 700 species of bacteria have been identified in the oral cavity by culture-dependent and culture-independent tests (Paster, Olsen, Aas, & Dewhirst, 2006). The oral bacterial species associated with healthy gingivae consist of Gram-positive cocci (S. mutans, S. mitis, S. sanguis, S. oralis, Rothiadentocariosa, S. epidermidis), a few Gram-positive bacilli (A. viscosus, A. Israelis, A. gerencseriae, C. spp.), and very few Gram-negative cocci (Veillonellaparvula, Neisseria spp.) (Listgarten, 1976). Socransky grouped oral microbes into several groups, in which orange and red complexes are associated with periodontal diseases, for example, P. gingivalis, P. intermedia, T. forsythia, and T. denticola (Socransky, Haffajee, Cugini, Smith, & Kent, 1998). Next-generation sequencing (NGS) technique, also known as high-throughput sequencing, differs from traditional Sanger sequencing in that it enables us to sequence hundreds of genes at one time and has a deeper coverage of microbial community (Koboldt, Steinberg, Larson, Wilson, & Mardis, 2013). By using NGS technique, we can sequence either 16S rRNA or the whole genome (metagenomics). 16S rRNA exists in all bacteria and archaea and is a phylogenetic marker, which can be utilized to determine taxonomic composition in the oral cavity. The 16S rRNA amplicon sequencing technique is typically based on the amplification of small fragments usually covering one or two hypervariable regions (such as V1-V3, V4, or V4-V5 regions) of the 16S rRNA genes of bacteria/archaea (Ju & Zhang, 2015). The DNA sequences of these amplicons are then mapped to a reference 16S sequence database for taxonomic identification and abundance estimation. While the high-throughput sequencing of 16S rRNA genes can usually profile taxonomic composition at the genus or species level, the whole genome sequencing can potentially provide species-or even strain-level taxonomic resolution in the human microbiome analysis. Moreover, the whole genome sequencing provides more metabolic features and enables us to further understand functional capacities of microbial communities. The shotgun strategy based on NGS identifies the sequence of entire genomes by producing random fragments of DNA (25-500 bp) and assembling them by computers via overlapping ends (Quince, Walker, Simpson, Loman, & Segata, 2017). In modern-day studies employing NGS, the terms of metagenomics and 16S rRNA are often used interchangeably.
The cornerstone of periodontal disease treatment is mechanical debridement of supra-and subgingival plaque and calculus (scaling and root planning, SRP) (Pihlstrom, Michalowicz, & Johnson, 2005).
The effectiveness of those interventions has been evaluated largely in clinical setting or by traditional culture methods/specific sequencing (e.g., 16S RNA probe) before. It is insufficient to get a comprehensive picture as how patients with periodontal diseases benefit from those treatments at the level of compositional and metabolic pathway change in oral microbial communities. Since increasingly more studies use NGS to study the effects of periodontal interventions, this systematic review aimed to summarize and elucidate the changes in the compositional profile and metabolic pathways in the microbial community associated with these interventions.

| Search methods
The search strategy used in this review was based on the search terms, which were related to either "metagenomics" or "oral diseases." The purpose of using search terms of "oral diseases" was to get broad search results of metagenomic studies in oral diseases, including caries, gingival diseases, and periodontitis. This paper only reports on the findings of the studies on periodontitis. According to the specific requirements in PubMed, EMBASE, and Web of Science (updated until December 2019), search terms were modified accordingly (Appendix S1). In order to expand the scope of the search, manual search was performed on the list of references in relevant papers and some registry of clinical trials (such as ClinicalTrials.gov by US NIH).

| Selection criteria
The inclusion criteria for selection of papers in this review were as follows: original studies with the implementation of periodontal interventions (chemical agents, mechanical plaque removal, or systemic administration of antibiotics); used NGS-based metagenomic approach (16S RNA or/and whole genome) before and after the intervention to analyze oral microorganisms; conducted on generally healthy human subjects; and published in English.

| Data collection
Two reviewers first independently screened papers based on titles and abstracts according to the above selection criteria to identify potentially eligible papers. The full text of potentially eligible papers was then obtained. Disagreement between the reviewers was resolved through discussion. The key information of each selected paper was recorded, including study design, sample size, intervention, microbial sampling methods, metagenomic analysis methods, and main findings.

| Quality assessment
Two investigators assessed the quality of randomized clinical trials (RCTs) and non-RCTs included in this systematic review, according to the Agency for Healthcare Research & Quality (AHRQ) Evidencebased Practice Center (EPC) Methods Guide (Viswanathan et al., 2008). This systematic review was written according to PRISMA guidelines checklist (Table S1).

| Search results
A total of 63,900 item/papers were retrieved from the three searched databases, including 30,893 articles in PubMed,19,699 in EMBASE,and 13,308 in Web of Science. After removing duplicates, 47,586 papers were left. In the first round of screening based on titles and abstracts, 42,429 papers not using metagenomic approach, 1,702 papers not related to oral health, 1,723 papers of review/conference proceedings/letters/editorial comments, 1,498 papers of animal studies, and 17 papers not in English were excluded. In the second round of screening based on full text, 91 papers related to periodontitis, 7 related to gingivitis, and 119 related to caries were identified. Among the 91 papers related to periodontitis, 77 observational studies and 2 studies without baseline data were excluded. The remaining 12 papers were included in this review paper (Belstrom et al., 2018;Bizzarro et al., 2016;Califf et al., 2017;Chen et al., 2018;Hagenfeld et al., 2018;Han, Wang, & Ge, 2017;Junemann et al., 2012;Laksmana et al., 2012;Liu et al., 2018;Schwarzberg et al., 2014;Shi et al., 2015;Yamanaka et al., 2012). The screening process is shown in Figure 1. Details of the included studies and summary of their findings can be found in Table 1. The quality of the selected studies can be found in Table S2. The quality assessment showed that the following information is missing commonly in the F I G U R E 1 Process of identification and selection of studies for inclusion in review Both the alpha diversity of subgingival plaque and saliva significantly decreased after 2 weeks and 6 weeks, and completely reversed after 12 weeks. 3. There was an overall significant correlation between relative abundance of putative periodontitis-associated pathogens in subgingival and saliva samples at baseline and each postinterventional time point.
SRP had greater impact in decreasing relative abundance of periodontitisassociated genera in the subgingival plaque than in saliva. Significant correlation of putative periodontitisassociated pathogens between subgingival and saliva samples indicated that those specific pathogens in saliva samples could reflect also subgingival colonization.
1. Alpha diversity: Shannon index 2. Beta diversity: PCA based on Bray-Curtis 3. Relative abundance 4. Correlation network 1. Between-group comparison showed that there was no difference in alpha diversity between two groups at any time point. Within-group comparison showed that there was no significant difference in alpha diversity over time. 2. PCA showed there was only significant difference between intervention and control groups at the 3-month follow-up. Within group, there was a difference between pre-and postinterventional samples at baseline and all follow-up periods. 3. Relative abundance showed that Neisseria, Rothia, Capnocytophaga, Streptococcus increased and Filifactor, Tannerella, Fusobacterium, Porphyromonas, Treponema, Syntrophomonas in both groups. 4. Among patients who did not respond well to interventions clinically, the topology of correlation network after intervention was similar to the one at baseline, with disease-associated genera highly interconnected. Among patients who responded well to interventions clinically, the topology of correlation network included a fully connected network of health-associated genera and four networks of disease-associated genera at baseline, and changed to one network consisting of one subnetwork of health-associated genera and one subnetwork of disease-associated genera.
The change of bacteria composition structure was correlated with intervention and clinical treatment results. The change of alpha diversity and co-occurrence of specific microorganisms can be useful in predicting the resolution of diseased sites after intervention. 1. Higher diversity and number of metabolites were both significantly positively correlated with the maximum pocket depth (MPD). 2. PCoA showed that there was no clear compositional separation of microbiomes across different groups of disease severity in any set of metagenomic data. 3. There was a change in the taxa correlated with MPD among the top 20 relative abundant taxa after intervention compared to baseline (based on both 16S rRNA and shotgun metagenomics). The change was even more significant in metabolites. 4. Greater instability was found in periodontal pockets that did not improve compared to pockets that improved from the data of 16S rRNA, while the finding is the opposite from the data of metabolic features 1. Higher alpha diversity and metabolites were indicators of deep pockets. 2. No periodontitis-associated pathogen group was found based on beta diversity analysis. 3. Metabolic dynamism was more indicative of the effectiveness of treatment than shifts in composition in community. 4. The positive correlation between taxonomic instability and diseases status further proved the polymicrobial etiology of periodontal diseases, and the negative correlation between metabolites and diseases status may reflect the postinterventional metabolites level of host rather than that of bacterial community. (Continues)

Metagenomic analysis Main Findings Authors' Conclusion
1. Alpha diversity: Shannon index, Simpson index 2. Evenness: Pielou's index 3. Beta diversity: PCoA, hierarchical clustering 4. Relative abundance (also illustrated in heatmap) 5. Network analysis (random matrix theory) 6. Community ecological process 1. Alpha diversity was similar among all types of plaque samples, while richness and evenness were significantly less in saliva samples of healthy subjects compared to subjects with chronic periodontitis (both pre-and postintervention). 2. PCoA revealed a clear separation between plaque and saliva samples. Heatmap showed disease-and health-associated taxa. 3. Relative abundance showed that 24 taxa, including periodontitisassociated bacteria, decreased after treatment in plaque samples. Six taxa, part of which were disease-associated taxa, decreased after treatment in saliva samples. 4. Network analysis showed that the topology of the post-treatment samples (both saliva and plaque samples) was different compared to the corresponding pre-intervention samples. Less correlation was found in plaque samples after intervention, while more correlation was found in saliva samples after intervention. 5. Undominated and homogenizing dispersal were the two major factors that dictated bacterial community turnover after treatment.
Microbiota were distinct between saliva and subgingival plaque sample, as well as between disease and healthy sites. SRP was effective in decreasing relative abundance of periodontitis-associated genera in the subgingival plaque and saliva samples, and changing correlation between bacteria in both subgingival plaque and saliva.
1. Richness: number of ribosomal sequence variants (RSV) 2. Evenness: Pielou's index 3. Alpha diversity: Shannon index 4. Beta diversity: PCoA based on Bray-Curtis 5. Relative abundance: mean read counts 1. Within the group of SRP plus placebo, there were no significant differences in richness, evenness, and alpha diversity between baseline and after intervention. 2. Within the group of SRP plus antibiotics, there were no significant differences in evenness and diversity between baseline and after intervention; however, richness decreased and dissimilarity increased significantly after intervention. 3. PCoA showed that there was a clear compositional change of microbiota after intervention in the group of SRP plus antibiotic, which could not be seen in the group receiving SRP plus placebo. 4. Within the group of SRP plus placebo, there was no significant difference in high-abundant aRSVs after intervention, while within the group of SRP plus antibiotics, there was significant decrease in 10 high-abundant aRSVs which belong to the complex associated with periodontitis.
SPR plus antibiotics was effective in reducing richness, decreasing periodontitis-associated genera, and changing bacterial composition structure, while SPR plus placebo did not have such effects, but both interventions improved clinical outcomes significantly 1. Richness: Chao1 2. Alpha diversity: Shannon index 3. Beta diversity: PCoA based on weighted UniFrac distance, hierarchical clustering 4. Relative abundance 1. Alpha diversity decreased, while richness increased significantly after treatment. 2. PCoA showed that pre-interventional samples of the two subjects aggregated in one group, and were different from the postinterventional samples. The two postinterventional samples also differ from each other. 3. Relative abundance showed among the six most abundant bacteria phyla Actinobacteria and Proteobacteria increased significantly, while Bacteroidetes, Spirochetes, and Fusobacteria decreased significantly.
SRP was effective in decreasing alpha diversity, changing bacteria composition structure, and reducing relative abundance of dominant bacteria in subgingival plaque samples.

| Study type
Of the 12 included papers, four were interventional studies with concurrent comparison group (Bizzarro et al., 2016;Califf et al., 2017;Hagenfeld et al., 2018;Junemann et al., 2012) (three were RCT (Bizzarro et al., 2016;Hagenfeld et al., 2018;Junemann et al., 2012)), and the remaining eight were pre-post 1. Shannon index showed that there was no significant difference in alpha diversity 2. PCoA indicated there was difference in the bacterial composition between the pre-and post-treatment groups. 3. Core microbiome (prevalence > 70% and relative mean abundance > 2%) at baseline and after intervention was different. 4. Network of OTU with prevalence > 50% showed that after intervention, correlation nodes in the pathogen component increased significantly, high-connectivity nodes only belonged to the pathogen component rather than distributed in health and pathogen components, highconnectivity nodes had relatively lower abundance compared to those in pre-interventional samples.
SRP was effective in decreasing alpha diversity, changing bacteria composition structure, reducing relative abundance and prevalence of certain bacteria species, and also resulted in a less coordinated microbial community in subgingival plaque samples.
1. Beta diversity: UniFracbased PCoA 2. Relative abundance 1. PCoA did not show clear separation between the pre-and postintervention groups. 2. Relative abundance showed that Fusobacterium was significantly correlated with pocket depth in all samples. The flora after interventions differed between individual subjects.
The species-level analysis of certain species found that after intervention, the relative abundance of diseaserelated and health-related bacteria showed great individual differences and no clear trend could be observed. 2. PCoA showed that the microbial composition was significantly different between diseased state at baseline and resolved states after intervention. Heatmap and cluster analysis showed disease-and health-associated taxa. 3. Relative abundance showed that eight bacteria genera significantly abundant belong to previous red complex, while four bacteria genera in yellow and purple complex were significantly abundant in resolved state. 4. Correlation network showed that in the diseased state, both diseaseand health-associated bacteria correlated more than in the resolved state. 5. Functional pathway analysis showed significant difference between diseased state and resolved state after intervention, including flagellar assembly and chemotaxis.
SRP was effective in decreasing alpha diversity, changing composition structure toward health-associated bacteria genera, decreasing microbial correlation, and reducing diseaseassociated functional pathway.
1. Richness: ACE, Chao1 2. Alpha diversity: Shannon index 3. Beta diversity: PCoA based on unweighted UniFrac distance 4. Relative abundance: mean read counts 1. The microbial richness estimated by the Chao I and ACE, and alpha diversity assessed by the Shannon index were significantly lower in supragingival plaque samples after intervention, while no change in the above indices was observed in saliva samples after intervention. 2. PCoA showed the strong distinct clustering between plaque and saliva samples. The compositional difference following periodontal therapy was smaller than between the two bacterial communities. There was a more significant compositional change after intervention in supragingival plaque than in saliva sample.

Relative abundance showed that predominant genera Fusobacterium and
Kingerlla significantly decreased after intervention in supragingival plaque samples, while no such change was observed in saliva samples.
SRP had greater impact in reducing richness, diversity, and changing composition structure in the supragingival plaque samples than saliva samples.

| Alpha diversity, richness, and evenness
When alpha diversity is used to assess richness and evenness within a certain sample, habitat, or ecosystem (within-sample diversity), it does not only take into account how many species are present but also how evenly each species is distributed. In the included studies, Shannon index ( that although the alpha diversity in dental plaque samples was significantly lower at 2 weeks and 6 weeks after SRP, it bounced back to an even higher level than baseline after 12 weeks (Belstrom et al., 2018).
Two RCTs had the same design of intervention (SPR plus 500 mg amoxicillin and 400 mg metronidazole, three times daily for 7 days) and comparison groups (SRP) (Hagenfeld et al., 2018;Junemann et al., 2012). Within the interventional group of SRP plus antibiotics, the change of alpha diversity in dental plaque samples was not consistent: Alpha diversity was higher in one study (Junemann et al., 2012) while similar in the other study (Hagenfeld et al., 2018).
Within the comparison group of SRP, the change of alpha diversity in dental plaque samples was also not consistent: Alpha diversity was higher in one study (Junemann et al., 2012) while similar in the other study (Hagenfeld et al., 2018). In the remaining RCT study, 0.12% chlorhexidine rinse plus antibiotics was implemented in the intervention group, while 0.12% chlorhexidine rinse was implemented in the comparison group (Bizzarro et al., 2016). No significant difference in alpha diversity was found within each group over time after baseline. In an interventional study with concurrent comparison group (randomization not mentioned), the effectiveness of the intervention on alpha diversity was not reported, but it found that alpha diversity (Faith's PD) could be used as an indicator of pocket depth (Califf et al., 2017).
ACE (Junemann et al., 2012;Yamanaka et al., 2012), Chao 1 (Califf et al., 2017;Han et al., 2017;Liu et al., 2018;Yamanaka et al., 2012), and the number of RSVs (Hagenfeld et al., 2018) were used to assess richness in the selected studies. Three pre-post studies (Han et al., 2017;Liu et al., 2018;Yamanaka et al., 2012) and one RCT (Hagenfeld et al., 2018) reported the change of richness after baseline. For pre-post interventional studies with the intervention of SRP, the findings regarding the effectiveness of SRP on richness in dental plaque samples were not consistent: Compared to baseline, richness was significantly lower in one study (Yamanaka et al., 2012), significantly higher in one study (Han et al., 2017), and similar in one study (Liu et al., 2018). The only RCT, which reported the change of richness, found that within the invention group of SRP plus antibiotics, richness decreased significantly compared to baseline, while within the comparison group of only SRP, there was no significant change in richness (Hagenfeld et al., 2018).
Pielou's index was used to assess evenness in two RCTs with the same design of intervention and comparison groups (Hagenfeld et al., 2018;Junemann et al., 2012). Between-group analysis in one of the RCTs showed that there was no significant difference in evenness between the intervention and control groups after intervention (Hagenfeld et al., 2018). There was also no significant change within each group after the intervention (Hagenfeld et al., 2018). In the other RCT, no such between-group analysis was made and Pielou's index showed that the abundance of OTUs was more equally distributed within each group after the intervention (Junemann et al., 2012).

| Beta diversity
Beta diversity is a measure for comparing microbial composition between different samples, habitats, or ecosystems (between-sample diversity), or detecting the overall shift in microbial community after a certain period of time. Among the included studies, eight studies used principal coordinates analysis (PCoA) to assess beta diversity (Califf et al., 2017;Chen et al., 2018;Hagenfeld et al., 2018;Han et al., 2017;Liu et al., 2018;Schwarzberg et al., 2014;Shi et al., 2015;Yamanaka et al., 2012) and one study used principal component analysis (PCA) (Bizzarro et al., 2016). Beta diversity matrix can be calculated in different ways: incorporating quantitative information of sequence abundance (e.g., Bray-Curtis (Bizzarro et al., 2016;Hagenfeld et al., 2018)) or incorporating information of phylogenetic distances between species (weighted UniFrac (Califf et al., 2017;Han et al., 2017;Liu et al., 2018;Shi et al., 2015) or unweighted UniFrac (Yamanaka et al., 2012)). The advantage of PCoA and PCA is that they reduce the dimensionality of microbiome data so that a lowdimensional graphical plot could be generated where distance between each dot represents the dissimilarities between two samples.
For dental plaque samples, four pre-post studies with the intervention of SRP revealed a clear compositional change after interventions compared to baseline (Han et al., 2017;Liu et al., 2018;Shi et al., 2015;Yamanaka et al., 2012), while two pre-post studies did not find such a clear compositional separation after SRP Schwarzberg et al., 2014). It is noteworthy that beta diversity of dental plaque sample was more significantly different between individuals in resolved status than that in diseased status (Han et al., 2017;Shi et al., 2015). Two pre-post studies with the intervention of SRP found that for saliva samples, there was no compositional separation between samples collected after intervention and at baseline Yamanaka et al., 2012).
Two RCTs reported the effectiveness of intervention on beta diversity (Bizzarro et al., 2016;Hagenfeld et al., 2018). In one RCT, between-group analysis showed there was a clear compositional separation between intervention (0.12% chlorhexidine rinse plus antibiotics) and control groups (0.12% chlorhexidine rinse) at the 3-month follow-up (Bizzarro et al., 2016). Within-group analysis found that there was a clear compositional separation between samples at baseline and all follow-ups within each group. In the other RCT, PCoA showed that there was a clear compositional change after intervention within the group of SRP plus antibiotic, which could not be seen within the group only receiving SRP (Hagenfeld et al., 2018).

| Relative abundance
The dominant genera in pre-and postinterventional samples and the major change of relative abundance of some genera in the included studies are demonstrated in Table 2. Some studies used statistical approach to assess the change of relative abundance (Bizzarro et al., 2016;Hagenfeld et al., 2018;Liu et al., 2018;Schwarzberg et al., 2014;Shi et al., 2015;Yamanaka et al., 2012), while some studies made the descriptive conclusion as to which species increases or decreases (Belstrom et al., 2018;Chen et al., 2018;Han et al., 2017;Junemann et al., 2012;Laksmana et al., 2012).
Some of the included papers also employed heatmap with hierarchical clustering or cladogram to visually display distributive and quantitative change in microbial community Han et al., 2017;Liu et al., 2018;Shi et al., 2015). This kind of visual display helps to identify the "core microbiome" and how these species are related in phylogenic tree. As shown in Table 2, the most commonly found dominant genera at baseline in the included studies were Porphyromonas, Treponema, Fusobacterium, Tannerella, Prevotella, and Filifactor, and they were replaced by Streptococcus, Actinomyces, Rothia, and Villonella after intervention.

| Correlation network analysis
Correlation network indicates synergic or antagonistic interactions between genera. Among the included papers, four studies used network analysis to investigate coordinated interaction in microbial community (Bizzarro et al., 2016;Chen et al., 2018;Liu et al., 2018;Shi et al., 2015). Two papers showed that the number of connection nodes between microorganisms decreased significantly after intervention Shi et al., 2015). One paper did correlation network of OUT with prevalence > 50% and found that after intervention, the number of correlation nodes increased significantly in the pathogen component, but the density of the network became lower compared to that of pre-interventional network (Liu et al., 2018). Also, after intervention those high-connectivity nodes only distributed in the pathogen component rather than in both health and pathogen components. In the study by Bizzarro et al., the subjects were divided into two subgroups, those who had relatively good clinical outcomes and those not (Bizzarro et al., 2016). Among patients who did not respond well to interventions clinically, the intervention did not change the topology of correlation network significantly. Among patients who responded well to interventions clinically, the topology of correlation network changed significantly, from a fully connected network of health-associated genera and four relatively separated networks of disease-associated genera at baseline to one network consisting of one subnetwork of health-associated genera and one subnetwork of disease-associated genera (Bizzarro et al., 2016).
This indicates that well-connected commensals at baseline could be a predictor for better clinical outcomes.

| Functional pathway analysis
Only one study performed functional pathway analysis and reported that among the 90 functional bacterial pathways found in the microbiome, 24 were overexpressed at baseline (Shi et al., 2015). Among those overexpressed pathways, the most noteworthy ones are flagellar assembly and chemotaxis. These two pathways were significantly more abundant at baseline compared to after intervention.
These two functional pathways favor flagellated motile microbial species to grow, colonize, and penetrate oral epithelial cells.

| D ISCUSS I ON
Previous studies showed that detection of specific pathogens by using targeted microbial techniques (DNA probes, microarrays) was poorly predictive of the prognosis of periodontitis. Therefore, analyzing entire disturbed microbial community rather than several putative pathogens might be the key in understanding periodontitis (Berezow & Darveau, 2011;Friedrich, 2008). NGS technique enables us to unveil the whole picture of microbial communities of different niches in oral cavity (Lazarevic et al., 2009). The papers included in this review all adopted the NGS technique, although at different levels. Most of the included papers adopted NGS technique at the 16S RNA level, while NGS technique in the papers of Shi et al. (2015) and Califf et al. (2017) were at the whole genome level, which could explore functional pathways as well.
By using the search strategy in this systematic review, a large number of papers were retrieved from three major biomedical databases, which yielded a very comprehensive search result. The small number of eligible papers reveals the challenge in real world to employ metagenomic approach in interventional studies among human subjects, from recruiting subjects, implementing interventions, obtaining microbial samples, controlling confounding factors, to conducting large-scale metagenomic analysis. There is a huge variance between the selected studies in terms of the demographic characteristics of the study subjects, the baseline periodontal status, the types of interventions, and the length of the follow-up, all of which collectively affect to the outcomes. Due to the heterogeneity of the aforementioned, it is not possible to conduct any quantitative analysis to summarize the results.
Alpha diversity consists of two aspects, richness and evenness.
In a particular environment, richness measures the number of different species, while evenness measures how individual species are distributed. Shannon index, Simpson index, and Faith phylogenetic diversity each has its strength: Shannon index emphasizing more on rare species, Simpson index giving more weight to evenness, and Faith phylogenetic diversity incorporating phylogenetic information (Kim et al., 2017). Although it is intuitive to infer that periodontal interventions should increase alpha diversity as a more diverse community is associated with greater resilience and healthier status from the ecological point of view (Proulx et al., 2010), most of the included studies did not found such a trend of increasing alpha diversity after intervention. First, periodontal disease is caused by complex alteration of entire microbial community rather than a few dominant pathogens; therefore, the microbial community does not simply become more diverse after periodontal intervention. Secondly, in a healthier state after intervention, the number (richness) and the loads (abundance) of pathogenic species may decrease, while those of commensal species may increase. Considering that both pathogens and commensals change simultaneously, it would be difficult to determine the magnitude and direction of change in richness and evenness. Therefore, it is inconclusive as to how alpha diversity changes after intervention.
Compared to alpha diversity, beta diversity is more robust to noise introduced by PCR and sequencing errors (Ley et al., 2008). It is also more meaningful to assess beta diversity than alpha diversity in interventional studies as it can demonstrate whether periodontal interventions resulted in compositional rearrangement in microbial community, which likely contributed to the recovery of periodontal tissue. The findings of a clear separation in PCoA/PCA scatterplot between samples at baseline and after interventions from the majority pre-post studies with SRP confirm that non-surgical periodontal therapy (SRP) is effective in disrupting biofilm in periodontal niche (Cobb, 2002). Beta diversity in the included studies also highlights additional interesting findings. The increased difference in beta diversity between individual samples after intervention suggests that healthy periodontal environment welcomes a variety of health-associated microorganisms to co-exist (Shi et al., 2015). In the reviewed studies, the beta diversity of saliva samples after intervention remained relatively stable and was significantly different from that of the plaque samples, indicating that salivary sample is not an ideal substitute for dental plaque because it cannot accurately reflect the dynamic change in periodontal niche (Yamanaka et al., 2012).
More detailed information regarding how much taxonomic change was after intervention in the included studies was provided by relative abundance at the genera level as well as visually in the heatmap and cladogram. It is not surprising to find that in most of the included studies, Gram-negative genera Porphyromonas, Treponema, Fusobacterium, Tannerella, Prevotella, and Filifactor were predominantly enriched in samples at baseline. This is consistent with some landmark works, which show that "red and orange complexes" are considered to play leading roles in the development of periodontitis (Curtis, Zenobia, & Darveau, 2011;Darveau, Hajishengallis, & Curtis, 2012;Socransky et al., 1998).
Porphyromonas gingivalis is considered as a key pathogen in periodontitis, altering the amount and composition of oral commensals (Hasturk et al., 2007;Kumar et al., 2006), manipulating complements and leukocytes (Liang et al., 2011), finally leading to destructive inflammation in periodontal tissue. Fusobacterium is considered to be a critical bacterium in forming biofilm as it bridges early colonizing species and late colonizers such as Porphyromonas gingivalis (Kolenbrander et al., 2006). Prevotella includes several well-known periopathogens (Preveotella nigrescens, Preveotella intermedia, and Prevotella melaninogenica) (Socransky & Haffajee, 2005). The majority of studies also demonstrated that Streptococcus, Actinomyces, Rothia, and Villonella increased and became dominant in the samples after interventions, which were collectively known as commensals (Kolenbrander et al., 2006). Other than the findings in "core microbiome" before and after interventions, a plethora of microorganisms, which could not be detected by conventional culture or probes technique, were also found in the included studies (Hagenfeld et al., 2018;Liu et al., 2018;Shi et al., 2015). Although they were present at a low abundance, they collectively accounted for a larger proportion than the classical disease-and health-associated genera. There are some limitations in the included studies that prevent further studying those low-abundant microorganisms. First, the majority of the included studies sequenced 400-500 bp at the 16S rRNA level, which might not be able to produce a taxonomic resolution down to the species level. As a result, it is not always possible to determine the pathogenic potential of the low-abundant genera. Also, correlation network and functional analyses focus only on high-abundant species. How the low-abundant microorganisms interact with one another as well as the "core microbiome" and function in the entire community under the effect of the intervention is a missing piece of the puzzle. It should be also noted that the due to the small sample size, some of the selected studies were only able to sum-  Bizzarro et al., 2016;3. Califf et al., 2017;4. Chen et al., 2018;5. Hagenfeld et al., 2018;6. Han et al., 2017;7. Junemann et al., 2012;8. Laksmana et al., 2012;9. Liu et al., 2018;10. Schwarzberg et al., 2014;11. Shi et al., 2015;12. Yamanaka et al., 2012 a Using statistical approach to compare the change of relative abundance TA B L E 2 (Continued) (Hagenfeld et al., 2018), and only one study used LEfSe analysis (Liu et al., 2018). In order to make more accurate inference regarding the change of relative abundance, more sophisticated statistical methods should be employed in future studies.
Findings of the included studies show that the study interventions did not only influence subgingival microbial community in the taxonomic composition but also the interactions between microorganisms. A preliminary finding generated from the limited number of studies with network analysis is that the interventions can reduce either the numbers of nodes/links or the density of network. This indicates that the interactions in the microbial community (especially the disease-associated component) can be disrupted by the interventions, shifting from a symbiotic state to a dysbiotic state. The metagenomic changes can be used as outcome measures in future studies to investigate the effectiveness of different treatment modalities for periodontitis. For instance, current clinical evidence is inconclusive as whether full-mouth scaling within 24 hr is more effective compared to conventional quadrant SRP (Eberhard, Jepsen, Jervoe-Storm, Needleman, & Worthington, 2015). In this situation, evaluating the metagenomic outcomes to compare the effectiveness between different treatment modalities seems more sensitive than clinical outcomes.
It is noteworthy that some of included studies adopted adjunctive use of systemic antibiotics. Systemic administration of antibiotic may be a double-edged knife in the treatment of periodontitis.
Antibiotics can suppress commensals and lead to a dysbiotic status.
A good example of dysbiotic alteration of gut commensal microbiome following long-term use of antibiotics is Clostridium difficile infection (Bartlett, 2017). Findings of this review do not show that adjunctive use of systemic antibiotics in interventions of SRP leads to an additional significant difference in metagenomic outcomes.
Therefore, its standard application in the treatment of periodontitis may not be warranted.

| CON CLUS ION
Existing evidence from metagenomic studies depicts a complex change in microbiome after periodontal intervention. However, due to the heterogeneity of the methods and outcomes adopted, only descriptive and preliminary findings could be summarized in this review. As periodontitis is currently viewed as a multifactorial disease caused by the interaction between the dysbiosis of the entire subgingival microbial community and the host inflammatory response to it, future studies evaluating the effectiveness of periodontal treatment should shift from merely assessing clinical outcomes and several classical periopathogens to the metagenomic outcomes relating to the entire microbial community.

ACK N OWLED G EM ENT
The study was supported by National Natural Science Foundation of China (NSFC No. 81701036) and Anhui Province Natural Science Foundation (AHNSF No. 1808085QH247).

CO N FLI C T S O F I NTE R E S T
None to declare.